We randomly assign 1250 unemployed men and women to be invited once or twice to a specific-skills training program offered by a large multinational firm as part of its social responsibility efforts. Individuals with the lowest opportunity cost self-select into attending the program, which implies that the ability distribution of those invited but not registered, from which we draw our second group of invitees, is truncated from below. The impact of the first invitation is to decrease labor market participation, whereas the second invitation increases participation. We suggest that the program helps participants learn their own type (through interactions with other students), which is relatively low on average for the first group and relatively high for the second. Firstinvite treatments declare themselves to be less prepared for the job market relative to controls whereas second-invites feel better prepared as a result of the program. Socialization and selection effects may be important for job training programs.