The impact of sampling variation on peer measures: a comment on a proposal to adjust estimates for measurement error

Investigation of peer effects on pupil’s achievement with survey data on samples of schools and pupils within schools may mean that only a random sample of peers is observed for each individual pupil. This generates classical measurement error on peer variables. Hence under OLS model fitting the estimated peer group effects in a regression model are biased towards zero (attenuation). A simple adjustment for this kind of measurement error was proposed by Neidell and Waldfogel (2008). We review the derivation of the simple adjustment and suggest that it is not properly justified.