Research

Citation Data (4/20/2009)

Total Citations: 1535 (ISI); 2288 (Google Scholar)           

Latent Class and Categorical Data Models

My current research in this area deals primarily with latent class and grade-of-membership models and how they can be used to identify important customer and patient profiles (Earl, Cooil, Rubin , & Chari, 2008; Cooil, Keiningham, Aksoy & Hsu, M., 2007; Cooil, Aksoy, & Keiningham, 2007; Aksoy & Cooil, 2006; Cooil & Raggi, 2006; Aksoy, Bloom, Lurie, & Cooil, 2006) and to assess instrument accuracy, and data reliability.  The instruments may be the questions on a marketing survey, multivariate scales developed for market segmentation, coders who summarize open-ended customer replies or complaints, doctors who diagnose patients, risk indices for a disease, or customers who evaluate the quality of products or services on a nominal scale.

 Many commonly used data reliability measures can be interpreted as proportional reduction in loss (PRL) measures (Cooil & Rust, 1994, 1995).  For categorical data, a natural extension of this idea is to define reliability as the proportional reduction in the misclassification probability that is achieved by the empirical Bayes estimator of the correct category (Cooil & Rust, 1995; Rust & Cooil, 1994).  Grade-of-membership models provide an even more general setting in which to measure classification probabilities and reliability (Cooil & Varki, 2003; Varki, Cooil & Rust, 2000). 

&nsp    This research includes the development of new approaches to the screening, diagnosis and treatment of coronary heart disease (CHD) that reduce mortality and morbidity rates, and also provide significant cost savings.  These issues have enormous health care implications:  CHD is the greatest single cause of morbidity and mortality in the U.S., and in the world. For example, over a million Americans have a myocardial infarction annually.  The human and economic losses are staggering.  Consequently, it is crucial that more effective statistical models be used to identify those that are at the highest risk, and that decision models are developed to effectively apply new medical technology. Most of this research has appeared in medical journals where it has already had a substantial impact on the delivery of medical care in cardiology.  Additional research has appeared in statistics and management journals (e.g., Cooil and Raggi, 2005).

            We have developed a significantly more accurate measure of coronary calcification, the ‘volume score’ (Callister, Cooil, Raya, et al., 1998), which can be used to follow the progression of atherosclerosis, and to monitor the effects of treatment, in individual patients (Raggi, Cooil, Ratti & Callister, 2005; Raggi, Cooil, Shaw, et al., 2003; Callister, Raggi, Cooil, et al., 1998).  We have also shown that a patient’s coronary calcification, as measured by a conditional percentile (relative to the appropriate age-sex cohort with coronary calcification) is preeminent as a predictor of myocardial infarctions (e.g., Cooil and Raggi, 2005; Raggi, Cooil, Callister, et al., 2003;  Raggi, Cooil & Callister, 2001; Raggi, Callister, Cooil, He, et al., 2000). In fact, screening and treatment procedures based on measures of calcification reduce mortality and morbidity rates while providing expected cost saving of at least 33% in theory (when prevalence levels are high and savings are minimal) and at least 56% in empirical studies of populations with low to moderate prevalence levels (Raggi, Callister, Cooil, Russo, et al., 2000).