I ran a regression model with a change score (post minus pre manipulation) predicting my dependent variable. Initially, I thought it might be also interesting to control for the baseline score (pre manipulation) and noticed then a stronger effect for the change score (though it was also visible when not controlling for baseline).

I read some discussions and papers that raise some concerns about this procedure, stating that this might lead to bias in the estimated effect of the change (some say it can inflate effects, not sure if it could also reduce it), but all arguments I read were mostly about designs with different groups of subjects that were compared (between-subjects design, e.g. treatment vs. control group). I wonder whether adjusting for baseline is also problematic in the analysis of change scores in a within-subject design in which all subjects undergo the same treatment/manipulation (but this may be the same situation since it is again about controlling for between-subject variability)?

And if it can be problematic, is anyone aware of situations where it still could be meaningful to control for baseline scores when dealing with change scores (since at least intuitively such a model would be interesting)? And are you aware of some literature in this regard?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.