Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 6

Question: Exercise 9.

Refer to Table 7.1 on high school students’ use of alcohol, cigarettes, and marijuana. Consider
the data as matched triplets.

a. Construct the marginal distribution for each substance. Specify a marginal model that can
compare them and explain how to interpret its parameters. State the hypothesis of marginal
homogeneity in terms of model parameters.

b. Specify a corresponding subject-specific model. Explain how parameter interpretation differs


from the marginal model.

c. Construct a data file to fit the marginal model. (Hint: You could mimic the Abortion data file,
ignoring gender.) Use GEE methods to fit the model and to test the hypothesis of marginal
homogeneity.

Table 7.1
Solution: Exercise 9.1

The data is shown below:

Alcohol Use Cigarette Use Marijuana Use Total

Yes No

Yes Yes 911 528 1449

No 44 456 500

No Yes 3 43 46

No 2 279 281

Total 960 1316 2276

A) The sample proportion of students who used alcohol (A) is calculated below:

1449+ 500
=0.86
2276

The sample proportion of students who used cigarettes (C) is calculated below:

1449+ 43
=0.66
2276

The sample proportion of students who used Marijuana (M) is calculated below:

960
=0.42
2276

B) It is given that t = 1, 2, 3 that refers to the A, C, and M.

Then marginal model can be written as shown below:

logit [ P ( Y t =1 ) ]= β1 z 1+ β 2 z2 + β 3 z 3

Here, z 1=1

If t =1
And, z 2=1

If t = 2

And, z 3=1

If t = 3

Consider that, e β is the odds of using alcohol (A).


1

Marginal Homogeneity hypothesis can be written as:

H 0 : β 1=β 2=β 3

C) To fit generalized estimating equations approach (Marginal Model) in SAS, use PROC

GENMOD. First, write the referred table in SAS editor window where female is denoted

by 1, Ause (Alcohol yes) is denoted by 1, Cuse (Cigarette yes) is denoted by 1, and Muse

(Marijuana yes) is denoted by 1.

Use ‘data’ keyword to assign a filename to the data file in SAS editor window. By using

“input” statement, introduce all variables and their data types. The SAS code for entering

the data is shown below:


Run

The data is analyzed for 32 cases. The binary response is Marijuana status of 32 people

who are categorized under white and other. Here, explanatory variables are race, Ause,

Cuse, gender, and Count. Now, use PROC GENMOD to implement GEE approach on

data. The ‘class’ and the ‘model’ statement will specify the model of the data. The

‘repeated’ statement invokes GEE approach and ‘subject’ must be the case numbers

because each and every observation is individual. For exchangeable working correlation

use “type = exch” and use “covb” to dislay covariance matrix. The SAS code for PROC

GENMOD is shown below:

You might also like