2

THE STANDARD DEVIATION

“I may venture to say that his observations have stretched much farther…”
-- Sense and Sensibility

2.1 INTRODUCTION
It is sometimes said that “physics is an exact science.” But physics is based on experiment,
and experiment on measurement, and measurement is inexact by its very nature. Imperfect (or
simply just finite) instruments, necessary approximations, and inevitable background noise make
all measured values impossible to pin down exactly. Even so, clever experiments that coax even a
very uncertain but crucial measurement from a sea of noise can lead to great advances in physics. A
recent article in a science magazine trumpeted the fact that, with the help of data from the Hubble
Space Telescope, the various scientific groups measuring the Hubble Constant (a quantity whose
value determines the age and fate of the universe) were now getting results that agreed to within
roughly 10% of each other. This was considered great news, partly because it was a dramatic im-
provement over the situation just a few years ago (when measurements disagreed by more than a
factor of two), but partly because the age of the universe computed from this quantity was finally
settling down toward values that were consistent with other physical estimates of the age of the
universe, soothing physicists’ worries that whole areas of physics might be poorly understood.
The point is that, while the greatest precision is always desirable, even inexact measurements
can lead to important scientific progress. But a crucial part of making good scientific use of inexact
results is being able to quantify how inexact (the technical word is uncertain) a measurement is.
We can draw scientific conclusions from an inexact value only if we know something about how
sharp or fuzzy the value is.
One of the major goals of this laboratory program is to teach you how to extract the maxi-
mum possible scientific meaning from uncertain measurement results. The next chapter of this ref-
erence manual will begin discussing the meaning of the concept of uncertainty in the context of
measured quantities. The purpose of this chapter is to lay some mathematical foundations for that
discussion by exploring ways that we might quantify the spread in a set of values that represent
imperfect measurements of the same quantity.

2.2 POSSIBLE MEASURES OF SPREAD IN A DATA SET

Imagine that N teams of scientists measure a physical quantity, and each group ends up with
a value that is somewhat different from the values obtained by other groups. This process gives us
a data set consisting of N values, each of which represents an imperfect estimate of the “true value”
of that quantity. As we will see in the next chapter, it is useful when discussing the physical mean-
ing of these results to be able to quantify the “spread” in these measurement values. How might we
quantitatively express the “spread” in a data set like this in a meaningful and useful way?
The simplest way to express the spread in a data set is to use its range R, which in this con-
text we define to be the difference between the highest and lowest values in the set. Clearly, the
more “spread out” a set of data values is, the larger its range is likely to be.
A potential problem with the range as a measure of the spread in a data set is that its value is
determined entirely by the two most extreme values in the set (all the other values are ignored!) The
extreme value may be the result of improbable flukes in the measurement process (or even outright
2. The Standard Deviation 18

errors), and are not representative of the whole by their very nature. One might imagine one data
set where the values are mostly very closely clustered around some central value with only a hand-
ful of wildly different values at the extremes, and another data set having the same range whose
values are fairly evenly spread out between the extremes. Would we really want to quantify the
“spread” in these two data sets by the same number?
A more meaningful way to quantify the spread is to compute the average deviation of the
data points from the data set’s mean (= average) value. If we use the symbols x1, x2 , x3 , º to
represent the first, the second, the third, etc. measurement values in our set of N values, and the
symbol x to stand for the mean of those values, then the average deviation is defined to be:

average deviation ∫
1
N
[ x1 - x + x2 - x + x3 - x + º + x N - x ] (2.1)

You can see that this definition gives you exactly what the name implies, the average (summed
over the entire data set) of the deviations of the measurements (which for a given measurement is
the absolute value of how far that measurement is from the mean). Note that the absolute value is
essential in this expression: since generally a given measurement is as likely to be above the mean
as below it, if we did not take the absolute value of each difference, they would sum to zero (in fact
the definition of the mean implies that if we remove the absolute value symbols in equation 2.1, the
sum would be exactly zero, as you can check).
The average deviation, in contrast to the range, nicely takes each measurement value equally
into account in its value, and thus is probably a better representation of the spread in a data set. It
also has an easily understood meaning. However, it turns out that we often want to do some calcu-
lus with the quantity we use to characterize the spread in data. The absolute value presents a minor
complication in doing calculus with the average deviation, because the derivative of x is unde-
fined when x = 0.
The standard deviation does not suffer from this problem. If we use the same symbols that
we used in equation 2.1, we can write the standard deviation as follows:

standard deviation ∫ s ∫
1
N -1
[
( x1 - x )2 + ( x2 - x )2 + ( x3 - x )2 + º + ( x N - x )2 ] (2.2)

If we ignore for a moment the fact that we are dividing by N–1 instead of N, the standard deviation
is thus the square root of the average squared deviation. Dividing by N–1 instead of N is im-
portant for deep mathematical reasons beyond our level here, but note that for a data set consisting
of a single measurement, the average deviation states that the “spread” in this data set is zero (sug-
gesting that we know the measurement value perfectly), whereas the standard deviation states that
the spread is undefined ( 0 ∏ 0 ), which more meaningfully suggests that we know nothing about
the spread in a data set when that set only has one value.
Equation 2.2, though it superficially looks more complicated than that for the average devia-
tion, is actually simpler in a number of respects. Doing calculus with this expression is easier be-
cause squares and square roots are easier to deal with mathematically than the absolute value. Since
many calculators are specially set up to calculate the standard deviation with a few button presses,
it is generally much easier to calculate than the average deviation, and is really no more difficult to
calculate even if one is reduced to using equations 2.1 and 2.2 directly. However the main reason
that the scientific community typically uses the standard deviation (rather than the range or the
average deviation) to characterize the spread in the values of a data set is that it has an important
property that the range and average deviation do not have.
2. The Standard Deviation 19

Imagine that we model each of the N measurements in our data set as being what happens
when we add tiny random errors from fairly large number of sources to the measured quantity’s
“true value” (whatever that might be). This “random error” measurement model is only a model of
the measurement process (and a pretty simplified one at that), but it does seem to be useful and rea-
sonably accurate in many cases. When this model does adequately describe a measurement pro-
cess, then we find that if we plot the probability of getting a certain measurement value versus that
value, we get the famous “bell-shaped curve” (which indicates that measurement values close to the
quantity’s “true value” are common and extreme values are rare). A wide variety of measurement
process tend to yield data sets that are distributed this way.
So here is the payoff. If the “random error” measurement model is a good model for the
measurement process giving rise to given data set, then it can be shown mathematically that the
value one gets by computing the standard deviation for this data set is fairly independent of N. This
means that in the context of this “random error” measurement model, the standard deviation better
isolates the effect of the measurement process itself on the spread in the data from effects that re-
sult from including more or fewer data points in the data set. In contrast, the range of a data set
typically increases as N increases (because the more measurements we take, the more likely we
are, by chance, to encounter some really extreme measurement results), while the average devia-
tion tends (more subtly and for more subtle reasons) to decrease with increasing N. It is this spe-
cial relationship between the standard deviation and the most common and useful mathematical
model for the measurement process that makes the standard deviation the most widely accepted
measure of the spread of the values in a data set.

2.3 CALCULATING THE STANDARD DEVIATION

One can fairly easily calculate the standard deviation of a list of measurement values directly
from equation 2.2 by going through the following steps:
1. Write the values in a vertical column.
2. Sum the values and divide by the number of measurements N to find the mean.
3. Subtract this mean from each value and square the result. Write each “squared deviation”
to the right of the corresponding measurement value.
4. Sum the squared deviations in this new column and divide by N–1.
5. Take the square root of the result to get the standard deviation.
Almost any scientific calculator (which we recommend you have for this course) will provide
an easier way for you calculate the standard deviation. In general, the process looks like this:
1. Initialize the calculation somehow (varies from calculator to calculator).
2. Type in the first number.
3. Press a key that is usually marked Â+ or M+.
4. Repeat steps 2 and 3 for each of the measurement values (after you press Â+ each time, the
calculator typically displays the number of values that you have entered so far).
5. When all values have been entered, calculate the standard deviation by pressing one of the
following: s, sx , s , s n-1 (if the choice is between that and s n ), s x or s x n-1 (instead
of s y or s y n-1 ). The choice, unhappily, varies from calculator to calculator.
For example, on my HP-32S calculator, one initializes the calculation using the CLEAR button and
selecting “clear all” from the choices presented. To get the final result, I have to push the STAT
button and then select “s” and then “ sx ” from the choices presented. On a TI-36X calculator, one
initializes the calculation by selecting the STAT 1 function (to indicate we are going to do statistics
of one variable), and one gets the final result by pressing the s x n-1 button.
2. The Standard Deviation 20

On a TI-82 or TI-83 calculator, select the STAT key, used the arrow keys to get to EDIT if
you aren’t there already, and enter your data in one of the data lists (L1, L2, L3, …). Then press
the STAT key again, use the right-pointing arrow key to choose CALC, the down-arrow key (if
necessary) to get to 1-Var Stats, and press ENTER to get the mean and standard deviation, as well
as some sums used to calculate their values.
You should look in the instruction booklet that came with your calculator to determine how to
calculate the standard deviation on your calculator. (If you don’t have your booklet, do some ex-
periments, ask a friend, or check your calculator manufacturer’s web site. If you come a bit early
to lab, your lab instructor or a lab assistant may be able to help you.) To test that you are actually
calculating the standard deviation as defined by equation 2.2, note that the standard deviation of the
data set 2, 4, 6 should be 2.

EXERCISES

Exercise 2.1
Compute the standard deviation of the following data set using the direct method (the first
method outlined in section 2.3). Show your work in the space below (be sure to keep track of
units!). Then figure out the particular short-cut approach that works on your calculator and use
it to check your work.

0.56 s
0.52 s
0.59 s
0.48 s
0.51 s

Exercise 2.2
Show your interviewer that you can calculate the standard deviation of the data set 2, 4, 6 using
whatever short-cut approach works on your calculator.
 3
EXPERIMENTAL UNCERTAINTY

“ ‘I am no matchmaker, as you well know,’ said Lady Russell, ‘being much too
aware of the uncertainty of all human events and calculations.’”
--- Persuasion

3.1 UNCERTAINTY AS A “95% CONFIDENCE RANGE”

We generally assume in physics that any quantity we measure has a “true” value, which is the
result that we would get if we had a perfect measuring apparatus. Fifteen minutes in any labora-
tory, regardless of the sophistication of the equipment, will rapidly disabuse you of the notion that
any measurement apparatus is perfect. Real measurement devices suffer from a variety of imper-
fections that limit our knowledge of the “true” value of any measurement. Devices may be poorly
made, out of adjustment, subject to noise or other random effects, or hard to read accurately, and
all devices read to only a finite number of digits. These problems mean that the exact value of any
measured quantity will always be uncertain.
Uncertainty is therefore an unavoidable part of the measurement process. We will (of course)
always seek to reduce measurement uncertainty whenever possible, but ultimately, there will re-
main some basic uncertainty that cannot be removed. At this point, our task is to estimate thought-
fully the size of the uncertainty and clearly communicate the result.
How can one quantify uncertainty? For our purposes in this course, we will define a value’s
uncertainty in terms of the range centered on our measured value within which we are 95% confid-
ent that the “true value” would be found if we could measure it perfectly. This means that we ex-
pect that there is only one chance in 20 that the true value does not lie within the specified range.
This range is called the 95% confidence range or 95% confidence interval.
The conventional way of specifying this range is to state the measurement value plus or mi-
nus a certain number. For example, we might say that the length of an object is 25.2 cm ± 0.2 cm:
the measured value in this case is 25.2 cm, and the uncertainty U in this value is defined to be
±0.2 cm. The uncertainty thus has a magnitude equal to the difference between the measured value
and either extreme edge of the uncertainty range. This statement means that we are 95% confident
that the measurement’s true value lies within the range 25.0 cm to 25.4 cm.
Now (as you may have already noticed), this definition of uncertainty is rather fuzzy: one
person may be more confident about a value’s precision than the next. In fact, the definition seems
almost devoid of objective meaning except as a description of the experimenter’s frame of mind.
We will see that the definition is not as subjective as it seems, particularly in certain cases to be
discussed shortly, where it is possible to make an educated and generally agreed-upon estimate of
the uncertainty. But the fact remains that “uncertainty” is itself an uncertain concept, and uncer-
tainties should only be taken to be rough estimates, good to one or at most two significant digits.
In spite of this problem, knowing the uncertainty of a measured value is essential if one is to
correctly interpret the meaning of a measured value. For example, imagine that you measure the
period of a (very long) pendulum to be 12.3 sec. Imagine that someone’s theory predicts that the
period should be 11.89275 sec. Is your result consistent with that theory or not? The answer to
this question depends entirely on the uncertainty of your result. If your result has an uncertainty of
±0.5 s, then the true value of your measured duration could quite easily be the same as the theoreti-
3. Experimental Uncertainty 22

cal value. On the other hand, if the uncertainty in your result is ±0.1 s, then it is not very likely
(less than one chance in 20) that the true value behind your measurement is the same as the predict-
ed value, meaning that the theory is probably wrong. What a measurement means, therefore, can
depend crucially on its uncertainty!

3.2 SYSTEMATIC ERRORS

Why aren’t measurements perfect? The causes of measurement errors can be divided into
three broad classes: systematic problems, limited precision, and random effects. The focus of
this chapter will be on the last of these, but the first two causes need to be discussed briefly.
Systematic errors occur when a piece of equipment is improperly constructed, calibrated
or used. For example, suppose that you measured lengths with a meter stick that you failed to no-
tice had been cut off (for some reason) at the 5 cm mark. This would mean that all of your meas-
ured values would be 5 cm too long. Or suppose that the stopwatch that you are using runs proper-
ly at 20°C, but you happen to be using it where the temperature is closer to 30°C, which (unknown
to you) causes it to run 10% too fast.
One does not generally include systematic errors in the uncertainty of a measurement: if you
know that a systematic problem exists, you should fix the problem. For example, you would use a
complete meter stick in the first case above or reduce your stopwatch measurements by 10% in the
second case. The most appropriate thing to do with systematic problems in an experiment is find
them and eliminate them. Unfortunately, no well-defined procedures exist for finding systematic
errors: the best that you can do is be clever in anticipating problems and alert to trends in your data
that suggest their presence. You might come to suspect the short meter stick, for example, if you
noticed that your data would agree with theoretical predictions if all of your length measurements
were about 5 cm shorter.
In some cases, it is appropriate to estimate the magnitude of possible systematic errors and
include them in the uncertainty of a measurement result. For example, it is possible to read most
automobile speedometers to a precision of about 1 mph. But it is well known that variations in the
manufacture and calibration of speedometers mean that the reading of a typical speedometer may be
off by as much as 5%. Therefore, the uncertainty of a speedometer reading of about 60 mph would
have to be taken to be ± 5% of 60 mph, or about ± 3 mph. The uncertainty in this case is called a
calibration uncertainty. We will deal with calibration uncertainties only rarely in this course.

3.3 LIMITED PRECISION

No measurement device can read a value to infinite precision. Dials and linear scales, such as
meter sticks, thermometers, gauges, speedometers, and the like, can at best be read to within one
tenth of the smallest division on the scale. For example, the smallest divisions on a typical metric
ruler are 1 mm apart: the minimum uncertainty for any measurement made with such a ruler is
therefore about ± 0.1 mm. This statement is not an arbitrary definition or convention: rather, it is a
rule based on experience. If you try using a ruler to make as precise a measurement as you can,
you should be able to see that it is really quite difficult to do better than the stated limit.
For measuring devices having a digital readout, the minimum uncertainty is ± 1 in the last
digit. For example, imagine that you measure a time interval with a stopwatch, and find the result
to be 2.02 s. The measurement’s “true value” in this case could be anywhere from 2.010...01 s to
2.0299...99 s. We cannot narrow this range without knowing details about the design of the stop-
watch: does it round up to 2.02 s just after the true elapsed time exceeds 2.01 s, or does it round to
the nearest hundredth of a second, or does it not register 2.02 s until at least 2.02 s have passed, or
what? So in this case, we must take the uncertainty to be at least ± 0.01 s.
3. Experimental Uncertainty 23

In both of the cases described above, these rules are meant represent the minimum possible
uncertainties for a measured value. Other effects might conspire to make measurements more un-
certain than the limits given (as we shall see), but there is nothing that one can do to make the un-
certainties smaller, short of buying a new device with a finer scale or more digits.

3.4 RANDOM EFFECTS

This chapter is mainly focused on the analysis of random effects. It is commonly the case
that repeated measurements of the same quantity do not yield the same values, but rather a spread
of values. For example, in the speed of sound lab, you measured the time interval between the in-
stant the sound was sent into the tube and the instant that it arrived at the other end. If you measure
this interval five times, you are almost certain to get five different results (for example, 0.54 s,
0.52 s, 0.55 s, 0.49 s, and 0.53 s).
Why are these results different? In this case, the problem is that it is difficult for you to start
and stop the stopwatch at exactly the right instant: no matter how hard you try to be exact, some-
times you will press the stopwatch button a bit too early and sometimes a bit too late. These un-
avoidable and essentially random measurement errors cause the results of successive measurements
of the same quantity to vary.
Random perturbing effects, which are sometimes human and sometimes physical in origin,
are a feature of almost all measurement processes. Sometimes a measuring device is too crude to
register such effects: for example, a stopwatch accurate to only one decimal place might read 0.5 s
for each of the measurements in the case described above. But laboratory instruments are often
chosen to be just sufficiently sensitive to register random effects, because while one always desires
as precise an instrument as possible, there is no point in buying an instrument much more precise
than the limit imposed by unavoidable random effects. For example, while a stopwatch that reads
to a hundredth of a second is better than one that registers to only a tenth, a stopwatch that reads to
a thousandth of a second not really any better than one that reads to a hundredth, because human
errors in operating the stopwatch are typically on the order of a few hundredths of a second, mak-
ing the value indicated in the thousandth’s place meaningless.
The point is that random effects will be an important factor of many of the measurements that
you will encounter in any scientific experiment. Now, it should be clear that such effects increase
the uncertainty in a measurement. In the stopwatch case, for example, that different trials lead to
results differing by several hundredths of a second implies that the uncertainty in any given meas-
urement value is larger than the basic ± 0.01 s uncertainty imposed by the digital readout.

