Click to edit Master title style

Chapter 3: Assessing and

Presenting Experimental Data
Mohammad Suliman Abuhaiba, Ph.D., PE
The Islamic University of Gaza
Mechanical Engineering Department

1
Click to edit Master title style
INTRODUCTION

• Before a data set can be used in engineering or

scientific application, its quality must be
established
• Measurements should not be compared to a
theory in order to assess their quality
• Error = measured value - true physical value

2 2
Click to edit Master title style
INTRODUCTION

• Error cannot be calculated exactly unless we know

true value of quantity being measured.
• We can never know the true value of a physical
quantity without first measuring it
• Some error is present in every measurement, true
value is something we can never know exactly.
• We can never know the error exactly, either.

3 3
Click to edit Master title style
INTRODUCTION

• Error or uncertainty may be estimated with statistical

tools with a large number of measurements are taken.
• Experimentalist must bring to bear his own knowledge
of how the instruments perform and of how well they
are calibrated in order to establish the possible errors
and their probable magnitudes.

4 4
Click to edit
COMMON TYPES
Master
OF ERROR
title style

• Error in measuring a quantity x: deference between

measured value and true value,

• A primary objective in designing and executing an

experiment is to minimize the error.
• After the experiment is complete, we must estimate
a bound on error with some level of confidence.

5 5
Click to edit
COMMON TYPES
Master
OF ERROR
title style

• u = uncertainty estimate at odds of n : 1

• Only one measurement in n will have an error
whose magnitude is greater than u

• Two general classes of error:

1. bias (systematic) errors
2. precision (random) errors

6 6
COMMON
Click to edit
TYPES
Master
OF ERROR
title style
Bias (Systematic) Errors

• Bias errors occur the same way each time a

measurement is made.
• Example: if the scale on an instrument consistently
reads 5% high, then the entire set of measurements
will be biased by +5% above true value.
• The scale may have a fixed offset error, indicated
value for every reading of x is higher than true value
by an amount Xoffset

7 7
COMMON
Click to edit
TYPES
Master
OF ERROR
title style
Bias (Systematic) Errors

• Bias errors cannot be treated using statistical

techniques - they are fixed and do not show a
distribution
• Bias error can be estimated:
 By comparison of instrument to a more accurate standard
 From our knowledge of how the instrument was calibrated
 From our experience with instruments of that particular type

8 8
COMMON
Click to edit
TYPES
Master
OF ERROR
title style
Precision (Random) Errors

• Precision errors are different for each successive

measurement - have an average value of zero.
• Example: mechanical friction or vibration may cause
the reading of a measuring mechanism to fluctuate
about the true value, sometimes reading high and
sometimes reading low.
• This lack of mechanical precision will cause sequential
readings of the same quantity to differ slightly, creating
a distribution of values surrounding the true value.
9 9
COMMON
Click to edit
TYPES
Master
OF ERROR
title style
Precision (Random) Errors

• Successive readings will generally cluster about a

central value and will extend over a limited interval
surrounding that central value.
• We may use statistical analysis to estimate:
 likely size of the error
 likely range of x in which the true value lies

1010
COMMON
Click to edit
TYPES
Master
OF ERROR
title style
Precision (Random) Errors

1111
Click to edit Master
CLASSIFICATION OF ERRORS
title style

1. Bias or Systematic Error

a. Calibration errors
b. Certain periodic human errors
c. Certain errors caused by defective equipment
d. Loading errors
e. Limitations of system resolution

1212
Click to edit Master
CLASSIFICATION OF ERRORS
title style

2. Precision or Random Error

a. Errors caused by disturbances to the equipment
b. Errors caused by fluctuating experimental conditions
c. Errors from insufficient measuring system sensitivity

1313
Click to edit Master
CLASSIFICATION OF ERRORS
title style

3. Illegitimate Errors
a. Mistakes during an experiment
a. Computational errors after an experiment

1414
Click to edit Master
CLASSIFICATION OF ERRORS
title style

4. Errors That Are Sometimes Bias And Sometimes Precision

a. Errors from instrument backlash, friction, and hysteresis

b. Errors from calibration drift and variation in test or
environmental conditions
c. Errors resulting from variations in procedure or definition
among experimenters
1515
 (c)

Click to edit Master

CLASSIFICATION e

Of ERRORS
title style

Calibration Errors
• Occur when an instrument's scale has not
been adjusted to read the measured value
properly.
• Typical calibration errors:
 Zero - offset: cause all readings to be
offset by a constant amount Xoffset
 Scale errors in slope of output relative to
input which cause all readings to error by
a fixed percentage

1616
 (c)

Click to edit Master

CLASSIFICATION Of ERRORS
e

title style

• Human errors may be systematic, as when an experimenter

constantly tends to read high when synchronized readings
are to be taken.
• Equipment itself may introduce built-in errors resulting
from incorrect design, fabrication, or maintenance.
 Such errors result from defective mechanical or electrical
components, incorrect scale graduations, and so forth.
 Errors of this type are often consistent in sign and
magnitude, and because of their consistency they may
sometimes be corrected by calibration.

1717
 (c)

Click to edit Master

CLASSIFICATION Of ERRORS
e

title style

Loading Error
• Refers to influence of measurement procedure on system being tested.
• Measuring process unavoidably alters the characteristics of both the
source of the measured quantity and the measuring system itself; thus,
the measured value will always differ by some amount from the
quantity whose measurement is sought.
• Example: sound - pressure level sensed by a microphone is not same as
sound - pressure level that would exist at that location if microphone
not present.
• Minimizing influence of the measuring instrument on a measured
variable is a major objective in designing any experiment.
1818
 (c)

Click to edit Master

CLASSIFICATION
t
Of ERRORS
title style

• Variations in the actual quantity being measured may

also appear as precision error in the results.
• Sometimes these variations are a result of
experimental design, as when a system designed to
run at a constant speed instead has a varying speed.

1919
 (c)

Click to edit Master

CLASSIFICATION
t
Of ERRORS
title style

• Sometimes variations are an inherent feature of the

process under study, as when manufacturing variations
create a distribution in the operating lifetimes of a
group of light bulbs.
• Variation in the measured quantity is not a
measurement error; however, it is possible to apply
same statistical techniques to variations in the
measured variable and to treat them as if they were
errors.

