Activity Number:
|
45
|
Type:
|
Contributed
|
Date/Time:
|
Sunday, August 11, 2002 : 4:00 PM to 5:50 PM
|
Sponsor:
|
Social Statistics Section*
|
Abstract - #301434 |
Title:
|
Latent Class Models with Continuous and Categorical Indicators
|
Author(s):
|
Brian Flaherty*+ and Joseph Schafer and Linda Collins and Nancy Darling
|
Affiliation(s):
|
Pennsylvania State University and Pennsylvania State University and Pennsylvania State University and Pennsylvania State University
|
Address:
|
S-159 Henderson Bldg., University Park, Pennsylvania, 16802, USA
|
Keywords:
|
EM algorithm ; general location model ; latent class ; missing data ; mixed variables ; parenting style
|
Abstract:
|
The traditional latent class model assumes conditional independence among a set of observed categorical indicators given latent class membership. We extend the latent class model by including continuous indicators. The model is a special case of the general location model. The latent class variable is treated as ignorable missing data, and parameters are estimated via an EM algorithm. Continuous indicators are modeled as conditionally independent and normally distributed given latent class membership. The means of the continuous indicators are freely estimated within each latent class, but a single, diagonal covariance matrix of residuals is assumed across the latent classes. The procedure accommodates arbitrary patterns of missing data on all indicators and allows one to test for violations of conditional independence. The model is illustrated with an analysis of parenting style in a sample of U.S. high school students. Parenting style is a typology characterizing how parents interact with their children. Data measuring parenting style are typically analyzed with factor analysis. The model presented here provides a more intuitive analytic method for this construct.
|