3.5 THE DISTRIBUTION OF VALUES DUE TO RANDOM EFFECTS

The first step towards describing the magnitude of the uncertainty due to random effects is to
understand more precisely what these effects do to a set of measurement values. Consider, for ex-
ample, a simple experiment where 25 different people measure the length of a soft-drink can with a
ruler. Assume that the “true” length of the can is 8.51293... cm. None of the 25 people will meas-
ure the object to have exactly this length, of course, because people will view the ruler and can
from slightly different angles, make different judgments about the exact reading, and so on.
Nevertheless, we would expect the measurements to cluster around the value 8.51 cm, with
most of them agreeing with that result within a few hundredths of a centimeter or so (that is, tenths
of a millimeter). Results different from the true value by roughly 0.1 cm will be less common, but
not really rare. Results that differ much more dramatically from the true value will be rarer, and the
more that a measurement value differs from the true value, the less likely it is to occur.
3. Experimental Uncertainty 24

If one were to plot the number of

Histogram of length
measurements obtained versus the
measurements, N = 25
measurement value, we might obtain a
7

Frequency of occurence
graph looking something like the graph
shown in Figure 3.1. Note that the 6
range of measurement values has been 5
divided into “bins”, each 0.02 cm wide,
so, for example, measurement values of 4
8.50 cm and 8.51 cm would both be 3
counted as being in the central bin. (The
2
purpose of grouping values into bins
like this is to show more clearly the 1
characteristic shape of the distribution: a
graph where each bin was only 0.01 cm

8.42

8.44

8.46

8.48

8.50

8.52

8.54

8.56

8.58

8.60
wide would be flatter and less reveal-
ing.) A graph of this type is generally Length of can (cm)
called a histogram.
Figure 3.1: Distribution of 25 measurements
This graph roughly sketches what of the length of a can.
is often called a “bell-shaped curve.” If
we were to plot 100 or 1000 measurements on this graph instead of just 25, the curve would be
more smooth, symmetrical, and bell-like. Measurement values subject to random effects are almost
always distributed in such a pattern. In fact, it is possible to show that a bell-shaped distribution of
values having specific and well-defined characteristics is the mathematical consequence of perturb-
ing effects that are truly random in nature and continuously variable in size. We call the specific
bell-shaped distribution of values caused by such random influences a normal distribution.
Simply by looking at this graph, we can make a rough estimate of the uncertainty of any indi-
vidual measurement value. The definition of “uncertainty” that we have adopted implies that the
uncertainty range should enclose the true value (in this case 8.5129... cm) about 19 out of 20
times. In the case shown above, a range of ±0.06 cm attached to any of the measurements would
include the true value, except for the one case in the rightmost bin. One out of 25 is roughly equal
to one out of 20, so ±0.06 cm would a reasonably good estimate of the uncertainty of a given
measurement in this (hypothetical) case.

3.6 THE MEANING OF THE STANDARD DEVIATION

In chapter 2 of this manual, we defined the standard deviation s of a set of N measure-
ments x1, x2 , x3 , º, x N with mean x to be given by the expression

s ∫
1
N -1
[
( x1 - x )2 + ( x2 - x )2 + ( x3 - x )2 + º( x N - x )2 ] (3.1)

We also discussed the mathematical fact that s for a population of measurements whose distribution
can be modeled by a normal distribution (in the sense that the phrase is used in the last section)
Let the symbol xi stand for an arbitrary “i-th” measurement in our set. Recall that we have
defined the uncertainty U of any measurement xi to be the value such that we are 95% confident
that the “true value” of the measured quantity lies within the range xi ± U. If we have happened to
take a large number of measurements of this quantity, our otherwise somewhat subjective “95%
confidence” can be given a directly quantitative meaning: the measurement’s true value (which
should correspond to the value at the peak of the bell curve) should lie within the range xi ± U for
3. Experimental Uncertainty 25

95% of the measurements xi . Given a set of measurement values, then, we can use this criterion to
determine the value of U. The only problem is that we need hundreds (if not thousands) of meas-
urements to accurately estimate U this way (to accurately determine U, N must be large enough that
the number of measurements in the 5% that exclude the true value is more than just a handful).
Fortunately, mathematicians have shown that it is possible to accurately estimate the value of U that
would have this property for a very large set of measurements from a much smaller set. The uncer-
tainty U of any given single measurement can be estimated using the standard deviation of a small
set of similar measurements as follows:
U ª ts (uncertainty of a single measurement) (3.2)
where t is the so-called Student t-factor, a number that depends somewhat on N, the number of
measurements in the set used to calculate s. A table of t-values as a function of N is shown below.
While uncertainties are generally accurate
only to one significant digit, this table states val- TABLE OF STUDENT t-VALUES
ues to two or three significant digits to show
clearly the difference between adjacent values. N t-value N t-value
Note that for N > 30, the t-value is within a few 2 12.7 10 2.26
percent of being 2.0: for this reason, some books 3 4.3 12 2.2
will tell you that the 95% confidence range for a 4 3.2 15 2.15
given measurement xi is simply xi ± 2s. Howev- 5 2.8 20 2.09
er, this is not a good estimate of that range for the 6 2.6 30 2.05
small values of N that we will commonly en- 7 2.5 50 2.01
counter. In using the table, you should also keep 8 2.4 100 1.98
in mind that it is really only valid for measure- 9 2.3 • 1.97
ments that are randomly distributed. Specifically,
if the uncertainty of your measurement is limited
by the precision of the apparatus rather than random effects, you should not use equation 3.2: you
should instead use one of the strategies outlined in section 3.3
Please note that equation 3.2 estimates the uncertainty of any given single measurement in the
set. As we’ll see later, though, if we have already bothered to take a set of measurements required
to determine s, we might as well compute the mean of the set, which is a better estimate of the
measurement’s true value (that is, it has a smaller uncertainty) than any arbitrarily chosen single
measurement is.
Note in addition that the table is telling you indirectly something about the number of meas-
urements you need to get a good estimate of the uncertainty. In particular, two measurements are
not enough! Three measurements are a bare minimum, and five is a good compromise between
getting a good estimate of the uncertainty and spending too much time on a single measurement.

3.7 THE RepDat PROGRAM

In chapter 2 of this manual, you learned how to calculate the standard deviation of a set of re-
peated measurements using your calculator. Since this is a situation you will encounter often in the
lab, we have written a simple computer program called RepDat (short for “Repeated Data Analyz-
er” to make this even easier. The program is pre-installed on the computers you will use for lab,
and may be downloaded for your personal use from the Physics 51 web site. Figure 3.2 shows a
screen shot of this program in operation. To calculate the mean, standard deviation, and the uncer-
tainty (95% confidence range) of any repeated set of measurements, simply enter the measurement
values as a list of numbers (separated by commas) in the main window and press “Calculate.”
3. Experimental Uncertainty 26

Figure 3.2: Using R e p D a t .

Ignore the “Uncertainty of the Mean” entry for now: we will talk about this in Chapter 11.
Note that this program simply does what you already know how to do: it computes the stan-
dard deviation using to equation 2.2 (as you can easily check), and then multiplies by the appro-
priate Student t-factor from Table 3.1 to get the “uncertainty of the data” (as you can also check.)
Note also that this simple program, like your calculator, is too dumb to round the quantities
to the appropriate number of significant digits. Remembering that uncertainties are generally good
to two significant digits at best, you should read the standard deviation as being 1.6 and the uncer-
tainty of the data as being ±4.4.

EXERCISES

Exercise 3.1
A standard household thermometer has one mark for every two °F. What is the minimum un-
certainty that you should assign to the temperature that you read from such a thermometer?
What do you think is the best uncertainty to assign to this reading, do you think larger than
this? Explain your reasoning (there is no absolute right answer to this last question).

Exercise 3.2
According to your digital bedside clock, it took you exactly 12 minutes to dress for class some
morning. What uncertainty should you assign to this result? Explain your reasoning.
3. Experimental Uncertainty 27

Exercise 3.3
Imagine that you are one of ten different people who measure the time of flight of a thrown
baseball. Assume that these ten measured speeds are as listed below and to the left. On the grid
below and to the right, plot a histogram of this data. (Choose a “bin” size that displays this data
as a pseudo bell-curve rather than scattered data or only one or two columns.)
2.53 s 2.58 s
2.67 s 2.63 s
2.59 s 2.60 s
2.62 s 2.56 s
2.66 s 2.61 s

Exercise 3.4
Compute the standard deviation of the data from the previous problem, and estimate the uncer-
tainty of your particular measurement.

Exercise 3.5
If you have access to a copy of RepDat, check that the results it yields for the data in Exercise
3.3 are consistent with the results you calculated by hand in exercise 3.4.
3. Experimental Uncertainty 28
 4
PRESENTING DATA GRAPHICALLY

“ ‘A crowd in a little room -- Miss Woodhouse, you have the art

of giving pictures in a few words.’”
-- Emma

4.1 INTRODUCTION
“Draw a picture!” is an important general principle in explaining things. Frank Churchill’s re-
mark to Emma Woodhouse notwithstanding, “the art of giving pictures in a few words” is not
nearly as useful as a good diagram or graph, because most people process visual information much
more quickly than information in other forms. Graphing your data shows relationships much more
clearly and quickly, both to you and your reader, than presenting the same information in a table.
Typically you use two levels of graphing in the lab. A graph that appears in your final report
is a higher-level graph. Such a graph should be done very neatly, following all the presentation
guidelines listed at the end of this chapter. It's made primarily for the benefit of the person reading
your report.
A lower-level graph is a rough graph you make for your own benefit; they're the ones the
lab assistants will hound you to construct. These lower-level graphs tell you when you need to take
more data or check a data point, since any strange measurements really stand out in a graph.
They're most useful when you make them in time to act in response to what you see. This means
that you should graph your data roughly before you leave the lab so you still have the chance to
make more measurements. (That's one reason we recommend that you leave every other sheet in
your lab notebook free, so you can use that blank sheet to graph your data.) In graphing your data
in the lab, you don't need to be too fussy about taking up the whole page or making the divisions
nice, but you should label the axes and title the graph to remind yourself later what it shows.
Graphing your data right after you have completed a set of measurements also flags regions
in your data range where you should take more data. Typically people take approximately evenly
spaced data points over the entire range of the controllable variable (the “independent” variable),
which is certainly a good way to start. A graph of that “survey” data will tell you if there are re-
gions where you should look more closely: regions where your graph is changing rapidly, going
through a minimum or maximum, or changing curvature, for example. The graph helps you identi-
fy interesting sections where you should get more data, and saves you from taking lots of data in
regions where little is happening.
Graphing each point as you take it, though, is not a good idea. Doing so is inefficient and,
worse, can prejudice you about the value of the next data point. So take five or six data points and
then graph them all.
For example, Figure 4.1 on the next page shows the original data taken on a phenomenon
called mechanical resonance. All you need to know about resonance for our purposes right now is
that the “amplitude” (a measure of the response of an oscillating system) depends on the frequency
at which that system is perturbed (or “driven”) by an external oscillating force.
Notice that the experimenter initially chose driving frequencies in the first run that were ap-
proximately evenly spaced across the range shown in the graph. For this particular apparatus, the
highest and lowest frequencies attainable with the equipment are easy to find, and the experimenter
chose to space the frequencies evenly to get roughly 10 different frequencies over the range in fre-
4. How To Present Data Graphically 30

quencies. You can see from the graph of the original data that the response doesn't change very
much at either very high or very low frequencies, but near some intermediate frequency, between 4
and 6 cycles/s, something strange and interesting happens.

Mechanical Resonance
50

40
Amplitude (cm)

30

20

10

0 1 2 3 4 5 6 7 8 9 10
Frequency (cycles per second)

Figure 4.1: First set of data (black dots) for amplitude response versus driving
frequency in an oscillating system. Notice that the data points are evenly spaced.

Mechanical Resonance
50

40
Amplitude (cm)

30

20

10

0 1 2 3 4 5 6 7 8 9 10
Frequency (cycles per second)

Figure 4.2: Amplitude vs. driving frequency in a resonant system, after adding
more data points (white dots) between 4.0 and 6.0 cycles per second.
4. How To Present Data Graphically 31

The experimenter noticed this, too, and went back to take more data in the interesting range
of frequencies. The frequency spacing used in the second round is smaller than used in the first set
by about a factor of ten, yielding 15 more measurements in the critical region. The result of adding
the second set of measurements is shown in Figure 4.2. As you can see, the shape of the graph is
now much better defined. Furthermore, the new data show that the anomalously high amplitude at
4.5 cycles/s is not a mistake (as one might think considering the other values). The experimenter
could, of course, have taken data with the closer spacing over the entire frequency range, but that
would waste time on measurements at both low and high frequencies where nothing much is hap-
pening. The strategy of taking coarsely spaced data and then backing up to take more data in inter-
esting regions is a good compromise between completeness and efficiency. But remember that you
usually can’t identify the “interesting” regions if you don't graph your data to begin with!

4.2 ANALYZING YOUR GRAPH

Graphical data analysis is typically used as a euphemism for “find the slope and intercept of
a line.” You will find this semester that you spend a lot of effort manipulating your data so that the
resulting graph is a straight line. As you will find throughout the semester, the slope and/or inter-
cept of such a line often gives useful information about the physical system under investigation.
Determining the slope and intercept of a linear graph is such a common and important task
that we have developed a computer program (described in Chapter 10 in this manual) to help you
do it accurately. Many scientific calculators can also do this (although they almost never give the
uncertainty in the slope and intercept). It is good to be able to estimate roughly the slope and inter-
cept of a lower-level graph by hand, so that you can see if your measurements are at least roughly
correct before you enter them all into the computer. Manual estimates of the slope and intercept also
give you a check on the computer's results (allowing you to catch simple errors like entering the
data in the wrong columns, for example). This process is so important that, although we have this
fond hope that you learned how to do this in high school, we're going to review it anyway.
Imagine that you have constructed a lower-level graph of your data by hand, and it looks
pretty linear. Start by drawing in by eye the line that you think best matches your data. The analytic
procedure called linear regression (described in Chapter 10) gives the optimum result, but in
fact an eyeballed “best fit” line will generally be quite close to the line found by linear regression.
Your job now is to find the slope and the intercept of that line you've drawn.
The slope of a line is defined as the “rise over run,” the change in the vertical coordinate val-
ue divided by the change in the horizontal coordinate value. To determine the slope, you must first
choose two points on your line. They need not be actual data points but they must lie exactly on
your line. They should be about as far apart on the graph as possible to minimize the effects of the
inevitable experimental uncertainty in their position. Mark each of those points with a medium-
large dot or ¥ and/or draw a circle around it. Read the coordinates ( x1, y1 ) and ( x2 , y2 ) of each
point off the graph. (As is the convention, the symbol x here represents the independent variable,
plotted along the horizontal axis, and y is the dependent variable, plotted along the vertical axis.)
With these two coordinate pairs, you can calculate the slope m using the equation
rise Dy y –y
slope = m = = = 2 1 (4.1)
run Dx x2 – x1
substituting your values for ( x1, y1 ) and ( x2 , y2 ) . You can call either point ( x1, y1 ) , as long as you
assign the correct y-value to each x-value. Now that you have m, you can find the y-intercept from
y-intercept = b = y1 - mx1 (4.2)
4. How To Present Data Graphically 32

Again, you can call either point ( x1, y1 ) as long as they both lie on the line. Since the y-intercept is
defined as the value of y where a line intersects the y-axis (defined to be the x = 0 line), you can
also read the intercept directly off the graph as long as the graph shows the x = 0 line.
The graph in the second sample lab notebook in section 1.5 illustrates the analysis of a lower-
level graph. Note the use of ¥’s to mark the points used to compute the slope.

4.3 UNCERTAINTY BARS

Individual data points plotted on any graph should include uncertainty bars, sometimes
called error bars, showing the uncertainty range associated with each data point. You should show
both vertical and horizontal uncertainty bars, if the uncertainties are large enough to be visible on
the graph. If they aren't large enough, you should mention this in your report so we don't think
you've forgotten them. You draw uncertainty bars by indicating the “best guess” value (typically
either a single measured value or mean of a set of measurements) with a dot, and then drawing an
“I-bar” through the dot, whose length spans the 95% confidence range of that value. An example
of such an uncertainty bar is shown in Figure 3.3 below. The single data point plotted corresponds
to a measured pendulum period T of 1.93 s ± 0.03 s for an initial release angle q of 20° + 2°. (The
dotted lines are not part of the graph, but are included to show you how the point and the uncer-
tainty bars are related to the axes. Notice also that the T-axis does not begin at T = 0.)
2.00

1.95

1.90

1.85
0 10 20 30 40

Figure 4.3. A graph of pendulum period T ver-

sus initial angle q, showing how uncertainty bars
indicate the uncertainty ranges associated with the
displayed measurement value.

4.4 PRESENTATION GUIDELINES for "higher level" graphs

You will create "higher-level" graphs for any written work that you submit for a project lab.
These graphs should be more carefully and formally drawn and labeled than the lower-level graphs
that appear in your lab notebook. Here are some guidelines for constructing these graphs:
1. Draw your graphs in pencil; mistakes are easy to make. If you wish, go back later and touch
them up in ink. High-quality graph paper may be purchased from Connie Wilson, the phys-
ics department secretary for 10¢ a sheet. Computer-drawn graphs are fine as long as they
comply with the remaining guidelines. The program described in Chapter 10 of this manual
makes it very easy to produce graphs that automatically comply with all these guidelines, but
the graphs produced other programs (such as Excel or Cricket Graph) may require extensive
modification to fit the remaining guidelines.
4. How To Present Data Graphically 33

2. Scale your axes to create as large a graph as possible consistent with the constraint that the
divisions on the axes correspond to some nice interval like 1, 2, or 5 (times some power of
10). If you must make the graph smaller than full size to get nice intervals, OK, but check
that you've picked the interval that gives you the largest possible graph (which will display
your data in as much detail as possible). When using log-log or semi-log paper, choose paper
with the number of cycles that gives the largest possible graph (see Chapters 5 and 13).
3. The lower left-hand corner need not be the point (0,0). Choose the range of values for each
axis to be just wide enough to display all the data. If (0,0) does not appear on a hand-drawn
graph, it is customary to mark the break in the axis or axes with two wavy lines (ª).
4. Mark the scale of each axis along each axis for the entire length of the axis.
5. Label both axes, identifying the quantity being plotted on each axis and the units being used.
6. Give each graph a title that summarizes the information contained in the axes and provides
any additional information needed to distinguish this graph from other graphs in the report.
7. Give each graph a number (e.g., "Figure 2"), which you use in the body of the report or
summary to refer quickly to the graph. (You can write such a number on a computer graph.)
8. Draw points and uncertainty bars as discussd in section 4.3.
9. If you calculate the slope and intercept of the graph from two points (rather than using the
method of linear regression described in chapter 10), indicate the two points you used on the
graph. Draw the line through the two points, label it "Best-fit line" (or something similar),
and give its slope and intercept on the graph in some large clear space.

4.5 CHECKLIST FOR EACH GRAPH IN A WRITTEN REPORT

Use this checklist to make sure that any higher-level graph that you include in a submitted
written report (as opposed to your lab notebook) has the correct features and format.
The axes are scaled correctly with divisions equal to "nice" intervals.
The graph is drawn as large as possible so that it fills the page.
The scales on the axes have tick marks that run for their entire length.
The axes have labels describing the variables they represent (including units).
The measured data points are clearly plotted, including uncertainty bars.
The graph has an appropriate title and figure number.
The points used to calculate the slope and intercept are clearly marked
(if that method is used).

Table 4.1: Kinetic energy per unit mass of a

EXERCISES falling object as a function of distance fallen.

