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 (with standard errors clustered at the level of the individual survey respondent).

We test H1 by estimating the AMCE for “PfP” (vs. fixed pay) on job attraction. We test H2a/H2b/H2c by estimating the AMCE for interaction terms for “PfP” (vs. “fixed pay) x race/gender/age on job attraction.  

Moreover, we perform three sets of supplementary analyses. First, we consider the AMCE when a higher proportion of the pay compensation is PfP-based—i.e., using the range of PfP job attribute values (not collapsing the three PfP job attribute values). 

Second, we test for moderating effects for how job performance is measured and evaluated. In particular, we examine how the attribute values for the “job performance evaluation” job attribute—i.e., “goal-based” vs. “supervisor-based”—moderate the AMCE of PfP on job attraction (H1) and interact with the moderating effects of race/gender/age on the AMCE of PfP on job attraction (H2a-H2c) (three-way interaction).

Finally, we probe how individuals’ self-reported risk aversion moderate the AMCE of PfP on job attraction (H1). Since risk aversion may be a mechanism by which demographic groups differ in their response to PfP (H2a-H2c), we also try including interaction terms for PfP x race/gender/age at the same time we are estimating an interaction term for risk aversion x PfP. This allows us to see if estimated moderating effects of demographic variables change after accounting for any moderating effect of risk aversion. We perform these tests using two separate measures of risk aversion: A single-item measure capturing general risk aversion (Dohmen et al. 2005) and a single-item measure capturing risk aversion in relation to a lottery prize scenario (Xiaohao et al., 2010).