Scribd
Upload
3 views

Module-5-2015

The document discusses the application of statistical methods for quality assurance, focusing on measurement error and one-sample inference. It outlines the use of confidence interval formulas for estimating the mean and standard deviation of measurements, emphasizing the importance of understanding measurement error and bias. Examples are provided to illustrate how to apply these concepts in practical scenarios involving multiple measurements and different batches.

Uploaded by

Dame Yenyetou
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
3 views

Module-5-2015

The document discusses the application of statistical methods for quality assurance, focusing on measurement error and one-sample inference. It outlines the use of confidence interval formulas for estimating the mean and standard deviation of measurements, emphasizing the importance of understanding measurement error and bias. Examples are provided to illustrate how to apply these concepts in practical scenarios involving multiple measurements and different batches.

Uploaded by

Dame Yenyetou
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 8

IE 361 Module 5

Measurement and One Sample Inferences Part 1

Reading: Section 2.2.1 Statistical Methods for Quality Assurance

ISU and Analytics Iowa LLC

(ISU and Analytics Iowa LLC) IE 361 Module 5 1/8


Measurement Error and One Sample Inference
"Ordinary" con…dence interval formulas (taught in Stat 231 or any other
basic statistics course) based on a model for y1 , y2 , . . . , yn of sampling
from a normal distribution with mean µ and standard deviation σ, are
s
y tp for estimating µ
n

and s s !
n 1 n 1
s ,s for estimating σ
χ2upper χ2lower

(It is your responsibility to review these on your own if they do not look
absolutely familiar.) The "news" here is that sources of physical variation
(and in particular, sources of measurement error and item-to-item
variation) and data collection plans interact to give practical meaning to
"µ" and "σ" and to govern what of practical importance can be learned
from the application of these formulas.
(ISU and Analytics Iowa LLC) IE 361 Module 5 2/8
Measurement Error and One Sample Inference
Let us consider what these elementary inference formulas and the data
collection of the previous module, beginning with the situation of repeat
measurements on a …xed measurand as again pictured below.

In this context, if n repeat measurements of a single measurand,


y1 , y2 , . . . , yn , have sample mean y and sample standard deviation s
applying the t con…dence interval for a mean, one gets inference for
x + δ = measurand plus bias
(ISU and Analytics Iowa LLC) IE 361 Module 5 3/8
Measurement Error and One Sample Inference

So
in the event that the measurement device is known to be
p is sure that δ = 0, there is no systematic error),
well-calibrated (one
the limits y ts/ n based on ν = n 1 df are limits for x
in the event that what is being measured is a standard for which x is
known, one may use the limits
s
(y x) tp
n

(once again based on ν = n 1 df) to estimate the device bias, δ


applying the χ2
con…dence interval for a standard deviation, one has
an inference for the size of the device "noise," σdevice

(ISU and Analytics Iowa LLC) IE 361 Module 5 4/8


Measurement Error and One Sample Inference
Next consider what can be inferred from single measurements made on
n di¤erent measurands y1 , y2 , . . . , yn from a stable process with sample
mean y and sample standard deviation s. as pictured below.

p
The limits y ts/ n (for t based on n 1 degrees of freedom) are
limits for
µx + δ = the mean of the distribution of true values plus bias
(ISU and Analytics Iowa LLC) IE 361 Module 5 5/8
Measurement Error and One Sample Inference
q
The quantity s estimates σy = σ2x + σ2device , that really isn’t of
fundamental interest. So there is little point in direct application
of the χ2 con…dence interval for a standard deviation in this context.

Note in passing (we’ll say more about this later) that since
q q
σx = (σx + σdevice ) σdevice = σ2y σ2device
2 2 2

estimates of item-to-item variation (free of measurement noise) might be


based on estimates of σdevice (based on data like those on panel 3) and σy
(based on data like those on panel 5). See display (2.3), page 20 of
SQAME in this regard.

Example Below are m = 5 "measurements" made by a single analyst on


a single physical sample of material using a particular assay machine.
1.0025, .9820, 1.0105, 1.0110, .9960
(ISU and Analytics Iowa LLC) IE 361 Module 5 6/8
Measurement Error and One Sample Inference

These have mean y = 1.0004 and s = .0120. Consulting a χ2 table using


ν = 5 1 = 4 df, we can …nd a 95% con…dence interval for σdevice
r r !
4 4
.0120 , .0120 i.e. (.0072, .0345)
11.143 .484

(One moral here is that ordinary small sample sizes give very wide
con…dence limits for a standard deviation.) Consulting a t table also using
4 df, we can …nd 95% con…dence limits for the true value for the specimen
measurand plus instrument bias (x + δ)

.0120
1.0004 2.776 p i.e. 1.0004 .0149
5

(ISU and Analytics Iowa LLC) IE 361 Module 5 7/8


Measurement Error and One Sample Inference

Suppose that subsequently, samples from n = 20 di¤erent batches are


analyzed and y = .9954 and sy = .0300. The t con…dence interval

.0300
.9954 2.093 p i.e. .9954 .0140
20
is for µx + δ, the process mean plus any assay machine bias/systematic
error.

(ISU and Analytics Iowa LLC) IE 361 Module 5 8/8

You might also like