Effect Heterogeneity and Bias in Main-Effects-Only Regression Models