\magnification=1200
\baselineskip=20pt
\nopagenumbers
\font\big=cmr12 scaled \magstep2

\centerline{\bf STANFORD UNIVERSITY}
\centerline{\bf DEPARTMENT OF STATISTICS}
\centerline{\big DEPARTMENTAL SEMINAR}
\bigskip

\baselineskip=12pt
\centerline{4:15 p.m., Tuesday, March 15, 2005}
\centerline{Sequoia Hall Room 200}
\centerline{(Cookies at 3:45 in 1st Floor Lounge)}

\bigskip
\baselineskip=15pt
\centerline{\sl Simon Jackman}
\centerline{Department of Political Science}
\centerline{Stanford University}
\bigskip
\centerline{\bf Generalized Latent Variable Modeling}
\centerline{(with an application assessing applicants to a Ph.D.~program)}
\bigskip

Abstract:

Latent variables are ubiquitious in the social sciences: micro-level
examples include traits such as ideology, partisanship, and
intelligence, while macro-level examples include support for political
candidates or a regime, levels of democracy, or military or terrorist
threat.  In many applications, interest focuses on not only the latent
variables, but also on assessing the quality of the indicators of the
latent variables.  I state the problem in general terms, and adopt a
Bayesian approach to estimation and inference.  The approach adopted
here overcomes the restrictions imposed by traditional methods for
multivariate analysis used in the social-sciences and is easily
extended in interesting directions (e.g., inferences over rank
orderings of the latent variables).  The application considered in the
talk is the problem of assessing 279 applicants to a political science
Ph.D. program: ``applicant quality'' is the latent trait, with GRE
scores and an eight-member admissions committee's (incomplete) ordinal
ratings serving as the indicators, with the goal being to (more or
less) authoritatively determine which subset of applicants to admit to
the Ph.D. program.
\bye