2020
 t

Click to edit Master

CLASSIFICATION
t
Of ERRORS
title style

• Illegitimate Errors: errors that would not be expected

• These include mistakes which can be eliminated through
exercise of care or repetition of the measurement, such as:
 incorrectly writing down a number
 failing to turn on an instrument
 miscalculating during data reduction
• Sometimes a statistical analysis will reveal such data being
extremely unlikely to have arisen from precision error
2121
 t

Click to edit Master

CLASSIFICATION Of ERRORS
title style
r
t

• Backlash and mechanical friction are

important sources of variation in measuring
systems.
• Example: friction may cause a mechanical
element, such as a galvanometer needle, to
lag behind advances in its intended
position, thus reading low while the
measured variable is increased and reading
high while the measured variable is
decreased.
2222
CLASSIFICATION Of ERRORS
Click to edit Master title style

• Drift of an instrument's calibration may occur if the response

of an instrument varies in time.
• Often drift results from the sensitivity of instrument to
temperature or humidity fluctuations.
• If changes in environmental conditions occur between the
time an instrument is calibrated and the time it is used, a bias
error may appear in the readings.
• If the test duration is long, environmental conditions may
fluctuate throughout the test, causing different calibration
errors for each successive measurement. In this case, the
fluctuations create a precision error.
2323
Click to edit Master
CLASSIFICATION Of ERRORS
title style

• When an experiment is repeated using different equipment or by

different experimenters, bias errors of successive experiments
are unrelated.
• If enough different experiments are performed, bias errors are
effectively randomized, and they become another form of
precision error in the set of all experiments.
• For example, speed of light in vacuum has been measured by
many experimenters, each using different techniques and
apparatus to obtain what is supposedly a unique quantity.
• Each experiment included its own bias and precision errors; but
taken together their results show a distribution about a mean
value, which may be estimated statistically. 2424
TERMS
Click toUSED
edit Master
IN RATING
title style
INSTRUMENT PERFORMANCE

• Accuracy: difference between measured and true values

• Precision: difference between instrument's reported
values during repeated measurements of the same
quantity. this value is determined by statistical analysis of
repeated measurements.

2525
 TERMS
Click toUSED
edit Master
IN RATING
title style
INSTRUMENT PERFORMANCE

Accurate vs precise
Density measured by four students (g/cm3)
Mohammad Sameer Abdallah Noor
Four students perform experiments to measure
2.924 2.316 2.649 2.701
density of Aluminum (True density = 2.7 g/cm3).
2.923 2.527 2.731 2.699
Classify results in terms of accuracy and
2.925 2.941 2.695 2.702
preciseness.
2.926 2.136 2.742 2.698
Mean of Experimental Data 2.925 2.480 2.704 2.700
Standard deviation from the mean 0.00129 0.346 0.0419 0.00183
True value 2.700 2.700 2.700 2.700
Error = Mean - True value 0.224 0.220 0.004 0.000
Order in terms of accuracy (highest to lowest) 4 3 2 1
Order in terms of precesion (highest to lowest) 1 4 3 2
Not Accurate Not Accurate Accurate Accurate
2626
Precise Not Precise Not precise Precise
TERMS
Click toUSED
edit Master
IN RATING
title style
INSTRUMENT PERFORMANCE

• Resolution: smallest increment of change in the

measured value that can be determined from the
instrument's readout scale
• Resolution is often on the same order as the precision;
sometimes it is smaller.

2727
TERMS
Click toUSED
edit Master
IN RATING
title style
INSTRUMENT PERFORMANCE

• SENSITIVITY: change of an instrument output per unit

change in the measured quantity (input).
• A more SENSITIVE instrument's reading changes significantly
in response to smaller changes in the measured quantity.
• An instrument with higher sensitivity will also have finer
resolution, better precision, and higher accuracy.

2828
TERMS
Click toUSED
edit Master
IN RATING
title style
INSTRUMENT PERFORMANCE
• Reading error: error introduced when reading a number from the display
scale of an instrument.
• This type of error may sometimes be a bias error caused by truncation or
rounding of the actual value to one within the resolution of the display.
• Reading error will also include error from inadequate instrument
sensitivity if the instrument does not respond to smallest fluctuations of
the measured quantity.
• Example: a digital display may truncate an actual value of 10.4 to a
displayed value of 10. The reading error of the digital display is thus ±
1/2 of the last digit read. This is a bias error in the sense that 10.4 will
always be displayed as 10. When many different values are to be read,
the error may be thought of as a precision error if the many values have
no particular relation to one another.

2929
TERMS
Click toUSED
edit Master
IN RATING
title style
INSTRUMENT PERFORMANCE

• For a needle display on a galvanometer or the

scale on a micrometer one may be able lo
estimate to ±1/2 or even ±1/5 of the finest
graduation.
• Depending on the particular experimenter, such
error may be either bias or precision error.

3030
Click to edit Master
INTRODUCTION TO UNCERTAINTY
title style

When estimating uncertainty, we usually concerned

with:
1. Two types of error:
a. precision error
b. bias error
2. Two classes of experiments:
a. single-sample experiments
b. repeat sample experiments

3131
Click to edit Master
INTRODUCTION TO UNCERTAINTY
title style

• A sample refers to an individual measurement of a specific

quantity.
• When we measure strain in a structural member several times
under identical loading conditions, we have repeatedly sampled
that particular strain.
• With such repeat sampling, we can statistically estimate the
distribution of precision errors in the strain measurement.
• If we measure that strain only once, we have instead a single
sample of the strain, and our result does not reveal the
distribution of precision error.

3232
Click to edit Master
INTRODUCTION TO UNCERTAINTY
title style

• After determining individual bias & precision

uncertainty in a measurement of x, we must
combine them to obtain total uncertainly in our
result
• If bias uncertainty is Bx and precision uncertainly
is Px , then total uncertainty, Ux is given by:

3333
Click to edit Master
INTRODUCTION TO UNCERTAINTY
title style

3434
Click to edit Master
INTRODUCTION TO UNCERTAINTY
title style

Assignment # 1
3, 4, 5, 6, 7, 8

3535
ESTIMATION
Click to edit OF
Master
PRECISION
title style
UNCERTAINTY

Two fundamental concepts form the basis for analyzing

precision errors:
1. Distribution of errors: it characterizes probability that an
error of a given size will occur.
2. A population from which samples are drawn. Usually, we
have only a limited set of observations, our sample, from
which to infer characteristics of the larger population.

3636
ESTIMATION
Click to edit OF
Master
PRECISION
title style
UNCERTAINTY
• Statistical analysis of error: model for the
distribution of errors in a population, Gaussian or
normal distribution.
• Using this assumed distribution, we may estimate
the probable difference between, say, the average
value of a small sample and the true mean value of
the larger population.
• This probable difference, or confidence interval,
provides an estimate of the precision uncertainty
associated with our measured sample.
3737
ESTIMATION
Click to edit OF
Master
PRECISION
title style
UNCERTAINTY
Sample Versus Population
• Manufacturing variations in a production lot of marbles
will create a distribution of diameters.
• To estimate mean diameter, we may take a handful of
marbles, measure them, and average the result
• The handful our sample, drawn from the production lot.

