*
* Example HETER.DAT from pp 430-433
*
open data heter.dat
data(format=free,org=columns) 1 50 mo x1 x2 y
*
scatter(hlabel="X1",vlabel="Y")
# x1 y
scatter(hlabel="X2",vlabel="Y")
# x2 y
*
linreg y
# constant x1 x2
*
scatter(footer="Figure 10-19 Heteroscedasticity of residuals for X1 using OLS",hlabel="X1",vlabel="Residuals")
# x1 %resids
*
* Weighted least squares is done with LINREG with the option SPREAD=series to
* which variances are assumed proportional. Here, that's the square of x1. Note
* that the "weighting" is all done internally.
*
linreg(spread=x1**2) y
# constant x1 x2
*
* However, the residuals are in the natural scale, so they have to be scaled to
* get figure 10-20.
*
set uscaled = %resids/x1
*
scatter(footer="Figure 10-20 Residuals vs X1 with weighted least squares",hlabel="X1",vlabel="Residuals")
# x1 uscaled