2
$\begingroup$

I'm performing a psych study on a group of single people . My binary dependent variable is their relationship status one year in the future (still single/in a relationship). My variables include their gender (male/female) and whether they have a crush on someone (yes/no). My chi square tests show no significance for either variable individually, but when I subgroup (women with a crush, men without a crush, ...), I achieve significance.

I applied logistic regression to my data, but it showed pretty much what my chi square tests showed -- no significance for the group as a whole, but significance when I subgroup. I believe that this is because my variables are not independent? Which is a requirement of logistic regression?

My question: is there a test to demonstrate that my variables are not independent? Or maybe results showing no significance for my larger group demonstrates this? If so, should I just stick with chi square and state in my methods section that logistic regression was investigated but the variables were found to lack independence? I could really use some advice for next steps.

EDIT:

Finally occurred to me I had collected their relationship status at 3 and 12 months and was treating these as separate outcomes when I should OR the values to create a dependent variable that relates whether participants were in a relationship at either of these two points in time. Getting much better results.

$\endgroup$
1
$\begingroup$

"is there a test to demonstrate that my variables are not independent?" - Yes, this is the chi-squared test.

"If so, should I just stick with chi square and state in my methods section that logistic regression was investigated but the variables were found to lack independence?" - No, you can still build a logistic regression using an interaction term.

$\endgroup$
  • $\begingroup$ Thanks! Reading up on interaction terms now. If I build a logistic regression using an interaction term, what will it tell me that my current chi square results showing significance for some subgroups won't? $\endgroup$ – buttonsrtoys Nov 28 '16 at 14:34
  • $\begingroup$ Using a logistic regression you don't just detect a relationship, you also predict the outcome. The chi square test is a tool to identify if a relationship exists in the first place. $\endgroup$ – Arun Jose Nov 28 '16 at 15:28
  • $\begingroup$ Got it. Thanks again. In my reading I see that a benefit of logistic regression over chi sq. is that logistic regression allows for continuous and categorical predictors. However, I don't benefit from this as all my predictors are binary (e.g., gender, or have yes/no values). Are there still advantages to logistic regression for my case? $\endgroup$ – buttonsrtoys Nov 28 '16 at 20:41
  • $\begingroup$ As I mentioned, if you want to predict relationship status, then you need logistic. If you only want to state you found a relationship use chi square $\endgroup$ – Arun Jose Nov 29 '16 at 2:03

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.