Probability and Statistics

Chapter 4

Apr. 2020
1
 Behavior of a random variable: Probability density
function => Provide exact information about its mean
value and variability.

The purpose of the measurements: Estimate the density

function based on the acquired limited data set.

Normal scatter of data about some central mean value:

Spatial distribution under nominally fixed operating
conditions + Random errors in MS & Meas.procedure.

2
 Finite number of measurements of a variable => estimating
the behavior of an entire population of values.

Estimates based on a limited data set introduce another

random error into predicting the true value of the
measurement.

Statistics is a powerful tool used to interpret and present

data

FIND: Methods to estimate the true mean value based on a

limited number of data points and the random uncertainty;
3
were presented along with treatment of data curve fitting
 4.2 STATISTICAL MEASUREMENT THEORY
From a statistical analysis of the data set an analysis of sources of
error that influence these data:

x’: The most probable estimate of based on the available data

𝑢𝑥ҧ the uncertainty interval at some probability level P%.
 Uncertainties are numbers that quantify the possible range of the
effects of errors.
 Uncertainty interval: Range about 𝑥ҧ within which we expect x'
to lie: Combines the uncertainty estimates of random &
systematic errors in the measurement of x.
4
 Probability Density Functions

5
 Histogram of the variable
 Plot the data of Table 4.1:
Consists of N individual measurements: 𝑥𝑖 , i = 1, 2, . . . , N
each measurement taken at random but under identical test
operating conditions.
 The measured values of this variable are plotted as data
points along a single axis

 The abscissa will be divided between the maximum and

minimum measured values x into K small intervals.
 𝑛𝑗 : number of times, measured value assumes a value within
an interval 𝑥 − 𝛿𝑥 ≤ 𝑥 ≤ 𝑥 + 𝛿𝑥 be plotted on the
6 ordinate.
 Histogram of the variable
 Histogram: The resulting plot of 𝑛𝑗 versus x => view both the
tendency and the probability density of a variable.
 The ordinate can be nondimensionalized as 𝑓𝑗 − 𝑛𝑗 /N
converting the histogram into a frequency distribution.
 An estimate for the number of intervals K

𝐾 = 1.87 𝑁 − 1 0.40 +1

 As N becomes very large, a value of K: 𝐾 𝑁

7 good rule that 𝑛𝑗 >= 5 for at least one interval

 Standard Statistical Distributions
Distribution Applications Density Function Shape

Normal

Lognormal

Rectangular

8
 Standard Statistical Distributions
Distribution Applications Density Function Shape

Triangular

Poisson

Binormal

9
 Histogram of the variable
 As N → ∞, the probability density function p(x), of the population
of variable x developed. In the limit as x → 0

 True mean value (central tendency) of a random variable

countinous
time

space

discrete data

10
 Histogram of the variable
 Width of the density function reflects the data variation.
For a continuous random variable, the true variance

equivalent to

for discrete data

Standard deviation, : statistical parameter = Square root of

the variance
𝜎= 𝜎2

11
 4.3 DESCRIBING THE BEHAVIOR OF A POPULATION
Normal (or gaussian) distribution

Given p(x), how can we predict the probability that any future measurement will fall ?

Probability that x will assume a value within the interval x ± x is given by the area
under p(x), which is found by integrating over the interval.

transformations

for a normal distribution,

p(x) is symmetrical about x'

12 Table 4.3
 4.3 DESCRIBING THE BEHAVIOR OF A POPULATION
Probability that the ith measured value of x will have a value between x'±x is
2 x P(z) x 100% = P%.
This is written as

13
 4.4 STATISTICS OF FINITE-SIZED DATA SETS

 N measurements of x (N repetitions), each measurement

represented by 𝑥𝑖 , i = 1, 2, . . . , N; N is a finite value.
 Finite-sized data sets provide the statistical estimates: Sample
mean value (ഥ 𝒙); Sample variance (𝒔𝟐𝒙 ), its outcome: Sample
standard deviation (𝒔𝒙 )

14
 4.4 STATISTICS OF FINITE-SIZED DATA SETS

 Deviation of 𝑥𝑖 : 𝑥𝑖 − ഥ
𝑥
ഥ provides a most probable estimate of the true mean value x'
 𝒙
ഥrepresents a probable measure of the variation found in a data
 𝒙
set.
 The degrees of freedom, : Number of data points minus the
number of previously determined statistical parameters used in
estimating that value (in the sample variance:  = N-1)

15
 4.4 STATISTICS OF FINITE-SIZED DATA SETS

 Relation between probability and ifinite statistics can be

extended to data sets of finite sample size with only some
modification.
 When data sets are finite or smaller than the population:
z variable does not provide a reliable weight estimate of the
true probability.
 Sample variance can be weighted to compensate for the
difference between the finite statistical estimates and the
statistics based on p(x):

16
 4.4 STATISTICS OF FINITE-SIZED DATA SETS

Student's t variable: t,P provides a coverage factor used for finite data sets
(replaces z)

The interval ± t,P represents a precision interval, given at probability P%, within
which one should expect any measured value to fall.

