1
$\begingroup$

Can I use a difference score as the dependent variable and use the baseline score as the independent variable when doing correlation, linear regression and multivariate linear regression? Thank you.

$\endgroup$
2
  • $\begingroup$ Are you asking whether you can use the baseline score to predict the difference in scores? I don't know what type of data and domain you're working with -- but just based on the information you've provided in the question, I don't see any problems in doing this. $\endgroup$
    – Vishal
    Apr 5, 2016 at 19:39
  • $\begingroup$ Thank you. yes , I would like to use the baseline score to predict the difference in score. The score is continuous data. actually I am using the functional level score for my patient to predict the improvement of their functional score in the hospital. $\endgroup$
    – Harry
    Apr 6, 2016 at 16:57

2 Answers 2

1
$\begingroup$

If the variables are measured less than perfectly, this is a bad idea because you will be predicting the error of the baseline measure. E.g. suppose person 1 had a "good day" at BL and person B had a "bad day". Then, even if neither changes, it will look like they did.

You should use a multilevel model.

$\endgroup$
6
  • $\begingroup$ thank you, yes you are right What do you mean by multilevel model?? $\endgroup$
    – Harry
    Apr 6, 2016 at 16:58
  • $\begingroup$ A multilevel model is one way of analyzing data that is not independent. $\endgroup$
    – Peter Flom
    Apr 6, 2016 at 19:12
  • $\begingroup$ @PeterFlom, but isn't that true for all regression models? We assume that error in measurement is present only in the $y$ but ignore the error in measurement for all $x$'s? $\endgroup$
    – Vishal
    Apr 6, 2016 at 19:50
  • $\begingroup$ Not to the same extent. $\endgroup$
    – Peter Flom
    Apr 6, 2016 at 22:10
  • $\begingroup$ Should I still run the multivariate linear regression using the baseline score and other independent that are significant predictors? $\endgroup$
    – Harry
    Apr 9, 2016 at 18:22
1
$\begingroup$

Given you have the same thing measured twice you could either do

lm(y2 ~ y1 + x, data = df)

or treat it as multilevel/repeated measures to account for the nesting of the two occasions within subjects as @Peter Flom noted)

m1 = nlme::lme(y ~ x, random = ~ 1|SID, data = df)

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.