We have a complicated design that is presenting issues in terms of an appropriate analysis. I'll present our hypotheses first and then our methods to clarify what our issues are.

H1: Group leader gender will impact individual liking of leader.

H2: Task type will interact with group-leader gender to impact individual liking of leader.

H3: Group perceptions of fairness will interact with task type and group-leader gender to impact individual liking of leader.

Methods: In a nutshell, we took 5 surveys over the course of a semester of individuals working on team projects. T1 was a demographics survey. T2-T5 were surveys that assessed our focal variables. At T2, participants were working on a trading task (weakly competitive), at T3 they were working on a selling task (strongly competitive) and at T4 & T5 they were working on a video project (non-competitive). We assessed emergent team leader gender, perceptions of fairness, and liking at each time point except T1.

We are interested in focusing on the different between T3 & T4 to compare liking between the two time points. We focus on these two time points because T3 is strongly competitive and T4 is non-competitive, and we are most interested in this particular comparison. We are not interested in change in time beyond that, as time is just a proxy for task type. Our variables are therefore:

DV: Liking (Individual level, continuous)

IV: Task type (competitive vs. non-competitive, which occurred for all groups at T3 & T4, respectively, and therefore time acts as a proxy for)

IV: Leader gender (group-level, dichotomous)

IV: Group perceptions of fairness (group-level, continuous)

Repeated measures ANCOVA doesn't seem appropriate because we want to assess the interaction of group-perceptions of fairness with time/task type and leader gender. It also doesn't capture the nested nature of the data (i.e. within groups). Multilevel models don't seem appropriate because we only have two time points and we want to compare liking at the individual level between the two time points. We also thought about running two separate HLM analyses for each time point and comparing the coefficients of our effects, but worry that that does not tap into the non-independence between those two samples. Any help would be appreciated here.