TAM R Statistics Rasch analysis software and Winsteps

TAM is an R Statistics package with some of the capabilities of Winsteps.

 

Look at the item estimates in https://cran.r-project.org/web/packages/TAM/TAM.pdf section: tam.jml Example 1.

 

 

TAM item estimates are not zero-centered. TAM Zero column adjusts the TAM item estimates to be zero centered. These are almost the same as Winsteps item estimates, which are zero-centered. The difference is probably due to slightly different implementation of JMLE and the convergence criteria.

 

The person "measure" in Winsteps is the TAM theta value. The range of theta estimates in TAM & WINSTEPS are

 

  TAM  original    zero   Winsteps

score 0: -4.404   -4.35  -4.35

score 5: -0.070   -0.02  -0.01

score 10: 4.315    4.37  -4.38

 

The TAM Zero column applies the TAM Zero item adjustment to the TAM thetas. The TAM Zero thetas and Winsteps person measures are essentially the same.

 

So, TAM and Winsteps estimates differ mainly because:

 

(1) Estimation bias correction: in Winsteps set

STBIAS=Yes

 

(2) Winsteps zero-centers the items: to zero-center the person thetas:

UPMEAN=0

UEXTREME=Yes