Regression adjustment—using linear regression methods, generalized linear models, binary regression, and other regression models—adjusts for differences across units in measured characteristics that might be correlated with outcomes and with other factors that change outcomes. Propensity score matching and reweighting methods accomplish a similar adjustment for measured characteristics but can dispense with certain functional form assumptions.

Various panel regression techniques can also adjust for differences in unmeasured characteristics (such as native ability or intrinsic perseverance or patience) by comparing outcomes and predictors only within units and not across units. Instrumental variables and regression discontinuity models can also adjust for differences in unmeasured characteristics both across and within units under certain assumptions, and they can offer impact estimates with internal validity close to that of experimental methods.