3838
ESTIMATION
Click to edit OF
Master
PRECISION
title style
UNCERTAINTY

Sample Versus Population

• No two samples from same population will yield precisely
same average value; however, each should approximate
the average of population to some level of uncertainty.
• Difference between sample characteristics and those of
population will decrease as sample is made larger.

3939
ESTIMATION
Click to edit OF
Master
PRECISION
title style
UNCERTAINTY
Sample Versus Population
• Because our handful of marbles includes a number of
members of the population, it may be regarded as a
repeated sample.
• If we had drawn only one marble, we would obtain a
single sample, which would give no direct evidence of
distribution of marble diameters.

4040
ESTIMATION
Click to edit OF
Master
PRECISION
title style
UNCERTAINTY
Sample Versus Population
• Experimental errors can be viewed in terms of population and
sample.
• If we measure diameter of a single marble repeatedly, the set of
measured diameters gives a sample of the precision error in
measuring the diameter. In this case, we could measure diameter
as many times as we liked, and each measurement would include a
slightly different precision error.
• Population of precision errors is theoretically infinite. (Note that
this particular repeat sampling of the precision error is performed
on a single sample from the marble population.)
4141
Click to edit OF
ESTIMATION Master title style
PRECISION
UNCERTAINTY
Sample Versus Population
Two classes of sampling:
1. A sample of size n is drawn from finite population of size p. The
sample is used to estimate properties of the population. Sample
size is assumed to be small compared to population size: n<<p.
2. A finite number of items, n, is randomly drawn from what is
assumed to be a population of indefinite size. The properties of
the assumed population are inferred from the sample.

4242
Click to edit OF
ESTIMATION Master title style
PRECISION
UNCERTAINTY
Sample Versus Population
• Sample must be randomly selected from population.
• If we select only the largest marbles from the bag, our
sample will not accurately represent the whole
population of marbles.

4343
Click to edit OF
ESTIMATION Master title style
PRECISION
UNCERTAINTY
Probability Distributions
• Probability: likelihood of a particular event taking place,
measured with reference to all possible events.
• Suppose that one of n equally likely cases will occur and
that m of these cases correspond to an event A. The
probability that event A will occur is m/n.
• A penny is tossed. Total number of possible outcomes is
two - heads and tails. If we choose heads as event A,
then the probability of A is 1 in 2, or 50% .
4444
 Number of possibilities Sum of Spots
1 2
2 3
Click to edit OF
ESTIMATION Master title style
PRECISION 3
4
4
5
UNCERTAINTY 5
6
6
7
5 8
Probability Distributions 4
3
9
10

• Example: throwing a pair of dice 2

1
11
12
36
• One possible outcome yields a sum
of 2, six outcomes yield a sum of 7,
and remaining outcomes are as
distributed in the histogram.
• If we divide ordinate of chart by total
number of possible outcomes (36),
we obtain a graph of probability
distribution.
4545
 x,

THEORY
Click to edit
BASEDMaster
ON THE
x
title style
POPULATION
• Consider an infinite population of data
• Each datum representing a measurement of a single
quantity, and assume each datum, differs in magnitude
from the rest only as a result of precision error.
• Each time a different member of the population is
randomly selected and measured, a value of x is
determined and, a different precision error occurs.
• Probability of obtaining a specific value of x depends
magnitude of x, and probability distribution of x-values is
described by a probability density function.
4646
 x,

Click to edit
THEORY BASED
x
Master
ON THE
titlePOPULATION
style

4747
 x,

THEORY
Click to edit
BASEDMaster
x
ON THE
title style
POPULATION
• Population is infinite, PDF is a continuous curve
• PDF: probability of occurrence per unit change of x
• probability of measuring a given x is not f(x)
• probability of measuring an x in the interval Dx = x2 – x1
is the area under the PDF curve between x1 and x2

4848
 x,

THEORY
Click to edit
BASEDMaster
x
ON THE
title style
POPULATION

• In any measurement, some value of x will be

observed, so total area under PDF curve is unity
(probability of measuring some x value is 100%)
• Precision error in an experimental data is often
distributed according to the bell - shaped curve given
by the Gaussian or normal distribution (Fig. 7)

4949
 x,

THEORY
Click to edit
BASEDMaster
ON THE
x
title style
POPULATION
• For an infinite population, the mathematical
expression for Gaussian PDF is

• where
x = magnitude of a particular measurement
µ = mean value of the entire population
s = standard deviation of the entire population
5050

•
 x,

THEORY
Click to edit
BASED
x
Master
ON THE
title style
POPULATION
• Arithmetic average of n measurements (most probable
single value of the quantity, µ) is given by:
•

• Amount by which a single measurement is in error is

termed the deviation d:
5151
 x,

THEORY
Click to edit
BASEDMaster
ON THE
title style
x

POPULATION

Summary of probability estimates based on the

normal distribution
Common Name for Error Level in % Confidence that deviation of Odds that deviation
Error Level Terms of s x from mean is smaller of x is greater

Standard deviation ±s 68.3 1 in 3

Two-sigma error ± 1.96s 95.0 1 in 20
•
Three-sigma error ± 3s 99.7 1 in 370

5252
 x,

THEORY
Click to edit
BASEDMaster
x
ON THE
title style
POPULATION
• Mean standard deviation

• When s is smaller, data are more precise

• One common criterion for discarding a data point as illegitimate is
that the data point exceeds the 3 s level: since the probability of
an error is larger than this is 1 in 370. Such values are unlikely in
modest-sized data sets. Such data are sometimes called outlier
5353
 x,

THEORY
Click to edit
BASED
x
Master
ON THE
title style
POPULATION

• For purposes of tabulation, Gaussian PDF may be transformed by

introducing the variable z:

5454
 x,

Click to edit
THEORY BASEDMaster
x
ON THE
titlePOPULATION
style

Standard normal distribution curve (z- distribution)

5555
 u = the standard deviation of the entire population
x,

Click to edit
THEORY BASEDMaster
ON THE
titlePOPULATION
style x

(x
)

Table 2: Areas under the Standard normal curve

5656
 x,

THEORY
Click to edit
BASEDMaster
x
ON THE
title style
POPULATION
EXAMPLE 2
a) What is te area under the curve between
z = - 1.43 and z = 1.43?
b) What is the significance of this area?

5757
 x,

THEORY
Click to edit
BASEDMaster
x
ON THE
title style
POPULATION
(x
)

EXAMPLE 3
What range of x will contain 90% of the data?

