Topics in simple linear regression
Examining lack of fit
Calibration: predicting x for a given y value
Use of SSPexp to study lack of fit
When we have more than one observation per xi value, we can partitiion SSE:
SSE = SSPexp + SSlack = SS for pure error + SS for lack of fit
which leads to an F test of lack of fit: F = MSlack / MSPexp
Multiple regression
Often a simple linear regression model is not sufficient as a model for y .
Other possibilities are:
yi = b0 + b1xi + b2xi2 + ei (polynomial in x)
yi = b0 + b1x1i + b2x2i+ ei (linear in two variables x1 and x2 )