You are on page 1of 1

Stat 502

Homework 8
Assigned 11/29/10
Due 12/06/10

Germination: The dataset “ag.dat” comes from an agricultural experiment carried out in Rotham-
sted, UK. The factors are high and low levels of water, fertilizer, planting density, and herbicide.
The response is % of seeds germinated.

1. Write out the columns of +’s and −’s corresponding all two way, three way, and four way
interactions (tedious, I know). Pick any two columns and show they are orthogonal.
P P P
2. Note that if Y1 , . . . , Yn are independent, then Cov[ ai Yi , bi Yi ] = ai bi Var[Yi ]. Assuming
variances are constant across treatment assignments and equal to σ 2 , use this result to find
the covariance of the estimated main effect of water and the estimated main effect of fertilizer.
What is the covariance between any two estimated effects?

3. Get estimates for all factors and interactions (i.e. 15 estimated effects). Using a qqnorm plot,
determine which if any of the effects or interactions seem significant.

4. Now assume all 3rd order and higher interactions are negligible and use them to form an
estimate of error (i.e., fit an ANOVA which includes up to second-order interactions in the
model). Based on the ANOVA, determine if any effects are significant.

5. Now assume all 2nd order and higher interactions are negligible (i.e., fit an ANOVA which
includes main effects only). Based on the ANOVA, determine if any effects are significant.
Observe the change in MSE and error degrees of freedom in going from (b) to (c).

6. Give a suggestion as to what variables to increase or decrease in order to increase % of seeds


germinated.

Hydrocarbon: The dataset “hydro.dat” is from an experiment trying to determine the amount
of unburned hydrocarbon in engine exhaust. Read in the data using the “header=T” option in
“read.table”. The four factors are A: carburetor tension (low -, high +), B: air gap (narrow -, open
+), C: aperture size (small -, large +), D: rate of gas flow (rapid -, slow +). The response y is
amount of unburned hydrocarbons.

1. What kind of experimental design is this?

2. What is the generator of the design? Write out the confounding pattern for this design, i.e.
write out 24 effects and the aliasing relations between them.

3. Estimate the seven distinct contrasts, and determine significance with a qqnorm plot. The
goal is to find levels of the variables that give low amounts of unburned hydrocarbons. What
would you recommend for a future experiment?

You might also like