0
$\begingroup$

I hope all of us get well during the pandemic. I have conducted an analysis using binary logistic regression to investigate the interaction between gender (male, female) and official language efficiency (English, French, both English and French) with the outcome: youth's sense of belonging. However, I am unsure of two considerations that needed your comments.

First, I found a significant contribution of the interaction between gender and official language efficiency with the outcome (e.g. youth' sense of belonging). Male, speaking English as the reference group. However, the results showed no significant differences among the variable's levels. In such a case, is it fine if we still report it as a significant result?

Second, in the same situation, I found nonsignificant contributions to the entire interaction (visible minority status interacting with official language efficiency) with the outcome. However, I found a significant difference among its levels. For example, compared to youth not experiencing visible minority status speaking English (the reference group), youth with visibility status speaking French are more likely to have a strong sense of belonging. In this case, is it appropriate to report it?

Interactions Sig. Exp(B)
Gender * official language * visible minority .062
Male by English by not visible minority Reference
Female by French by visible minority status .019 19.748
Female by both English and French by visible minority status .711 1.160
Gender_1 * official language .076
Male by English Reference
Female by French .396 .673
Female by both English and French row
Official language * Religious affiliation .047
English by Not religious affiliation Reference
French by Religious affiliation .051 .045
Both French and English by Religious orientation .101 .442

Hello, thank you for reaching the question and comments. I would like to add more detail here. My question is that experiencing multiple social locations (e.g., gender, official language, visible minority status, and religious affiliation) impacted the sense of belonging? Here is the excerption from SPSS output as my concern. For coding: male, English, not a visible minority status, not religious affiliation = baseline. So, the first and second interactions were found nonsignificant but significantly different in levels (p= .019 and = .30). For the third one, the entire interaction was significant, but its levels were found nonsignificant different.

Improve this question
$\endgroup$
7
  • 1
    $\begingroup$ Agree with @DemetriPananos: a table with output would really help. One thing to keep in mind with statistics is what your ultimate goal is. If you want to do statistical inference, then you should have specific hypotheses that relate to what you would observe in the main effects and interactions. If you do indeed have a priori hypotheses about how these variables relate to your outcome, then you absolutely have to report their results. Ultimately, however, you need to be sure you understand your model and why some of these things are or are not significant when other variables are entered $\endgroup$
    – Billy
    Commented Sep 2, 2021 at 16:36
  • 1
    $\begingroup$ Many similar Qs that might help you: stats.stackexchange.com/questions/24298/…, stats.stackexchange.com/questions/135184/…, stats.stackexchange.com/questions/245256/…, stats.stackexchange.com/questions/31690/… $\endgroup$
    – kjetil b halvorsen ♦
    Commented Sep 2, 2021 at 17:31
  • 1
    $\begingroup$ Note that the actual coding used is important, especially the choice of reference level. male can be significant with one coding, but not with another, it depends on base level. A better example might be etnicity: chinese might be insignificant if reference level is jew , but significant with some other reference level! Remember that levels are shorthands for some specific comparison to some other level. $\endgroup$
    – kjetil b halvorsen ♦
    Commented Sep 2, 2021 at 17:35
  • 1
    $\begingroup$ Can you clarify what you mean by having "multiple social locations"? I would think that everyone has data for the variables you listed there, so I don't know how there could be differences in the number of "social locations." Do you mean that people who belong to "majority" social groups (e.g., male, English speaking, Christian, etc) have higher belonging? If so, see @kjetilbhalvorsen comment regarding contrasts because you'll need to think about what should be the reference level in that case and what effect/contrast coding you want $\endgroup$
    – Billy
    Commented Sep 4, 2021 at 15:12
  • 1
    $\begingroup$ Can you also share your sample size? p-values aren't very useful in this case because I'm more concerned about the precision of your estimates (the standard errors). These are a lot of interactions, so I worry about having small n's in the cells (e.g., what's the n for male, French-speaking, visible minority, Jewish participants). If you sample size is too small, then that might explain why certain variables aren't significant when interactions are included (though it's also not uncommon for interactions to "soak" up the variance of their main effects) $\endgroup$
    – Billy
    Commented Sep 4, 2021 at 15:14

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.