reference group + interaction

Welcome to the forum for runmlwin users. Feel free to post your question about runmlwin here. The Centre for Multilevel Modelling take no responsibility for the accuracy of these posts, we are unable to monitor them closely. Do go ahead and post your question and thank you in advance if you find the time to post any answers!

Go to runmlwin: Running MLwiN from within Stata >> http://www.bristol.ac.uk/cmm/software/runmlwin/
Forum rules
Post Reply
yfgeogcmm
Posts: 2
Joined: Mon May 19, 2014 9:12 pm

reference group + interaction

Post by yfgeogcmm »

Hello George or Chris,
I tried to use runMLwin for my research (house price), but have some basic questions to ask you - I have searched to find the answer but could not quite find where I have done wrong

1) I have following the instruction to change my reference category using the following command, but it does not seem it omitted the defined catergory though.
char type[omit] 4
char new[omit] 2

2) I wish to put interaction of categorical variable (eg. type and old/new house). I remembered that we could just put the a*b, and the main effect will be correctly specified (without duplication). I therefore used the following command, but I have an error message saying "_Itype_2 _Itype_3 _Itype_4 _Inew_2 _Iduration_2 are duplicated in the fixed part". My command is as below

xi: runmlwin logeprice cons i.type*i.new i.type*i.duration i.new*i.duration level2(neighbourhood:cons) level1(house:cons) nopause

Thank you for your help!
Top
ChrisCharlton
Posts: 1384
Joined: Mon Oct 19, 2009 10:34 am

Re: reference group + interaction

Post by ChrisCharlton »

1) I'm not sure why the omit syntax isn't working for you as the following seems to work as expected:

Code: Select all

. use http://www.bristol.ac.uk/cmm/media/runmlwin/tutorial, clear

. xi: runmlwin normexam cons i.schgend*i.vrband, level2(school: cons) level1(student: cons) nopause
i.schgend         _Ischgend_1-3       (naturally coded; _Ischgend_1 omitted)
i.vrband          _Ivrband_1-3        (naturally coded; _Ivrband_1 omitted)
i.sch~d*i.vrb~d   _IschXvrb_#_#       (coded as above)
 
MLwiN 2.31 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)   =       1.45
Number of iterations =          3
Log likelihood       = -4816.6846
Deviance             =  9633.3691
------------------------------------------------------------------------------
    normexam |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        cons |   .5713802   .0601215     9.50   0.000     .4535441    .6892163
 _Ischgend_2 |   .1975847    .129877     1.52   0.128    -.0569695     .452139
 _Ischgend_3 |   .3030197   .0994147     3.05   0.002     .1081706    .4978689
  _Ivrband_2 |  -.7717841    .038821   -19.88   0.000    -.8478719   -.6956962
  _Ivrband_3 |  -1.572875   .0568076   -27.69   0.000    -1.684215   -1.461534
_IschXvr~2_2 |  -.1270987   .0886994    -1.43   0.152    -.3009464    .0467491
_IschXvr~2_3 |  -.2003226   .1248478    -1.60   0.109    -.4450198    .0443745
_IschXvr~3_2 |  -.0974494   .0625693    -1.56   0.119    -.2200829    .0251842
_IschXvr~3_3 |  -.0409024   .0941537    -0.43   0.664    -.2254402    .1436355
------------------------------------------------------------------------------

------------------------------------------------------------------------------
   Random-effects Parameters |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
Level 2: school              |
                   var(cons) |   .0883445   .0175274      .0539913    .1226977
-----------------------------+------------------------------------------------
Level 1: student             |
                   var(cons) |   .6065081   .0135719      .5799077    .6331085
------------------------------------------------------------------------------

. char schgend[omit] 3

. xi: runmlwin normexam cons i.schgend*i.vrband, level2(school: cons) level1(student: cons) nopause
i.schgend         _Ischgend_1-3       (naturally coded; _Ischgend_3 omitted)
i.vrband          _Ivrband_1-3        (naturally coded; _Ivrband_1 omitted)
i.sch~d*i.vrb~d   _IschXvrb_#_#       (coded as above)
 
