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
- /
- time varying covariates and fixed time outcome

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

06-16-2010 02:00 AM

Hi,

I have 3 waves of panel data. The dv is measured at wave 3. There is one covariate (continuous data) of particular interest that i would like to consider it's impact across all three waves of data. That is, rather than use dummy variables for wave1, wave 2 and wave3 separately i was wondering if it is possible in some way use the data to determine the relationship between the patterns of covariate and the outcome variable. Rather then use something like a cluster analysis (which would profoundly decrease my sample size) is it reasonable for instance to estimate the mean and slope of each subject and model this with respect to the outcome variable. Is it possible to do this within 1 model or would I need to run 2 models.

thanks in advance.

I have 3 waves of panel data. The dv is measured at wave 3. There is one covariate (continuous data) of particular interest that i would like to consider it's impact across all three waves of data. That is, rather than use dummy variables for wave1, wave 2 and wave3 separately i was wondering if it is possible in some way use the data to determine the relationship between the patterns of covariate and the outcome variable. Rather then use something like a cluster analysis (which would profoundly decrease my sample size) is it reasonable for instance to estimate the mean and slope of each subject and model this with respect to the outcome variable. Is it possible to do this within 1 model or would I need to run 2 models.

thanks in advance.

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

06-16-2010 09:16 AM

You can always string out panel data into one long observation for analysis. That way you avoid the homoscedasticity problems often associated with repeated measures analyses.

You can look at the slope in one model (in GLM) using a TEST statement and coefficients that reflect the shape that you want to test (linear, here). This will only use complete observations.

You can address the complete observation challenge (which is sounds like is substantial) with PROC MI to get a better estimate of the behavior than simple linear interpolation (which is what would happen if you used the within-subject slope).

Doc Muhlbaier

Duke

You can look at the slope in one model (in GLM) using a TEST statement and coefficients that reflect the shape that you want to test (linear, here). This will only use complete observations.

You can address the complete observation challenge (which is sounds like is substantial) with PROC MI to get a better estimate of the behavior than simple linear interpolation (which is what would happen if you used the within-subject slope).

Doc Muhlbaier

Duke