On Tue, Jul 29, 2008 at 4:44 PM, Frederik Hjorth
<[log in to unmask]> wrote:
> I am working with data from a survey of Danish MP's. The data consists of
> answers from each MP on about two dozen items. Each item consists of a
> statement to which the respondent places him/herself according to a classic
> five-point symmetric Likert scale (strongly agree, agree, neutral, etc.).
> The data thus differ from roll call data in significant ways: there are no
> NA's, and the answers contain more information than an up/down vote. By
> design each statement is relatively straightforward to classify as left-wing
> or right-wing. What I am looking for is a way to use this data to estimate
> the ideological ideal points of the MP's surveyed.
>
> My question to the PolMeth list, then, is: what would be the best way to use
> this data for ideal point estimation?
>
> Any inputs are greatly appreciated - I have virtually no experience with
> ideal point estimation, so references to good introductory texts are also
> very much welcome. It is a plus if an approach has an associated R package.
This would appear to be an ordinal factor analysis problem that you
could tackle using MCMCpack's MCMCordfactanal procedure. The
legislators' responses are the observed manifest variables; fit a
single factor and the posterior factor scores for each legislator will
correspond with their expected ideal points.
Now, if you sit down with the R source of MCMCpack, you'll find
something very interesting: MCMCirtKd simply calls MCMCordfactanal to
do the work. So if you want to make this easier to wrap your brain
around as being an IRT problem, just use MCMCirtKd, even though it
says x has to be 0,1, or missing(!).
For an extended discussion, see the workshop of AJPS 52(1) for Treier
and Jackman's piece that applies the methodology (albeit with more
sophistication than the "canned" MCMCpack procedure, to deal with
local independence problems in the indicators - a problem you
shouldn't have in your data) to POLITY:
http://www3.interscience.wiley.com/journal/119397791/abstract
Chris
--
Christopher N. Lawrence, Ph.D. <[log in to unmask]>
Assistant Professor of Political Science
Texas A&M International University
313 LBVSC, 5201 University Blvd
Laredo, Texas 78041-1920
Website: http://www.cnlawrence.com/
**********************************************************
Political Methodology E-Mail List
Editors: Melanie Goodrich, <[log in to unmask]>
Delia Bailey, <[log in to unmask]>
**********************************************************
Send messages to [log in to unmask]
To join the list, cancel your subscription, or modify
your subscription settings visit:
http://polmeth.wustl.edu/polmeth.php
**********************************************************
|