Hi,  Mike,

I am conducting a standard setting involve items and raters. I used theYes/No Angoff method.  Each rater was asked whether the items can be answered correctly by a borderline students.  If this item can be answered correctly by the borderline student, then the rater wrote yes, otherwise, wrote no.

I want to evaluate both item difficulty and rater's severity. Thus, I think the two facet model is suitable for my study and the FACET software was used.  My model is specified as follows:

ln(P1 /P0 )=B-D-W-T                
  P1= the probability of a ˇ¨Yesˇ¨ being rated on item i by panelist j.
  P0= the probability of a ˇ§Noˇ¨ being rated on item i by panelist j.
  B= panelistˇ¦s view of item difficulty,
  D= difficulty of item i,
  W= the severity of panelist j
  T= difficulty of rating a ˇ§Yesˇ¨ relative to ˇ§Noˇ¨.

However, when I submit this paper, I get the reviewer's comment as follows:

"With two facets you are not really using the many-facet Rasch model but a simple rating scale Rasch model, where people answer to items using either 1 or 0 (i.e. yes at this level they can answer the item correctly or no they cannot answer correctly); this analysis could have been performed with Rasch computer programs such as Winsteps or RUMM, which do not implement the many-facet Rasch model. "

I am wondering whether the Winsteps can perform two facet analysis?   In my case, can I still use FACET to conduct my analysis? or I have to use winsteps? Thanks!

Hm7523, Facets can definitely analyze your data, It can analyze almost every dataset that is analyzable with RUMM or Winsteps. Winsteps or RUMM are used in preference to Facets, where possible, because conceptualizing the dataset and communicating findings are usually easier when the dataset is a rectangle.

But some details  of your analysis need clarification.
1) With the usual definition of item difficulty, T is zero.
2) B and W appear to be the same thing, or else B and D are the same thing, or is B an interaction term between D and W?

In Winsteps, we would expect the main model to be:
ln(P1 /P0 )=B-D  where B = rater leniency and D = item difficulty
Then we would investigate interactions between B and D with a secondary analysis.

Hi, Mike,
   Thanks for your quick response. You are right, it seems that B and D are the same thing. If I change B to be the interaction term between D and W, should I still use Winsteps?
    I am just wondering whether there are any advantages using FACET over Winsteps in my case (so I can keep my original analysis results). Or maybe add one more facet (such as gender)?  

Hm7523, Winsteps and Facets give the same results for similarly configured analyses.
For interactions, both do a post-hoc analysis. In Facets, this is Table 13. In Winsteps, this is Table 30.

My advice: agree with your reviewer's statement that "this analysis could have been performed with ....", and then point out that your findings would have been the same.