I have a very basic statistical question about the use of quantiles to tease
out policy impacts. I am attempting to use quantile regressions (QREG) to
detect the impact of state-level prevailing wage law repeals (the
elimination of wage floors for construction workers in publicly funded
projects) on very specific groups of workers at different points in their
wage distribution -- defined by state and
occupation. I have just over 30 observations who I code as "treated"
(treatment is not directly observed -- I am using the CPS, and in fact, I
can only assume that a subset of these 30 observations are actually
treated).
Is there any rule of thumb about when your sample is too small to reliably
estimate percentiles? Are there any guidelines similar to the general rule
that 30 observations is (more or less)
sufficient to estimate a mean? Any articles that anyone can point me toward?
