Consistent with social comparison theory (SCT), Blanton, Buunk, Gibbons, and Kuyper (1999) and Huguet, Dumas, Monteil, and Genestoux (2001) found that students tended to choose comparison targets who slightly outperformed them (i.e., upward comparison choices), and this had a beneficial effect on subsequent performance-a behavioral assimilation effect (BAE). We show (Studies 1 and 2) that this apparent BAE is due, in part, to uncontrolled measurement error in pretest achievement. However, using simulated data (Study 3), these phantom BAEs were eliminated with latent-variable models with multiple indicators. In Studies 4 and 5, latent-variable models were applied to the Blanton et al. and Huguet et al. data, resulting in substantially smaller but still significantly positive BAEs. More generally in personality research based on correlational data, failure to control measurement error in pretest/background variables will positively bias the apparent effects of personality variables of interest, but widely applicable approaches demonstrated here can correct for these biases.