JMLE iterations may take a long time for big data sets, so initially set this to -1 for no JMLE iterations. Then set MJMLE= to 10 or 15 until you know that more precise measures will be useful. The number of PROX iterations, MPROX=, affects the number of JMLE iterations but does not affect the final estimates.

 

MJMLE= (or MUCON=) specifies the maximum number of JMLE iterations to be performed. Iteration will always cease when both LCONV= and RCONV= criteria have been met, see CONVERGE=. To specify no maximum number limitation, set MJMLE=0. Iteration always be stopped by Ctrl with F, see "Stopping Winsteps".

 

Example 1: To allow up to 4 iterations in order to obtain rough estimates of a complex rating (or partial credit) scale:

  MJMLE=4 4 ; JMLE iterations maximum

 

Example 2: To allow up to as many iterations as needed to meet the other convergence criteria:

  MJMLE=0 ; Unlimited JMLE iterations

 

Example 3: Perform no JMLE iterations, since the PROX estimates are good enough.

  MJMLE=-1 ; No JMLE iteration

 

Example 4: Run as quick estimation as possible to check out control options.

  MPROX=-1 ; minimal prox estimation iterations

  MJMLE=-1 ; no JMLE iterations