How can I handle missing values when fitting IRT model? I am conducting a study for analyzing male involvement in Family planning. It is of interest to develop an index for the involvement of male. I am currently having difficulties fitting an IRT model since my data has a lot of missing values and answers like "don't know" and some confusing answers. Note that my variables are all dichotomous in nature. 
Any suggestions for handling this kind of problem regarding missing values in IRT model?Thanks
 A: Item Response Theory is perfect for handling data with missing values.  In classical test theory, scores tend to be summed and subjects not answering questions wind up in their observations being removed.  IRT and Rasch programs are designed to handle missing data.  As the analysis is based on the probability of a subject with a determined ability selecting the response, a missing response just does not contribute to the item measure.
The more blank responses a subject submits, the greater will be the size of the standard error associated with that subject's ability estimate.  Higher missing observations for items will also increase the standard error associated with the difficulty of those items. It is up to the analyst to determine when the standard errors are too high to be useful.  
With your dichotomous scoring, correct answers should be scores a 1, incorrect answers a 0, and missing answers as blank (or whatever value the software is programmed to accept for missing).  The missing cells will not contribute to the estimates; that is to say, they will not be scored as a zero.    
Hope this helps!