Exercise 4.1 Distance fallen Kinetic energy

from rest (m) per unit mass (J/kg)
Using the blank graph paper on the next page, 0.200 ± 0.003 2.00 ± 0.10
create a higher-level graph of the data provided in
0.400 ± 0.003 3.78 ± 0.15
Table 4.1. Use the checklist in section 4.5 to
0.600 ± 0.003 5.65 ± 0.20
make sure that you have included everything.
0.800 ± 0.003 8.01 ± 0.25
Exercise 4.2 1.000 ± 0.003 9.82 ± 0.30

Draw what you think is the best possible line through the data points in the graph you just
created in the last problem, and find the slope and intercept of this line.
4. How To Present Data Graphically 34
 5
POWER-LAW FITTING AND LOG-LOG GRAPHS

“She had taken up the idea, she supposed, and made everything bend to it.”
--- Emma

5.1 DEALING WITH POWER LAWS

Although many relationships in nature are linear, some of the most interesting relationships
are nonlinear. Power-law dependences, of the form
y( x ) = kx n (5.1)
are particularly common. In many cases, we might suspect that two experimental variables are re-
lated by a power-law relationship, but are unsure of what k or n are. For example, we might sus-
pect that the period T of a simple harmonic oscillator might depend on the mass m of the oscillating
object in some kind of power-law relationship, but we might be unsure of exactly what the values
of either n or k . If we knew n, then we could plot y vs. x n to get a straight line; the slope of that
line would then be k. But if we don’t know n, it would seem at first glance that the best we could
do would be simply try different possible values of n to see what works. This could get old fast.
We can, however, take advantage of the properties of logarithms to convert any relationship
of the type given by equation 5.1 into a linear relationship, even if we do not know either k or n!
The most basic property of logarithms (for any base, but let’s assume base-ten logs) is that:
log( ab) = log a + log b (5.2a)
From this it follows that
n times n times
log( a n ) = log( a ◊ a ◊º◊ a) = log a + log a + º + log a = n log a (5.2b)
(this is actually true even for non-integer n). Also, in the case of base-ten logarithms, this means

log(10 b ) = b log(10) = b ◊ 1 = b (5.2c)

which shows that raising 10 to a power is the inverse operation to taking the (base-ten) log.
With this in mind, let us take the (base-ten) logarithm of both sides of equation 5.1 and use
the properties described by equation 5.2. If we do this we get

log y = log(kx n ) = log k + log x n = log k + n log x (5.3)

Now define log x ∫ u and log y ∫ v. Substituting these into 5.3 and rearranging, we get
v = nu + log k (5.4)
Now, this is the equation of a straight line. This means that if we graph v vs. u (that is, log y vs.
log x), we should end up with a straight line, even if we do not know what n and k are.
Furthermore,the slope of this line is the unknown exponent n in equation 5.1. We can there-
fore find the value of n by calculating the slope in the usual way. That is,
Dv v –v log y2 – log y1
n = = 2 1 = (5.5)
Du u2 – u1 log x2 – log x1
5. Power-Law Fitting and Log-Log Graphs 36

The value of the intercept (which is the value of v = log y when u = log x = 0) is log k , so if we
can find the intercept and its uncertainty, we can find k and its uncertainty.
In summary, we can take any relationship of the form given in equation 5.1, take the loga-
rithm of both sides, and convert it to a linear relationship whose slope and intercept are related to
the unknown values of n and k. This is, therefore, a very powerful way of learning about un-
known power-law relationships (and displaying them). A graph that plots log y versus log x in
order to linearize a power-law relationship is called a log-log graph.

5.2 AN EXAMPLE OF A LOG-LOG GRAPH

As an example, consider a hypothetical experiment testing how the period of an object oscil-
lating at the end of a spring depends on the object’s mass. Table 5.1 gives a set of measurements
taken during this experiment, Figure 5.1 shows a graph of period vs. the mass. Table 5.2 shows
the base-10 logs of the measurement data, while Figure 5.2 shows a graph of logT vs. logM.
Table 5.1 Table 5.2
mass (M) period (T) log M log T
0.125 kg 2.42 s –0.903 0.384
0.325 kg 3.76 s –0.488 0.575
0.525 kg 4.75 s –0.280 0.677
0.725 kg 5.52 s –0.140 0.742
0.825 kg 5.87 s –0.084 0.769

0.8

6 0.6
Period (seconds)

log(period)

4 0.4

2 0.2

0 0
0 0.2 0.4 0.6 0.8 –1.0 –0.8 –0.6 –0.4 –0.2 0
mass (kg) log(mass)
Figure 5.1: Graph of the oscillation period as a Figure 5.2: Graph of the log of the oscillation
function of mass. period as a function of the log of the mass.

We can see that the graph of the period vs. the mass does not yield a very good straight line:
(if the uncertainties are smaller than the dots representing the data points, we would have to say
that a straight line is inconsistent with the data). On the other hand the plot of logT vs. logM is a
very nice straight line, suggesting that the period T and the mass M have a power-law relationship.
What are n and k according to this experiment? We can easily get a quick estimate from the
graph. For the sake of round numbers, consider the points marked with ¥s on the graph above.
The slope of this graph (rise over run) is thus
5. Power-Law Fitting and Log-Log Graphs 37

log T2 – log T1 0.72 – 0.43

n = = = 0.48 (5.6)
log M2 – log M1 0.80 – 0.20
(The two points that you choose to compute the slope need not correspond to actual data points:
simply choose convenient points on your drawn line near the ends of the line. I chose the points so
that the denominator of the expression above would be simple.) Since exponents in physical situa-
tions are most often integers or simple fractions, we might guess that the actual value of n is 1/2
(this turns out to be the theoretical value as well as we will see chapter N11 in the class text).
The intercept is the place where the line crosses the logM = 0 grid line. According to the
graph, this is roughly where logT = 0.82 (note that the logM = 0 line is the right edge of the graph
here, not the left!). So the intercept is 0.82 = log k, which means that k = 10 0.82 = 6.6 s/kg1/2 .

5.3 A COMMENT ABOUT LOGARITHMS AND UNITS

You may also have noticed that units seem to spring in and out of these calculations rather
haphazardly. This behavior conflicts with a general rule about special functions: the log function,
like sine, cosine, and the exponential function, is supposed to have a dimensionless arguments.
That is, the number you take the logarithm of is supposed to be a pure number, with no units or di-
mensions (like seconds or centimeters), attached to it. What's going on?
One can think of a number with units as the product of the number part and the units part; that
is, 10 cm is 10 (a pure number) multiplied by the unit of centimeters. When you take the logarithm
of a number with units, then, you are taking the logarithm of this product, which behaves like the
logarithm of all products: the result is the sum of the logarithms of the two numbers making up the
product. The logarithm of 10 cm is therefore log (10) + log cm = 1 + log cm. You can't give a nu-
merical result for log cm, which may seem rather disturbing.
Fortunately, this all works out fine anyway, because you typically find the logarithm of a
quantity with units only when you in the process of finding the difference between two logarithms
of quantities with the same units. Consider, for example, the situation where we compute the slope
using equation 5.6. If we attach to each logarithm in the numerator the appropriate term of log s
(where s is the unit of seconds here), we see that two log s terms cancel out when we subtract,
leaving a unitless number in the numerator. The same thing happens in the denominator. There-
fore, the slope ends up being a unitless number, as it should be.
To find the units of k, note that equation 5.4 in this situation really ought to read
log t + log s = n log m + n log kg + log k (5.7)
(where in this expression we are considering t and µ to be the unitless parts of T and M respec-
tively). The intercept is where the line crosses the vertical grid line corresponding to log m = 0: we
see from the graph that logt has the value log t 0 = 0.82 there. Therefore equation 5.7 implies that
log t 0 + log s = n log kg + log k fi log k = log t 0 + log s – n log kg (5.8)

When we take the antilog of (that is, 10 to the power of) both sides of this, all the items in logs get
multiplied together, so we get (assuming that n is really 1/2):

10 log t 0 ◊ s 10 0.82 ◊ s s
k= n
= 1/2
= 6.6 1/2 (5.9)
kg kg kg
Keeping track of these unit terms when working with logarithms involves a lot of work,
however, and less often pays off the way that keeping track of units in normal equations does.
5. Power-Law Fitting and Log-Log Graphs 38

Therefore, people generally ignore the units associated with logarithmic quantities, and fill in the
units of quantities after taking the antilog (as I did with k in the last section) to make them consist-
ent across the master equation 5.1. But if you ever get confused about units and want to make sure
that things work out correctly, this is how to do it.

5.4 A PROCEDURE FOR EXPLORING POWER-LAW RELATIONSHIPS

Log-log graphs are most useful when you suspect your data has a power-law dependence
and you want to test your suspicion. Sometimes your suspicion is based on a theoretical predic-
tion, sometimes a previous low-level Cartesian plot. Figures 5.3 through 5.5 are typical Cartesian
graphs that could be power laws. Whatever the source of your suspicion, your next step is to plot
the logarithms of your data as a low-level graph. If this graph looks like a pretty good straight line
(within your experimental uncertainties) you can proceed to the next steps.
Once you have plotted the points, you should use a ruler to draw the straight line that you
think best fits your data. You can then use this line to estimate the slope n using equation 5.6. You
can also estimate the value of the constant k in equation 5.1 by extrapolating your straight line back
to the vertical grid line where the value of the independent variable (let’s call it x) is equal to 1 (and
thus log x = 0 ): the value of log k is the vertical scale reading where your best fit line crosses this
vertical grid line where log x = 0 . Note that this vertical line may not correspond to the left edge of
your graph! In Figure 5.2, for example, it happens to be at the right edge of the graph, and on a
general log-log graph, it could be almost anywhere. (If the line log x = 0 is off the edge of your
graph, you can often bring it onto the graph by changing the units of x. For example if x is a dis-
tance ranging from 20 cm to 200 cm, the place where log x = 0 is when x = 1 cm, which will
probably be off the left side of your graph. But if we change the units of x to meters, then the place
where log x = 0 is where x = 1 m, which is right in the middle of your data.)
To estimate the uncertainties of these quantities, draw a new line with the largest slope that
you think might be consistent with your data, and another line with the smallest slope consistent
with your data, and find the slope and intercept for each of these lines. The greatest and least slope
will then bracket the uncertainty range of n and you can use the greatest and least values of log k
determine the greatest and least values of k, which bracket the uncertainty range of k.
To go further than crude estimates, one needs the help of a computer. The program LinReg,
which is discussed in Chapter 10 of this manual, makes it very easy to plot log-log graphs and find
the best-fit slope (with its uncertainty) and the best-fit intercept (and its uncertainty).
40 40 40

30 30 30
y y y
20 20 20

10 10 10

0 0 0
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
x x x
Figure 5.3: y = x a , a > 1. Figure 5.4: y = x a , a < 1. Figure 5.5: y = x a , a < 0.
5. Power-Law Fitting and Log-Log Graphs 39

EXERCISES

Exercise 5.1
The table to below gives the orbital periods T (in years) of the planets known to Newton as a
function of their mean distance R from the sun in AUs (where 1 AU = the earth’s mean orbital
radius). Plot plot a log-log graph of the period versus the distance on the graph paper provided
as Figure 5.6 on the next page.

Table 5.3 Planetary Periods

vs. Mean Orbital Distances
Planet Distance (AU) Period (y)
Mercury 0.39 0.24
Venus 0.72 0.62
Earth 1.00 1.00
Mars 1.52 1.88
Jupiter 5.20 11.86
Saturn 9.54 29.46

Exercise 5.2
Assuming that the period and distance are related by a power-law of the form T = kRn , where
n is an integer or simple fraction, what does your graph suggest is the likely value of n?

Exercise 5.3
Find the value of k (with appropriate units) for the data of Table 5.3 from the intercept of your
log-log graph. Combine this with the result of exercise 5.1 to find the power-law equation (of
the type given in equation 5.1) that seems to fit this data.
5. Power-Law Fitting and Log-Log Graphs 40

Figure 5.6
5. Power-Law Fitting and Log-Log Graphs 41

5.5 (OPTIONAL) USING LOG-LOG PAPER

If you have more than five or ten data points, calculating the logarithms quickly gets tedious
even with a calculator. To reduce this tedium (which would have been particularly gruesome in the
era before computers and calculators), someone invented a special kind of graph paper called log-
log paper. In effect, this kind of graph paper calculates the logarithms for you.
Imagine that you have data for an independent variable x that ranges from, say, 0.01 m to
about 10 m. The values of log x would then range from about –2.0 to 1.0, and a useful horizontal
scale might look something like this:
log x
–2.0 –1.5 –1.0 –0.5 0.0 0.5 1.0
Now, imagine that we were to also draw a scale immediately above this scale that showed the cor-
responding values of x. The two scales together would look like this:
0.01 0.1 1.0
0.02 0.03 0.05 0.2 0.3 0.5 2.0 3.0 5.0 10 x
log x
–2.0 –1.5 –1.0 –0.5 0.0 0.5 1.0
(Note how the pattern of the spacing between marks on the upper scale is identical for each power
of 10.) Now, note that if we had graph paper that had its axes pre-labeled as shown in the upper
scale, then we could locate points on the plot directly according to their value of x rather than hav-
ing to compute the value of log x for each data point.
The way that the pattern repeats for each power of 10 makes it possible to create general and
flexible graph paper with essentially pre-labeled scales. Figure 5.7 on the next page is an example
of log-log paper which has 3 cycles of the pattern horizontally and vertically, making it possible to
display data points whose x and/or y values span up to three powers of 10 (or three decades). If
you compare this graph paper to the double-axis shown above, you will see that only the equival-
ent of x scale is displayed on this paper: the log x scale has been suppressed for the sake of clarity,
but should be considered implicit. Also you will note that the publishers of the graph paper do not
commit you to particular powers of 10: each decade is labeled as if it spans from 1 to 10. You can
cross out the numbers shown to adapt the graph paper to the particular ranges of your data points
(this is conventionally done for just the numbers labeled 1 or 10 on the graph paper). Figure 5.7
shows how you would do this for the planetary orbit data given in Table 5.3.
The point is that with just a little relabeling you can use graph paper like this to construct
quickly a log-log graph without having to do any actual calculations of logarithms. This is great for
doing low-level, quickie graphs of a set of data that you think might reflect a power-law relation.
y = kx n . You can even read the value of k directly from the graph by finding the point where your
best-fit line crosses the vertical line corresponding to x = 1 unit and reading the vertical coordinate
of this point according to the vertical axis.
But how can you compute the slope n of data drawn on such graph? Remember each axis has
an implicit linear scale that reflects the log of the value displayed. “Linear” means that the change
in the value of log x or log y is proportional to the physical distance on the sheet of paper. So to
find the slope, all that you have to do is measure the rise of your line (in cm on the sheet of paper!)
and divide by the run (in cm).
Using log-log paper is optional in this course. But you can purchase sheets of log-log paper
from Connie (the department secretary) for 10¢ per page if you would like to use it for quick low-
level graphing.
5. Power-Law Fitting and Log-Log Graphs 42

(OPTIONAL) Exercise 5.4

Plot the planetary orbit data in Table 5.3 on the log-log paper shown below, and use the meth-
ods described in the previous section to find k and n assuming that T = kRn . Check that these
agree with the values you found before.

Figure 5.7: An example of 3-cycle ¥ 3-cycle log-log paper.

 8
HOW TO WRITE A LAB REPORT
“… it was in plain, unaffected English, such as Mr. Knightly
used even to the woman he was in love with…”
--- Emma
8.1 INTRODUCTION
Science is fundamentally a communal process, in which individual scientists develop ideas
and then seek through the medium of scientific journal articles to convince the scientific community
of their validity. Learning how to communicate your ideas effectively is therefore a crucial skill for
a working scientist (and is useful in many other callings as well).
Consequently, you need know how to describe the science that you do in a way that convinc-
es the reader that your work is interesting and should be taken seriously. You may feel that com-
paring your lab work and the resulting report to “real” science that appears in journals is a bit pre-
tentious, since we’re probably not going to have you do much cutting-edge physics in an introduc-
tory laboratory. The purpose of the lab reports, though, is not so much to see if you did bold,
original work as it is to give you practice in writing scientific reports, so that you'll be able to do it
well when you do do bold original work.
Most articles in scientific journals (physics journals, at least) follow at least approximately a
standard format, which looks something like this:
ABSTRACT
I. INTRODUCTION
A. Motivation
B. Summary of the experiment
II. THEORETICAL BACKGROUND
III. EXPERIMENTAL DESIGN AND PROCEDURE
A. Description of the apparatus
B. Description of the experimental procedure
IV. ANALYSIS
A. Method of analysis
B. Presentation of results
C. Discussion of results
D. (Optional) Suggestions for future improvements
V. CONCLUSIONS
A. Summary of the results
B. Pertinence of the results to the questions raised in the introduction

This format has evolved to answer the general questions a potential reader will ask:
What did you do? (Procedure)
Why did you do it? (Introduction, Theoretical background)
How did you do it? (Procedure, Analysis)
What happened? (Analysis, Conclusions)
The format also provides some shortcuts for busy (or lazy) people. Most scientific prose
tends to be fairly dense, and readers like to find out in a hurry if a paper is actually of interest or
8. How to Write a Lab Report 56

importance to them. The abstract section provides a concise summary of the article and its most
important results, so the reader only has to read a few sentences to determine if the entire article is
relevant. The introduction and conclusions contain a little more information; usually the reader
goes to the introduction for more information about the motivation and the method of the experi-
ment, and the conclusion for more detail on the results summarized in the abstract.
Each of these report sections is discussed in a separate section of this chapter. You will prob-
ably find it helpful to read over the entire chapter the first time you are asked to write a lab-report
section (to get some sense of how the pieces of a lab report fit together). At the end of the semes-
ter, when you will write a full report, you should go back and read the entire chapter again.