5858
Click to edit Master title style

Assignment # 2

1, 2, 15

5959
 x,

Click to edit
THEORY BASEDMaster
x
ON THE
titleSAMPLE
style

• Objective: use average values from sample to

estimate mean or standard deviation of population
• Sample mean and sample standard deviation

6060
 x,

Click to edit
THEORY BASEDMaster
x
ON THE
titleSAMPLE
style

Symbol Designation: Population vs Sample

Population Sample

Mean m or mx

Standard Deviation s or sx

6161
Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style

Example of Sampling
During a 12-hour test of a steam generator, inlet pressure
is to be held constant at 4.00 MPa. For proper
performance, pressure should not deviate from this value
by more than about 1 %. Inlet pressure was measured 100
times during the test. Various factors caused the readings
to fluctuate. The resolution of the digital pressure gauge
used was 0.001 MPa. The number of results, m, is the
number of readings falling in an interval of ± 0.005 MPa
centered about the listed pressure.
6262
Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style

Pressure, p (MPa) Number of results, m

3.970 1
3.980 3
3.990 12 Example of Sampling
4.000 25
4.010 33
4.020 17
4.030 6
4.040 2
4.050 1
100 6363
Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style

Example of Sampling
Plot Histogram of the
readings to assess the
distribution pattern

6464
 Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style - Example of Sampling
Sample mean & standard deviation mean pressure = 4.008 MPa
Standard Deviation = 0.014 MPa

Pressure, p (MPa) Number of results, m p.m deviation, d d2 m.d2

3.970 1 3.970 -0.038 0.00142 0.00142
3.980 3 11.940 -0.028 0.00077 0.00230
3.990 12 47.880 -0.018 0.00031 0.00376
4.000 25 100.000 -0.008 0.00006 0.00148
4.010 33 132.330 0.002 0.00001 0.00017
4.020 17 68.340 0.012 0.00015 0.00257
4.030 6 24.180 0.022 0.00050 0.00298
4.040 2 8.080 0.032 0.00104 0.00209
4.050 1 4.050 0.042 0.00179 0.00179
Sum 100 400.770 0.000 0.0000605 0.0186
6565
Click to edit
THEORYn
BASEDMaster
ON THE
o
n
titleSAMPLE
style

Example of Sampling

• Interval containing 95% of pressure readings:

µ ± 1.96s = p ± 1.96 Sp = 4.008 ± 0.027 MPa
• One objective of the pressure test was to verify
that pressure did not deviate from 4.00 MPa by
more than 1% = 0.04 MPa.
• In terms of standard deviation, 0.04 MPa = 2.86s
• For a Gaussian distribution, probability of a 2.86s
fluctuation is about one chance in 240. 6666
Assignment #3
Click to edit Master title style

21

6767
Click to edit
THEORYn
BASEDMaster
ON THE
o
n
titleSAMPLE
for
style

Confidence Intervals for Large Samples, n ≥ 30

• Previous example showed how we can assess the dispersion of

sample values about the mean value of the sample.
• It did not yield an estimate for the uncertainty in using as an
approximation to the true mean µ.
• If pressure test were repeated and another 100 points acquired,
the new mean value would differ somewhat from first mean value.
• If we repeated the test many times, we would obtain a set of
samples of the mean pressure.
6868
 e

Click to edit
THEORYn
BASEDMaster
ON THE
o
n
titleSAMPLE
style
for

Confidence Intervals for Large Samples, n ≥ 30

• Samples of the mean also show a dispersion about a

central value.
• If n for each sample is very large, distribution of mean
values is Gaussian that has a standard deviation of

6969
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
.

o
n

Confidence Intervals for Large Samples, n ≥ 30

• Since is Gaussian distributed, with standard deviation

• With c% confidence, true mean value, µ, lies in the following interval

7070
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
.

o
n

Confidence Intervals for Large Samples, n ≥ 30

• For large samples, set in estimating the confidence interval for :

7171
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
.

o
n

Confidence Intervals for Large Samples, n ≥ 30

Example 4: Determine a 99% confidence interval for the mean pressure
calculated in the previous example

7272
Assignment #4
Click to edit Master title style

18

7373
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
.

o
n

Confidence Intervals for Small Samples, n<30

• The distribution of a quantity t

• As n (or v = n – 1) becomes
large, t-distribution approaches
standard Gaussian PDF
• For n > 30, the two
distributions are identical.

7474
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
.

o
n

Confidence Intervals for Small Samples, n<30

• Table 5: Student’s t-distribution (ta,n)

• Area a between ta & t = ∞ is listed
• Two-sided confidence interval for the
mean of a small sample: both upper and
lower bounds on the mean are stated
• With a confidence level of c%, true
mean values lies in the interval

7575
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
.

o
n

Confidence Intervals for Small Samples, n<30

• This confidence interval defines the precision

uncertainty in the value of the mean.
• Precision uncertainty in the mean is

7676
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
.

o
n

Confidence Intervals for Small Samples, n<30

7777
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. ar
e
o
n

Confidence Intervals for Small Samples, n<30

Example: A simple postal scale of the equal-arm balance

type, is supplied with 0.5, 1, 2, and 4-oz machined brass
weights. For a quality check, the manufacturer randomly
selects a sample of 14 of the 1-oz weights and weighs
them on a precision scale.
The results, in ounces, as follows: 1.08, 1.01, 0.97, 1.03,
0.98, 1.00, 0.96, 0.99, 0.98, 0.95, 1.05, 1.01, 1.04, 1.08
Based on this sample and the assumption that the
parent population is normally distributed, what is the
95% confidence interval for the population mean?
7878
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. ar
e
o
n

Confidence Intervals for Small Samples, n<30

7979
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a

Small Sample Size , n < 30
• Employ statistics in order to make certain decisions
regarding a measurand.
• Example: consider pressure measurements listed in Table 3.
These measurements represent single sample containing
100 measurements obtained during a 12- hour period which
produce a mean pressure of 400.77 MPa.
• If additional samples were taken, and the mean pressure of
each sample obtained, the mean pressure can be assumed
to be normally distributed.
8080
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a

Small Sample Size , n < 30

• Often we are concerned whether the pressures

observed are representative of a set point pressure of
4.00 MPa.
• One of the most commonly used methods for making
such decisions is hypothesis testing.
• Typically there are two hypotheses in a hypothesis test:
1. The null hypothesis
2. The alternate hypothesis
8181
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a

Small Sample Size , n < 30
• The first step in setting up a hypothesis test is to choose
the null hypothesis.
• Since hypothesis testing often refers to a simple mean, it
takes the form:

