Instrumental variable (IV) techniques represent a key tool for estimating causal effects. However, identifying plausibly exogenous instruments can be difficult.  We provide a simple test to bolster arguments for instrument validity by considering the difference in 2SLS estimates using linear and polynomial first stages.  In a homogeneous effects model, a large difference in these estimates can be cast as evidence of an invalid instrument.  In a heterogeneous effects framework, such differences may result from estimating a different average partial effect.  Namely, the two estimates place different weight on effects at different values of the instrument.  We derive an estimate of the ratio of the weights for the two cases under very general assumptions. These ratios allow for a description of the relative magnitudes of the partial effects across the instrument distribution.  If the patterns contradict economic reasoning, this provides further evidence against the validity of the instrument.  Otherwise, they help to uncover heterogeneity in the effect of interest.  We apply our method to two influential papers: Becker & Woesmann (2009) and Acemoglu, Johnson, & Robinson (2001).  In each case we find that the conclusions are sensitive to the choice of first stage, raising concerns over the validity of, or the degree of heterogeneity driving, the original results.  We also apply the approach to a set of thirteen objectively chosen papers and find many cases where the economic conclusions change.