The strategy of building a regression model can be done by following these steps:

First the data has to be collected. The requirements for collecting the data vary with the types of studies which are: (The basic types of studies are later explained).  

1. Controlled experiments
2. Controlled experiments with supplemental variables
3. Confirmatory observational
4. Exploratory observational

The investigator should study and choose the data carefully to reduce data errors. You can look at the following then make the necessary corrections.

• descriptive statistics
• box plots 
• other graphs to help check for errors

If necessary, after the data is collected and prepared the explanatory variables may need to be reduced. In some cases having too many explanatory variables results in an over-fitted model which has larger variances of the estimated parameters than a simpler model with  smaller variances.  The model has to be refined and the best one has to be selected depending on the diagnostics. Diagnostics, such as the distribution, fits, dfits, residuals, Cook's distance, and checks for multicollinearity are good tools to refine the model and help select the best model.

After the ultimate model is selected, it has to be validated. Validity is based on the

•  stability and correct reasoning of regression coefficients
•  the efficiency of the function
•  and the ability to generalize conclusions and predictability.

Procedure                     Return to index

Regression Tutorial Menu    Dictionary

STATS @ MTSU