8.2 THE SHORT SECTIONS: The Abstract, Introduction, and Conclusions

Most published scientific papers are not read in their entirety by everyone who looks at them.
It's not that they are poorly written (although some certainly are), and it’s not that scientists don't
care; there are just so many hours in a day. The short sections of a technical paper -- the abstract,
introduction, and conclusions sections -- identify the important results of your work, and per-
suade a reader that really reading the paper is worth the time. Typically a reader will look first at the
abstract, to find out what the paper is about. If the abstract looks promising, the reader will look at
the conclusions. If they look interesting (and especially if they're unexpected) the reader will then
check the introduction to see if the experimental method looks good. If the introduction suggests
that you knew what you were doing, then reader will read the rest of the article for the details.
8.2.1 The Abstract
An abstract is an extremely terse summary of the entire paper, about three to six sentences
long, which in a journal appears in small print just below the article’s title and list of authors. (The
abstract is also often published separately and distributed more widely than the article itself.) The
purpose of an abstract is to provide readers with a brief glimpse into the subject of the article, to
help them decide whether to read the whole thing. One of the first things that one does when be-
ginning a research project is to search recent publications for articles that might be helpful: good
abstracts make it possible to determine quickly which articles are relevant.
The structure of the abstract is essentially a miniature version of the structure of the article,
except that each of the five major sections (introduction, theory, experimental design, analysis,
and conclusions) might be represented in the abstract by only a sentence or even a phrase. Often
the theory section is omitted completely from the abstract unless the paper is theoretical (which will
not be the case for your lab reports!) Even so, the outline for the whole article is a pretty good
starting point for the outline of the abstract as well. The abstract should always summarize the in-
troduction and conclusion sections; this means that it will always include a short summary of what
question you were seeking to answer, what your results were and what they imply. Although the
abstract is the first section of a lab report, you may want to write it last because it is a summary.
In particular, a physics abstract should include a summary of any quantitative results you re-
port in your conclusions. Apparently including quantitative results in the abstract is not standard in
chemistry and biology articles (or so some chem and bio majors say when we criticize them for
omitting this information), but it is standard in physics. Remember, the abstract is the "hook" you
use to get people to read the rest of the paper, and you can best capture their attention with a nice
juicy quantitative result with a promisingly small experimental uncertainty.
8.2.2 The Introduction
The introduction section is meant to provide the reader with the answers to two very im-
portant questions: What is the question that your experiment is supposed to answer, and why
8. How to Write a Lab Report 57

is answering this question interesting (and/or important)? In a published journal article, this sec-
tion often begins with a brief summary of previous related research, a statement of a problem that
this research has raised, and a brief description of the experiment in question and how it addresses
the problem. Detailed descriptions are not appropriate in this section; the point is to provide a con-
cise picture of your purposes and a broad survey of your approach. This section should capture the
interest of your readers, provide them with some general orientation, and convince them that what
you are doing is interesting and worth reading about.
After you motivate the experiment, you should give a brief summary of the experimental
method you will use. This need not be extensive; the detailed description goes in the procedure
section, which is separated from the introduction only by the theory section. You need to give
enough information so that a reader who is interested primarily in your method, perhaps to du-
plicate your experiment or apply it to a related problem, can see if that method is appropriate.

8.2.3 The Conclusions

A conclusions section should, in one or two paragraphs, review the purpose of the lab and
summarize the implications of your experimental results. That is, you should remind the reader of
the basic question that the experiment was to address (as presented in the introduction), and then
briefly explain how your results bear on that question or problem. This section should be a sum-
mary of information presented elsewhere rather than a place to present new information: the pur-
pose of this section is to close the report with a review that highlights the most important results.
As with the abstract, you should report quantitative results and their experimental uncertainties.
Students often ask, “What's the difference between the conclusions and the abstract?” The
answer is, “Not much.” Both are summaries of the rest of the report, and both contain quantitative
results. The main differences have to do with location: the abstract is the “hook” at the beginning,
and should contain hints of the wonders to come. It also summarizes the entire report. The con-
clusion comes at the end, and should give some sense of finality or closure. It will emphasize your
deductions from your data analysis, describing them in more detail than is given in the abstract.
Both the abstract and the conclusions should report comparisons between predictions, presumably
made in your theory section, and your measurements or their consequences.

8.2.4 Appropriate Detail in the Short Sections

By referring to the abstract, introduction, and conclusions sections as being the “short” sec-
tions, we imply that these three sections shouldn't be long enough to contain much detail. That's
right. The abstract and conclusions sections in particular should be the least detailed, giving the
broadest look at the purpose of the experiment and the implications of the results. The introduction
should be a bit more detailed, but not much: its focus should be on a general statement of the prob-
lem to be considered and the experimental method used to study that problem. Too much detail in
any of these sections will obscure the reader's view of the main issues in the report.

8.2.5 A Checklist for the Short Sections

(All checklists in this chapter are summarized on the inside front cover of this reference manual.) In
your short sections, you should
Summarize the entire paper in the abstract
Discuss quantitative results in both the abstract and conclusions
State the problem or question under investigation in the introduction
Summarize the experimental procedure in the introduction
8. How to Write a Lab Report 58

8.3 THE THEORY SECTION

The theory section is meant to provide the reader with enough mathematical or theoretical
background to understand how the experiment works, what assumptions have been made, and
how the experiment is related to the physics being studied. This section may be very short (or even
non-existent) if the theory is well-understood and the connection between the theory and the meas-
urements are straightforward and obvious. It can be quite extensive, however, if the experiment is
complex or the actual measurements being made are related in a complicated way to the results be-
ing compared to the theory.
If, for example, you were measuring the average velocity over some interval for your experi-
ment, your theory section would be very short: you measure a distance and a time, divide the first
by the second, and there’s your average velocity. Suppose, on the other hand, that your experi-
ment was the determination of an acceleration in a situation where you couldn’t be sure the object
was starting from rest. It is still possible to find the acceleration, but you have to measure two time
intervals over two distances, and the connection between those measurements and the final result
involves a fair amount of algebra. In that case, you would be expected to derive the connection for
your theory section, which you could expect to be one or two pages long. You don't need to show
each algebraic step, but you should show some intermediate results, especially if they involve
complicated algebra, a substitution, or some trick of manipulation.
The amount of theoretical background that you provide also depends on the expertise of your
intended audience. For the purposes of this course, you should imagine your typical reader to be a
classmate (not a professor or a lab assistant), who for some reason has not done the lab in question
and knows nothing about it. This situation is analogous to that of a researcher whose audience has
quite a bit of general knowledge about physical principles and experimental techniques, but no ex-
perience with the specific experiment.
For this semester, at least, you should start your theory section with first principles, or at
least the equation (such as the law of conservation of momentum or energy) that defines the phe-
nomenon you'll be studying. In a journal article you wouldn't go this far back, because starting
from first principles to get to the result would take up too much space. Doing so in the introductory
lab is a good idea, though, because you're probably just learning how to write a theory section.
And since you're doing experiments that are usually close to the basic principles, starting with
those principles helps you to examine your assumptions carefully.

8.3.1 Writing down equations

Theory sections tend to involve equations. There are three general rules about equations in text.
Rule 1: Don't write equations in the body of the text. Give each equation a line of its
own. (Set aside three or four lines in your printout if you write in equations by hand.) You may
break this rule for very simple equations you will not need later. For example, “When L = 1.0 m,
the period of a pendulum is about 2 s.” Setting equations apart from the text makes the text read
more smoothly, and also signals to the reader that it's time to go into Math Mode. You also get
more room for writing the equation.
Rule 2: Give every equation a number (except the simple ones mentioned in Rule 1). This
way you and the reader can find them easily later on.
Rule 3: Don’t try to typeset equations without an equation editor. If your word pro-
cessing software doesn’t have a integrated equation editor that will let you typeset equations in
standard form, don't try to type in an equation or parts of it; instead write the entire equation in by
hand, instead. Faked Greek letters are almost never recognizable, and the time required to get frac-
8. How to Write a Lab Report 59

tions to print out correctly isn't worth it. Most readers don't instantly recognize “**” or “^” as
meaning exponentiation, either, and they look terrible. You can get "+" by underlining the "+"
sign; don't use "+/-," because it looks terrible, too.
If you do write in equations by hand, don't forget to enter them after you print out your re-
port! Missing equations are a sure tip-off that you forgot to proofread your report. People seem es-
pecially prone to forgetting the Greek letters and special symbols in partially typeset equations,
whereas they usually notice those big blank spaces set aside for entire equations.
(Note: Recent versions of Microsoft Word, WordPerfect, and many other word-processing
programs for both the Mac and Windows operating systems have integrated solid equation editors,
and one can buy good stand-alone equation editors relatively cheaply. Dr. Moore likes MathType,
which is easy to use and can be used with any word processor: see www.mathtype.com. There is
therefore no excuse any more for attempting to typeset equations without an equation editor. Writ-
ing equations in by hand, however, is perfectly acceptable and will not lower your report grade.)
Look at the class text for examples of good style regarding equations: the book was typeset
according to McGraw-Hill’s professional standards for science texts (as described in a long docu-
ment that Dr. Moore has). Note in particular that variables should always be set in italics:
this helps set them apart from the text and identifies them as variables as opposed to just letters.

8.3.2 Checklist for a Theory Section

Your theory section should:
Start with the basic defining equations
Show all non-obvious intermediate algebraic steps
Clearly describe any assumptions and/or approximations involved in the model
Display each equation on its own line
Give each equation an equation number

8.4 THE PROCEDURE SECTION

Your job in the procedure section is to convince your reader that you carried out an experi-
ment carefully and knowledgeably enough that the reader should take your experimental results
seriously. In describing your experimental procedure, you should think of the reader as someone
who is unfamiliar with the particular experiment you are doing but who is familiar with the pitfalls
of working with the equipment you will be using. Furthermore, to keep you on your toes, you
should think of this reader as being someone who is inclined to be skeptical about your results and
hence will be picky about your procedure. (This doesn't sound very friendly, but professional sci-
entists act just this way reading other authors’ papers, especially about experiments they wish
they'd thought of doing, or about experiments they were about to do themselves.)
Consider, for example, an experiment you will do later, measuring the period of a pendulum
as a function of several variables. Simply saying, “We measured the pendulum period as a function
of mass hanging from the end” doesn't do justice to what's really a rather elaborate procedure.
Making this measurement carefully requires multiple measurements, timing several periods for
each measurement, and choosing a particular starting and stopping point in the swing, all to reduce
the uncertainty in your results, and you should say so. You should also explain why you went to
all that trouble; doing so enhances your credibility with the reader, providing evidence that you
thought carefully about the experiment. (It also justifies going to all that trouble.)
Most procedure sections have a fairly standard format, which (as usual) you should feel free
to modify. A typical description of experimental procedure starts with a list and description of the
8. How to Write a Lab Report 60

equipment. The equipment description should state the precision to which measuring devices read.
Anything that isn't a standard device should be described somewhat quantitatively. (For example,
in the pendulum experiment you would give the approximate length of the string, and say someth-
ing that would tell the reader whether to look in the stockroom for lightweight fishing line or big,
hefty twine for wrapping packages.) Identify your lab station by its number if it has one. Large
pieces of equipment should be identified by manufacturer's name, model, and serial number,
which you should have written down in your lab notebook. Giving this information in your report
tells the reader what performance is possible from the equipment you used.
It is very important that you also give the reader a sketch of the apparatus. A good and com-
plete sketch may be able replace a text list of equipment, and if so, it should be used instead.
Sometimes this sketch will be schematic in nature, like a block diagram or a circuit diagram; in that
case, a computer-drawn sketch is fine. In cases where you need to show fine detail, or where it's
important to show the geometry accurately, a carefully hand-drawn sketch is usually better (and
takes much less time to do well). Unless you are very skilled or have very good drawing software,
computer drawings don't normally look enough like the objects they represent to be useful.
The list and/or sketch of the apparatus tell the reader what equipment was available to you,
and to some extent whether you set it up in an appropriate fashion. Next, you tell what you did
with the equipment. You should do this in a logical order, but not be too "step-by-step" about it.
Specifically, avoid a numbered list of steps, which are difficult to read and hence inappropriate ex-
cept for the rare reader who intends to repeat your experiment exactly. At the other extreme, you
should avoid narratives like this: "First we did (whatever), but that didn't work, so then we tried
(something else) to fix the problem with the first measurements." Refine your procedure to remove
these false steps, and present it in enough detail so that the reader can clearly understand what you
did without being overwhelmed by irrelevant tiny details.
If you've made some revision in some seemingly obvious procedure that significantly im-
proves the accuracy of your results, though, make sure you take credit for it. For example: “At the
longest pendulum lengths (L > 1 m), the pendulum frequently hit the wall before completing ten
swings. For those lengths we only timed five swings. This gave satisfactorily consistent results."
You can also refer to the lab manual if its description of the procedure is sufficiently detailed
(many articles in professional journals refer to other papers for details regarding equipment or pro-
cedure), but be especially sure to include a complete description of any procedural details that do
not appear in the lab manual! In referring to a lengthy source like the lab manual, state the author,
title, year of publication, and page number. (For example, a reference to this booklet should look
like: Moore and Zook, Laboratory Reference Manual for Physics 51a, 2001, p. 16.) A reference
to a journal article would state the author, journal name (but not the article title), volume number,
number of the first page, and year of publication. Instructions from the lab instructor or lab assis-
tant can be cited as A. C. Zook, 1999, private communication. (This format is used in journal arti-
cles to refer to a conversation, unpublished letter, or e-mail message from the person cited.)
You might also consider the following questions as you write this section:
1) How did you determine the experimental uncertainties that you chose?
2) What (if anything) did you do to reduce them?
3) Did you experience any difficulties with the apparatus? If so, how did you resolve them?
4) Did you encounter any problems or difficulties in following the lab manual's procedure?
If so, how did you resolve them?
5) Did you modify that procedure in any way, and, if so, how and why?
8. How to Write a Lab Report 61

Standard techniques, such as the correct use of a stopwatch or a vernier caliper, need not be
described in your procedure section. Unless you've given us some reason to be wary of your abili-
ty to use a device that you’ve presumably either used before (for example, the stopwatch) or re-
ceived some instruction about (for example, a caliper) we'll assume that you used it correctly.
One detail you should definitely include, at this stage in your career, is the number of times
you repeated any given measurement. Every year, we’re surprised at the number of students who
don't seem to remember the importance of repeated measurements. Remember that repeating re-
peatable measurements is essentially the only way to determine the measurement’s uncertainty!
Although you will formally calculate the experimental uncertainty in the analysis section, it's good
to mention the uncertainty ranges of your basic, unprocessed measurements in the procedure sec-
tion, or at least state whether a given measurement was repeatable or not.
Finding the appropriate level of detail is difficult. You don’t need to tell the reader every-
thing, but you do have to say enough. The ideal procedure section is one that provides just enough
so that the reader to go into the lab stockroom, pick out the right equipment, repeat the experiment,
and get results consistent with yours based only on the information in your report and the lab man-
ual. Providing just the right amount of detail requires practice, and probably the most aggravating
comments you'll get on your lab reports will be in this section.

8.4.1 A Few Comments on Style...

Procedure sections are right up there with theory sections for putting the reader to sleep. In
procedure sections, the culprit is usually excessive use of the passive voice. (“The ball was hit by
the batter” rather than “The batter hit the ball.”) In the natural sciences, we have this fond hope that
the identity of the experimenter should not affect the result of the experiment, except insofar as one
person may be a more skilled than another. Writing in the passive voice became standard in the sci-
entific community partly to move the experimenter one step back. Unfortunately, the passive voice
is really boring to read, partly because it wordier and partly because it dilutes the sense of action.
The procedure section is a place where the historical convention especially required the pas-
sive voice. In other parts of the report, the spring exerts a force, or some results suggest an inver-
se-square law; since you’re out of the picture, the active voice was acceptable. But in the procedure
section, you describe what you did, except that your identity isn’t supposed to important.
However, the times are changing! We here at Pomona are not the only people who have trou-
ble staying awake reading technical literature, and we’ve noticed that authors of scientific papers
are more often saying things like “We observed NGC 253 on seven consecutive nights looking for
supernovae,” and even (gasp) “I measured the activity of the radioactive sample at 15-minute inter-
vals.” So go ahead, be on the cutting edge -- every so often, admit that human beings with names
and faces make those measurements. If you're describing a division of labor, the standard phrase
seems to be “One of us (TAM) calibrated the Heisenberg compensators while the other (ACZ) car-
ried out the tachyon-beam efficiency measurements.”

8.4.2 Checklist for a Procedure Section

Your procedure section should:
Provide a sketch or schematic diagram of experimental setup
Provide a textual list and/or descriptions of equipment (when needed for clarity)
Describe all measurements, in roughly the order in which they were made
Describe any departures from procedure described in the lab manual, if any
Describe any steps taken to reduce experimental uncertainty
(the last two descriptions should follow the description of the measurement in question)
8. How to Write a Lab Report 62

8.5 ANALYZING YOUR DATA AND WRITING AN ANALYSIS SECTION

8.5.1 Data Reduction

The general task you have to accomplish in an analysis section is this: You start with a
bunch of numbers (your measurements). You want to wind up with a few numbers (maybe only
one) that characterize those measurements. Those few numbers in turn presumably tell you some-
thing about a theoretical prediction you or someone else has made; typically you have to make a de-
cision about the validity of a theory based on your results.
You get to the few numbers from the many numbers through your data analysis. In your
analysis section, you show the reader how you got from the many numbers to the few, in enough
detail that the reader can decide if you used the appropriate methods and carried them out correctly.
Then you present your case for the implications of your numerical results. For example, in the first
lab you measured the time it took sound to move through a long tube. Presumably you made
several measurements of the time, and you must have made at least one measurement of the length
of the tube. Then you made some calculations and came up with a result for the speed of sound.
Human nature being what it is, you probably compared your result to the accepted result.

8.5.2 Graphing
Your analysis section could more accurately be called your “data presentation and analysis”
section, because the first thing you must do in an analysis section is display the data you are ana-
lyzing. You should not, however, display your original or “raw” data (the numbers you wrote
down in your lab notebook) in tables in your report, because it's very difficult to pick out data
trends from a large table. Instead, you should present your data graphically, plotted on Cartesian
paper. Even this graph (or set of graphs) will probably not simply be a graph of your unprocessed
data: you will more likely plot averages (or means) of sets of data with appropriate uncertainty
bars. (See Chapter 4, Presenting Data Graphically, for details about setting up graphs.)
Just drawing the graph isn't enough, though. You must tell the reader that it exists, what it’s
about, and where it is. A typical first sentence in an analysis section reads something like this: “The
dependence of falling time on distance from the initial position is given in Figure 1.” (Obviously
you should give the dependent and independent variables for the experiment you're actually de-
scribing!) Notice that you have identified the graph both by the data being displayed and by stating
a figure number. Identifying the graph by the data tells the reader why this graph is part of your
logical argument about the meaning of your data and results. Identifying the graph by a number
makes it easy to find, especially if you put all your graphs at the end of your report. If your word
processor lets you display a graph on the same (or at worst the next) page as the text discussing
the graph, then do that; the next best thing is to put all your graphs at the end. Either way, the read-
er knows exactly where to look for them, which is better than having a figure located at the nearest
convenient empty space several pages away. (A word of caution about positioning graphs: you can
use up an enormous amount of time trying to put a graphic in just the right location while keeping
section and page breaks where you want them. If you find your word processor driving you mad
while positioning figures, a particular problem with Word for Windows, put all your figures at the
end. It is OK to do this, really!)
You must, of course, show error bars on your graphs, unless they’re too small to be visible.
If this is the case, say so explicitly so that your reader does not assume that you have simply for-
gotten about them (which could have deleterious effects on your grade). If your error bars are large
enough to be visible, you should also state explicitly whether they represent one standard devia-
tion, the 95% confidence interval, or some other range. (The 95% confidence range is standard.)
8. How to Write a Lab Report 63

The details of your analysis from here depend on exactly what question you are trying to an-
swer with your data. Often in your theory section you have worked out an expected relationship
between the variables that you are measuring. If the expected relationship is linear, you can check
that the data you have graphed are consistent with that prediction. If the expected relationship is not
linear, you will generally have to draw another graph of your data using one of the linearization
techniques described elsewhere (Chapters 5, 12 and 13) to make the expected relation linear. If
this second graph is necessary, refer to it by title and number in your report. It's usually a good
idea to put the linearized plot right after the Cartesian plot, and comment briefly on the relation bet-
ween the Cartesian and non-Cartesian plots in the report. For example, in a write-up of a pendulum
lab, you might say something like this: “The curve in Figure 1 and the predicted L1 / 2 dependence
suggest a power-law relation between pendulum length and period. Figure 2 shows a log-log
graph of the data of Figure 1. The data in Figure 2 lie on a straight line, indicating that period and
pendulum length are in fact related by a power law.”
The result you are after in an experiment is often related to the slope and/or intercept of this
final straight-line graph. Early in the semester you may find the slope and intercept by eyeballing
the best-looking straight line. (You may also use this method later when you want a quick estimate
of the slope.) If you do this, indicate on the graph the two points you used for the slope and inter-
cept calculations, and give the numerical results in your Analysis section. Later on, after you be-
come familiar with a technique known as linear regression (see Chapter 10), you will use that
method, usually with the program called LinReg.
If you did some calculations to extract the value you want from the slope or the intercept of
your final graph, please go through these calculations in enough detail that the reader can duplicate
your work if necessary. If you have to do a series of very similar calculations (and they're more
complicated than dividing by 2 or p), show one such calculation in some detail as an example and
then state that the other calculations are similar.

