Different procedures exist for finding nonparticipants who match each participant and creating matched samples (for an overview, see Caliendo and Kopeinig 2008); we applied propensity score matching (Rosenbaum and Rubin 1983), which has proved advantageous in many settings (Dehejia and Wahba 2002; Wangenheim and Bayón 2007a). First, we conducted a binary logistic regression to calculate the propensity for participation in the firm's CRP. Second, we built an artificial control group by matching each customer from the treatment group (i.e., CRP participants) with a customer who did not participate but achieved a similar participation propensity score. Third, we evaluated the quality of the matching. Fourth, we compared the loyalty of the treatment and control groups