Some of my statistical papers, presentations and computer programs.

Articles and Chapters

• Uebersax, J. S. (1988). Validity inferences from interobserver agreement. Psychological Bulletin, 1988, 104, 405-416.

• Uebersax, J. S., Grove, W. M. (1990). Latent class analysis of diagnostic agreement. Statistics in Medicine, 9, 559-572. [request reprint]

• Uebersax, J. S. (1992). A review of modeling approaches for the analysis of observer agreement. Investigative Radiology, 17, 738-743. [1]

• Uebersax, J. S. (1993). Statistical modeling of expert ratings on medical treatment appropriateness. Journal of the American Statistical Association, 88, 421-427. [2]

• Uebersax, J. S., & Grove, W. M. (1993). A latent trait finite mixture model for the analysis of rating agreement. Biometrics, 49, 823-835. [3]

• Uebersax, J. S. (1997). Analysis of student problem behaviors with latent trait, latent class, and related probit mixture models. In: Rost J, Langeheine R, eds. Applications of Latent Trait and Latent Class Models in the Social Sciences. New York, NY: Waxmann; 1997:188-195.

• Uebersax, J. S. (1999). Probit latent class analysis: conditional independence and conditional dependence models. Appl Psychol Measmt, 23(4), 283-297, 1999. [request reprint]

• Uebersax, J. S. (2004). Breast cancer risk modeling: an application of Bayes networks. Technical Report. RavenPack Intnl.

Reprints by kind permission of [1] Wolters-Kluwer Publishers, [2] the American Statistical Association, and the [3] the International Biometric Society.

Presentations

• Uebersax, J. S. (1994). On the dimensionality of a latent class model. Paper presented at the Annual Meeting of the Classification Society of North America. Summary: principal components or factor analysis of probit-transformed conditional response probabilities lets one:
  • assess the dimensionality of the latent classes;
  • plot latent classes and manifest variables in 1 to K dimensions;
  • approximate a discrete latent trait model model; and
  • interpret and/or diagnose the LCA solution.

Web Articles

These are mostly online articles and do not include my other articles publications.

• The tetrachoric and polychoric correlation coefficients
• Estimation of a latent trait model by factor analysis of tetrachoric correlation coefficients.
• Bayesian unconditional power analysis   Power estimation for clinical trials often chooses as the alternative hypothesis the effect size observed in a previous study (e.g., a Phase 2 trial). This is a bad idea -- it neglects to consider the effect of sampling variation on the previously observed effect size. Understimation, often substantial, of actual statistical power is typically the result. This paper examines the nature of the problem and describes methods to estimate true statistical power. Every statistician involved with power estimation and sample size choice for clinical trials needs to understand this issue. (July 2007)
• Likert scales: dispelling the confusion.   What are Likert scales and Likert-type items? What aren't? (June 2006)
• A practical guide to local dependence in latent class models
• A brief study of local maximum solutions in latent class analysis.
• Rasch models FAQ

An old FAQ on Rasch models and software. Also has information on IRT and latent trait models.

• Get the most from Command Prompt!   Most of my programs run in a Command Prompt (DOS) window. This explains how to use Command Prompt. Command Prompt isn't an anachronism or second-rate in any way. It's powerful and efficient--like having unix on your PC--and easy to use. (June 2006)

Programs

• LTA-2 (zipped)

Latent Trait Analysis and IRT modeling with dichotomous items or manifest variables. This program will estimate both normal-ogive and logistic-ogive IRT models using the marginal maximum likelihood method. Bayesian (expected a posteriori; EAP) scores are calculated for each respondent or response pattern. The program can be used to estimate a one-factor common factor analysis model for dichotomous items. The program is very user friendly.

• LLCA (zipped)

Located Latent Class Analysis---a discrete latent trait model, with latent classes located on a latent trait dimension, as described in Uebersax JS, JASA, 1993. Includes excecutable program, FORTRAN source, User's Guide, and examples.

• LTMA (zipped)

Latent Trait Mixture Analysis, as described in Uebersax JS, Grove WM, Biometrics, 1993. This model can be understood as a mixture of latent traits, or a form of conditional-dependence LCA. A common use is to estimate accuracy of diagnostic tests in the absence of a "gold standard." Includes excecutable program, FORTRAN source, User's Guide, and examples.

• glc

  Estimates the latent correlation for pairs of dichotomous variables with a user-specified skewed latent trait distribution; very easy to use.

• TetMat (zipped)

Very user-friendly program to produce a matrix of tetrachoric correlations. Also supplies standard errors, confidence intervals, p-values, and item thresholds. Provisions for smoothing a potentially improper correlation matrix.

• POLYCORR (web page)

POLYCORR 1.1, a program with advanced features to estimate the polychoric correlation coefficient. Calculates the polychoric correlation for a single table, but has many options.

• tcorr (zipped)

Calculate the tetrachoric correlation coefficient. For a single table. Very easy to use--just enter the four frequencies and get the answer. Also reports standard error, item thresholds, etc.

• MH (web page)

Simple program to test marginal homogeneity in 2-way tables using McNemar and Stuart-Maxwell tests. The Bhapkar test is also performed.

• MVN (web page)

An easy-to-use but technically solid program to generate random multivariate-normal data.

• CONDEP (zipped)

Simple program to perform conditional dependence diagnostics on a latent class analysis solution. Includes executable code, QuickBASIC source, instructions, and example.

• MIXBIN (zipped)

User-friendly program to estimate a mixture of binomials. Includes executable program and short documentation.

• probit (zipped)

Simple utility to calculate the probit function.

• PAIR (zipped)

Performs unidimensional scaling based on pair-comparison ratings.

• CLUSBAS (zipped)

Extremely easy to use program for average-linkage hierarchical cluster analysis. Executable code and source code (in both fortran and QuickBasic versions) included. User supplies similiarity/dissimilarity matrix. This version limited to 100 objects, but the limit can be increased. Ideal for cluster analysis of variables/items.

Last updated: 20 Apr 2010 (reprints)