3
$\begingroup$

I'm performing a study on a large group of single participants, testing my categorical independent vars against future relationship status. My main IV has three categories: (type 1) no attraction, (type 2) romantic or sexual attraction to an acquaintance or stranger, or (type 3) attract to a current or former partner (girlfriend, husband, ex-spouse).

My other two categorical variables include Gender and whether they have a crush/"waking attraction" (WA) to someone in their life.

I tested all 8 subgroups using C.S., and also tested their aggregate groups. E.g., all participants at 3 months, all Women at 12 months (both with and w/o WA). My dichotomous DV is future relation status (single | in a relationship) and my binary IVs are gender and WA. Below are my SPSS results for Block 0 and Block 1.

enter image description here

I'm not getting L.R. significance with IV variable ("Attracted_1Yes2No"), but it appears that there's interaction between my variables?

EDIT:

It occurred to me that I was treating my data at 3- and 12- months as separate outcomes and really should OR them together to create a DV that relates whether participants were in a relationship at either of these two time points. Getting much more interesting results now with both my C.S. and L.R. tests.

rerprocessed LR

$\endgroup$
3
  • 1
    $\begingroup$ No responses yet. Did I put in too much info? $\endgroup$ Nov 30, 2016 at 23:37
  • 1
    $\begingroup$ Generally low activity in december ... people are to preoccupied with exams and christmas preparations! $\endgroup$ Dec 2, 2016 at 14:57
  • $\begingroup$ If you think there is an interaction between person's gender and waking attraction need to be entered into the logistic regression model in addition to the interaction terms $\endgroup$ Dec 2, 2016 at 15:33

2 Answers 2

2
$\begingroup$

The bar charts which you present do seem to show interaction. Just looking at the 3 month data the patterns for women with and without WA seem very similar in contrast to the men where the patterns are different with and without WA and different from the women's pattern.

You ask whether you are fitting the interaction correctly. What you have done is something which I sometimes advise people to do when they cannot understand their output but it is not the canonical way. I do not use SPSS myself but there must be a way of specifying interaction directly so you can see the two-way and three-way interaction. I have found in the past that typing into your favourite search engine UCLA followed by the name of what you want to do in your case logistic regression interaction followed by the name of your favourite statistical software in your case SPSS usually find a very helpful page on their web site. Try it and see what happens.

$\endgroup$
1
  • $\begingroup$ Thanks @mdewey. I'll update with post with by adding my gender and WA variables, and will read up on what you mean by two- and three-way interaction. $\endgroup$ Dec 3, 2016 at 0:05
1
$\begingroup$

Some points: the spss output shows you are doing a stepwise regression (could be that is spss's default?) You shouldn't do that, tell spss to keep in all variables. Second, the wald statistics reported by spss are based on large-sample approximations which often are way off with logistic regression. I believe that spss logistic regression offers bootstrapping, try that. Or get some better software, such as R, and try likelihood profiling. And third, ask a local statistician, logistic regression isn't for amateurs! And, search this site, a lot of useful information about logistic regression here.

$\endgroup$
2
  • 3
    $\begingroup$ I believe that is indeed a default way that SPSS presents the info, but it is not "stepwise regression" per se. The last time I used SPSS was to help a doctoral student for her qualifying exams ~4 years ago, but I seem to remember that SPSS shows the "Null model" results first and then whatever model the user wants to fit is shown in "Block 1" $\endgroup$ Dec 2, 2016 at 15:32
  • 1
    $\begingroup$ @kjetil, I appreciate the advice for amateurs like me! LR certainly isn't for the faint of heart. $\endgroup$ Dec 3, 2016 at 0:07

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.