Accurately evaluating yield response to nitrogen can increase crop management profitability and sustainability. Many studies estimate yield response by fitting a regression model to data collected from different fields. But analysing such combined data requires that heterogeneity across fields be accounted for in the regression analysis along with the variation in input rates. This study uses data from 27 large-scale on farm experiments to test the potential danger of getting biased estimates of yield response functions. Models with and without field fixed effects are run. The yield response functions from the two models showed different slopes, which provides a visual representation of the bias resulting from the pooled estimation. Use of the Mundlak approach indicated that ignoring the endogeneity of regressors with respect to field effects leads to an unreliable estimation of yield response to N.