#75 Please state the obvious/nonobvious

Tim Nall

I haven't studied Econ/Stats since 1991, and have forgotten everything. Downloaded gretl for work with linguistic data. Have spent weeks now furiously trying to catch up on what I forgot decades ago. While I admit it has been genuinely pleasant and gratifying to re-learn about autocorrelation, Durbin-Watson etc., there comes a point where I have to ask, why do you force me to spend weeks studying, when you have all the info to necessary to draw the relevant conclusions? Why don't you just display the conclusions? When I run White's test for heteroskedacity, since you know the p-value, the comparison statistic, and the sample size, why don't you display a string of text that say there is/isn't heteroskedacity? For DW, you could say is/isn't/indeterminate, etc.

Why make it hard on users? It would be one thing if the answers weren't available for display, but they are. Not everyone knows (or remembers) the relavant stats.


  • Sven S.

    Sven S. - 2015-03-12

    I guess there are several answers. For one, in statistics there are no conclusions without user-defined significance levels. And there are good reasons to choose different sig levels for different types of tests. So your expertise is an input, the program cannot all do it by itself. Which brings me to the next point, gretl like most econometrics packages is what IT people I think call an "expert system". So the target audience is people who know (or are learning to know) what they're doing.
    If some of the output is confusing due to the lack of clear information about what the test really tested etc., then please point this out on the mailing list or on the bug tracker here.


  • Sven S.

    Sven S. - 2015-03-12
    • status: open --> closed
  • Allin Cottrell

    Allin Cottrell - 2015-03-12

    Actually, I believe a so-called "expert system" in the AI world is not
    one designed to be operated by experts, but rather one that mimics
    human expertise. So the request here is for an econometric "expert
    system" in gretl. I think Hendry and co have gone some distance in
    that direction. It's not a silly request, but it would take some work.
    I can see a case for something like a "Please tell me about potential
    problems with this model" option, though it would have to come with
    a strong disclaimer since some serious problems are not detectable
    via "mechanical" analysis of residuals.


Log in to post a comment.