Consumer Research • Choice Models • Additional Examples of Logistic Regression Using the Choice Platform
Previous
 • 
Next
Additional Examples of Logistic Regression Using the Choice Platform
Example of Logistic Regression Using the Choice Platform
Use the Choice Platform
1.
Select Help > Sample Data Library and open Lung Cancer Responses.jmp.
Notice this data table has only one column (Lung Cancer) with two rows (Cancer and NoCancer).
2.
Select Analyze > Consumer Research > Choice > Select Data Table > Lung Cancer Responses.jmp > OK.
3.
Select Lung Cancer as the Profile ID and Add Lung Cancer as the model effect. The Profile Data window is shown in Profile Data for Lung Cancer Example.
Profile Data for Lung Cancer Example
4.
Click the disclosure icon for Response Data > Select Data Table > Other > OK.
5.
Open the sample data set Lung Cancer Choice.jmp.
6.
Select Lung Cancer for Profile ID Chosen, Choice1 and Choice2 for Profile ID Choices, and Count for Freq. The Response Data launch window is shown in Response Data for Lung Cancer Example.
Response Data for Lung Cancer Example
7.
Click the disclosure icon for Subject Data > Select Data Table > Lung Cancer Choice.jmp > OK.
8.
Add Smoker as the model effect. The Subject Data launch window is shown in Subject Data for Lung Cancer Example.
Subject Data for Lung Cancer Example
9.
Uncheck Firth Bias-adjusted Estimates and Run Model.
Choice Modeling results are shown in Choice Modeling Logistic Regression Results for the Cancer Data.
Choice Modeling Logistic Regression Results for the Cancer Data
Use the Fit Model Platform
1.
Select Help > Sample Data Library and open Lung Cancer.jmp.
2.
Select Analyze > Fit Model.
Automatic specification of the columns is: Lung Cancer for Y, Count for Freq, and Smoker for Add under Construct Model Effects. The Nominal Logistic personality is automatically selected.
3.
Click Run.
The nominal logistic fit for the data is shown in Fit Model Nominal Logistic Regression Results for the Cancer Data.
Fit Model Nominal Logistic Regression Results for the Cancer Data
Notice that the likelihood ratio chi-square test for Smoker*Lung Cancer in the Choice model matches the likelihood ratio chi-square test for Smoker in the Logistic model. The reports shown in Choice Modeling Logistic Regression Results for the Cancer Data and Fit Model Nominal Logistic Regression Results for the Cancer Data support the conclusion that smoking has a strong effect on developing lung cancer. See the Logistic Regression chapter in the Fitting Linear Models book for more details.
Example of Logistic Regression for Matched Case-Control Studies
This section provides an example using the Choice platform to perform logistic regression on the results of a study of endometrial cancer with 63 matched pairs. The data are from the Los Angeles Study of the Endometrial Cancer Data in Breslow and Day (1980) and the SAS/STAT(R) 9.2 User's Guide, Second Edition (2006). The goal of the case-control analysis was to determine the relative risk for gallbladder disease, controlling for the effect of hypertension. The Outcome of 1 indicates the presence of endometrial cancer, and 0 indicates the control. Gallbladder and Hypertension data indicators are also 0 or 1.
1.
Select Help > Sample Data Library and open Endometrial Cancer.jmp.
2.
Select Analyze > Consumer Research > Choice.
3.
Click the Select Data Table button.
4.
Select Endometrial Cancer as the profile data table. Click OK.
5.
Assign Outcome to the Profile ID role.
6.
Assign Pair to the Grouping role.
7.
Add the following columns as model effects: Gallbladder, Hypertension.
8.
Deselect the Firth Bias-Adjusted Estimates check box.
9.
Select Run Model.
A pop-up dialog window asks whether this is a one-table analysis with all the data in the Profile Table.
10.
Click Yes.
11.
On the Choice Model red triangle menu, select Utility Profiler.
The report is shown in Logistic Regression on Endometrial Cancer Data.
Logistic Regression on Endometrial Cancer Data
Likelihood Ratio tests are given for each factor. Note that Gallbladder is nearly significant at the 0.05 level (p-value = 0.0532). Use the Utility Profiler to visualize the impact of the factors on the response.