[Rasch] fit analysis

Mike Linacre (RMT) rmt at rasch.org
Mon Jun 5 13:05:30 EST 2006

Dear Mary:

You wrote: "N= 1054.  Mean square fit statistics for items are good, but 
the standardized stats for three items are bad (outfit standardized MNSQs 
of 3.8, -2.9, -5.4)."

Reply: Thank you for being detailed ... we can tell you what your 
mean-squares are from the plot on www.rasch.org/rmt/rmt171n.htm

t = 3.8, Mnsq = 1.2; t = -2.9, Mnsq = 0.9; t = -5.4, Mnsq =  0.8

Conclusion: your fit tests are too sensitive due to the large sample size. 
You have turned the fit "magnification" up so high that you can see every 
scratch on the pain of glass. The fit statistics tell you: "these data 
aren't perfect". The mean-squares tell you "but they are good enough for 
government work!" (As Ben Wright often used to say.)

The useful sample size range for the type of significance test you are 
using is 30 to 300 (or so).



Mike Linacre
Editor, Rasch Measurement Transactions
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailinglist.acer.edu.au/pipermail/rasch/attachments/20060604/31c79fbf/attachment.html 

More information about the Rasch mailing list