Prediction, Goodness-of-Fit, and Modeling Issues
Several extensions of the simple linear regression model are now considered. First, conditional predictions are generated using results saved by gretl. Then, a commonly used measure of the quality of the linear fit provided by the regression is discussed. We then take a brief detour to discuss how gretl can be used to provide professional looking output that can be used in your research.
The choice of functional form for a linear regression is important and the RESET test of the adequacy of your choice is examined. Finally, the residuals are tested for normality. Normality of the model’s errors is a useful property in that, when it exists, it improves the the performance of least squares and the related tests and confidence intervals we’ve considered when sample sizes are small (finite).