Statistical Methods - Text Mining Techniques for Healthcare Provider Quality Determination - page 90

We also want to look at a simple linear regression of total charges versus length of stay. Then Y (total
charges)=α+βX+ε, where ε represents the error term. In order for the regression model to be valid, there
are some assumptions that need to be considered. The residuals need to be independent and identically
distributed such that the mean is zero and the variance is equal to the population variance divided by n.
We estimate the population variance by the sample variance.
The best way to examine the assumptions is to look at the residuals. Figure 10 shows the actual
versus predicted values; Figure 11 shows the residuals by the total charges. Figure 11 should have no
Figure 10. Actual versus predicted values for total charges
Figure 11. Residuals versus total charges