MLwiN 2.31 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)   =       1.48
Number of iterations =          3
Log likelihood       = -4816.6846
Deviance             =  9633.3691
------------------------------------------------------------------------------
    normexam |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        cons |   .8743999    .079175    11.04   0.000     .7192198     1.02958
 _Ischgend_1 |  -.3030197   .0994147    -3.05   0.002    -.4978689   -.1081706
 _Ischgend_2 |  -.1054349   .1397216    -0.75   0.450    -.3792842    .1684143
  _Ivrband_2 |  -.8692334   .0490698   -17.71   0.000    -.9654085   -.7730584
  _Ivrband_3 |  -1.613777   .0750854   -21.49   0.000    -1.760942   -1.466612
_IschXvr~1_2 |   .0974494   .0625693     1.56   0.119    -.0251842    .2200829
_IschXvr~1_3 |   .0409024   .0941537     0.43   0.664    -.1436355    .2254402
_IschXvr~2_2 |  -.0296493   .0936395    -0.32   0.752    -.2131794    .1538808
_IschXvr~2_3 |  -.1594203   .1341554    -1.19   0.235    -.4223601    .1035196
------------------------------------------------------------------------------

------------------------------------------------------------------------------
   Random-effects Parameters |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
Level 2: school              |
                   var(cons) |   .0883445   .0175274      .0539913    .1226977
-----------------------------+------------------------------------------------
Level 1: student             |
                   var(cons) |   .6065081   .0135719      .5799077    .6331085
------------------------------------------------------------------------------
2) The documentation for xi (see http://www.stata.com/help.cgi?xi) indicates that the syntax i.varname1*i.varname2 will create dummies for all main effects and interactions of these categories. As your categorical variables are all involved in more than one interaction it is generating variables for the main effects more than once, hence the error message that you are seeing. If your second variable was continuous you could avoid duplicating the main effect with the i.varname1|varname3 syntax, but I don't see an equivalent syntax for categorical variables.

runmlwin does support the newer factor variable syntax (see http://www.stata.com/help.cgi?fvvarlist) for univariate models, which I believe does allow what you are wanting. Here is the second model from above fitted using this syntax:

Code: Select all

. runmlwin normexam cons i.schgend i.vrband i.schgend#i.vrband, level2(school: cons) level1(student: cons) nopause
 
note: 1b.schgend omitted because of collinearity
note: 1b.vrband omitted because of collinearity
note: 1b.schgend#1b.vrband omitted because of collinearity
note: 1b.schgend#2o.vrband omitted because of collinearity
note: 1b.schgend#3o.vrband omitted because of collinearity
note: 2o.schgend#1b.vrband omitted because of collinearity
note: 3o.schgend#1b.vrband omitted because of collinearity
MLwiN 2.31 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)   =       1.84
Number of iterations =          3
Log likelihood       = -4816.6846
Deviance             =  9633.3691
------------------------------------------------------------------------------
    normexam |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        cons |   .5713802   .0601215     9.50   0.000     .4535441    .6892163
  _2_schgend |   .1975847    .129877     1.52   0.128    -.0569695     .452139
  _3_schgend |   .3030197   .0994147     3.05   0.002     .1081706    .4978689
   _2_vrband |  -.7717841    .038821   -19.88   0.000    -.8478719   -.6956962
   _3_vrband |  -1.572875   .0568076   -27.69   0.000    -1.684215   -1.461534
_2_schgend~d |  -.1270987   .0886994    -1.43   0.152    -.3009464    .0467491
_2_schgend~d |  -.2003226   .1248478    -1.60   0.109    -.4450198    .0443745
_3_schgend~d |  -.0974494   .0625693    -1.56   0.119    -.2200829    .0251842
_3_schgend~d |  -.0409024   .0941537    -0.43   0.664    -.2254402    .1436355
------------------------------------------------------------------------------

------------------------------------------------------------------------------
   Random-effects Parameters |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
Level 2: school              |
                   var(cons) |   .0883445   .0175274      .0539913    .1226977
-----------------------------+------------------------------------------------
Level 1: student             |
                   var(cons) |   .6065081   .0135719      .5799077    .6331085
------------------------------------------------------------------------------
I believe that the following syntax should be what you require for your model:

Code: Select all

runmlwin logeprice cons i.type i.new i.duration i.type#i.new i.type#i.duration i.new#i.duration level2(neighbourhood:cons) level1(house:cons) nopause
or if you want to use ## to avoid specifying a couple of the main effects:

Code: Select all

runmlwin logeprice cons i.type##i.new i.duration i.type#i.duration i.new#i.duration level2(neighbourhood:cons) level1(house:cons) nopause
This also allows an alternative method for specifying the base category via the ib#. syntax.
Top
Post Reply

Return to “runmlwin user forum”