Homework 04
Assigned: 19 September 2005
Due: 26 September 2005
C:\baileraj\Classes\Fall 2004\sta402\hw\Homework-04-19sep05.doc
1. Repeat the regression analysis of the Manatee – Boat registration data
a. Fit the model MANATEE = 00 +1 NBOATSi + i
b. Fit the model MANATEE = 01 +1 [NBOATSNBOATSi-min(NBOATS)] + ] + i
c. Fit the model MANATEE = 02 +1[NBOATSNBOATSi-average(NBOATS)] + ] + i
Prepare a table summarizing the estimated regression coefficients, standard errors, P-values for
these three models. Hint: use ODS output to manipulate the coefficient information.
2. Refer to the 4 group (packaging method)- log-bacterial growth study. In this exercise, you will
compare the one-way anova model to regression with coding.
a. Fit the anova model and obtain the estimated effects (estimates of , 1, 2, 3, 4) using
SOLUTION option to the GLM model statement.
proc glm data=meat order=data;
title2 fitting the one-way anova model via GLM;
class condition;
model logcount = condition/solution;
means condition;
run;
b. Define 4 indicator variables, one for each of the conditions. For example, let ICO2 = 1 if
condition=”CO2” and ICO2=0 otherwise. Define the other 3 indicator variables similarly.
c. Can you fit a regression model with all indicator variables included? If so, how?
d. Fit 4 different regression models with 3 of the 4 indicator variables. Which fit matches the
solution from part a? What does this tell you about the coding in the model fit from part 2a?
model logcount = IPlastic IVacuum IMixed;
model logcount = IPlastic IVacuum ICO2;
model logcount = IPlastic IMixed ICO2;
model logcount = IVacuum IMixed ICO2;
3. Fit a regression model with a linear and a quadratic term to the nitrofen data using the square
root of “total” as the response variable. As an aside, the square root transformation is a
normalizing transformation for Poisson data and is often performed prior to fitting models
that assume normal error terms. By the way, an alternative that might be even better than
ordinary regression with square root transformed counts is the generalized linear model with
