Scores increment in integers so that 0.1 is about as precise a recovery of observed data as can be hoped for.

 

Specifies what value the largest score residual, corresponding to any person measure or item calibration, must be less than in the iteration just completed for iteration to cease. The current largest value is listed in Table 0, and displayed on your screen. In large data sets, the smallest meaningful logit change in estimates may correspond to score residuals of several score points. See convergence considerations.

 

The standard setting is CONVERGE="E", so that iteration stops when either LCONV= or RCONV= is satisfied. (Note: this depends on Winsteps version - and may explain differences in converged values.)

 

Example: To set the maximum score residual, when convergence will be accepted, at 5 score points and maximum logit change in estimates of .01 logits. Your data consists of the responses of 5,000 students to a test of 250 items.

  NI=250  ; 250 items

  RCONV=5 ; score residual convergence at 5 score points

  LCONV=.01 ; this is the standard.

  CONVERGE=Both ;convergence when both RCONV= and LCONV= are met