Dear Allin, I'm working on a function that involves iterative WLS in order to correct OLS estimators in the presence of outliers. As a 1st step, the function has to leave out observations with a Cook's distance (D) > 1. Cook's D can be easily computed from the leverage series, what I do through the following code within the function: <hansl> ols myY myX --quiet ... # retrieve some statistics here leverage --quiet --save series l = lever ... # some code for Cook's D based on l, OLS residuals and OLS se) </hansl> Now, the function works smoothly...unless a "lever" series already exists in the dataset. In that case the following error message is generated, The symbol 'lever' is undefined *** error in function robust, line 22 I first suspected that this is probably due to the fact that before executing my function I had already saved a lever series through the "leverage" command. So, I deleted the existing "lever" series and successfully re-run the function. Then, I created a new "series lever = normal(0,1)" which obviously has nothing to do with the "leverage" command. Re-run the function...and the same error came out. Deleted the "lever" series and the function runs perfectly. Is there anything I'm missing here? Best, Artur

Le mer. 30 déc. 2020 à 13:59, Sven Schreiber <svetosch(a)gmx.net&gt; a écrit : Thank you Sven. I'll manage to send it over. 2020g - git build 26/12 Best, Artur

Am 30.12.2020 um 17:32 schrieb Artur Bala: Ah OK, I see what you mean. Of course the error message is misleading, but see below. I guess it's a gretl bug. Here's my hypothesis: - As per the doc 'leverage --save' uses 'lever' for the saved series only when such a series doesn't already exist. Otherwise it uses a different name (lever_aaa I think). - So in your second invocation gretl looks in the outer-scope dataset (which it shouldn't do!), sees a 'lever' series there, and saves the series under a different name. Then when you try to access 'lever' inside your function gretl doesn't find it because it only looks in the function scope (which is correct!). Hence the unknown object message. So the bug might be that gretl mixes up the outer and inner scopes here. As a workaround perhaps you could use the series ID numbers instead: Before you run 'leverage' check what's the highest existing ID number, for example with nelem(dataset); then the new lever-style series should have this ID number +1, no matter what its name is. cheers sven

Thank you Sven, for figuring this out. Probably I'm repeating Riccardo's suggestion but wouldn't be better just to: - discard the "--save" option from the leverage command so nothing is saved automatically - and provide $-style accessors for leverage, influence and Dffits if one wanted to save anything from the leverage command. Artur Le mer. 30 déc. 2020 à 18:15, Riccardo (Jack) Lucchetti < p002264(a)staff.univpm.it&gt; a écrit :

Le mer. 30 déc. 2020 à 18:51, Sven Schreiber <svetosch(a)gmx.net&gt; a écrit : ...which has come in the current snapshot. The 'leverage' command within the function works now as it was meant to work. Best, Artur