Factor analysis is arguably not necessary but can be helpful.

Factor analysis and Rasch analysis typically have somewhat different assumptions and use somewhat different approaches, so you may get slightly different answers based on each approach. A purist might say that they are incompatible paradigms, except under very narrow circumstances (which you probably have to use confirmatory factor analysis with a number of constraints to achieve), but I think practically speaking they complement each other well.

I think that factor analysis is a nice way to get a quick look at the structure of the data - it is easy to do, usually interpretable, and it can be done with minimal computational power.

You can also apply multidimensional IRT, but you might first want to do factor analysis to refine the way you set up the mutldimensional IRT, especially given that multidimensional IRT might run slowly on your computer, especially with multiple dimensions using a maximum likelihood estimation method.

Having said that: 

I would probably do the factor analysis first. If you have a multidimensional outcome, then you cannot apply a unidimensional rasch model in a meaningful way. So, factor analysis after Rasch analysis for the purpose of item clustering is not worth doing, because if you find multiple clusters and really believe your factor analysis, then you will have to redo the Rasch analysis.

You can also use factor analysis to look for violations to Rasch analysis assumptions: e.g. local independence. Do the factor analysis and then look for high residual correaltions.

Based on the results of the factor analysis, you might decide how you want to set up the IRT model (multidimensional IRT, copulas to account for local dependence, etc...)

You can used Rasch Model to reduce item or remove outlier. The data will fit to Rasch Model.

Rasch analysis first or factor analysis first - thoughts at http://www.rasch.org/rmt/rmt114l.htm

Thanks all for the answers!

I am working with a multidimensional problem, and previously working with a Biostatisician, he was strongly in favour of Factor analysis rather than RASCH.....but now that I learned more I think that RASCH approach is better for the developping the PROs. I already know the dimensions of the problem because of the analysis of Phase 1 (Qualitative Study), so clustering the items in factors is not so useful as the RASCH parameters of difficulty and discriminant.

Anyone know about an online curse for IRT? or maybe, Anyone with skill on IRT analysis would like to co-author me in this work? If the answer is yes, please email me to [email protected]