We test our hypotheses using the commonly used quantity of interest in conjoint analysis: the average marginal component effect (AMCE) which represents the causal effect of changing one attribute of a profile while averaging over the distribution of the remaining profile attributes (Hainmuller et al. 2014). We use the responses for all three sets of paired profiles conjoints. AMCE is estimated using OLS regression with standard errors clustered at the level of the individual survey respondent.

We test the PSM hypotheses by estimating the AMCE for interaction terms for PSM x relevant job attributes (dummies for total pay, target community income level, target community racial makeup, target community prior involvement, workload requirements, and main job task) on job attraction. We estimate seven model specifications (all including direct effects for PSM and all eight job attributes): One for each PSX x job attribute interaction term and a joint model including all six interaction terms. 

We test the self-efficacy hypotheses by estimating the AMCE for interaction terms for self-efficacy x relevant job attributes (performance pay and job performance evaluation) on job attraction. We estimate three model specifications (all including direct effects for self-efficacy and all eight job attributes): One for each self-efficacy x job attribute interaction term and a joint model including both interaction terms.

Moreover, we estimate a full model including direct effects of PSM, self-efficacy and all eight job attributes, as well as all eight relevant interaction terms (the six for PSM x job attributes and the two for self-efficacy x job attributes).