CONVERGE= select convergence criteria = Either

Up  Previous  Next

This selects which of LCONV= and RCONV= set the convergence criterion. See convergence considerations.

 

CONVERGE=L                LCONV= for "Logit change size" controls convergence.

Iteration stops when the biggest logit change is less or equal to LCONV=, or when the biggest logit change size increases (divergence).

 

CONVERGE=R                RCONV= for "Residual size" controls convergence.

Iteration stops when the biggest residual score is less or equal to RCONV=, or when the biggest residual size increases (divergence).

 

CONVERGE=E                Either LCONV= for "Logit change size" or RCONV= for "Residual size" controls convergence.

Iteration stops when the biggest logit change is less or equal to LCONV=, or when the biggest residual score is less or equal to RCONV=, or when both the biggest logit change size increases and the biggest residual size increases (divergence).

 

CONVERGE=B                Both LCONV= for "Logit change size" and RCONV= for "Residual size" controls convergence.

Iteration stops when both the biggest logit change is less or equal to LCONV= and the biggest residual score is less or equal to RCONV=, or when both the biggest logit change size increases and the biggest residual size increases (divergence).

 

CONVERGE=F                Force both LCONV= for "Logit change size" and RCONV= for "Residual size" to control convergence.

Iteration stops when both the biggest logit change is less or equal to LCONV= and the biggest residual score is less or equal to RCONV=.

 

Example 1:        We want to be take a conservative position about convergence, requiring both small logit changes and small residual sizes when iteration ceases.

       CONVERGE=Both

 

Example 2:        We want to set the convergence criteria to match BIGSTEPS version 2.59

       CONVERGE=B        ; the rule was LCONV= and RCONV=

       RCONV= 0.5        ; the BIGSTEPS standards or whatever value you used

       LCONV= .01

 

Example 3:        We want to set the convergence criteria to match Winsteps version 3.20

       CONVERGE=E        ; the rule was LCONV or RCONV

       RCONV= 0.5        ; the 3.20 standards or whatever value you used

       LCONV= .01

 

Example 4:        We want the convergence criteria to match Winsteps version 2.85

       CONVERGE= F        ; force both LCONV and RCONV to be met

       RCONV= 0.5        ; the 2.85 standards or whatever value you used

       LCONV= .01

       You may also want:

       WHEXACT=NO        ; centralized Wilson-Hilferty was the default

 

Example 5:        Question: With anchored analyses, iterations never stop!

 

|----------------------------------------------------------------------------|

|    JMLE     MAX SCORE   MAX LOGIT    LEAST CONVERGED     CATEGORY    STEP  |

| ITERATION   RESIDUAL*    CHANGE    EXID   BYCASE   CAT   RESIDUAL   CHANGE |

|----------------------------------------------------------------------------|

|       1      -239.04       .5562    1993    392*      6      85.70   -.5960|

|       2      -105.65      -.1513    1993    392*      4     -28.92    .2745|

.....

|      18        -5.35      -.0027    2228    352*      3       2.35    .0146|

|      19        -5.16       .0029    2228    352*      3       2.31    .0106|

|      20        -5.05       .0025    2228    352*      3       2.28    .0055|

|      21        -5.00       .0010    2228    352*      3       2.26    .0075|

|      22        -4.99      -.0008    2228    352*      3       2.25    .0025|

.....

|     170        -5.00      -.0011    1377    352*      3       1.96    .0109|

|     171        -5.00       .0018     187    352*      3       1.96    .0019|

.....

 

The standard convergence criteria in Winsteps are preset for "free" analyses. With anchored analyses, convergence is effectively reached when the logit estimates stop changing in a substantively meaningful way. This has effectively happened by iteration 20. Note that the logit changes are less than .01 logits - i.e., even the biggest change would make no difference to the printed output (which is usually reported to 2 decimal places)

 

To have the current Winsteps do this automatically, set

       CONVERGE=L

       LCONV=.005   ; set to stop at iteration 22 - to be on the safe side.


Help for WINSTEPS® Rasch Measurement Software: www.winsteps.com.