if two regressions use different sets of observation, then we can tell how the r-squareds will compare, even if one regression uses a subset of regressors