8.5.3 Experimental Uncertainty

An essential part of any analysis is a discussion of experimental uncertainty. Careful treat-
ment of uncertainty is essential if you are to draw meaningful conclusions from your data. If you
have to estimate the uncertainty of any measured quantities, describe how you did your estimate,
unless you already did this in your procedure section. If you computed the uncertainty of a value,
describe how you did that calculation and show an example calculation. Also make sure that you
specify explicitly (where relevant) whether the uncertainty you are quoting is the uncertainty of a
single observation or the uncertainty of the mean.
Report uncertainties with units and in the same form and to the same precision as your re-
sults: for example, 3.98 ± 0.07 N, not 3.98 ± 6.8 ¥ 10 –2 N. If you are reporting a result (with un-
certainty) whose magnitude requires the use of scientific notation, report both numbers written
with the same exponent: (1.10 ± 0.06) ¥ 10 –6 meters, not 1.10 ¥ 10 –6 ± 6.2 ¥ 10 –8 meters. Com-
paring the precision of your uncertainty to your result is much simpler with the preferred format.
This might be a good place to point at that “uncertainty analysis” or “error analysis” does not
mean, “Explain what went wrong and how you'd do it differently next time.” Certainly, if in ana-
lyzing your data you realize that you carried out some part of the procedure in a way that gave
poorer results than you had expected, and you don’t have the time do redo that part of the experi-
ment, you should say so: thinking carefully about your procedure after you've done the experiment
is an important part of improving your experimental technique, and can be critical for eliminating
systematic errors from your results. The term “uncertainty analysis,” however, refers to the quan-
titative estimation of the experimental uncertainty in your numerical results.
8. How to Write a Lab Report 64

8.5.4 Results
Earlier we said that you should not give tables of your raw data in your analysis section (or
anywhere else). There are occasions, however, when reporting processed results in tabular form
is appropriate, when a graph is difficult or meaningless. Suppose, for example, that you repeated
the collisions experiment in a number of different ways (using magnetic pucks that repel each oth-
er, using velcro to make the pucks stick together, using pucks with varying masses), and generate
from class data the mean ratio of the total momentum magnitude after the collision to that before the
collision in each case (and its uncertainty). In this case, there is no independent variable to plot
these results against, so a tabular display of these processed results would be appropriate.
At some point you will draw some conclusions about whether the data you have obtained are
consistent with the expected relationship between your variables. If you predicted a straight line in
your theory section and your experimental results support your prediction, you should say so. You
should, however, avoid comments like, “Our results prove that the theory is correct.” You can
never prove a theory; to do so, you would have to perform all possible experimental tests of that
theory, and you don’t have time for that in a three-hour lab period. On the other hand, it is possible
to disprove a theory with a single contradictory measurement (provided that the experiment has
been done correctly, which may be a matter of debate!). The accepted phrase in both cases is less
rashly assertive: "Our results are consistent (or inconsistent) with the theory.
Often your discussion of the implications of your results will be straightforward; if you're
working with a well-known physical system and you follow the treatment in a textbook to develop
a theory, your results will be probably consistent with the theory. We have tried to slip in a few
curve balls just to keep the lab from being “verify what's in the book,” though. Your discussion of
the implications of unexpected results will show your strength as a physicist most clearly. You
should be creative, but also very careful. Don't allow yourself to indulge in empty speculation
about an unexpected result; test your speculations. If you come up with an explanation, try to
show that it could indeed have caused an effect of the same magnitude and in the same direction as
the effect you observed. That is, if your explanation predicts a greater-than-expected measurement,
you'd better observe a greater-than-expected measurement if your explanation is to be valid.

8.5.5 Checklist for an Analysis Section

Your analysis section should
Briefly describes the data
Include a Cartesian (unlinearized) graph of data
Includes linearized graphs of data, if appropriate
Discusses consistency or lack thereof with any theoretical predictions
Discusses how you calculated the slope and intercept of any linear graphs
Shows the calculation any derived quantities from slope or intercept
Completely discusses all uncertainties involved, showing sample calculations if needed
Discusses the results and their implications

8.6 PUTTING IT ALL TOGETHER

8.6.1 Proofreading
In principle, if you write the various sections of your report using the guidelines above, you
should be done. Before you turn in that masterpiece of scientific prose, though, you need to make
sure that it all hangs together. That is, do the links between sections that you imply in one section
actually appear in another? For example, did you test in your analysis section the equation that you
8. How to Write a Lab Report 65

derived in your theory section? If you made assumptions in your theory section, did you include
tests of those assumptions in your procedure section? Did the measurements you describe in the
procedure appear as graphs in analysis? Do your quantitative results support your discussion and
your conclusions? Is it clear that your theory and your procedure are about the same experiment?
You should really read your report over twice. The first time through is for proofreading, a
step we find people often omit. That word-processed output from the laser printer may look won-
derful at first glance, but it has to stand up to a careful reading. Remember, the computer may not
going to catch your mistakes in punctuation, and the spelling checker will probably not distinguish
between “there” and “their,” or “it’s” and “its.” (Now is a good time for you to make sure that you
know the difference between it’s and its.) It also won’t notice that you’ve left out the equations.
(Indeed, using a spelling checker with technical writing can be pretty annoying, as it chokes on
every technical word, symbol, and equation number.) Our experience with grammar checkers sug-
gests that they are not up to college-level English, so don't slavishly follow every instruction your
grammar checker makes, either. We're not suggesting that you turn your backs on some benefits
of modern computer technology and not use your spelling and grammar checkers at all, but you
should recognize that they have their limitations.
The second reading is for sense and continuity. Do the steps of your procedure follow each
other logically? Is the same true for your analysis? Do the sections of your report relate to each oth-
er as described above? If you can stand it, and if you can get yourself to write your report well
ahead of time (a good intention with which the road to hell is no doubt liberally strewn), get some-
one else (preferably not your lab partner) to read your report. The lab assistants will be prepared to
read over your reports for just such considerations as we've described above.

8.6.2 More on Good Writing Style

The mechanics of your presentation are arguably its least important aspect. Nevertheless, a
sloppy presentation can add to your reader’s difficulty in getting through your report, and hence
lower your credibility. (If you didn’t care enough about your report to run it through your spelling
checker, how much effort could you have gone to on the parts that needed some real work?) You
are presumably already familiar with the need for correct spelling and punctuation; here are some
mechanics of presentation that may be less familiar.
• Set apart the different sections of your report (abstract, introduction, etc.) with blank lines.
• Avoid breaking a section between the heading and the first paragraph; that is, don’t leave a
section heading dangling at the bottom of a page with the text of the section beginning at the
top of the next page.
You will be expected to write good, clear, English in your lab reports, using correct grammar
in complete sentences. The days when someone in a science course could wail, “But this is a phys-
ics course, not an English course!” are, thanks to the concept of writing across the curriculum,
long gone (if in fact they ever existed at classy liberal-arts colleges like Pomona). Remember that
the point of any report is communicating with someone else. If you keep distracting your readers
with grammatical mistakes or unclear prose, you will make it difficult for them to concentrate on
the meaning. You will be graded partially on the quality and clarity of your writing. As a general
guide to a good prose style, we recommend Strunk and White's The Elements of Style. It is a
small paperback, usually available at the bookstore. We think it will be a useful investment for
several of your classes. Also keep handy your copy of Hacker, A Writer’s Reference from your
ID 1 course; the lab staffers are likely to refer to it when pointing out grammatical mistakes.
8. How to Write a Lab Report 66

In spite of what Strunk and White say, however, you should use “inclusive” pronouns rather
than the generic “he.” That is, you should use constructions like, “When physicists make measure-
ments, they ...” rather than, “When a physicist makes a measurement, he ...” Strunk and White
wrote their book before inclusive language became standard. It's almost the 21st century now,
usage changes, and it's time to get with it. (You might also count the members of the lab teaching
staff who are left out by the generic “he” and think of inclusive language as simple self-defense.)
You should also avoid certain words and expressions. “Readings” (as in, “We took five
readings for each distance”) belongs on Star Trek,where it's used to avoid using the technical ter-
minology that a 23rd-century scientist would use, since the screenwriters don't have any idea what
that terminology might be. You’re using 20th-century equipment and a 20th-century vocabulary,
and you can describe exactly what you’re measuring: “For each distance between the source and
the timer, we measured the time interval for the sound wave to travel that distance five times.”
Other words and phrases that people often use incorrectly are:
• Defined as, in the sense of “found to be” or “may be described empirically by.” You can
define the length of a pendulum as “the distance from the pivot to the center of mass of the
bob,” if that is the correct definition, but you find or measure it to be 1 meter long.
• Calculated value, in the sense of “number we calculated from our measurements.” Usu-
ally the calculated value (or the theoretical value) is one you derive from some theoretical
calculation, and the measured value (or the experimentally-determined value) is the one
you calculate from your measurements.
• Approximate for “estimate” (as a verb). Estimates (as nouns) usually are approximations,
in the sense that you typically know them to one significant figure. But you estimate a
number (that is the process), and end up with an approximation (or better, estimate [noun])
of its value.
• Correlation for “simple relation.” Saying that two quantities are “correlated” only means
that they seem to be related in some way, so that if one changes, the other one changes as
well. The relationship between variables in many disciplines of natural and social science can
be extremely complicated, and although we often assume that some underlying cause is re-
sponsible for the relationship, this is often not the case: correlation does not imply causa-
tion. In physics, however, the variables that we generally will look at will be clearly related
by some simple relation. Saying that two quantities are “correlated” in physics is usually too
weak a statement: describe the relationship.
• Calibration. People really like this term, because it sounds so technical. It refers specifi-
cally to the comparison of one measuring instrument either against another or some reference
standard, to make sure the instrument is working correctly. If this is not what you're doing
(and you rarely will do this in this lab program), you are not “calibrating.”
• Prove meaning “support.” We talked about this already, but it's worth repeating. You can't
prove a theory with one experiment, although you can disprove a theory with one. Results
can only support or be consistent with a theory.
• Correct value in the sense of “a value published in a book.” In some experiments, you
might be measuring a value (like the speed of sound) whose value we can look up in a refer-
ence, and you may be tempted to call the value in the book the “correct” value. It is not the
correct value: it is the (currently) accepted value. The values of physical constants published
in books are summaries of experimental results, and new experiments can (and often do) lead
to modifications in these accepted values.
8. How to Write a Lab Report 67

An episode from the history of optics illustrates the last point. Albert Michelson (1852-1931)
was the first American to win the Nobel prize in physics, for his precision measurements in the
field of optics. He invented the Michelson interferometer, used in the famous Michelson-Morley
experiment to demonstrate (unexpectedly!) that the speed of light is the same in all inertial reference
frames. He also made several measurements of the speed of light using a method very similar to
the one you will use later on this semester, although with considerably longer baselines. (One of
his measurements was made between Mt. Wilson and Mount Baldy [no lie!], and Baseline Road in
northern Claremont was surveyed accurately as part of this measurement.) His last measurement,
made in an evacuated tunnel about a mile long (on what was then the Irvine Ranch) was accepted
as the standard for decades, and probably most physicists thought of his result as the “correct”
one. A 1941 review of fundamental physical constants (R.T. Birge, “The General Physical
Constants,” in Reports in Modern Physics,8, 90, 1941) weights this result the most heavily in
coming up with a weighted average of several contemporary measurements of the speed of light.
You can guess what's coming. Later measurements, mostly made in the 1950s, consistently
got results that disagreed with Michelson's. The disagreement wasn't very large, about 17 km/s
(out of 300,000 km/s). Their result and Michelson's differed by more than the sum of the experi-
mental uncertainties, though. Eventually a partial explanation for the discrepancy surfaced. Michel-
son died shortly before the experiment was actually performed, although he did see the apparatus
installed. His collaborators made the measurements (almost 3,000 altogether) at night, to reduce
temperature variations and human activity in the area as sources of experimental uncertainty. The
baseline distance was measured during the day, though, and only two or three times. (It’s difficult
to survey distances of more than a few tens of meters at night.) Apparently the thermal expansion
and contraction of the ground itself with temperature was large enough to have a systematic effect
on the speed of light they deduced from their measurements.
Lest Michelson’s collaborators seem inept, we should mention that they were quite alert to
some even more obscure possible sources of systematic error. In reporting their results, they men-
tioned an apparent weak dependence of the measured speed of light on the tides, but since they
couldn’t identify the cause of this dependence, they couldn’t figure out how to correct for it, or
even whether they should! The cause of this systematic effect is unclear even now. Michelson’s
collaborators and the authors of the review article from which most of this historical summary is
taken, mindful that “correlation doesn’t imply causation,” all hesitated to claim that the tides were
directly responsible for the apparent variation in the measured speed of light. (For more details, see
E.R. Cohen and J.W.M. DuMond, “Fundamental Constants in 1965,” Reviews of Modern
Physics, 37, 537, 1965, and the references therein.)
The moral of this tale, of course, is that there are no “correct” results in science, only accept-
ed ones. Even prominent scientists forget sources of systematic error, or run into systematic error
where do one would have expected it, or someone comes along with better equipment. It is true
that you're not likely to hit the frontiers of physics in an introductory laboratory, but you should
get into the habit now of regarding every scientific result as only one carefully designed experiment
away from revision.

8.6.3 How Long Should a Lab Report Be? (and Stuff to Leave Out)
A typical scientific journal article might be about ten pages long. Your full lab reports will
probably be shorter; try to limit yourself to the equivalent of four or five single-spaced typewritten
pages of text, not counting graphs or diagrams. This means that few of the five major sections (the
ones with Roman numerals on the outline) will exceed a page in length, and some may be shorter.
8. How to Write a Lab Report 68

There are also some items you should leave out of a lab report. Please don't complain about
the equipment; we already know that if we had an infinite budget, we could buy really frictionless
gliders and opto-electronic timers good to a microsecond. You won't have an infinite budget in real
life, either. Even if your equipment budget is large, you will always be making measurements that
require care and ingenuity to make; sometimes the equipment you would like doesn't even exist!
Experimental physics isn't about making really precise measurements so much as it is about mak-
ing the best measurements you can with the equipment you have. By practicing with the admittedly
limited equipment available now, you prepare yourself for those later measurements when you
can't improve the data simply by spending more money.
Don't editorialize about an experiment being a “success” or “failure” in the context of agree-
ment with accepted results or theories. It’s true that we have some expectation that your results will
be in agreement with established laws of physics, because normally you won’t be dealing with
particularly exotic (that is, poorly understood) physics in an introductory course. We also expect
that in the full report, in which you do write a draft for which you have presumably analyzed your
data, that if your results are in gross disagreement with established laws of physics, that you will
make some attempt to figure out the cause of that disagreement and fix it. You do, after all, have
most of that second lab period to collect more data if that should seem appropriate, and that’s ex-
actly why we arranged the lab schedule the way we did. In evaluating your work, though, we look
primarily for evidence that you understood how the equipment worked, how the measurements
you made were related to the theory discussed, and generally that you were thinking about what
you were doing. Some real physical effect could be present that the designers of the lab over-
looked, or have left in to keep you on your toes. (This happens more often than you might think.)
If you have been careful about your work, be confident in presenting what you have observed.
(The confidence should follow from being careful, though, and if the lab staff identify some sys-
tematic effect in data collection that you overlooked, go take more data!)

8.6.4 Example Lab Reports

Two sample lab reports are provided as appendices to this chapter. Each is mostly well-writ-
ten, but has problems with specific sections, as discussed in the exercises below. Except for these
problem sections, though, you can use these reports as examples of good report style.
Note again that a summary of all checklists appears on the inside front cover of this manual.

EXERCISES

Exercise 8.1
Read the lab report entitled “The Speed of Sound.” This report (which describes an older ver-
sion of the experiment you did during the first week of lab) is mostly well-written except for
the abstract and procedure sections. See what you think is lacking in these sections (according
to the checklists and other information in this chapter) and then compare with the comments on
the last page of this chapter. (There is no penalty for not spotting everything: just do the best
that you can.) Write your comments on the report itself.

Exercise 8.2
Read the lab report entitled “Gravitational Potential Energy”. This report is mostly well-written
except for except for the analysis section (where little superscripted numbers indicate problem
areas). See if you can figure out what these numbered problems are, and then check the an-
swers provided on the last page of the chapter. (There is no penalty for not getting everything
just right: just do the best that you can.) Write your guesses in the margin of the report itself.
8. How to Write a Lab Report 69

APPENDIX 8.1: FIRST SAMPLE LAB REPORT

Torrin Hultgren
Partner: Alix Hui
9/10/98

The Speed of Sound

Abstract:

In this lab we determined the speed of sound by timing the interval that it took for a loud
bang to echo off a surface a known distance away. Our average time interval was 1.28 s, and the
distance was 440 m, so our calculated value for the speed of sound was 343.8 m/s. This is con-
sistent within our experimental uncertainty with the accepted value at 30°C, which is 349.7 m/s.

Introduction:

The speed of sound has many practical applications, such as determining the distance from
lightning, knowing when jets will break the sound barrier, designing acoustical facilities like con-
cert halls and auditoriums, and literally thousands of others. The phenomenon of an echo is famil-
iar to most people, and it is a relatively easy way to measure the speed of sound.

We used two blocks of wood to create a loud and sharp bang. We determined the distance
using a counting wheel whose circumference we measured and we used hand stopwatches to time
the echo. We repeated the time measurement 20 times to reduce experimental uncertainty. We cal-
culated the speed of sound by dividing the distance measurement by the time measurement. In ad-
dition, because the speed of sound varies with the temperature of the air through which it propa-
gates, we measured the temperature with a mercury thermometer in order to calculate the accepted
value for the speed of sound.

Procedure:

We used the following pieces of equipment to do the lab.

