random assignment may not completely stand up either to Brady and McNulty’s careful
data analysis or to a priori reasoning -- elections supervisors, after all, are probably trying
to maximize turnout -- after consideration of potential confounders, Brady and McNulty
can convincingly argue that the costs of voting negatively influenced turnout.
Social scientists have also exploited such naturally-occurring phenomena as the
weather as a source of natural experiments. The study by Achen and Bartels (2004) can
be interpreted as a natural experiment in which shark attacks, droughts, and influenza
pandemics randomly assign poor economic conditions to some local jurisdictions; voters
in these jurisdictions appear to punish incumbents through retrospective voting, even
though these incumbents should reasonably shoulder none of the blame for the worsened
conditions. Miguel et al. (2004) use economic growth shocks stemming from bad
weather to study the sources of civil conflict; this study will be briefly discussed in
Section IV below.
Institutional rules that create sharp thresholds that assign subjects to treatment and
control groups may also be used to argue for “as if” random assignment. For example,
Angrist and Lavy (1999) exploit a rule in contemporary Israel (known as Maimonides’
Rule, after the 12
th
century Rabbinic scholar) that mandates that secondary schools have
no more than 40 students per classroom. In a school in which many classrooms are near
this threshold, the addition of a few students to the school through increases in grade
enrollment can cause a sharp reduction in class sizes, since more classrooms must be
created to accommodate the additional students. Students in classes that were just under
the threshold can then be compared to students in those just over the threshold; since the
latter group are reassigned to classrooms with smaller number of students, this natural
Convention