[Rasch] A FACETS problem
PetroskiG at health.missouri.edu
Sat Jan 23 16:05:46 EST 2010
I've had FACETS for a couple of years ago but have not made use of. However we may have a problem that will justify the effort. However I am not sure and so thought I would seek comments form others with more Rasch experience.
The data consists of many ordinal items from a survey that asks staff about their work environment. There are several scales with names like Staff Coherence, Leadership, etc. As part of the demographic information respondents are asked for their job title which is collapsed to just 4 or 5 categories. The survey was also given to the same institutions at two points in time, but there is no way to link individual responses across time. For each institution there are multiple respondents per job title but obviously no individual "rates" any workplace but their current one.
Interest centers on if there are differences in perception of the workplace by staff type (job title), and is there a shift in perception over time (and possibly was that shift different for different staff types, i.e. a time by job-title interaction).
Ignoring job-title and using data from only the first survey occasion the items seem to have pretty good fit to the partial credit model -- actually surprisingly good fit. So this seems like a FACETS type problem but since employees rate only their own workplace it seems that measures of rater effect and work environment are confounded -- the data lacks the "connectedness" described in Linacre & Wright (2004).
The obvious alternative is to skip the Rasch stuff and used linear mixed-model methods with the observed scores, but I want to consider some alternative approaches first.
U. of Missouri - Columbia
Please consider the environment before you print
More information about the Rasch