For generating pseudo-random numbers, used in simulating data, Winsteps implements a FORTRAN 77 adaptation of this "Mersenne Twister" code:

 

c  A C-program for MT19937, with initialization improved 2002/1/26.

c  Coded by Takuji Nishimura and Makoto Matsumoto.

c  Copyright (C) 1997 - 2002, Makoto Matsumoto and Takuji Nishimura,

c  All rights reserved.

c  Copyright (C) 2005, Mutsuo Saito,

c  All rights reserved.

 

Their website is www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html

 

The generator is seeded with the system time or a value chosen by the analyst.

 

Simulated data observations.

 

Winsteps simulates data using the estimated values of person ability, item difficulty and rating-scale thresholds. Winsteps implements www.rasch.org/rmt/rmt213a.htm for dichotomous and polytomous data.

 

If you use these simulated data to estimate the parameter values, then the estimated values will be somewhat wider than the generators. This is due to JMLE bias. It can usually be corrected with STBIAS=YES.