© 1996 and 2002
P. J. Ludovice
Experimental Statistics
P. J. Ludovice
School of Chemical & Biomolecular Engineering
Georgia Institute of Technology
Atlanta, Georgia 303320100
Contents
Introduction
1
Statistical Inference & the Normal Distribution
4
Confidence Limits or Confidence Intervals
8
tTests for Differences of Means
12
Generalized Least Squares Regression
14
Propagation of Error
18
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
© 1996 and 2002
P. J. Ludovice
1
Introduction
The purpose of this guide to experimental statistics is to familiarize the student with a few useful
techniques in experimental statistics that will assist in the analysis of experimental data.
After
introducing the critically important concept distributions we will discuss the concept of statistical
inference for analyzing data.
This concept is used to infer something about the distribution of all
possible data (population) from a single small set (sample) of data.
We will learn how to place error
bars on data as well as determine of two measurements are statistically different.
Next we carry out
correlation to see if the behavior of one variable of interest changes in a fashion similar to another
variable (correlates).
Since many of the parameters of interest to us come from fitted parameters as a
result of some correlation we will also learn how to put an error bar on these fitted parameters.
Finally
we will learn how to propagate error through a calculation to determine the error on a derived parameter
that utilizes a measured parameter and its associated error.
In general the goal of most experiments is to summarize the data in single quantity and then associate
some statistical error with this quantity.
These summary quantities are typically called measures of
central tendency and the include averages and fitted parameters of from correlations.
The most
commonly used measure of central tendency is the arithmetic mean (
X ), which is the average of various
X values or all N values
N
X
X
N
i
i
∑
=
=
1
.
[1]
Because this measure of central tendency is derived from several points, there is a spread of the data that
is also associated with this measure.
The spread or width of the data describes the precision of the data
and is called the dispersion.
The most commonly used measure of dispersion is the standard deviation
(
σ
) which measures the root mean square deviation from the mean of a population (
µ
) averaged over all
the data in the population (N
p
)
p
N
j
j
N
X
p
∑
=
−
=
1
2
)
(
µ
σ
.
[2]
Note that that equation 2 describes the standard deviation for an entire population or a complete set of
elements.
In science and engineering we rarely have this complete set or entire population so we often
use a sample or subset of the population.
When the entire population is accessible then it is preferable to
use this, however often this is not the case.
For example if one wishes to take data on an entire school
this is usually accessible, while accessing the population of an entire country is often cost prohibitive
(this is why the U.S. census costs so much).
This is the end of the preview.
Sign up
to
access the rest of the document.
 Spring '07
 Gallivan
 Normal Distribution, Mole, Peter J. Ludovice

Click to edit the document details