• Two small blocks of wood

• 2 stopwatches
• 1 measuring wheel
• 1 meter stick
• Thermometer
• A small piece of masking tape

We set up on the concrete bench closest to the grass on Marston Quad. We chose this spot
because it lined up with the small wall at the end of Stover Walk (which we could see through the
trees) which gave us an easy reference point for beginning our distance measurement. One of us
held the stopwatch and the other hit the blocks together. Because we could see the blocks coming
together we could anticipate when they would hit. Then we stopped the stopwatch when we heard
the echo, without anticipating it. This gave us a slight delay in timing the echo because of our reac-
tion time, but we were able to correct for this as described below. Both of us made 10 time meas-
urements and hit the blocks together 10 times.
8. How to Write a Lab Report 70

To account for the reaction time delay we devised this procedure. I started both stopwatches
at the same time. I then handed one stopwatch to Alix and kept the other. Behind my back she si-
multaneously stopped her stopwatch and hit one of the blocks against the concrete bench. When I
heard the sound I stopped my stopwatch. The difference between the two times on the stopwatches
was my reaction time. We repeated this measurement for each of us five times.

To calibrate the measuring wheel we put a small piece of tape at the edge of the wheel. We
put the meter stick on the ground and lined this piece of tape up with one of the ends of the meter
stick. We then rolled the measuring wheel along the ground next to the meter stick until the piece
of tape had traveled one full revolution. The point that it lined up with was our value for the cir-
cumference of the wheel.

For the distance measurement we began at the wall at the beginning of Stover Walk that lined
up with the place where we had taken our time measurements. We walked the measuring wheel
down the middle of Stover Walk, using the sidewalk lines to make sure we were traveling in a
straight line and not zigzagging excessively. We continued across the street, and then used the
sidewalk lines to line up perpendicularly so we could move over and roll the measuring wheel
across the wood chips and right up to the face of Carnegie that we believed the sound was echoing
off of. We then doubled this measurement to arrive at the total distance the sound had traveled.

Analysis:

The average of the measurements I took was 1.55 s, with a standard deviation of s = 0.05 s.
The uncertainty of this measurement, using the Student t-value, is

st = 0.05 s ¥ 2.09 = 0.10 s (1)

This measurement therefore had a fractional uncertainty of

0.10 s
= 0.064 = 6.4% (2)
1.55 s

The similar values for Alix’s measurements, which were different because she had a different reac-
tion time, were 1.43 s ± 0.13 s for a fractional uncertainty of 9.1%. Both of these fractional uncer-
tainties seem reasonable for the type of measurements we were doing. My average reaction time
was 0.27 s ± 0.02 s, and her average reaction time was 0.20 s ± 0.03 s. Our actual calculated
times of flight were therefore 1.28 s ± 0.082 s and 1.23 s ± 0.11 s.

Our measurement for the circumference of the wheel was 0.587 m. Our measurement for the
number of rotations of the wheel was 374.3. The distance from us to Carnegie was therefore

m
374.3 turns ¥ 0.587 = 220 m (3)
turn

Doubling this we arrived at a total distance of flight measurement of 440 m. We generously esti-
mated our uncertainty to be ± 1.0m. This gives us a fractional uncertainty for the distance meas-
urement of 0.2%. Compared to the uncertainty of the time measurement, this is tiny.
8. How to Write a Lab Report 71

My calculated value for the speed of sound was

440 m
= 344 m/s (4)
1.28 s

Propagating uncertainty using the weakest-link rule, my calculated uncertainty was ± 22 m/s.
Alix's value was 355 m/s ± 32 m/s.

The formula for the speed of sound as it varies with temperature is

m Ê m ˆ
vs = 331.3 + 0.6 T (5)
s Ë s ◊ C∞ ¯

where T is measured in Celsius degrees. Our measured value for the temperature was 30°C. Plug-
ging this into the above formula gives us an accepted value for the speed of sound of 349.3 m/s.
This value lies well within both of our experimental uncertainties.

Conclusion:

We measured the time it took for an echo to travel a measurable distance. Using our separate time
and mutual distance measurements we calculated two values for the speed of sound: my result was
344 m/s ± 22 m/s and Alix’s was 355 m/s ± 32 m/s. These values for the uncertainty are a reason-
able fractional amount. Our calculated accepted value for the speed of sound based on the observed
temperature was 349.3 m/s. This value lies well within the experimental uncertainty of both our
measurements.
8. How to Write a Lab Report 72

COMMENTS ON THIS REPORT:

Short sections:
These are fairly good, except that the abstract should include an estimate of the uncertainty in
their measurement of the speed of sound, not just their measured value. The introduction should
provide a clearer statement of the particular experimental question to be resolved here (that is, that
the goal of the experiment is to measure the speed of sound by measuring the round-trip time of an
echo from a distant object and compare the result with an accepted formula for the speed of sound).

Theory:
The Theory section is missing! This is obviously a simple experiment based on very simple
theory, but at the very least the author should state explicitly that he is assuming that the speed of
sound is constant, and give the appropriate equation for finding the speed from distance and time
measurements.

Procedure:
The equipment list does not include their stopwatch number or the number of the measuring
wheel. Consequently, if they needed to check their calibration of the wheel (or the accuracy of the
stopwatch, which is less likely), they would have no way of identifying it.
The procedure section does provide an equipment list but not a sketch or diagram. However,
this lab is a case where an equipment list is probably more useful than a sketch for helping the
reader understand how the lab works. Even though the guidelines strongly suggest that one should
include a diagram, the guidelines should not be followed slavishly if a diagram does not really add
much to the reader’s understanding. Do whatever makes things most clear to the reader!
It might have been nice to briefly discuss that the author is assuming that the “actual” flight
time of the echo that he will use to calculate the speed of sound is his measured flight time of the
sound minus his reaction time. This is implicit but should be stated more explicitly.
The calibration of the measuring wheel needs more discussion. For example, the piece of
tape mentioned presumably has a finite width, probably about 1 cm. If they weren’t careful to
identify a particular reference point on the tape (such as a penmark on the tape, or one of the two
edges), this would introduce a systematic error into their calibration, which would carry over into a
systematic error in their value for the distance.
The author also doesn’t state the precision of their measurement of the circumference of the
measuring wheel. Without this, the reader has no way of knowing if the later estimate in the uncer-
tainty of the distance is reasonable. It is also unclear if they repeated the circumference measure-
ment or the distance measurement.

Analysis:
The main problem with this section is the uncertainty analysis. To begin with, the author
mentions combining the uncertainties of their average time measurements for the echo time and the
reaction time, but does not identify the method used to combine the uncertainties. Next, no uncer-
tainty estimates are given for either the measurement of the wheel’s circumference or the number of
revolutions of the wheel. Finally, the author invokes the weakest-link rule in finding the uncertain-
ty in the final value for the speed of sound, but does not justify the use of the weakest-link rule by
explicitly locating the weakest link in the calculation and then showing a sample calculation using
that weakest link.
8. How to Write a Lab Report 73

APPENDIX 8.2: SECOND SAMPLE LAB REPORT

GRAVITATIONAL POTENTIAL ENERGY

Maria Goeppert-Meyer
(Lab partner: Irene Curie)
Sept. 26, 1995

ABSTRACT
In this experiment, we determined the change in the gravitational potential energy V of the
system consisting of the earth and a dropped plastic slab as a function of the distance h through
which the slab falls. We found this change in potential energy to be consistent with the expression
Vi – Vf = mgh , where m is the mass of the object and g is the gravitational field strength. We
found the value of g to be 9.81 ± 0.02 m/s2 , consistent with results obtained in other laboratories.

INTRODUCTION
Consider the change Vi - Vf of the gravitational potential energy of a system consisting of
the earth and a falling object, where Vi is the system’s initial potential energy, Vf is its final poten-
tial energy after the object has fallen a certain distance h. In section C7.4, the text claims that this
change in potential energy is given by Vi - Vf = mgh, where m is the object’s mass and g is the
gravitational field strength near the earth, a constant that is purportedly equal to 9.8 m/s2.
This result, which is stated without justification in the text, is a basic and important result that
subsequently used many times in the text. It would be valuable, therefore, to supply the empirical
foundation for this assertion. Our goals in this experiment were to demonstrate for a specific object
interacting with the earth that (1) for a given value of h, the value of Vi - Vf does appear to be pro-
portional to m, (2) for a given value of m, the value Vi - Vf increases linearly with h, and (3) the
value of g is what it is purported to be.
In this particular experiment we dropped a plastic slab (released from rest at a known initial
height) past a photodetector connected to a computer. A series of equally-spaced opaque bands
painted on the slab interrupted the light falling on the photodetector, and the computer measured the
time that it took each band to pass the photodetector. From this information, we could determine
slab’s speed as each band passed the photodetector, and thus determine its kinetic energy after it
had fallen whatever distance h was required to bring that particular band past the photodetector.
Given the object’s kinetic energy as a function of h, we could find Vi - Vf as a function of h. By
attaching various weights to the bottom of the slab, we could vary the mass of the falling object
and thus check how Vi - Vf depends on mass.

THEORY
As the plastic slab drops under the influence of the gravitational interaction between it and the
earth, the total energy of the earth-slab system must be conserved:
Ki + KE,i + Vi = Kf + KE,f + Vf (1)

where Ki and Kf are the initial and final kinetic energies of the slab, KE,i and KE,f are the initial and
final kinetic energies of the earth, and Vi and Vf are the initial and final gravitational potential ener-
gies of the system. According to the argument presented in section C7.3 of the text, we can con-
8. How to Write a Lab Report 74

sider the earth to be essentially at rest throughout the experiment (since it is so much more massive
than the slab) and thus KE,i and KE,f are negligible. If we drop the slab from rest, then Ki = 0 also,
and equation (1) becomes simply

Vi - Vf = Kf = 1 mv 2
2 f (2)

So, to measure the system’s potential energy change Vi - Vf after the slab has fallen a dis-
tance h, all that we have to is measure the slab’s mass m and its final speed v f . We can easily
measure its mass using a balance. We can measure its final speed as follows. Imagine that we paint
an opaque band across the width of the slab perpendicular to the direction that the slab falls. As the
slab falls, imagine that this band interrupts a horizontal beam of light between a light source and a
detector. We can use a computer to register the time Dt that the beam is interrupted. If the height of
the band is Dd, then the speed of the slab as the band crosses the beam is approximately given by:
v ª Dd/Dt (3)
This most closely approximates the slab’s speed halfway through the time interval and thus rough-
ly as the center of the band passes the light beam. This speed, therefore, can be used to determine
the slab’s kinetic energy after it has fallen a distance h equal to the change in the slab’s position
from its release point to the position where the band is centered on the photocell beam.
Finally, note that the claim is that Vi - Vf = mgh, where m is the slab’s mass and g is the
constant gravitational field strength. If this is true, then plugging this into equation (2) yields

mgh = 1 mv 2
2 f fi gh = 1 v2
2 f (4)

Therefore, if Vi - Vf is proportional to m as claimed, the slab’s final speed after falling through a
given distance h should be completely independent of its mass, which should be easy to check.
Also, if this is true, the slab’s mass is not really relevant and we do not need to measure it.

PROCEDURE
In this experiment, our falling object was a clear plastic slab about 1.1 m tall and 8 cm wide,
with five opaque bands 5.0 cm tall and vertically separated (center to center) by 20 cm. We could
vary the mass of the slab by attaching one to four weights to the bottom of the slab. We dropped
this slab past a photogate consisting of a paired infrared light source and a photodetector mounted
on a lab table so that the line connecting the source and detector was horizontal (perpendicular to
the motion of the slab). The output of the photodetector was connected to a small box which in turn
was connected to a Universal Lab Interface (ULI) circuit board (sold by Vernier Software, Inc.),
which processed the signal for the photogate before passing it on to a Macintosh Centris 610 (serial
number 3255967). A program called ULI Timer (also from Vernier Software) monitored the out-
put from the ULI and displayed time intervals on the computer screen (see Figure 1 for a sketch of
our experimental setup.) The program was configured to display the length of time that each of the
five dark bands on the slab blocked the photogate beam as the slab fell past it.
8. How to Write a Lab Report 75

CLOSE-UP and SIDE

VIEW of the slab:

photogate
(attached to connector box
the desk with 20 cm
a clamp)

Macintosh
ULI 20 cm
table

20 cm
slab
5 cm

FRONT VIEW 20 cm

20 cm

receptacle for mark

catching the slab
holes for attach-
ing weights
floor

Figure 1: Sketch of the apparatus

After our lab instructor gave a brief demonstration of the equipment, each of the seven lab
teams in our particular afternoon session did a run. When our turn came, one of us (M-G.M.) held
the center of the upper end of the slab between his thumb and forefinger and adjusted its vertical
position until a certain mark inscribed on the slab edge was aligned precisely in the middle of the
photogate as reported by IC. We waited until the slab had stopped swinging back and forth and
was completely at rest. I.C. then triggered the ULI Timer program to start taking data and M-G.M.
dropped the slab. The computer then automatically recorded and displayed the time Dt that it took
each of the five opaque bands to pass the photogate. We wrote these five numbers on the black-
board, filling in a table already started by other teams.
Once all the data was taken, each pair of lab partners calculated the means and uncertainties
of the mean (using the techniques in chapters 3 and 5 of the lab reference manual) of the results for
Dt for each of the five bands. We discussed the results as a class and decided that these results ap-
peared to be uncertain by very roughly ± 0.002 s.
While we were calculating the means and uncertainties, our pair of partners took turns doing
a total of seven more runs, three runs with two weights attached to the slab and four runs with four
weights attached to the slab. These runs were also recorded on the blackboard.
Finally, each pair worked individually to analyze the data. As we did this, we passed the slab
from pair to pair so that each could check that the opaque bands were 5.0 cm tall and separated
from center to center by 20 cm. We did this using an ordinary meter stick turned on its edge so that
the scale was right next to and perpendicular to the bands. We estimated that the height of the
bands was equal to 5.0 cm to within ± 0.05 cm and that the distances between the centers of the
band was 20.0 cm to within ± 0.1 cm (we actually measured these from bottom edge to bottom
edge).
8. How to Write a Lab Report 76

ANALYSIS
A table of the mean values of the time intervals appears below:1

Dt Dt Dt
Band Number (no added weight)2 (one added weights)2 (four added weights)2
1 0.0252 0.0250 0.0256
2 0.0179 0.0181 0.0179
3 0.0146 0.0149 0.0144
4 0.0126 0.0124 0.0126
5 0.0112 0.0113 0.0110

It is clear from these results that the speed of the slab is independent of its mass3, so (as we argued
in the theory section) Vi - Vf must be4 directly proportional to the slab’s mass m.
From the values of Dt for the slab with no added weight, we calculated 12 v 2f for each of the
heights5. Figure 2 shows a graph of these results. According to LinReg,6 the slope of the line is
9.81023 and the intercept is 0.012865.7 This proves8 that Vi - Vf = mgh (though our value of g is
a bit high due to experimental error).9

CONCLUSION
In this experiment, we showed that the final kinetic energy per unit mass 12 v 2f of a plastic
slab dropped from rest through a distance h is independent of the mass of the slab and seems to be
proportional to h (within experimental uncertainty), with the constant of proportionality being equal
to 9.81 ± 0.03 m/s2 . These results are completely consistent with the assertion made in the text
that Vi - Vf = mgh, where g = 9.8 m/s2 .
8. How to Write a Lab Report 77

10.0

8.0

6.0

4.0

2.0

0
0 0.20 0.40 0.60 0.80 1.00

Figure 2.10
8. How to Write a Lab Report 78

COMMENTS ON THE ANALYSIS SECTION

In general, the problem with this section is that it is far too short and thus does not provide us
with some information that we need to understand the results and how the authors analyzed them.
There are also several statements made that are not or cannot be supported by the data.
Here are specific comments about the places in the analysis section where specific errors
were flagged with numerical superscripts. (The simulated errors in this report reflect the most
common types of errors that people make when writing analysis sections.)

(1) This table is nicely laid out, but a table of processed data like this should state the units of the
quantities and state the uncertainties of the means as well as the means themselves. The writers
should have also included a description of how many measurements went into calculating the mean
and how the uncertainties were calculated. Also, if the uncertainties are really on the order of
±0.002 s (as stated in the procedure section), then the last digit in the tabulated data is totally mean-
ingless. Are the uncertainties really more like ±0.0002 s (this would be consistent with the varia-
tion appearing in the table data)?
(2) One could include the units of the data in the column heading like this: “Dt in seconds”.
(3) Is this really clear? Without knowing the uncertainties, the small variations in the values are
impossible to interpret.
(4) We really can’t say that Vi - Vf must be independent of m, only that our data is consistent
with this interpretation.
(5) This needs to be explained in much more depth. How did the writers calculate 12 v 2f from this
data? What are the uncertainties of these speeds, and how were they calculated (one has to use
something like the weakest link rule)? How were the heights determined and their uncertainties es-
timated? It would have also helped greatly if the writers had listed the calculated values and uncer-
tainties for 12 v 2f and h for each row of the table (or better yet, on a separate table).
(6) A brief description of LinReg and what it does would be appropriate here.
(7) The quantities quoted here have units and uncertainties: what are they? Also what is the signif-
icance or meaning of the slope and the intercept here?
(8) An experiment can never prove that any theoretical assertion is true. The best that we can say
here is that our results are consistent (or inconsistent) with this assertion. See the conclusion for
better language.
(9) How is g is related to something we have calculated in this lab? Also, the value is a bit high
compared to what? Saying that the difference is due to “experimental error” says nothing. What
kind of experimental error? Is the result within our uncertainties or not? If so, what does it mean to
say that this is “a bit high”?
(10) What are the experimental uncertainties of the data points? Are they not shown because they
are too small to appear on the graph or did the writers simply forget about them them? What is be-
ing plotted against what here? (The axes should be labeled.) What is this graph about? (It should
have a title!) What are the units of quantities displayed? What does the line mean? This graph is
missing many of the features that a higher-level graph should have.
 9
PROPAGATION OF UNCERTAINTY

“She drew up plans of economy, she made exact calculations…”

--- Persuasion
9.1 INTRODUCTION
In many kinds of physics experiments, one would like to know the uncertainty in a quantity
(call it f) that is calculated from directly measured and uncertain quantities a, b, c, ... , that is, f is a
function f(a,b,c,...) of the measured quantities a, b, c, ... . For example, this problem would arise
in an experiment where we want to determine the uncertainty in an object’s speed if that speed is
calculated from uncertain time and distance measurements The general problem of determining a
calculated quantity’s uncertainty is called the problem of propagation of uncertainties (expres-
sing the idea that uncertainties in measured quantities beget uncertainties in quantities calculated
from them). The goal of this chapter is to explore means for intelligently addressing this problem.

