[Gretl-devel] small new feature

Comments invited on a new feature in CVS/snapshot. It's a Frisch-Waugh-Lovell type thing, maybe useful in pedagogy. Look under "/View/Graph specified vars/X-Y with control...". You get to specify an X variable, a Y variable, and a Z or "control" variable, whose effects will be taken out of both X and Y. That is, the result is a scatterplot of (residuals from regression of Y on Z) against (residuals from regression of X on Z). (If you want to control for more than one "Z" variable, do it yourself with a script!) A nice example, I think, is Ramanathan's data4-3. Try plotting housing starts against gnp, and housing starts against "intrate" (mortgage interest rate). There's nothing to see (hint: there's a strong positive correlation between gnp and intrate in this sample). Now plot housing starts against gnp, controlling for intrate, and housing starts against intrate, controlling for gnp. Ah, now there _is_ something to see. (The title of the "controlled" graph still needs to be changed to properly reflect what's going on.) Allin.