turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Differences between Wald test vs. score test for T...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

08-26-2015 11:40 PM

My logistic regression model has a binomial response and 3 categorical predictors, A, B and C.

- A is binary
- B is ternary
- C is ternary

The observations are clustered under a factor R. There are 50 clusters based on R. Because of this clustering structure, I am using PROC GENMOD in SAS with the REPEATED statement. Thus, I am using generalized estimating equations to estimate the regression coefficients.

When I ask for the Type-3 analysis in the MODEL statement, I can use the default score test or the Wald test. Unfortunately, they give very different P-values.

Here are the P-values from the score test.

- A 0.2625
- B 0.0481
- C 0.0823

Here are the P-values from the Wald test.

- A 0.1705
- B <.0001
- C 0.0338

I want to determine if B and C are significant predictors of the response, but these contrasting Type-3 results confuse me.

My questions:

1) Which test should I use to evaluate the significance of the predictors?

2) Could these contrasting results be a sign that the model is invalid for my data set? If so, are there other ways to check the validity of my model?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

08-27-2015 07:34 AM

According to documentation. Wald test is based on Normal distribution approximation ( which usually unbias), while Score test is based on Likelihood (bias).

When you have a big data, both of them are going to have the same value .but if your data is small , they are going to be different .