- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Posted 11-06-2008 05:41 AM
(796 views)
Hi, I'm tryng to fit a regression model and I have several subjects by tratment. In order to reduce variation I want to try using the average subject value inside a treatment.
There is a way to do this in PROC REG or I have to process data to create a new variable with the avg value? Thanks
There is a way to do this in PROC REG or I have to process data to create a new variable with the avg value? Thanks
2 REPLIES 2
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
The short answer is "no", you have to do it outside of REG.
However, imputing the mean for all observations in a treatment group will result in no residual variability if treatment is a predictor in the model.
There are problems even if the treatment isn't in the model. Though the mean estimates are unbaised, the variability of the imputations is too small (that's the variance reduction you referenced), so the estimated precision of regression coefficients will be wrong and inferences will be misleading.
However, imputing the mean for all observations in a treatment group will result in no residual variability if treatment is a predictor in the model.
There are problems even if the treatment isn't in the model. Though the mean estimates are unbaised, the variability of the imputations is too small (that's the variance reduction you referenced), so the estimated precision of regression coefficients will be wrong and inferences will be misleading.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Ok, I'll try to improve the model with all data. Thanks a lot