µo = some constant specified value

8282
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a

Small Sample Size , n < 30

• The second step in hypothesis testing involves specifying

an alternate hypothesis.
• The choice of the alternate hypothesis should reflect on
what we are attempting to show

8383
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a

Small Sample Size , n < 30

Three possibilities for the choice of the alternative

hypothesis:

1. Two-tailed test

2. Left-tailed test

3. Right-tailed test
8484
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a

Small Sample Size , n < 30

Figure 12: Criteria for rejection of the null hypothesis

8585
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a

Small Sample Size , n < 30

Steps of hypothesis testing:

1. Define null & alternate hypothesis
2. Define level of confidence, c%
3. Calculate value of texp from data
4. Table 5: determine proper value of ta, m using degree of freedom n
5. If texp falls in the reject Ho region, reject Ho and accept the alternate hypothesis, Ha
6. If texp falls ls in the do not reject Ho region, we do not have sufficient data to reject Ho
at the level of confidence specified. (Strongest statistical statement occurs when we
can reject Ho and accept Ha)
8686
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a

Small Sample Size , n < 30

Example 7: Using data of Example 6, determine if

the sample of 14 of 1-oz weights comes from a
population of weights whose true mean weight is
greater than 1.00 oz, assuming a confidence level
of 99%.
Solution:
Right tailed test and texp falls in the do not reject
Ho region, the 99% confidence level that the
population mean was not significantly different
than 1.00 oz 8787
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a large

Sample Size , n ≥ 30

The six steps listed in previous section are followed

exactly except:
1. Replace ta,n by Zn
2. Replace texp by Zexp

8888
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a large

Sample Size , n ≥ 30

Example 8: From the pressures in Table 3, can we

conclude that the experimental data obtained indicate a
target pressure of 4.00 MPa at a confidence level of
99%?
Solution

8989
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a large

Sample Size , n ≥ 30

Example 8

9090
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

Hypothesis Testing for a Single Mean for a large

Sample Size , n ≥ 30

Example 8:
Since Zexp is in the reject Ho region, we reject Ho and
accept Ha
The data indicate that the target pressure of 4.00 MPa is
not being accurately controlled based on a confidence
level of 99%.

9191
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

The t-Test Comparison of Sample Means

• If we wish to compare two samples solely on the

basis of their means, use the following Eq:

• Value of t is compared to the interval ±ta/2,n found

in table 5
9292
 µ, e

Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n

The t-Test Comparison of Sample Means

• Degree of freedom n is approximated by:

9393
 Bulb A Bulb B
µ, e
7.2 7.5
Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n 7.6 8.7
The t-Test Comparison of Sample Means 6.9 7.7
8.2 7.5
7.3 6.7
Example 9: an apartment manager wishes to 7.8 11.2

determine if the lifetimes are different, under similar 6.6 7.0

6.9 10.7
conditions, for two major brands of light bulbs. In the
5.5 7.0
following sample data, the lifetime is in months. 7.4 8.6
Solution: 5.7 6.1
6.2 6.3
7.8
8.7
6.1

9494
 Bulb A Bulb B
µ, e
7.2 7.5
Click to edit
THEORY BASEDMaster
ON THE
titleSAMPLE
style
. a

I o
n 7.6 8.7
The t-Test Comparison of Sample Means 6.9 7.7
8.2 7.5
7.3 6.7
7.8 11.2
Solution:
6.6 7.0
6.9 10.7
5.5 7.0
7.4 8.6
5.7 6.1
6.2 6.3
Value of texp falls within the do not
7.8
reject Ho region. There is not a
8.7
significant difference in lifetimes of 6.1
bulbs A and B at 95% confidence
level 9595
Assignment #6
Click to edit Master title style

14, 19, 20, 22, 23,

24, 29

9696
ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style

• A variable is considered to have a chi-square

distribution if its distribution has the shape of a right-
skewed curve as shown.
• Total area under the curve = 1
• As degrees of freedom, v, becomes large, chi-square
distribution approaches a symmetric distribution which
resembles the normal distribution.
• Table 6 presents values of X2 for various values of a
and n
9797
ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style

• Figure 13: Probability distributions for the (X2) statistic

9898
THE CHI-SQUARE
Click to edit Master
(X2title
) DISTRIBUTION
style
is
a
s
various values of a and v.

9999
ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style

The chi-square goodness-of-fit statistic is defined as

Oi = observed frequency
Ei = expected frequency
K = total number of variables being compared
n=k-1

100
100
ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style

• Since the chi-square statistic, x2, is positive, a right-

tailed test is applied for hypothesis testing
• When the observed frequencies approach the expected
frequencies, the value of x2 approaches zero

101
101
 Face-Up Observed
ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style sum frequency
EXAMPLE 10 2 8
3 25
A pair of dice are tested to determine if they are 4 29
"true“. When the dice are tossed, the face-up
5 42
sum total is read. After 360 such tosses. the
6 53
results shown in lhe following table were
obtained. Determine if the dice are true at the 7 55
99% confidence level. 8 46
9 39
10 29
11 22
12 12
360
102
102
 Observed Expected
Face-Up
sum
frequency, frequency, X2Exp
Oi Ei
ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style 2 8 10 0.4000
EXAMPLE 10 3 25 20 1.2500

4 29 30 0.0333
Solution:
5 42 40 0.1000
Ho: frequencies are the same
6 53 50 0.1800
Ha: frequencies are different
7 55 60 0.4167

8 46 50 0.3200

9 39 40 0.0250

10 29 30 0.0333

11 22 20 0.2000

C= 0.99, a = 1-0.99=0.01, k = 12 12 10 0.4000

11, n = k-1=11-1=10
360 360 3.358

The dice are true at the 99%

103
103
confidence level
Assignment # 7
Click to edit Master title style

30

104
104
ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian
Distribution

Since a given set of data may or may not follow the

assumed distribution and since, at best, the degree of
adherence can be only approximate, some estimate of
goodness of fit should be made before critical decisions
are based on statistical error calculations

105
105
ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian
Distribution

Method 1: Histogram
• Plot a histogram to "eyeball" the result
• This approach can easily result in misleading conclusions.
• Appearance of histogram can sometimes be altered radically,
simply by readjusting the number of class intervals.

106
106
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian
Distribution

Method 2: Normal Probability Plot (NPP)

What is a Quantile?
• A quantile is where a sample is divided into equal-sized,
adjacent, subgroups.
• It also refer to dividing a probability distribution into areas of
equal probability.

107
107
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian Distribution
Method 2: Normal Probability Plot (NPP)
What is a Quantile?

