Section 17.5 Conclusion
The material we have covered is really only a taste of multiple regression and linear modeling. On the one hand, there are a number of additional factors that may be considered before deciding on a final model. On the other hand, there are a great number of techniques that may be used in specialized circumstances. For example, in trying to model attendance at the MCG, we have seen that the standard model fits the data some of the time but not others, depending on the selection of the explanatory variables.
In general, a simple model is a good model, and will keep us from thinking that we are better than we really are. However, there are times when we will want to find as many dependent variables as possible. Contrast the needs of a manager trying to forecast sales to set inventory with an engineer or scientist trying to select parameters for further experimentation. In the first case, the manager needs to avoid a falsely precise estimate which could lead her to be overconfident in the forecast, and either order too much stock or too little. The manager wants to be conservative about deciding that particular variables make a difference to prediction variable. On the other hand the experimenter wants to find as many variables as possible for future research, so is prepared to be optimistic about whether different parameters affect the variables of interest.