9.2 SOME NOTATION AND TERMINOLOGY

We will use the symbol U[f] to refer to the experimental uncertainty in any quantity f, wheth-
er that uncertainty has been directly measured or calculated from the uncertainties in other measured
quantities. If the quantity f depends on measured quantities a, b, …, then its uncertainty U[f]
should be related to the uncertainties in a, b,… (that is, on U[a], U[b], and so on) in some way that
we should be able to calculate knowing how f depends on these variables.
It turns out that the most useful quantity to know when dealing with the problem of propaga-
tion of uncertainty is a variable’s fractional uncertainty, which is defined to be ratio of the
variable’s uncertainty to its measured or best guess value:
U[ f ]
fractional uncertainty of f ∫ Q[ f ] ∫ (9.1)
f
(The symbol Q is meant to make you think “quotient.”). This quantity is very closely related to the
concept of percent uncertainty: to get the percent uncertainty from the fractional uncertainty,
simply multiply by 100. These ideas are so closely and simply related that we will often treat
“fractional uncertainty” and “percent uncertainty” as if they were the same.
As an example, say that the measured value of f is (5.96± 0.60) cm. The fractional uncer-
tainty of f is then Q[f] = 0.6 cm/5.96 cm = 0.10, and its percent uncertainty is 10% (that is, its
uncertainty is equal to 10% of its best-guess value).
Note that whatever units a quantity f might have, U[f] has the same units, so the ratio Q[f] of
these quantities (and thus the fractional or percent uncertainty) will always be a unitless number.
This observation is also a reminder that a quantity’s uncertainty U[f] and its fractional uncertainty
Q[f], while related, are not the same thing (if they were, they would have the same units).

9.3 A GENERAL APPROACH TO PROPAGATION OF UNCERTAINTIES

Think of the function f ( a, b, c,º) as a machine that has a handle (like a control stick) corre-
sponding to each of its input variables a, b, c, … and a big dial with a pointer that indicates the
output value f. Each of the input variables affects the final value shown on the dial, so adjusting the
positions of the handles individually or in combination will change the value shown on the dial.
9. Propagation of Uncertainties 80

Now if the input value a has an uncertainty U[a], then we can wiggle the handle correspond-
ing to the variable a back and forth from its most probable value a by a positive or negative amount
d a in the range d a £ U[a] and still be consistent with the experimental data. This wiggling will
cause the value of f indicated by the dial to wiggle back and forth from its central value by a certain
amount as well. Let us define d fa to be the (presumably small) change in the value of f from its
central when a is moved from its central value by d amax = +U[a] (corresponding to the upper ex-
treme limit of a’s uncertainty range), while the other handles are held constant. Similarly, let d fb
be the change in f when b is moved from its central value by an amount d bmax = + U[b] while the
other variables are held constant, and so on.
Now, what is the uncertainty in f when all of its variables are free to wiggle around within
their uncertainty ranges simultaneously? The maximum distance f could be from its central value is
d fmax = d fa + d fb + d fc + º if all of the input values happen to be simultaneously at whichever
edge of their uncertainty range causes them to shift f in the same direction. But this is fairly unlike-
ly, because there is only roughly a 5% chance that any single variable will be at or beyond either
limit of its uncertainty range; the likelihood that all of the variables are simultaneously at or beyond
their limits on the correct side to push the value of f in the same direction is vanishingly small.
It turns out that because of this, a statistically more accurate estimate of the uncertainty of f
due to the uncertainties in all of its variables is

U[ f ] ª d fa2 + d fb2 + d fc2 + º (9.1)

(The proof is beyond our scope here.) Quantities whose effects are “added” by squaring, adding,
and then taking the square root like this are said to be added in quadrature. Calculating U[f]
thus reduces to the problem of finding the changes d fa , d fb , º due to each variable separately.
This can be done easily in simple cases. Consider the special case where f(a,b) = a – b. If we
increase a to a + d a while keeping b fixed, then f changes to
f + d f = a + da - b fi d f = da fi d fa = d amax = +U[a] (9.2)

after subtracting f = a – b from both sides. Similarly, you can easily see that d fb = -U[b] (nega-
tive because when b goes up, f goes down). Therefore the total uncertainty in f in this case is

U[ f ] = (U[a])2 + (U[b])2 when f = a – b (9.3)

Therefore, if a and b have the same uncertainty then the best estimate of the uncertainty in f is not
2U[a] (as one might naively expect) but rather 2U[a] ª 1.4U[a].
On the other hand, if U[a] is more than about 3 times larger than U[b], then (U[a])2 is more
than 9 times larger than U[b], and thus dominate the expression for U[f] in equation 9.3. This is a
specific example of a more general feature of the original equation 9.1: that equation implies that
the uncertainty U[f] is typically dominated by one variable, the variable whose variation over its
uncertainty range causes the greatest change in the value of f.

9.4 THE WEAKEST-LINK RULE

The majority of calculated quantities f that arise in physics experiments can be put in the form

f(a,b,c, …) = ka m b nc j º (9.4)
where k is a constant and m, n, and j are exponents that may be positive or negative (these expon-
ents are usually integers or simple fractions). A dependence of this form on the variables a, b, c, …
9. Propagation of Uncertainties 81

is called a power-law dependence. For example, an object’s calculated speed v depends on the
distance D it had to travel and the time T that it took to travel that distance according to the power-
law relation v = D / T = D1T –1 (k = 1 here).
If equation 9.4 is true, then the weakest-link rule provides a fast and simple way of esti-
mating the uncertainty U[f] in the calculated quantity f:
The fractional uncertainty Q[f] of f = ka m b nc j º is approximately equal to the largest of
the values m Q[a], n Q[b], j Q[c], and so on.
The fractional uncertainties Q[a], Q[b], Q[c], … of the variables are typically quite different in a
real experiment, and so doing a few rough divisions in your head can quickly guide you to the
variable whose fractional uncertainty is largest.
We will look at why this rule is correct in a moment. First note that this rule says two inter-
esting things. The first is that the “weakness” (that is, the fractional uncertainty) of a calculated
quantity f is determined primarily by the “weakest” of the quantities on which it depends (the
“weakest” here being the quantity whose fractional uncertainty times its exponent is largest). This
is fundamentally because of the effect noted in the last paragraph of the last section: one variable’s
effect typically dominates the sum in equation 9.1. The rule’s name emphasizes this by bringing to
mind the old saying “the strength of a chain is determined by its weakest link.” (Note that contrary
to the currently popular game show, the “weakest link” is the one we keep in our calculation!)
The second interesting thing that the rule says is that it is the fractional uncertainty in f that is
related in a simple way to the fractional uncertainties of its variables. This was not the case when
we are talking about the simple sum or difference of variables (as equation 9.3 shows), but as we
will shortly see, it is the most natural way to deal with power-law relations.
Let’s see how this rule might work in a given situation. Imagine that we are computing the
magnitude of an object’s average velocity vavg = D1T –1 , where D is the distance it travels during a
time T. Say that we have measured D = (12.12 ± 0.02) m and T = (0.82 ± 0.05 s). The best guess
value of vavg = 12.12 m / 0.82 s = 14.8 m/s. The fractional uncertainties in D and T are:
U[ D] 0.02 m 0.05 s
Q[ D] = = = 0.0017, Q[T ] = = 0.061 (9.6)
D 12.12 m 0.82 s
The fractional uncertainty in T is more than 35 times larger than that for D so it dominates. Accord-
ing to the weakest link rule, the fractional uncertainty in vavg = D1T –1 is thus given by
Q[vavg ] ª - 1 Q[T ] = 0.061 fi U[vavg ] = 0.061vavg = 0.061(14.8 m/s) = 0.9 m/s (9.7)

Note that the calculations here are quick and simple: that is the beauty of the weakest-link rule.
The general “proof” of the weakest-link rule is somewhat beyond our mathematical means
here, but let’s see how we might “prove” it in the simple case where f = f ( a, b) = ka 2 b . If b
changes to b + d b while a remains the same, then f changes to

f + d fb = ka 2 (b + d b) = f + ka 2d b fi d fb = ka 2d b (9.8a)

If we now divide both sides of this by f = ka 2 b and set d b = + U[b], we find that

d fb ka 2d b db U[b]
= 2
= = ∫ Q[b] (9.8b)
f ka b b b
If we change a to a + d a while b remains the same, then f changes to f + d fa = k ( a + d a)2 b .
9. Propagation of Uncertainties 82

Writing out the square and subtracting f = ka 2 b from both sides, we get:

f + d fa = k ( a 2 + 2 ad a + d a 2 )b = f + 2 kabd a + kb(d a)2

fi d fa = 2 kabd a + kb(d a)2 (9.9a)

Dividing both sides of the result by f = ka 2 b yields:

2
d fa 2 kabd a kb(d a 2 ) da Ê daˆ
= + = 2 + (9.9b)
f 2
ka b 2
ka b a Ë a¯
Now, if we can assume that the variation d a due to a’s uncertainty is much smaller than the value
of a itself, then (d a / a)2 << d a / a , and we can ignore the second term in equation 9.9b in com-
parison to the first. Then, if we set d a = U[a] we find that
d fa U[ a]
= 2 = 2Q[a] (9.9c)
f a
If we now divide both sides of equation 9.1 by f and substitute in the results of equations 9.8b and
9.9c, we find that
2 2
U[ f ] 1 Ê d fa ˆ Ê d fb ˆ
Q[ f ] ∫ = d fa2 + d fb2 = Á ˜ +Á ˜ = (2Q[a])2 + (Q[b])2 (9.10)
f f Ë f ¯ Ë f ¯

Whichever of 2Q[a] or Q[b] is even marginally larger than the other will dominate inside the square
root and thus be essentially equal to Q[f]. Thus we have seen that the weakest link rule does indeed
adequately summarize the more exact calculation in this case as long as (1) the fractional uncertain-
ty in a is fairly small (so that we can ignore the complicating term in equation 9.9b), and (2) one of
2Q[a] or Q[b] is at least somewhat larger than the other.

9.5 WHAT IF THE WEAKEST-LINK RULE DOESN’T APPLY?

The weakest-link rule does not apply to situations where f’s dependence on its variables is
not a power-law relation (for example, the simple sum f(a,b) = a+b). The weakest-link rule is also
not very accurate in situations where the fractional uncertainties in the variables are large fractions
of 1, or when two fractional uncertainties are nearly the same. What do we do in such situations?
The first level of approximation is to use the weakest link rule anyway, and simply
recognize (and state in your lab notebook) that the estimate of the uncertainty might well be inaccu-
rate. The weakest-link rule will almost always yield estimates good to within a factor of two or so
unless your formula for f involves logarithms or exponentials. In situations where one is not inter-
ested in high degree of precision this may be acceptable, as long as you recognize situations when
the rule might not be expected to give accurate results and factor that into your conclusions.
A better way to determine the uncertainty of f would be to calculate many values of f
using values of its variables a, b, c, … that are randomly chosen from the raw data for these varia-
bles. Then one can determine the uncertainty of f in the usual way by evaluating the standard de-
viation of the set of values for f and so on. (This is essentially what we did in the speed of sound
lab: we computed speeds for many different time and distance values.) This method almost always
gives an excellent estimate of U[f] as long as the number of values of f that you generate is reason-
ably large (more than 20 at least!). However, because this method is so tedious, we cannot recom-
mend it unless you have a computer program to do the work.
9. Propagation of Uncertainties 83

Figure 9.1: The PropUnc program.

An approach of last resort is to apply the general method outlined in section 9.2. Cal-
culate (by hand) the variation in f when you vary each variable from its central position to the upper
edge of its uncertainty range while leaving the other variables constant. Then use equation 9.1 to
compute the total uncertainty in f from these individual variations. This will generally be pretty ted-
ious compared to the weakest link method, but does yield reasonably accurate answers in all cases.
This is the method that you must use if f involves a logarithm or exponential (unless you have a
computer program that can do the calculation outlined in the previous paragraph).
Of course, if f involves the simple sum or difference of two variables, one can apply equation
9.3, which we derived especially for the simple difference case. (You should be able to convince
yourself that equation 9.3 also applies to the case of a simple sum.)

9.6 THE PropUnc PROGRAM

PropUnc is a computer program that uses the “calculate many values” approach to generate
an accurate value of uncertainty of f in all cases involving five or fewer variables. A screen shot of
the program set up to calculate the uncertainty of the function f = ka 2 b is shown in Figure 9.1. All
that you have to do to use the program is type symbols, values, and uncertainties for your basic
variables in “variables” section and the symbolic expression for f in the “expression” section and
punch the “Evaluate” button. The program then calculates a randomly-chosen value for each vari-
able that lies within the uncertainty range you specified for that variable and calculates the value of f
using randomly-perturbed variable values using the formula you supply. The computer repeats this
9. Propagation of Uncertainties 84

process N times (N = 100 by default). Finally, the computer calculates the standard deviation of
the N values of f it has generated and the uncertainty in f from that.
In other words, the computer simulates having N teams of experimenters like yourselves
who have measured the same variables and have used them to calculate values of f. The uncertainty
in the value of f is clearly related to the spread in the values obtained by the N ficticious teams.
Note that in the case shown in the figure, the fractional uncertainty in f is 10%. Note that the
quantity a has by far the largest fractional uncertainty (5% compared to 1% for the other variables),
so the weakest link rule would say that Q[f] ª 2Q[a] = 2·5% = 10%. Thus the program agrees
with the weakest-link rule in this case. If you press the “Evaluate” button again, however, you may
get slightly different results, because of the random nature of the simulation. Choosing larger val-
ues of N will make the calculation more accurate, but could be slow on a old computer.
We actually would rather you use the weakest-link rule whenever you can; you will not al-
ways have PropUnc handy in real life, so it is good to practice using the weakest-link rule, which
is simple and usually gives good results. You may use PropUnc (1) to check a weakest-link calcu-
lation, or (2) whenever the weakest-link rule or equation 9.3 does not apply. PropUnc is installed
on all the lab computers and also may be freely downloaded from the Physics 51 web site.

9.6 SOME EXAMPLES

Example 9.6.1: Suppose that you want to find the uncertainty in the volume of a cylinder when
you have measured its diameter and height. The volume V of a cylinder in terms of its diameter d
and height h is given by V = p 14 d 2 h. Here the volume has power-law dependences on the varia-
bles d and h, so we should be able to apply the weakest-link rule. Suppose that our measurements
are d = (0.200 ± 0.002) m and h = (0.600 ± 0.003) m. The fractional uncertainties in d and h are:
0.002 m 0.003 m
Q[d ] = = 0.01, Q[h] = = 0.005 (9.11a)
0.200 m 0.600 m
Note that even though the absolute uncertainty of d is smaller than that for h (0.002 m compared to
0.003 m), the fractional uncertainty of d is larger. Moreover, since V µ d 2 , the weakest link rule
tells us that we should be comparing 2Q[d ] to Q[h]: we see that in this case the first is four times
larger than the second. Therefore, according to the weakest-link rule
Q[V ] ª 2Q[d ] ª 0.02 (9.11b)
The 1% uncertainty in d thus leads to a 2% uncertainty in V. Now that we have the fractional un-
certainty in V we can find the absolute uncertainty pretty easily. The central volume value that we
calculate from our best-guess estimates of d and h is
p
V = p 14 d 2 h = (0.200 m)2 (0.600 m) = 0.0188 m 3 (9.12)
4
This value is uncertain to 2%, so its absolute uncertainty must be

U[V ] = V ◊ Q[V ] = 0.02(0.0188 m 3 ) = 0.000377 m 3 ª 0.0004 m 3 (9.13)

where I have rounded the uncertainty to one significant digit. An uncertainty this size means that it
is pointless to include more digits than we already have in equation 9.12. So a statement of this
value and its uncertainty would be V = (0.0188 ± 0.0004) m 3 or (1.88 ± 0.04) ¥10 2 m 3 . Note
that in either case we have written both values so that they are multiplied by the same power of 10.
This makes the values much easier to compare.
9. Propagation of Uncertainties 85

Example 9.6.2: Imagine that the number of bacteria in a certain colony at a certain time is N =
305,000 ± 15,000. What is the uncertainty in f = ln N ? (You might need to know the uncertainty
of the logarithm if you want to draw an uncertainty bar for this data point on a log-log graph.)
Since f = ln N is not a power-law relation, we cannot use the weakest-link rule. If we can’t
use PropUnc, we can fall back on the general method. In this case, if we change N from its central
value of 305,000 to the upper limit of its uncertainty range which is 320,000, the value of ln N
changes from ln(305,000) = 12.6281 to ln(320,000) = 12.6761, so the change in f due to this
change is d f N = +0.0480. Since f only depends on N in this case, equation 9.1 implies that

U[ f ] = (d f N )2 = d f N = 0.05 (9.14)

where I again have rounded to one significant digit.

The result from PropUnc is shown in Figure 9.2. Note that in “computerese” ln N becomes
“log(n)”. I used a small n to distinguish it from the number N of trials the computer evaluates.

Figure 9.2: P r o p U n c ’s check of equation 9.14.

If we were to naively apply the weakest-link rule anyway we would estimate that since the
fractional uncertainty in N is 15,000 / 305,000 ª 0.05, the fractional uncertainty in f = ln N would
also be 5%. This would lead us to estimate the uncertainty of f to be 0.05(12.63) ª 0.63, which is
more than 10 times larger than the more correct calculation given by equation 9.14. This illustrates
our earlier statement that the weakest-link rule does poorly when f involves logarithms.
9. Propagation of Uncertainties 86

9.7 THE BOTTOM LINE

You will be expected to state uncertainties of all calculated quantities in this lab program. Use
the weakest-link rule to estimate these uncertainties whenever that rule applies; when it doesn’t,
either use one of the methods discussed in section 9.5 or use PropUnc to calculate the uncertainty.

EXERCISES

Exercise 9.1
A person is measured to run a distance of 100.00 m ± 0.05 m in a time of 11.52 s ± 0.08 s.
What is the person’s speed and the uncertainty of this speed according to the weakest link rule?

Exercise 9.2
A spherical balloon has a radius of 0.85 m ± 0.01 m. How many cubic meters of gas does it
contain, and what is the uncertainty in your result?
9. Propagation of Uncertainties 87

Exercise 9.3
Imagine that you want to estimate the amount of gas burned by personal cars every year in the
U.S. You estimate that there are an average of about 0.7 ± 0.4 cars per person in the U.S., that
there are 275 million ± 30 million people in the U.S. currently, that a car is driven on the aver-
age about 15,000 mi ± 3,000 mi a year, and that the average number of miles per gallon that a
car gets is about 23 mi/gal ± 5 mi/gal. What is the approximate amount of gas burned and what
is the approximate uncertainty of this estimate?