108
108
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian
Distribution

Method 2: Normal Probability Plot (NPP)

What is a Quantile?
• Median is a quantile; median is placed in a probability
distribution so that exactly half of the data is lower than
the median and half of the data is above the median.
• The median cuts a distribution into two equal areas and
so it is sometimes called 2-quantile.

109
109
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian
Distribution
Method 2: Normal Probability Plot
(NPP)
What is a Quantile?
• Quartiles: quantiles that divide the
distribution into four equal parts.
• Percentiles: quantiles that divide a
distribution into 100 equal parts.

110
110
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian
Distribution

Method 2: Normal Probability Plot

(NPP)
What is a Quantile?
• Deciles are quantiles that divide
a distribution into 10 equal
parts.

111
111
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian
Distribution

Method 2: Normal Probability Plot (NPP)

• NPP: determine if small sets of data come from a
normal distribution.
• This involves using the probability properties of the
normal distribution.
• The more nearly the data plots as a straight line and
the more nearly the mean corresponds to the 50%
point, the better the fit to normal distribution 112
112
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian
Distribution
•

113
113
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian Distribution
Method 2: Normal Probability Plot (NPP)
Example: Suppose we have ten samples for some
measurement: 100, 98, 101, 93, 123, 112, 85, 76,
119, 111
Determine if the data follow a normal distribution or
not?

114
114
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian Distribution
Method 2: Normal Probability Plot (NPP)
1. Sort data in ascending order Data Sorted Data
100 76
98 85
101 93
93 98
123 100
112 101
85 111
76 112
119 119
111 123

115
115
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian Distribution
Method 2: Normal Probability Plot (NPP)
2. Number sorted data from 1 to n
Data Sorted Data Number
n = 10 100 76 1
98 85 2
101 93 3
93 98 4
123 100 5
112 101 6
85 111 7
76 112 8
119 119 9
111 123 10
116
116
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian Distribution
•

Data Sorted Data Number Cumulative Probability

100 76 1 0.05
98 85 2 0.15
101 93 3 0.25
93 98 4 0.35
123 100 5 0.45
112 101 6 0.55
85 111 7 0.65
76 112 8 0.75
119 119 9 0.85
111 123 10 0.95

117
117
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian Distribution
Method 2: Normal Probability
Plot (NPP) Data Sorted Data Number Cumulative Probability z Value
100 76 1 0.05 -1.64485
4. Determine z value from 98 85 2 0.15 -1.03643
standard normal 101 93 3 0.25 -0.67449
distribution for each 93 98 4 0.35 -0.38532
123 100 5 0.45 -0.12566
cumulative probability
112 101 6 0.55 0.125661
85 111 7 0.65 0.385321
76 112 8 0.75 0.67449
119 119 9 0.85 1.036433
111 123 10 0.95 1.644853

118
118
 of

ClickCHI-SQUARE
THE to edit Master
(X2title
) DISTRIBUTION
style
Goodness of Flt Based on the Gaussian
Distribution

Method 2: Normal Probability

Plot (NPP)
5. Plot sorted data versus z
values

119
119
Assignment
Click to edit Master#8
title style

31, 32, 33, 34, 35

120
120
ClickAND
BIAS to edit
SINGLE-SAMPLE
Master title style
UNCERTAINTY

• The only direct method for uncovering bias error

in a measurement is by comparison with
measurements made using a separate more
accurate apparatus
• A second set of apparatus is seldom used due
to cost and time constraints.

121
121
ClickAND
BIAS to edit
SINGLE-SAMPLE
Master title style
UNCERTAINTY

• Rely on knowledge of our own equipment to

make estimates for likely sizes of bias errors.
• Estimation of bias uncertainly relies on
experience and understanding of calibration
accuracy and dimensional tolerances.
• Diligence, persistence, and careful examination
of one’s results are essential in identifying and
eliminating such errors.

122
122
ClickAND
BIAS to edit
SINGLE-SAMPLE
Master title style
UNCERTAINTY

• Estimates of bias uncertainty should be

accompanied by a confidence level.
• Level of confidence assigned is a product of our
knowledge of the system, reflecting our
assessment of the fraction of bias errors likely to
land within the uncertainty interval.
• Sometimes the term coverage is used in place of
confidence to reflect the empirical nature of these
estimates in contrast to those derived using
statistical methods.
123
123
ClickAND
BIAS to edit
SINGLE-SAMPLE
Master title style
UNCERTAINTY

• Data reduction often requires knowledge of physical

properties, system dimensions, or electrical
characteristics.
• For example, a flowmeter measurement may depend on
density of water and tube diameter.
• Differences between the assumed values of these
components and their actual values can systematically
shift all data taken, creating a bias error.

124
124
ClickAND
BIAS to edit
SINGLE-SAMPLE
Master title style
UNCERTAINTY

• To find water's density, temperature is specified

and a tabulated handbook value is taken, giving. for
example, 998 kg/m3.
• Inaccuracy in the specified temperature may cause
a bias uncertainty in density of 2 kg/m3, for
instance.
• This uncertainty may cover 95% of temperature
range we expect is possible for the system.
• This inaccuracy will systematically affect all data
reduced using the particular value of density.
125
125
ClickAND
BIAS to edit
SINGLE-SAMPLE
Master title style
UNCERTAINTY

• Similarly, nominal diameter of a pipe may differ from

production diameter by a percent or so
• Potential sources of error such as these remain
unchanged for each measurement made with the
system.
• If the uncertainty related to a manufacturing tolerance is
unacceptably large, taking our own measurement of the
specific part can usually reduce the uncertainty
substantially. The uncertainty of a pipe's diameter can
be measured to the accuracy of a set of calipers.
126
126
ClickAND
BIAS to edit
SINGLE-SAMPLE
Master title style
UNCERTAINTY

• Calibration uncertainty is another common source

of bias uncertainty.
• Calibration requires a reference or standard against
which system response can be compared.
• Reference may be fixed or one-valued, such as the
triple point of water or the other triple points and
melting points used to define the practical
temperature scale.

127
127
ClickAND
BIAS to edit
SINGLE-SAMPLE
Master title style
UNCERTAINTY

• Alternatively, the standard may be capable of

supplying a range of inputs comparable to the range
of the system, as do various commercially available
voltage references.
• Uncertainty of the standard should be considerably
less than that of the system being calibrated.
• A rule of thumb: uncertainty of standard should be no
more than one-tenth that of the system being
calibrated.

128
128
BIAS AND
Click SINGLE-SAMPLE
to edit UNCERTAINTY
Master title style

• Figure 17: Block diagram showing calibration procedure

• The indicated readout is compared to the reference standard and
a relation between the two is determined.
• Sometimes readout can be adjusted until agreement with the
standard is obtained; sometimes uncertainty appears in the
comparison and adjustment process.

