1
$\begingroup$

Please let me know of your opinions on the following:

A generalized least square model was used to compare a variable in 2 groups of subjects. Two covariates were included in the model. One of them is a continuous variable (age) and the other is categorical (sex). However there is no overlap in sex between the 2 groups of subjects.

Do you think the covariate (sex) adjustment is still valid with no overlap?

$\endgroup$
3
  • $\begingroup$ Does "no overlapping" mean that one group is all male and the other one is all female? $\endgroup$
    – Pere
    Jul 7, 2017 at 20:15
  • $\begingroup$ The meaning is unclear. Please edit your post to clarify. This can either be in words or with data. Do you mean, for example, that the ages do not overlap between the different sexes? $\endgroup$
    – Carl
    Jul 7, 2017 at 22:26
  • $\begingroup$ @james 1. Please don't use answers for comments. 2. Please merge your accounts so that you're commenting from the account you posted the question from; then you can comment. $\endgroup$
    – Glen_b
    Jul 8, 2017 at 0:59

1 Answer 1

1
$\begingroup$

You're going to run into a problem known as complete separation if you leave sex in the model, as it perfectly predicts the outcome.

$\endgroup$
3
  • 1
    $\begingroup$ Thanks for your answer. What if we include sex as a dummy variable and use ordinary least square (OLS) regression? I have seen people use the dummy varibales in social sciences. Is that going to help with the complete separation problem that you mentioned? $\endgroup$
    – james
    Jul 7, 2017 at 18:12
  • 1
    $\begingroup$ You may be able to make sense of the contrasts between groups if you make a heavy no-interaction assumption. $\endgroup$ Jul 7, 2017 at 19:08
  • 1
    $\begingroup$ @james 1. Please don't use answers for comments. 2. Please merge your accounts so that you're commenting from the account you posted the question from; then you can comment. $\endgroup$
    – Glen_b
    Jul 8, 2017 at 0:59

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