Using $\ell$EM

• $\ell$EM is a program for loglinear and related analysis of categorical data.
• Written by Jeroen Vermunt, Tilburg University.
• Freely available off the WWW, at http://www.kub.nl/faculteiten/fsw/organisatie/departementen/mto/software2.htmlP>
• It fits a very wide variety of models, from simple loglinear to log-multiplicative to latent class models. For anyone wishing to push beyond the bounds of SPSS it is worth examining carefully.

• It runs under Windows with a simple interface: a short syntax file is executed and the results are shown in another window.

• The format of the syntax file is straightforward, at least for simple models:

  man 2          * 2 'manifest' variables
  dim 3 3        * N cats, fastest changing last
  lab S V        * labels, default to A, B, . . .
  mod {S,V}      * Fit the main effects
  dat [43  6  9  * Data as a table.
       16 11 18
        3 10 16]
  

  This fits independence to the hospital visit data used in the practicals.

• This is the death penalty example, set up for the model with all the two way interactions:

  man 3
  dim 2 2 2
  lab V, D, P
  MOD {PV,VD,PD}
  dat [ 53  414 
        11   37 
         0   16 
         4  139]
  

• This is the VOTE*HIED*AGEG table used for the logit models in the previous session. The fastest changing variable is HIED followed by AGEG with VOTE changing last (the command that created these data was crostabs hied by ageg by vote). It fits the model with all two-way interactions.

  man 3
  dim 4 7 2
  dum 4 7 2
  lab V A E
  mod {VA,EA,VE}
  dat [ 9      75
        35     322
        64     349
        89     332
       149     257
     . . .   . . .
       113     197
        77     108
        58      59
        40      18
        54      12]
  

• The dum 4 7 2 statement over-rides $\ell$EM's usual effect coding of parameters and forces it to use dummy variables, with the last value as reference category (i.e., like SPSS).

• The dim statement defines the classifying variables according to the order of the counts: given the following order:

           c1  c2  c3
  l1  r1  111 112 113
      r2  121 122 123
      r3  131 132 143
      r4  141 142 133
  l2  r1  211 212 213
      r2  221 222 223
      r3  231 232 243
      r4  241 242 233
  

  the appropriate definitions would be:

  man 3
  dim 2 4 3
  lab l r c