Numerous shufflings of data produce a distribution of test-statistic v
alues that can be used to assess the degree to which the test-statisti
c value produced by the actual data is unusual. Because this controver
sial randomization-testing methodology, made partical by the computer
revolution, has begun to appear in applied econometric studies, econom
etricians should become familiar with its mechanics, rationale, and in
terpretation, all of which are quite different from the status quo. Th
is article exposits randomization tests in an econometric context, dis
cusses their advantages, and alerts practitioners to pitfalls.