zone, we were able to match 1,097 of the 1,116 (98.3%customers from the treatment group with customers whhad not participated in a CRP. We were unable to find a statistical twin within this tolerance zone for only 19 of th
1,116 participants.
Following Rosenbaum and Rubin (1985), we calculated
a percentage reduction in bias for all variables, a common tactic (Wangenheim and Bayón 2007a). As we show in Table 2, the matching procedure achieved a good bias reduction. Participants and nonparticipants exhibited different characteristics and behavior before the matching procedure but were relatively similar afterward. The average percentage reduction in bias for all variables was 86.3%.