I was recently told that it was not appropriate to mix psychological constructs (interval scaled personality traits) with socio-demographic variables (e.g. age) as multiple predictors in a linear regression model. Why is this inappropriate?
Asked
Active
Viewed 26 times
1
-
1What is the context of that conversation? This general statement invalidates a lot of the literature. However, I can see why if the said psychological construct is highly associated with age, or is in the causal pathway between age and the dependent variable. If you can clarify further by revising the question, it'd help getting better answers. Also, did the advice-giver suggest other methods? – Penguin_Knight Feb 10 '15 at 19:54
-
The specific comment received was, "In your regression analyses you mix variables that were assessed on 5 point Likert scales with the metric age variable. This approach is not appropriate!" The comment seems to indicate that the type of measurement is why the variables should not be estimated as predictors in the same model. The variables are personality traits and age, which I would not expect to be necessarily related. – Laura Feb 10 '15 at 20:00
-
Wow, in that case I am stumped. I have never heard of this rule. Hopefully someone here can help. :) – Penguin_Knight Feb 10 '15 at 20:01
1 Answers
0
You say you got the comment:
In your regression analyses you mix variables that were assessed on 5 point Likert scales with the metric age variable. This approach is not appropriate!
... and then no further reason. I simply do not understand where this comes from, and you should ask for a reason!
They might simply mean that using an ordinal variable (Likert) as if numerical is inappropriate, but that should not depend on the further use of a metric variable like age. Maybe see
kjetil b halvorsen
- 77,844