129
129
BIAS AND
Click SINGLE-SAMPLE
to edit UNCERTAINTY
Master title style

Instrument manufacturers supply calibration

data with their products, which can assist in
estimating the uncertainty of the instrument.

130
130
BIAS AND
Click SINGLE-SAMPLE
to edit UNCERTAINTY
Master title style
Single-Sample Precision Uncertainty

• When only one, two, or three repeat observations

are made, the confidence intervals calculated
statistically can be quite large.
• In that circumstance, you may determine a
narrower range for the mean value by treating
precision errors like bias errors and estimating a
standard deviation based on your knowledge of
the instruments.
131
131
BIAS AND
Click SINGLE-SAMPLE
to edit UNCERTAINTY
Master title style
Single-Sample Precision Uncertainty

• Begin by estimating s as a value that is based on our

knowledge of the experimental system.
• Thus, we are assuming that s of the population is known,
and, with Eq. (18), the precision uncertainty can be
estimated as

132
132
BIAS
ClickAND SINGLE-SAMPLE
to edit UNCERTAINTY
Master title style
Single-Sample Precision Uncertainty

• With a single reading, n = 1, so that no averaging is

performed to reduce the uncertainty. Taking a 95%
confidence level, Z0.95/2 = 1.96 and

133
133
BIAS
ClickAND SINGLE-SAMPLE
to edit UNCERTAINTY
Master title style
Single-Sample Precision Uncertainty
• The potential precision error underlying a
single measurement of a random variable can
usually be estimated from your knowledge of:
 how finely an instrument will resolve
 how precisely an instrument may repeat a
reading
 how much the test condition fluctuates
• Often, these estimates can be made in
advance of performing the experiment, in order
to gauge the expected uncertainly in the result
134
134
Click to edit Master
PROPAGATION OF UNCERTAINTY
title style

• Often several quantities are measured, and the

results of those measurements are used to
calculate a desired result.
• For example, experimental values of density
are usually determined by dividing the
measured mass of a sample by the measured
volume of that sample.
• Each measurement includes some uncertainty,
and these uncertainties will create an
uncertainly in the calculated result.
135
135
Click to edit Master
PROPAGATION OF UNCERTAINTY
title style
y

• Finding uncertainty in a result due to uncertainties

in the independent variables is called finding the
propagation of uncertainty.
• For uncertainties in the independent variables, the
procedure rests on a statistical theorem that is
exact for a linear function of several independent
variables xi with standard deviations si ; the
theorem states that the standard deviation of y is

136
136
Click to edit Master
PROPAGATION OF UNCERTAINTY
title style
y

• A calculated result y is a function of several independent

measured variables (x1, x2, …, xn)
• Each measured value has some uncertainty (u1,u2, …, un)

137
137
Click to edit Master
PROPAGATION OF UNCERTAINTY
title style
y

Example 12: Suppose that y has the form: Y = A x1 + B x2

and that the uncertainties in x1 and x2 are known with
odds of n:1. What is the uncertainty in y?

138
138
Click to edit Master
PROPAGATION OF UNCERTAINTY
title style
y

Example 13: Suppose that y has the form

and that the uncertainties in x1, x2 and x3 are known

with odds of n:1. What is the uncertainty in y?

139
139
Click to edit
EXAMPLE 15 Master title style
Obstruction meters: ventures and orifice plates are
used to measure steady flow rates of fluids.

140
140
Click to edit
EXAMPLE 15 Master title style
Obstruction meters: ventures and orifice plates are
used to measure steady flow rates of fluids.

141
141
Click to edit
EXAMPLE 15 Master title style

• Tables of empirical calibration coefficients, K:

• This provides a means of measuring flow rate in terms of

pressure drop across the obstruction.
• Published coefficients will yield approximate flow rates
• Accurate measurement requires careful experimental
determination of the coefficient for each specific installation.

142
142
Click to edit
EXAMPLE 15 Master title style

A proposed
arrangement for
calibration of a
thin-plate orifice
meter

143
143
Click to edit
EXAMPLE 15 Master title style

• Calibration consists of experimentally

determining coefficient K by collecting flowing
fluid in a weigh tank for some period of time.
• Record Pressure difference, DP = P2 – P1
• Flow rate, Q = measured weight, W, divided by
liquid density (r) and elapsed time (t) = W/rt
• Area, A2 = Pi D2 /4
• D = orifice diameter

144
144
Click to edit
EXAMPLE 15 Master title style

• Substituting these values into Eq. (38), we

may solve for the calibration constant

• Experimental value of K is obtained by inserting

observed values of W, and DP, along with a
measured value of D and tabulated data for r.

145
145
Click to edit
EXAMPLE 15 Master title style

Before undertaking the experiment, we'd like to

estimate the expected accuracy of the result to
identify those parts of the experiment that
contribute most of uncertainty and, if necessary,
improve them.

146
146
Click to edit
EXAMPLE 15 Master title style
Solution

• Both bias and precision error should be considered.

• Since we have no samples for statistical analysis, we can only
estimate the expected size of potential precision errors, using
estimates for the standard deviations.
• Single-sample uncertainty estimate.
• First estimate single-sample precision and bias uncertainly in each
measured variable, then propagate these uncertainties into K
147
147
Click to edit
EXAMPLE 15 Master
- Weight
title style

• Weight, W, is measured with a platform scale.

• What bias uncertainly exists in the scale measurement?
• Has the scale's calibration been checked? How recently
and against what standard?
• Presumably, we have made some sort of check, at a
minimum several point calibrations using reliable proof
weights. If not, the user's manual should include such
data.
• Let us say that, in our judgment, an uncertainty of ±1%
is justified, with a confidence of about 95% 148
148
 t,

Click to edit
EXAMPLE
th
e 15 Master
- Weight
title style

• Bias uncertainty is dependent to some degree

on magnitude of weight measured relative to
scale range; this effect could be taken into
account if warranted.
• Precision uncertainty in weight measurement
will be caused by:
 reading error in the scale
 Hysteresis
 Friction
 backlash in the scale mechanism 149
149
Click to edit
EXAMPLE 15 Master
- Weight
title style

• Size, or standard deviation, of the reading error will

depend on how finely the scale is graduated; perhaps
0.1% error covers one standard deviation in the scale
reading (so that ± 0.1 covers 68% of the reading).
• Hysteresis, friction, and backlash should be negligible if
the scale is in good condition; however, if these effects
are observed, their errors should also be included

150
150
Click to edit
EXAMPLE 15 Master
- Diameter
title style

• The diameter orifice must be determined.