Exercise 9.4
Equation 9.10 suggests that rather than dropping the other uncertainties entirely (as the weakest
link rule suggests) perhaps we would get a more accurate estimate of the fractional uncertainty
in a power-law relation by multiplying the fractional uncertainty of each variable by its power,
squaring the result, adding the squares and taking the square root of the sum. Do this for the
case described in Exercise 9.3 above. Is the answer you get from doing this careful way much
different from just using the weakest link result? Suppose that you do some research that ena-
bles you to reduce the fractional uncertainty in the all quantities but the worst one to 1%. Does
reduce the uncertainty much? If you really want to improve the uncertainty, what would be the
variable to focus on, and why?
9. Propagation of Uncertainties 88
 11
THE UNCERTAINTY OF THE MEAN

“ ‘Now, for instance, it was reckoned a remarkable thing that at the last party in my
rooms, that upon an average we cleared about five pints a head.’”
--- Northanger Abbey

11.1 USING A SET’S MEAN AS A “BEST GUESS”

Imagine that we are doing an experiment, and we want to make the best possible guess as to
the true value of a quantity whose measured values are subject to some kind of random error, either
due to human estimation error (as in the case of the stopwatch measurements in the speed of light
lab) or random perturbing physical effects (noise), or both. In order to get a sense of the uncertain-
ty of the measurement, you have already taken N individual measurements of the quantity.
If the effects that perturb the measured value away from its “true value” are truly random,
then they are equally likely to cause any given measured value to be higher than the true value as
they are to cause it to be lower than the true value. This means that if you take a set of N measure-
ments of the same quantity and those measurements are subject to random effects, it is likely that
the set will contain some values that are too high and some that are too low compared to the true
value. If you think of each measurement value in the set as being the sum of the measurement’s
true value and some random error, what we are saying is that the error of a given measurement is
as likely to be positive as negative if it is truly random.
One way to get a good estimate of the true value of a quantity is therefore to calculate the
mean (average) of the measurement values in the set. When you calculate the mean of a set of
measurements, you sum the measurements and then divide by N, the number of measurements in
the set. If we imagine each measurement to be equal to the true value plus some random error, then
the sum of N measurements will be equal to N times the true value plus the sum of all the random
errors. Since the random errors are as likely to be negative as positive, they will tend to cancel each
other out when summed, meaning that the sum of N measurements is likely to be very close to N
times the true value. Dividing the result by N to get the mean thus yields a number that is likely to
be close to the true value. The mean thus represents a better guess as to a quantity’s true value
than any of the individual values in set of measurements
The assumption that the errors cancel each other out gets better and better as the number of
measurements in the set increases. For example, if there are only three measurements in a set, there
is one chance in eight that all three measurements are too high, implying that the mean of these
measurements will not be a good estimate of the quantity’s true value. As the number of measure-
ments in a set increases, though, the probability that there is an approximately even mix of meas-
ured values in the set that are too high and too low becomes larger and larger.

11.2 THE UNCERTAINTY OF THE MEAN

The mean of any finite set of measurements is not going to be exactly equal to the quantity’s
true value: the random errors are not likely to perfectly cancel (especially if the number of measure-
ments is relatively small. Saying that the mean is a “better guess” than any single given measure-
ment is essentially saying that the uncertainty range of the mean is smaller than the uncertainty
range of any given single measurement. But how much smaller is it? How can we quantify the un-
certainty of the mean of the set of measurements?
11. The Uncertainty of the Mean 98

We define the uncertainty of the mean Um of a set of N measurements to be that value such
that we are 95% confident that ±Um encloses the “true value” of the measurement (where is the
mean of the set). If were to take enough data so that we have many sets of N measurements, (as
we did in the Speed of Sound lab), then this definition means operationally that if Um has the cor-
rect value, the range ± Um should enclose the true value for 95% of the sets’ means. According to
the mathematicians, we can estimate the uncertainty Um of the mean of a normally distributed set
of measurement using just one set of N measurements as follows:
ts
Um ª (11.1)
N
where s is the standard deviation of the set and t is the Student t-factor. In the Uncertainty of the
Mean lab, you should have verified that this expression was at least approximately consistent with
the definition of uncertainty just described.
Note that the estimate of the uncertainty of the mean given by equation 11.1 has two proper-
ties that we know (or at least intuitively expect) to be true. First, it implies that the uncertainty of
the mean is indeed smaller than the uncertainty of a single measurement (by the factor 1/ N ), as
we would expect from the argument given in the previous section. Second, the more measurements
we take, the smaller Um becomes, implying that the mean becomes a better estimate of the meas-
urement’s true value as the number of measurements it embraces increases (as we should expect).

11.3 ESTIMATING UNCERTAINTIES FROM THE RANGE

Calculating uncertainties of the mean through the TABLE FOR ESTIMATING Um
standard deviation can quickly become very tedious, FROM THE RANGE
particularly when an experiment involves a number of N Um N Um
sets of measurements. It would be nice to have a 2 10
6.40R 0.23R
method that allows us to determine the uncertainty of
the mean of a set of measurements at a glance. 3 1.30R 15 0.15R
4 0.72R 20 0.12R
Here is such a method. The first step is to comp-
ute the range R of your set of data, which is defined to 5 0.51R 25 0.09R
be the largest value minus the smallest value in your 6 0.40R 30 0.07R
set. (You can often compute R to one or two signifi- 7 40
0.33R 0.05R
cant digits in your head.) Then look in table shown to
the right for the number N of the measurements that 8 0.29R 50 0.04R
you took. An estimate of the value of the uncertainty 9 0.26R
(expressed as a multiple of R) appears in the next col-
umn, and again, you can probably do the multiplication (to one or two significant digits) in your
head. For example, if you have 10 measurements of a period of time ranging from 2.05 s to 2.22
s, the range is 0.17 s and the uncertainty of the mean of these measurements is (according to the
table) Um = 0.23R = 0.23(0.17 s) = 0.04 s. Easy!
The downside is that the value one reads from this table is only an estimate of the value that
we would get by calculating Um ª ts / N , and not a particularly great one for small N (because it
turns out that when N is small, the range of a set of measurements fluctuates from one set to anoth-
er even more than the calculated value of s does). Still, if we are after a quick and dirty estimate
that is still good to about one significant digit, this method is hard to beat.
We will call this method the range method of determining the uncertainty of the mean and
the method of directly computing the uncertainty of the mean using Um ª ts / N the formal
method. You may freely use the range method for quick estimates and for checking formal calcu-
11. The Uncertainty of the Mean 99

lations, but you should be sure to use the formal method when doing the final calculations that you
will report in an analysis summary or full report.
The program RepDat (see chapter 3) also calculates the uncertainty of the mean of a set of
measurements using the formal method (as you can easily check). This is probably the fastest way
to do the formal method if the program is available.

11.4 REPEATABLE AND UNREPEATABLE MEASUREMENTS

The discussion in the previous sections suggests that if you want to determine a measured
value as precisely as possible, it is better to take a set of measurements of that quantity and com-
pute the mean than it is to take just one measurement. IMPORTANT: In this lab program, when a
measurement is repeatable, you should always (automatically!) take a set of four to twenty meas-
urements of the quantity in question, compute the mean of the set to get a best guess for the meas-
urement value, and then compute the uncertainty of that mean.
But when is a measurement “repeatable”? There are three different kinds of measurements
that are not repeatable. The first are intrinsically unrepeatable measurements of one-time
events. For example, imagine that you are timing the duration of a foot-race with a single stop-
watch. When the first runner crosses the finish line, you stop the watch and look at the value regis-
tered. There is no way to repeat this measurement of the race duration: the value that you see on
your stopwatch is the only measurement that you have and could ever have of this quantity.
In the second category are subjectively influenced measurements where a subjective
human judgment is required to make the measurement and such a judgment can be influenced by
knowledge of past results or what the answer “should” be. In such cases, while we can technically
repeat the measurement process, our knowledge of previous results, or our expectation of a certain
outcome influences our judgment to such an extent that we cannot be sure that those measurements
will remain randomly distributed. For example, imagine measuring the size of an object using a
ruler. The first measurement that you take of this quantity may be subject to random effects (for
example, the way that you line up the ruler on the object, the orientation of your eye with respect to
the ruler, and so on). The second and subsequent measurements that you take of this quantity will
be strongly influenced by your knowledge of the first value that you obtained. It is very difficult to
be completely unbiased about such measurements after the first, particularly when it comes to read-
ing linear scales (after reading it once, it is hard to avoid looking at the scale the same way subse-
quently).
This is not generally a problem with digital readouts (where you do not make a subjective
judgment about the measurement) or when the random influences overwhelm the influences of
subjectivity. For example, if you are using a gauge to read the pressure of steam in a boiler and
clearly the reading is fluctuating dramatically from moment to moment, your memory of a previous
measurement is not likely to have much influence on subsequent measurements.
The final kind of unrepeatable measurement is a unvarying measurement whose value al-
ways comes out to be exactly the same if we do try to repeat it. This is a particular problem with
digital readouts, and is an indication that any random errors that might influence the measurement
are smaller than the precision of the instrument. In such a case, equation 11.1 cannot be used (it
would clearly yield a zero uncertainty): instead, we have to estimate the uncertainty from our
knowledge about the instrument’s precision.
Here is a general rule for distinguishing a repeatable from unrepeatable measurements:
A measurement is repeatable if (1) you are sure that you are really measuring the same
quantity each time you repeat the measurement and (2) you can be surprised by the result.
11. The Uncertainty of the Mean 100

The first condition excludes intrinsically unrepeatable one-time events, and the second excludes the
other two cases. (To decide whether you can be surprised in a given situation, you should prob-
ably take the measurement at least twice and find out.)

11.5 THE BOTTOM LINE

In this laboratory program, you should assign an uncertainty to every measured quantity as a
matter of course. If the measurement of that quantity is repeatable, take a set of measurements
(however many you have time for, but at least four) and compute the mean and the uncertainty of
the mean. Use the mean as the measurement value in any subsequent calculations.
If the measurement is not repeatable, estimate the uncertainty of your single measurement
based either on previous experience, on the principles that the uncertainty of linear scale or dial is
greater than ± one tenth of the smallest division and that the uncertainty of a digital readout is ± 1 in
the final digit, or your common sense. For example, you may remember that the uncertainty of
stopwatch measurements in the Inertial Mass experiment was a certain value: this would be a good
estimate of the uncertainty of a unrepeatable stopwatch measurement.
Remember that all uncertainty estimates (calculated or otherwise) are estimates good only to
about one significant figure. Do not waste a lot of time trying to calculate the uncertainty exactly:
use the range method (unless you really need the greater accuracy of the formal method).

EXERCISES

Exercise 11.1
Which of the following measurements are repeatable? Explain your response.
(a) Measuring the length of a sheet of paper with a ruler.

(b)Measuring the width of a building with a meter stick.

(c) Measuring the outside temperature with a thermometer.

(d) Measuring how long it takes a rock to drop 5 ft using a watch with a sweep second hand.

(e) Measuring the duration of a speech with a single watch.

Exercise 11.2
Consider the set of time measurements given in exercise 3.3 in chapter 3 of this lab reference
guide. What is the mean of these measurements, and what is its uncertainty? Calculate the latter
using both the formal method and the range method. Are the results about the same?
 13
EXPONENTIAL CURVE FITTING

13.1 INTRODUCTION
Many processes in nature have exponential dependencies. The decay with time of the ampli-
tude of a pendulum swinging in air, the decrease in time of the temperature of an object that is ini-
tially warmer than its surroundings, and the growth in time of an initially small bacterial colony are
all processes that are well-modeled by exponential relationships.
To better consider the issues involved in dealing with such relationships, let’s consider a very
specific case. The absorption of radiation by a given thickness of some material can be modeled by
the following simple exponential relationship:

R( x ) = R0 e b x (13.1)

Here R(x) is the count rate of radiation particles (typically measured as the number of clicks on a
Geiger counter that take place in some fixed time such as one minute), R0 is the count rate with no
shielding present, x is the thickness of the shielding material, and b is a negative constant that de-
scribes how rapidly the count rate decreases as the shielding thickness increases.
Some measurements on the rate at which radiation particles emitted by 55 Fe are detected
when a Geiger counter is shielded one or more thin sheets of aluminum foil appear in Table 13.1
and Figure 13.1. Note that the count rate decreases as the thickness of the aluminum shielding in-
creases. (Note also that the error bars in this case are just barely large enough to be visible.)

Figure 13.1: Count rate of 2000

radiation particles vs. thick-
ness of aluminum shielding.

Table 13.1: Counts/min vs. 1500

thickness of Al shielding
Count rate (counts/min)

Al thick- Count rate

ness (cm) (counts/min)
1000
0.00162 1850
0.00324 1250
0.00486 800
500
0.00648 450
0.00810 310
0.00972 165
0
0 0.002 0.004 0.006 0.008 0.01
Thickness of Al (cm)

While the count rate is clearly not a linear function of the shielding thickness x, you could not
(just by looking at the graph) tell the difference between the exponential dependence of equation
13.1 and certain power laws. Finding the value of the constant b would be difficult as well. You
could try different values of b and R0 , calculate an R(x) curve for each pair of values, and then see
which pair best matches your experimental data, but this approach would clearly be very tedious.
13. Exponential Curve Fitting 106

13.2 LINEARIZING EXPONENTIAL RELATIONSHIPS

Fortunately, a better strategy exists. If we take the (natural) logarithm of both sides of equa-
tion 13.1, we get
ln R( x ) = ln R0 + b x fi y( x ) = b x + ln R0 (13.2)

if we define y( x ) ∫ ln R( x ) . This is the equation of a straight line. Therefore, if you graph ln R( x )

vs. x, you should wind up with a straight line. Figure 13.2 shows such a graph.
9.0

ln(count rate) 8.0

Figure 13.2: Plot of the

natural log of the count rate 7.0
vs. aluminum thickness.

6.0

5.0
0 0.002 0.004 0.006 0.008 0.01
Thickness of Al (cm)

Furthermore, the slope of this line is b, the value of the constant in the original exponential
equation. You find the value of b by calculating the slope in the usual way. That is,
Dy y –y ln R( x2 ) – ln R( x1 )
b = = 2 1 = (13.3)
Dx x2 – x1 x2 – x1
Now, the data graphed above don't lie perfectly on a straight line, which is the result of ex-
perimental uncertainty. The line, however, looks like a good approximation to the data. The best fit
line, however, seems to pass through the points (0 cm, 8.0) and (0.01 cm, 5.1). [Remember that,
strictly speaking, both the y values should have the unit terms of ln(counts/minute) added, but
those terms will cancel when you do the subtraction.] The slope of this line is
5.1 – 8.0
b = = – 290 cm –1 (13.4)
0.01 cm – 0 cm
The point (0 cm, 8.0) where the line intersects the y axis, is (of course) the y intercept of the
line. So the value of y(0) is the constant ln R0 in equation 13.4. Therefore, R0 is the antilog
(exponential) of 8.0, or 2980 counts/minute. Therefore, we can use the ln R vs. x graph to deter-
mine both the constants in equation 13.2: that equation now reads:

R( x ) = (2980 counts / min) e bx , where b = 290 cm –1 (13.5)

In the expression above, x must be measured in cm, since the value of b has units of cm –1 . If you
wanted to measure x in mm instead, b would have to have the value –29 mm –1 .
13. Exponential Curve Fitting 107

13.3 CREATING SEMI-LOG GRAPHS USING LinReg

On can easily create a semi-log graph using the program LinReg: simply enter the basic data
into the data table and then check the “Show Ln(Vertical Data)” box. The program will then cal-
culate and plot the natural logarithms (and the uncertainties in those logarithms) of the measure-
ments displayed on your vertical axis. (See Chapter 10 for a more detailed description of LinReg.)

13.4 SEMI-LOG GRAPH PAPER

People have also invented special graph paper that can be used to create semi-log plots
without the help of a calculator or computer. This paper is called semi-log paper, and we have
included a sample semi-log plot of the radiation shielding data of Table 3.1 in Figure 13.5, and a
blank sheet of semi-log paper at the end of this chapter.
Fundamentally, the vertical axis on a piece of semi-log graph paper shows the values of y
corresponding to an implicit linear scale for ln y . To illustrate this, we have drawn the implicit lin-
ear logarithmic scale on the right edge of Figure 13.5. Therefore when we plot a point on the graph
according to the value of y shown on the left hand scale, we are really implicitly plotting the value
of ln y displayed on the right hand scale. This makes it easy to plot measured values directly with-
out having to compute the logarithms.
It is a property of such logarithmic scales that the pattern of y-marks on the left-hand scale
can be made to repeat exactly after y increases by a power of 10. The printers of semi-log paper
typically do repeat the pattern of marks for each power of 10, and make the paper more flexible by
not specifying the power of 10 that you will use at the bottom edge of the axis (the numerical labels
also simply repeat for each power of 10). You adapt the paper to the range of your own data by re-
labeling the 1s and 10s on the axis by a sequence of increasing powers of 10 that span your partic-
ular data. This is illustrated in Figure 13.5.
It really doesn’t matter what power of 10 you assign to the bottom edge of the axis. Chang-
ing the definition of the power of 10 at the bottom simply shifts the implicit linear logarithmic scale
up or down. Since when we compute the slope we are only really interested in the differences bet-
ween logarithmic values, shifting the logarithmic scale up or down doesn’t affect anything.

13.5 A PROCEDURE FOR EXPLORING EXPONENTIAL RELATIONSHIPS

Semi-log graphs are most useful when you suspect (for one reason or another) your data has
an exponential dependence of the form y = ke bx . To test your suspicion, do the following:
1. Plot a low-level semi-log graph of your data either using semi-log paper or by calculating
values of ln y and plotting your results using Cartesian graph paper. If the resulting graph is
in either case a pretty good straight line, this reinforces your suspicion that the data might re-
flect an exponential relationship, and it is worth continuing to the next step.
2. Find the constant multiplier k, by extrapolating your best fit line back to x = 0 and reading
either the value ln k off the vertical axis (if you used Cartesian graph paper) or the value of k
itself (if you used semi-log graph paper).
3. Calculate the value of b from the slope D(ln y)/Dx, as discussed in section 13.2. (If you used
semi-log graph paper, you will have to compute ln y by hand for the two points that you use
to define the slope.)
4. Finally, refine these values for intercept and slope by entering the data into LinReg. Be sure
to check your results against what you found using your low-level plot.
13. Exponential Curve Fitting 108

EXERCISES

Exercise 13.1
Verify that if b = 290 cm –1 , then b also equals 29 mm –1 , as claimed below equation 13.7.

Exercise 13.2
Figure 13.5 on the next page shows a semi-log graph of the radiation data, with error bars.
Draw the line that you think best fits your data, and then find values of b and R0 from your
line. Since Figure 13.5 is larger than Figure 13.2, your new estimates of b and R0 will prob-
ably be better than the earlier estimates summarized in equation 13.7. [Hint: You will have to
calculate ln R for two points points on the line to accurately determine the slope, but you can
read R0 right from the diagram.

Exercise 13.3
On the blank semi-log paper provided in Figure 13.6,
time t (min) Number of bacteria N
plot the data given in the table to the right. Determine
whether this data seems to reflect an exponential relation- 10 149,000 ± 15,000
ship N = N0 e b t , and if so, find the values of b and N0 20 215,000 ± 20,000
that best fit this data from both graphs. Also, plot in your 30 335,000 ± 35,000
lab notebook a graph of ln N versus t on ordinary graph 40 477,000 ± 45,000
paper and do the same analysis. (You can use equation 50 769,000 ± 75,000
9.16 in Chapter 9 of this manual to compute the uncer-
tainty in ln N .)
13. Exponential Curve Fitting 109

Figure 13.5: Semi-log plot of data from Table 13.1.

13. Exponential Curve Fitting 110

Figure 13.6: Blank piece of 2-cycle semi-log paper