On Sat, 14 Jan 2006, John Paravantis wrote: I'm not ruling that out, but could you (or anyone) explain why you want to see lags in the main screen? With first differences, logs, or other transformations, the transformed variables should (obviously) be shown, because you might want to select them for graphing, summary statistics, or whatever. With lags, on the other hand, I can't see any use for them other than as regressors (and they are visible OK in the dialog where you choose regressors). Suppose you have a monthly dataset with several variables, of which you want to have available 12 lags apiece for regression purposes. Showing dozens of such variables in the main window, it seems to me, is just filling the window with trash, making it harder to find variables you might actually want to examine individually. Am I missing something here? Allin Cottrell

On Sat, 14 Jan 2006, jack wrote: Good point. I've considered this, but haven't done anything about it yet because I had trouble envisaging exactly what the user interface would look like. Makes sense. I too have a vague recollection of how PcGive does it, but -- apart from the issue of gretl's regressor selection sub-window becoming overcrowded -- I feel that gretl's selector is easier to use. Anyway, if anyone has any brilliant ideas regarding the ideal interface, please post them! Allin Cottrell

jack schrieb: In PcGive, you select a variable, alter (if change is needed) the number of lags in an entry field, and add the bunch to the list of regressors. (Afterwards you can delete individual intermediate lags or the contemporaneous original variable itself if you want to.) The advantages over gretl are imho: - needed lags are created on-the-fly - less clutter/noise in the list of selectable variables A slight disadvantage could be that if you only want some non-contiguous lags, it might be a little more click-work than in gretl. Overall I think it's worth imitating. I also think there is a slight problem with the current state of affairs; imagine you want to run a quasi-dynamic regression: x_t = c + b y_{t-1} + u_t (One reason for that formulation could be to avoid simultaneity bias that could arise when you include contemporaneous y_t instead.) Doing that in gretl is possible, but since y_{t-1} is not visible in the main window, it might become slightly annoying to open the regression dialog, see that y_{t-1} does not yet exist, cancel the dialog, create y_{t-1}, go back to the dialog. Secondly, in this scenario you might want to graph x_t against y_{t-1}, which is difficult because lags are not selectable in the main window. So the solution would be (I think): First adopt the on-the-fly lag creation, and thus tell the users that explicitly creating lags is normally unnecessary => no clutter in the main window. Second, make created lags visible again in the main window to have them easily available for graphing, see above. This approach would become unnecessary after the above imitation of PcGive's behavior, wouldn't it? Just my 2c Cheers, Sven