Much of the previous literature on treatment effects assumed homogeneous responses to
treatment, meaning that based on certain observable characteristics, effects are constant across
individuals and that they would derive identical benefits from treatment. Recent studies have
given more attention to heterogeneous responses where the effects vary across individuals due to
their observable or unobservable characteristics. Much of the focus is now on the role of
unobservable characteristics in determining outcomes particularly in cases where individuals are
otherwise identical in their observed characteristics (Basu et al. 2007; Caliendo, 2006). Basu et
al. (2007) describe two instances in which heterogeneity (arising from unobservable
characteristics) may factor into treatment evaluation. The first instance is where individuals with
identical observable characteristics respond differently to treatment but do not opt for treatment
based on their idiosyncratic benefits or gains (non-essential heterogeneity). The second instance
is where individuals have identical observable characteristics and respond differently to
treatment and are aware of the benefits to be derived from treatment. In this latter case, their
treatment choices are influenced by anticipation of idiosyncratic gains (Basu et al. 2007). Basu
et al. (2007) and Heckman, Urzua and Vytlacil (2006a; 2006b) refer to the second instance as
essential heterogeneity.
In the context of this paper, heterogeneity of foreign farm worker responses to
legalization is maintained and subjected to a statistical test. In the presence of heterogeneity, it is
assumed that they obtained legalization because of individually perceived wage benefits, and that
in the future, workers without legal status would proceed similarly in the presence of a program
such as AgJOBS. The analysis follows a parametric approach developed by Heckman, Urzua and
Vytlacil (2006b) to estimate the choice and outcome models, and the treatment effects of
legalization; alternative non-parametric estimates are also evaluated. Their MTE algorithm was