st: factor variables vs xi

[Richard Palmer-Jones <rpjstatalist@googlemail.com>]

Dear Readers

I came across a problem using factor variables (with the prais comand).

I am wondering if the problem is that the problem occurs because the
variable from which you wish to produce factors has laerge numeric
values the i.var code fails even though it works with the older xi
i.var code and when I recode the large values to smallere ones.

I can't reproduce this using auto.dta so here are sections of the output.


* using xi
xi i.vilcd i.year

. .vilcd           _Ivilcd_330603001-332708149(naturally coded;
_Ivilcd_330603001 omitted)
. i.year            _Iyear_2002-2007    (naturally coded; _Iyear_2002 omitted)

prais  depvar ....   _Iyear_*     Ivilcd_*   .....


[output]

_Iyear_2003 |  -.3848006   .1138329    -3.38   0.001    -.6079923    -.161609
 _Iyear_2004 |  -.1880117   .0858427    -2.19   0.029    -.3563231   -.0197003
 _Iyear_2005 |   -.224122   .0593118    -3.78   0.000    -.3404144   -.1078295
_Iyear_2006 |  -.2366355   .0360867    -6.56   0.000    -.3073905   -.1658805
_Iyear_2007 |          0  (omitted)
....
_Ivilcd_330603003 |   13.83685   6.157102     2.25   0.025
1.764646    25.90906
_Ivilcd_330603004 |    11.0135   5.747892     1.92   0.055
-.2563655    22.28337
_Ivilcd_330603005 |  -2.277502   47.13286    -0.05   0.961
-94.69072    90.13571
_Ivilcd_330603006 |   5.394073    14.7922     0.36   0.715
-23.60894    34.39709
_



* using factor variables
. prais depvar .... i.year     i.vilcd   ...> lndist distyear

[output]

....
          year |
         2004  |   .1021677   .0276891     3.69   0.000     .0478827    .1564528
         2005  |  -.0220684    .051486    -0.43   0.668    -.1230076    .0788708
         2006  |  -.1549629   .0745452    -2.08   0.038    -.3011101   -.0088157
         2007  |  -.1560546   .0948655    -1.65   0.100    -.3420401     .029931
....
         vilcd |
    330603003  |          0  (omitted)
    330603004  |          0  (omitted)
    330603005  |          0  (omitted)
    330603006  |          0  (omitted)


* vilcd does not work with factor variables ?cos values too large?
gen villn = .
qui tab vilcd, matrow(row) matcell(cell)
forval i = 1/`r(r)' {
	qui replace villn = `i' if row[`i', 1] == vilcd
}

prais  depvar .... i.year   i.villn ....> lndist distyear

[output]

....
          year |
         2004  |   .1967895   .0342813     5.74   0.000     .1295743    .2640047
         2005  |   .1606798    .061101     2.63   0.009     .0408793    .2804804
         2006  |   .1481668   .0886042     1.67   0.095    -.0255591    .3218928
         2007  |   .3848028    .113833     3.38   0.001     .1616109    .6079946

.....
         villn |
            2  |   13.83682   6.157102     2.25   0.025     1.764615    25.90903
            3  |   11.01349   5.747892     1.92   0.055    -.2563812    22.28336
            4  |  -2.277797   47.13286    -0.05   0.961    -94.69101    90.13542
            5  |    5.39397    14.7922     0.36   0.715    -23.60905    34.39699
            6  |    18.0429   19.47698     0.93   0.354    -20.14554    56.23134
            7  |   19.83282   9.599951     2.07   0.039     1.010231    38.65541

.....

I took me a little time to work out "the" soultion, but it may be of
help to others

Cheers

Richard
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/