# lect9 - IE410 Design of Experiments Notes for Lecture 9...

This preview shows pages 1–12. Sign up to view the full content.

IE410 Design of Experiments Notes for Lecture 9: Remedial Measures What do we do if the assumptions are not met? Depending on which assumptions are violated, one has a few different options.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Often, a single cause (reason for the assumptions not being valid) may show up in several diagnostics. For example, non-constant variance often results in the residuals appearing non- Normal as well. Thus one remedial measure may cure more than one problem.
Transforms Variance stabilizing transforms are probably the most useful remedial measure, perhaps because non-constant variance is also quite common, indeed it is often quite natural. The most common phenomenon is that the variance increases with the mean.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Example 1. Use a 50 foot long tape measure to measure: i) something 1 inch long n=5 times ii) something 49 feet long n=5 times From each sample of n=5, calculate the sample variance
Would you not expect σ to be bigger for the 49 foot measure than for the 1 inch measure? (If not, then remember that tapes can stretch).

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Example 2 . We are testing two paper airplanes. Each plane is flown n=5 times. We measure Y= time aloft Suppose plane one averages 5 seconds aloft, while plane 2 averages 5 minutes aloft. Over the 5 runs, which do you expect will exhibit more variation?
A Common Occurrence: We have in general σ Y = f( μ Y ) in particular, suppose σ Y μ Y α Now consider transforming the data as follows:

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Y = original data Y * = transformed data using the transform: Y * = Y λ for λ 0 Y * = log(Y) for λ = 0 We see soon that: σ Y μ Y λ + α -1
Thus setting : λ = 1 - α will yield Y * with constant variance.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Mathematical Details Suppose σ Y 2 = g( μ Y ) and we propose the transform Y * = H(Y) Do a first order Taylor series expansion about the point Y= μ :
Y * = H( μ ) + (Y- μ )H'( μ ) where H'( μ ) = d/dy H(Y) evaluated at Y= μ So V(Y * ) = V[H( μ )]+V[(Y- μ )H'( μ )] + Rem = 0 + [H'( μ )] 2 V(Y)

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### Page1 / 48

lect9 - IE410 Design of Experiments Notes for Lecture 9...

This preview shows document pages 1 - 12. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online