• Sharp-edged orifice that we have
checked it with an inside micrometer.
• Did we check the micrometer against
gauge blocks, or are we accepting its
scale as is?
• How experienced are we in using a
micrometer?

151
151
Click to edit
EXAMPLE 15 Master
- Diameter
title style

• We intend to measure the office only

once, using this value in all
subsequent applications of the
meter, so the diameter's uncertainty
all appears as bias in the results.
• After these considerations. we
estimate a 95% bias uncertainty of ±
0.008 cm in the nominal 4-cm
diameter (±0.2% uncertainty).

152
152
Click to edit
EXAMPLE 15 Master
- Time title style

• How accurate is the determination of the time

period, t?
• If a hand-operated diverter is used, precisely
when did the flow start and stop? That is, how
well is the time period defined?
• If a stopwatch is used, how good is the
synchronization between the diverter and watch
actions?
• These uncertainties are essentially precision errors,
which likely lo vary from run to run.
153
153
Click to edit
EXAMPLE 15 Master
- Time title style

• If we carefully make a series of repeated runs of this

particular procedure, we could accumulate enough data
to perform a precision uncertainty analysis.
• In the present case, we simply estimate the likely standard
deviation to be ±3 s; total time period is 5 min, standard
deviation of t is ± I%.

154
154
Click to edit
EXAMPLE 15 Master
- Time title style

• Bias error in the time determination may arise if

the watch is fast or slow (probably a small error in
5 min) or if we systematically stop or start the
watch too soon.
• Without other information, we'll assume that bias
in the time reading is negligible compared to the
precision error

155
155
Click to edit
EXAMPLE 15 Master
- Density
title style

• Density will be read from a handbook table.

• Between 0°C and 38°C, density of water decreases by about 0.7%.
• Temperature may vary slightly between each experiment, leading
to a precision error if only one value of density is used.
• If we use a thermocouple to measure the correct temperature for
each experiment, precision error will still arise from the reading
error of the temperature measurement.
• If standard deviation in temperature is ±0.1°C, then corresponding
density variation is only 0.002%.
156
156
Click to edit
EXAMPLE 15 Master
- Time title style

• Bias error in temperature reading may be

higher, perhaps ±1°C, and the resulting bias
uncertainty in the density used would then be
±0.02% (at 95%).
• If we do not bother to measure the temperature,
bias uncertainty would be larger still, may be
0.2% if we just assume a standard room
temperature of 20°C.

157
157
Click to edit
EXAMPLE 15 Master
– Pressure
titleDrop
style

• Value of DP is measured using a manometer.

• Dominant uncertainty is in reading the difference
in heights of the manometer columns, essentially
a precision uncertainty resulting from reading
error.

158
158
Click to edit
EXAMPLE 15 Master
– Pressure
titleDrop
style

• This uncertainty has relatively constant size,

independent of the magnitude of DP, so that a
percentage uncertainty is a bit misleading.
• To keep uncertainty analysis simple, let us take a
representative value of ±1% for the standard
deviation in the pressure.
• Bias uncertainty in the manometer turns out to
be substantially smaller, about 0.1% at about 95%
confidence

159
159
Click to edit
EXAMPLE 15 Master title style

Summary of results

160
160
Click to edit
EXAMPLE 15 Master title style

• Apply Eq (37) to K:

161
161
Click to edit
EXAMPLE 15 Master title style

• First, calculate bias uncertainty in K:

162
162
Click to edit
EXAMPLE 15 Master title style

• The standard deviation of K:

163
163
Click to edit
EXAMPLE 15 Master title style

164
164
Click to edit
EXAMPLE 15 Master title style

• Parameters having greatest contribution to uncertainty.

• Improvement of timing and weighing procedures would
improve the results the most.
• Improvement of pressure and diameter measurements would
contribute significantly less improvement.
• Most of total uncertainty is caused by precision uncertainty.
We can reduce that uncertainty considerably by repeating the
calibration experiment several times and averaging the
results.
• Since PK will decrease as n^0.5, taking n = 4 experiments will
reduce PK to about 1.0%
165
165
Click to edit
EXAMPLE 15 Master title style

• Density contributes almost nothing to total

uncertainty.
• Even if we don't bother to measure temperature,
the contribution of density uncertainty remains
negligible; that is, (0.5x0.2)2 << (1) 2.
• Careful temperature measurement, in this case,
would be a waste of effort.

166
166
Click to edit
EXAMPLE 15 Master title style

• Are the various uncertainty estimates simply good guesses?

• To a degree, they are, but dismissing them as nothing but
guesses would be flippant. The specific considerations
leading to each estimate were not arbitrary; when properly
made, such "guesses" have a strong foundation in the
actual performance of the equipment and the method of
taking the data.
• Even the estimated confidence percentages (usually are a
quantitative assessment of our expectation for the variability
of the data, although they are essentially just educated
guesses. 167
167
Click to edit
EXAMPLE 15 Master title style

• But if we admit to guessing, why not simply

guess the overall uncertainty and skip all the
intermediate steps?
• Detailed analysis provides a means for
evaluating relative effect of each identifiable
source of error, thereby separating the more
important ones from the less important ones.
• Furthermore, one can evaluate uncertainties of
each of the individual variables with considerably
more assurance than one could judge the total
168
168
Assignment # 9
Click to edit Master title style

9, 10, 11, 12, 13,16,

17, 25

169
169
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
General Rules for Making Graphs
1. Independent variable goes on horizontal (X) axis.
2. Dependent variable goes on vertical (Y) axis.
3. Label units on each axis
4. Lines must be smooth and solid and should not be thicker than the
points they’re connecting.
5. Use an appropriate scale for each axis
6. Graphs don’t have to start at zero if your data doesn’t contain zero
7. All data must fit on the graph scale.
8. A key must be provided if there is more than one independent
variable
9. The title must include both variables.
170
170
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Chart Types
1. Column / Bar Graph

171
171
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Chart Types
2. Line Graph

172
172
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Chart Types
3. Pie Graph

173
173
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Chart Types
4. Bar Graph

174
174
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Chart Types
5. Area Graph

175
175
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Chart Types
6. X y Scatter Plot

176
176
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Chart Types
7. Surface Plot

177
177
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Chart Types
8. Combo Plot

178
178
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Chart Types
9. Polar Area Plot

179
179
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Types of Coordinates
1. Linear Coordinates

180
180
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Types of Coordinates
1. Semilogarthimic Coordinates

181
181
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Types of Coordinates
1. Log Log Coordinates

182
182
 ·

GRAPHICAL
Click to editPRESENTATION
Master title style
OF DATA
Straight line transformations

183
183
Assignment
Click to edit Master#title
10style

26, 27, 28
36 to 46

184
184