[Gretl-users] Suggestions for new features and some changes in gretl

[Andreas Karlsson]

Citerar Allin Cottrell <cottrell at wfu.edu>:

> > * Add a menu entry for creating cross tabulation tables, for 
> > cross tabulation of one variable against the other
> 
> This is available via the "xtab" command, which is now accessible 
> using a new GUI menu item, "/View/Cross Tabulation".  Note that 
> this works only for discrete variables.  It won't automatically 
> bin continuous variables for you.  On the other hand it's not 
> difficult to bin variables yourself using "genr".

Thanks, it looks nice. Would it be possible to add a checkbox for also showing
the row or column percentages for each row or column, respectively, together
with the frequency count, as for example SPSS does?




> > * Add the possibility to calculate bootstrapped confidence 
> > intervals for OLS estimates.
> 
> That's possible already, though you have to do it manually (no 
> GUI).  One of the functions in "genr" is resample().  See section 
> 5 of chapter 5 of the User's Guide, "Resampling and 
> bootstrapping".  May be worth adding to the GUI.

A GUI for this would be really great and useful.



> > * For OLS output, also show the ANOVA table (or add a check box 
> > where the user can choose to show the ANOVA table in the output)
> 
> I agree with Jack that this is redundant, but since redundancy can 
> sometimes serve a pedagogical purpose, I've done this: When you 
> estimate a model via the GUI, in the model window there's a menu 
> titled "Analysis" (with covariance matrix etc.).  I've added an 
> "ANOVA" entry to that menu, which displays the table.  This item 
> is available only for models that are estimated via OLS, and that 
> include a constant.

Thanks a lot. I know that it is redundant, but my students are used to getting
an ANOVA table with the output, as for example SPSS and MINITAB produces.
However, I noticed that the ANOVA table produced by gretl is not complete: The
general form of the ANOVA table my students are used to are as follows:

Source of   Sums of     Degrees of  Mean        F               P-value
Variation   squares     freedom     squares

Regression  Regression  k           Regression  Regression MS/  P
            SS                      MS          Residual MS

Residual    Residual    n-k-1       Residual
            SS                      MS

Total       Total SS    n-1

As for exampel this one produced by SPSS:
http://www.utexas.edu/its/rc/tutorials/stat/spss/spss2/images/Reg5.gif

and this one produced by MINITAB:
http://www.indiana.edu/~statmath/stat/minitab/win/ANOVA.gif


> > For the augmented Augmented Dickey-Fuller test, regarding the 
> > "testing-down algorithm" for choosing lag length...
> 
> Your suggestions on this are useful, but will take a bit longer to 
> implement.
> 
> > After estimating a model and choosing "Tests > Collinearity" 
> > from the output window, the variance infation factors (VIF) are 
> > given. Suggestion: Also show the value of the determinant of 
> > X'X...
> 
> Done: In this window we now show these characteristics of X'X:
> one-norm, determinant, and reciprocal condition number (the last 
> is actually a better guide than the determinant).

Could you also include the value of the X'X matrix when it is in correlation
form? In that case the determinant of X'X is 0<=|X'X|<=1, which I think is
useful because if |X'X|=1 the regressors are orthogonal, while if |X'X|=0 there
is an exact linear dependence among the regressors. Thus it is a measure of the
degree of multicollinearity that is easy for the students to interpret and
understand, which is useful for pedagogical purposes.



 
> > Some suggestions for cosmetic changes:
> 
> > * Move the "Augmented Dickey-Fuller test" and "KPSS test" into a 
> > new submenu entry "Variable > Unit root test", as follows: 
> > "Variable > Unit root test > Augmented Dickey-Fuller test" and 
> > "Variable > Unit root test > KPSS test"
> 
> We're thinking of adding the DF-GLS test before long.  If/when we 
> do that it make sense to add another menu layer as you suggest.
> 
> > * When adding Logs of selected varaibles call it ln_x or log_x 
> > instead of l_x, in line with the use of sq_x. It makes it easier 
> > to see what the variable is.
> 
> The trouble with longer prefixes is that you tend to run out of 
> space, if people are using long variable names close to the 
> 15-character maximum.  There's also a backward compatibility 
> issue, for example scripts may be in use that rely on the "l_x" 
> pattern for auto-generated logs.


Would it be hard to increase the 15-charcter maximum?


> > * In the outputs decimals (,) are used as separators (e.g., 
> > 3,14), but for input one have to use a full stop (.) as 
> > separator (e.g., 3.14). This is somwehat unintuitive, and should 
> > be changed so that the same is used for output and input.
> 
> As others have said, we debated this issue at some length before. 
> The outcome was this:
> 
> 1. Use of the locale decimal separator for output is there, but 
> can be turned off under Preferences.
> 
> 2. Numbers given in scripts _must_ use '.' as decimal separator. 
> Jack and I argued that the alternative is just too complex and 
> error-prone.
> 
> 3. When numbers are entered via gui dialog boxes, the expectation 
> for the decimal separator is in agreement with the choice made 
> under point 1.  (And if this is not the case in some instances, 
> it's a bug.)
> 
> > Some other suggestions for new features: Add menu entries for 
> > calculating
> > *1-sample sign test
> > *1-sample Wilcoxon signed rank test
> > *2-sample Wilcoxon rank sum test
> 
> Not very high priority (for me), but could be added at some point.

Thanks for giving it som thoughts, anyway.

Regards and thanks for a great software,
Andreas