# Logistic regression breaking due to an inverted interaction?

I have a three independent variables and am getting significance under chi squared for my main study variable in some of my subgroups, but not all, which I'm OK with because that mostly aligns with my predictions.

My editor suggested I use logistic regression, which I’m very new to. If I understand my results, I’m not getting significance for my main variable under logistic regression, I suspect because one of my variables (gender) inverts the effect of my main variable. Can that nullify significance under LR?

More specifically, my study is a dream study on a large group of single participants (N > 2000), testing for a relationship between dream content and future relationship status. My study questions is, are people who dream of attraction more likely to be in a future relationship compared to those who dream of other concerns, such as work?

My binary DV is future relationship status (single | in a relationship), my main IV is the binary of whether the participant reported no attraction or attraction (romantic and/or sexual) in their dream report. My other two IVs are gender and whether they have "waking attraction" (WA) to someone in their life (i.e., a crush on someone, which is known to affect dream content).

Below is a summary of some of my chi square results. I collected future relationship status at 3 and 12 months, so I have 8 subgroups: (3 mo., 12 mo.) x (Men, Women) x (no WA, WA). As described in the figure caption, the two shaded bars represent the likelihood of participants to be in a future relationship based on whether they dreamed of attraction (ns are along the x-axis and %-of-n is in the corresponding bars).

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). The groupings are indicated by the layered figure headings. Each subgroup and aggregate group's heading is accompanied by an asterisk if it achieved C.S. significance. E.g., significance was found for the groups of all participants at 3 months, women at 3 months, men at 12 months, women with WA at 3 months, women w/o WA at 12 months, men with WA at 12 months.

As witnessed in my C.S. analysis, my IVs appear to be strongly coupled and often inverted. E.g., at 12 months, women without WA were significantly more likely to be in a relationship if they dreamed of attraction, while the men with WA were significantly LESS likely to be in a relationship. This is what I was described earlier as “inverted” interaction, which brings me back to my question: would this mess with my L.R. tests?

Not sure if this is helpful, but below are my SPSS results, which include my attempt at an interaction term, which is a single variable with 8 categories, one for each subgroup.