Table 4.4 Student’s t Distribution

17
 Standard Deviation of the Means
 Take N measurements of x under fixed operating conditions.
 Duplicated this procedure M times => calculate the different
ഥ & 𝒔𝟐𝒙 for each of the M data sets.
estimates 𝒙
ഥ obtained from M replications follows a normal distribution
 𝒙
defined by p(x)
 Amount of variation in 𝒙 ഥ
depend on: 𝒔𝟐𝒙 & sample size N
 Standard deviation of the means 𝑠𝑥ҧ

18
 standard deviation & standard deviation of the means

𝑠𝑥ҧ represents a measure of how well a measured mean value

represents the true mean of the population.
the true
mean value
19
 the true
mean value

 The interval ± 𝑡,𝑃 𝑆𝑥ҧ expresses the mean value, with

coverage factor t at the assigned probability, P%, expected
the true value of x to fall.
 This confidence interval is a quantfied measure of the random
error in the estimate of the true value of variable x.
 The value 𝑆𝑥ҧ represents the random standard uncertainty in the
mean value
 The value 𝑡,𝑃 𝑆𝑥ҧ represents the random uncertainty in the
mean value at P% cofidence due to variation in the measured
data set.

20
 4.5 CHI-SQUARED DISTRIBUTION ( 2 )
 Develop the concept of the standard deviation of the means as a
precision indicator in the mean value; Estimate how well 𝑠𝑥2
predicts 𝜎 2
 Plotted the sample standard deviation for many data sets, each
having N data points: Probability density function 𝑝( 2 )

For the normal distribution,

21
 Precision Interval in a Sample Variance

 Formulated by the probability statement:

a probability of
: level of significance
 Combining

 For example, the 95% precision interval by which 𝑠𝑥2 estimes

𝜎 2 is given by

this interval is bounded by the 2.5% and 97.5% 𝟐

levels of significance (for 95% coverage). Table 4.5 Values for (𝜶 )
22
 4.6 REGRESSION ANALYSIS
 A measured variable is often a function of one or more
independent variables that are controlled during the
measurement.
 When the measured variable is sampled, these variables are
controlled, to the extent possible, as are all the other
operating conditions.
 One of these variables is changed and a new sampling is made
under the new operating conditions.
 Regression analysis to establish a functional relationship
between the dependent & independent variable

23
 • The dependent variable yi,j consisting of N measurements, i = 1, 2, . . . , N
of y at each of n values of independent variable, xj ; j = 1, 2, . . . , n.
• This type of behavior is common during calibrations and in many types
measurements in which the dependent variable y is measured under controlled
values of x.
• Repeated measurements of y yield a normal distribution with variance 𝑠𝑦2 (𝑥𝑖 ), about
some mean value, 𝑦(𝑥ത 𝑗)

24
 Least-Squares Regression Analysis

 For a single variable of the form y = f(x) provides an mth order

polynomial fit of the data in the form

 For n different values of the independent variable; the highest

order, m, of the polynomial: m  n -1.
 The values of the m coefficients 𝑎0 , 𝑎1 .... 𝑎𝑚 are determined
by the analysis.
 The method of least-squares attempts to minimize the sum
of the squares of the deviations between the actual data and
the polynomial fit of a stated order by adjusting the values of
the polynomial coefficients.
25
 Least-Squares Regression Analysis

GOAL: reduce D to a minimum for a given order of polynomial.

The total differential of D is dependent on the m + 1 coefficients through

minimize the sum of the squares of the deviations, we want dD to be zero.

26
 4.7 DATA OUTLIER DETECTION
 Data that lie outside the probability of normal variation
incorrectly offset the sample mean value estimate, inflate the
random uncertainty estimates, and influence a least-squares
correlation
 Chauvenet’s criterion, which identifies outliers having less
than a 1/2N probability of occurrence.
 Let suspected outlier in a data set of N values, the
data point is a potential outlier if

 The three-sigma test: data points that lie outside the range of
99.73% probability of occurrence, 𝑥ҧ ± 𝑡,99.7 𝑠𝑥 , as potential
outliers.
27
28