Factor variable and interaction in syntax command

[Adrien Bouguen]

Hi,

I am trying to add interaction terms into a simple new command. Something simple like this:

cap program drop newreg
program define newreg
syntax [varlist],  CONTrol(varlist fv)

xi: reg `varlist' `control'

end

Sine Stata 11 I think factor variable (fv) accept interactions term. Yet when I do:

newreg y x, cont(mu##nu)

or

newreg y x, cont(mu#nu)

with mu and nu numerical and categorical variable, I get the following message:

interactions not allowed. Surprisingly the command

newreg y x, cont(c.mu##c.nu)

runs but does consider mu and nu as continuous.

Similarly,

newreg y x, cont(i.mu*i.nu)

gives me:

variable mu*i.nu not found
(error in option control())

Any idea on how to specify interaction in a syntax command?

Thanks

[William Lisowski]

The problem is not with your syntax command but with the use of the xi: prefix

. sysuse auto, clear
(1978 Automobile Data)

. xi: reg price weight i.foreign##c.weight
interactions not allowed
r(101);

. reg price length i.foreign##c.weight

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(4, 69)        =     24.25
       Model |   371093867         4  92773466.8   Prob > F        =    0.0000
    Residual |   263971529        69  3825674.33   R-squared       =    0.5843
-------------+----------------------------------   Adj R-squared   =    0.5602
       Total |   635065396        73  8699525.97   Root MSE        =    1955.9

----------------------------------------------------------------------------------
           price |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-----------------+----------------------------------------------------------------
          length |  -96.49829   31.80898    -3.03   0.003    -159.9555    -33.0411
                 |
         foreign |
        Foreign  |  -2765.136   2684.043    -1.03   0.307    -8119.655    2589.382
          weight |   5.556982   .9319023     5.96   0.000     3.697887    7.416076
                 |
foreign#c.weight |
        Foreign  |    2.58161   1.063848     2.43   0.018     .4592919    4.703928
                 |
           _cons |   6565.929   3687.215     1.78   0.079    -789.8649    13921.72
----------------------------------------------------------------------------------

Is there a particular reason you are using the xi: prefix in your program? In Stata 16.1 the output of help xi tells us to use factor variable notation if a command allows it, and the reg command does, as the example above demonstrates.

[William Lisowski]

To those who come across this topic in the future, the problem was also discussed on the author's previous topic at

https://www.statalist.org/forums/for...syntax-command

where the discussants made essentially the same point about the irrelevance of the xi: prefix in most circumstances. Also, the output of help xi tells us that the "#" notation is now how interactions are specified when using the xi: prefix.

[Adrien Bouguen]

I understand. Thanks. I use xi: cause I wanted to be able to have the possibility to include fixed effects in my control(varlist fv ) option. It does not seem entirely coherent that xi: and ## would not work together.

[William Lisowski]

You do not need xi: to have fixed effects. If nu is a categorical variable, i.nu will give the fixed effects for the categories defined by the values of nu. xi: is old syntax that was made obsolete for most commands by the introduction of factor variables notation, which xi: was not designed to understand.

[Adrien Bouguen]

Oh yeah you are right! I am just old school I guess. Thanks for the help. Appreciate it.