Hello, To replicate Greene's example 9.3 (heteroskedasticity tests), in Gretl I run the script: open TableF9-1.txt square Income smpl Avgexp>0 --restrict ols Avgexp const Age Income sq_Income Ownrent modtest --white modtest --white-nocross With Gretl 1.9.11, everything goes fine: Gretl manages to drop the various combinations that are redundant in the auxiliary regression and prints the expected results. When I use Gretl 2016c (Lubuntu 16.04), both versions of the White test fail with the message "Exact or near collinearity encountered". Am I doing something wrong in Gretl 2016c, or has a bug creeped into modtest? Thanks, Pedro