categorical predictors in the random part

[marjolijndewilde]

Hi,

How should I include categorical predictor variables in the random part of my model? Just entering “i.variable” gives the following message: i: operator invalid. Do I need to dumify the variables that I want to enter? Does this gives the same result as the procedure in MLwiN?

Thank you very much in advance,
Marjolijn

[ChrisCharlton]

Using the i.variable syntax in runmlwin will just create the dummies automatically under the covers, so using dummies that you create yourself should be equivalent. Alternatively if you prefix your command with:

Code: Select all

xi:

then the i.variable syntax should work in the random part too.

[GeorgeLeckie]

Hi Marjolijn,

Here is an example of Chris' solution

Commands:

Code: Select all

* Load the data
use "http://www.bristol.ac.uk/cmm/media/runmlwin/tutorial.dta", clear

* Convert continuous standlrt into a 3-category oridnal variable called pass
generate pass = 2

replace pass = 1 if standlrt<-.75

replace pass = 3 if standlrt>.75

* Tabulate pass
tabulate pass

* Fit model with pass dummies and incorporate random slopes for the pass dummies
xi: runmlwin normexam cons standlrt i.pass, ///
	level2(school: cons i.pass) ///
	level1(student: cons) nopause

Output:

Code: Select all

. * Load the data
. use "http://www.bristol.ac.uk/cmm/media/runmlwin/tutorial.dta", clear

. 
. * Convert continuous standlrt into a 3-category oridnal variable called pass
. generate pass = 2

. 
. replace pass = 1 if standlrt<-.75
(876 real changes made)

. 
. replace pass = 3 if standlrt>.75
(890 real changes made)

. 
. * Tabulate pass
. tabulate pass

       pass |      Freq.     Percent        Cum.
------------+-----------------------------------
          1 |        876       21.58       21.58
          2 |      2,293       56.49       78.07
          3 |        890       21.93      100.00
------------+-----------------------------------
      Total |      4,059      100.00

. 
. * Fit model with pass dummies and incorporate random slopes for the pass dummies
. xi: runmlwin normexam cons standlrt i.pass, ///
>         level2(school: cons i.pass) ///
>         level1(student: cons) nopause
i.pass            _Ipass_1-3          (naturally coded; _Ipass_1 omitted)
 
MLwiN 2.29 multilevel model                     Number of obs      =      4059
Normal response model
Estimation algorithm: IGLS

-----------------------------------------------------------
                |   No. of       Observations per Group
 Level Variable |   Groups    Minimum    Average    Maximum
----------------+------------------------------------------
         school |       65          2       62.4        198
-----------------------------------------------------------

Run time (seconds)   =       2.06
Number of iterations =          5
Log likelihood       = -4658.0581
Deviance             =  9316.1162
------------------------------------------------------------------------------
    normexam |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        cons |   -.033233   .0564554    -0.59   0.556    -.1438836    .0774176
    standlrt |   .5294663   .0265595    19.94   0.000     .4774106     .581522
    _Ipass_2 |   .0079123   .0546539     0.14   0.885    -.0992075     .115032
    _Ipass_3 |   .0937119   .0903775     1.04   0.300    -.0834248    .2708485
------------------------------------------------------------------------------

------------------------------------------------------------------------------
   Random-effects Parameters |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
Level 2: school              |
                   var(cons) |   .0734045   .0208721      .0324959    .1143131
          cov(cons,_Ipass_2) |  -.0126694   .0153444     -.0427439    .0174051
               var(_Ipass_2) |   .0439243   .0182607      .0081339    .0797146
          cov(cons,_Ipass_3) |  -.0110608   .0205914     -.0514193    .0292977
      cov(_Ipass_2,_Ipass_3) |   .0689854   .0232667      .0233835    .1145874
               var(_Ipass_3) |   .1117496   .0360596      .0410741    .1824251
-----------------------------+------------------------------------------------
Level 1: student             |
                   var(cons) |   .5538607   .0125606      .5292423    .5784791
------------------------------------------------------------------------------

[marjolijndewilde]

Thank your so much!

The "xi"-solution, however, gave me a new error-message saying "r(101) interaction not allowed". I searched the web and found that this means that a command (I supose the xi-command) isn't adjusted at the stata-version I use. Inculding 'version 10.1" didn't solve the problem. Sorry for bothering you with this new pure-stata-problem.

I also tried the syntax with dummified variables and this did worked. So thank you!

Best regards,
Marjolijn

[GeorgeLeckie]

Hi Marjolijn,

Note that if you specify interactions then you will need to do more than just add the xi prefix when moving from Stata's current notation to their historic notation

See help xi for examples of how to specify interactions

If runmlwin still throws up an error then please post the do-file and dataset to replicate that error.

Best wishes

George

[marjolijndewilde]

Thank you very much!!
Marjolijn