Year 12s: what will be the first way you research your future options? >>

Conclusion from residuals

I've determined the best regression equation and have the residual plots of MPG versus CYL, BHP, and WGT
However, I need to use the residuals from the model to check the assumptions and adequacy of my regression model, and I'm not sure how to do so
Could anyone please help? ImageUploadedByStudent Room1459358623.427480.jpg

ImageUploadedByStudent Room1459358623.427480.jpg

Reply 1

Gregorius

Original post by Bruce Harrisface

Sure. The plot in the top left is a normal quantile-quantile plot (or qqplot). If the residuals lie along the straight blue line, it is showing you that your residuals are normally distributed - as required. This is confirmed by the plot in the bottom left, which is just a histogram of residuals.

The plot in the top right is a plot of residuals versus fitted values. Here you are looking for a random equally spread scatter across the graph. Trends would indicate a mis-specification of the functional form required (e.e quadratic versus linear); varying spread across the page would indicate heteroskedasticity. No such problems here.

The plot in the bottom right appears to be plotting residuals against observation order. This is looking for correlation between residuals, If there was correlation, you'd get runs of residuals on the same side of the zero line (positive correlation) or jumping too-and-fro across the zero line (anti-correlation). I would say there's a faint hint of the latter here; I would do a formal check for serial correlation just to check.