Model Selection Information Criteria for Non-Nested Latent Class Models

Lin, Ting Hsiang; Dayton, C. Mitchell

doi:10.3102/10769986022003249

Cited by 210 publications

(141 citation statements)

References 22 publications

Supporting

Mentioning

133

Contrasting

Order By: Relevance

“…Recent simulation studies have found that Bayesian information criterion (BIC) typically underestimates the number of classes, while the Akaike information criterion (AIC) overestimates the number of classes [28,1], but that the AIC statistic generally outperforms the BIC statistic [5]. Table 2 presents both, as well as the AIC3, which Lin and Dayton ( [28] and Vermunt et al [60]) suggest as a good compromise in estimating the number of classes that should be represented in latent class analysis.…”

Section: First Stage Analysis: Cluster Derivationmentioning

confidence: 94%

“…Table 2 presents both, as well as the AIC3, which Lin and Dayton ( [28] and Vermunt et al [60]) suggest as a good compromise in estimating the number of classes that should be represented in latent class analysis. The BIC statistic is lowest for the two-cluster model (3991.03), while the AIC statistic is lowest for the five-cluster model (3874.92).…”

Section: First Stage Analysis: Cluster Derivationmentioning

confidence: 98%

See 1 more Smart Citation

Relating Clusters of Adolescent Problems to Adult Criminal Trajectories: a Person-Centered, Prospective Approach

Bosick

Bersani²,

Farrington

2015

J Dev Life Course Criminology

View full text Add to dashboard Cite

Purpose Motivated by offender typology debates, we evaluate whether adult offending trajectories can be predicted from adolescent risk factors. Methods Drawing on data from the Cambridge Study in Delinquent Development (N= 411), we use person-centered, latent class cluster analysis (LCCA) to identify groups of respondents with similar behavioral, social, and psychological profiles measured in adolescence. We then use hierarchical linear models to estimate the criminal trajectories associated with each cluster using annual offending measures from 19 to 50 years of age. This offers a test of whether prospectively defined crime trajectories validate theoretical conceptions of qualitatively distinct offender groups. Results Our LCCA identified four clusters of boys with varying patterns of adolescent characteristics. The offending trajectories associated with these clusters differed in magnitude rather than shape. While we were able to identify a subgroup of offenders whose criminal offending remained relatively high over the life course, significant differences across subgroups were varied and dissipated after young adulthood. All offending trajectories followed the familiar age-crime curve, characterized by a sharp decline in offending in early adulthood. Conclusions Our findings support multidimensional interventions that would offset the constellations of behavioral, psychological, and social setbacks adolescents face. At the J Dev Life Course Criminology same time, our findings undermine the notion that qualitatively distinct patterns of offending can be prospectively identified and suggest that the processes behind criminal decline over the life course are generalizable across offenders.

show abstract

Section: First Stage Analysis: Cluster Derivationmentioning

confidence: 94%

Section: First Stage Analysis: Cluster Derivationmentioning

confidence: 98%

Relating Clusters of Adolescent Problems to Adult Criminal Trajectories: a Person-Centered, Prospective Approach

Bosick

Bersani²,

Farrington

2015

J Dev Life Course Criminology

View full text Add to dashboard Cite

show abstract

“…These studies have generally suggested that BIC outperforms AIC, in that it often exhibits lower error rates and greater power. However, these studies have also generally concluded that neither criterion is completely satisfactory, and that other fit criteria exhibit greater power to select true models (Celeux and Soromenho, 1996;Lin and Dayton, 1997;Yang, 1998). Interestingly, Monte Carlo evidence suggests that early MDL approximations (Rissanen, 1978), essentially corrected forms of BIC, perform well in selecting latent class models (Yang, 1998).…”

Section: Previous Studies Of Information-theoretic Criteriamentioning

confidence: 95%

An Empirical Comparison of Information-Theoretic Selection Criteria for Multivariate Behavior Genetic Models

2004

View full text Add to dashboard Cite

Information theory provides an attractive basis for statistical inference and model selection. However, little is known about the relative performance of different information-theoretic criteria in covariance structure modeling, especially in behavioral genetic contexts. To explore these issues, information-theoretic fit criteria were compared with regard to their ability to discriminate between multivariate behavioral genetic models under various model, distribution, and sample size conditions. Results indicate that performance depends on sample size, model complexity, and distributional specification. The Bayesian Information Criterion (BIC) is more robust to distributional misspecification than Akaike's Information Criterion (AIC) under certain conditions, and outperforms AIC in larger samples and when comparing more complex models. An approximation to the Minimum Description Length (MDL; Rissanen, J. (1996). IEEE Transactions on Information Theory 42:40-47, Rissanen, J. (2001). IEEE Transactions on Information Theory 47:1712-1717) criterion, involving the empirical Fisher information matrix, exhibits variable patterns of performance due to the complexity of estimating Fisher information matrices. Results indicate that a relatively new information-theoretic criterion, Draper's Information Criterion (DIC; Draper, 1995), which shares features of the Bayesian and MDL criteria, performs similarly to or better than BIC. Results emphasize the importance of further research into theory and computation of information-theoretic criteria.

show abstract

“…One limitation of both chi-square statistics is that they can only be used to compare nested models (Everitt 1988;Lin and Dayton 1997). Alternative methods are required for comparing models with a non-nested relationship.…”

Section: Assessing Model Fitmentioning

confidence: 98%

Estimating latent class model parameters for filter questions with skip patterns

Lin

2010

Qual Quant

Self Cite

View full text Add to dashboard Cite

Filter questions with skip patterns have been widely used in survey research, and latent class models (LCM) are often used to analyze this type of categorical data. The LCM parameters are usually estimated by means of an EM (expectation maximization) algorithm. When the pattern is present, the non-response of the skip pattern cannot be treated as random missingness. We thus propose a modified algorithm to estimate the latent class parameters when non-response is present, and the approach is attractive for two reasons. First, the latent class model with the algorithm is very flexible in the sense that it can model the association of variables with the skip patterns under study. Secondly, the algorithm can be easily implemented using any computer language. An empirical example is used to demonstrate the usefulness of the algorithm. The algorithm may also be flexibly generalized to more complex surveys, for example, polytomous responses.

show abstract

Model Selection Information Criteria for Non-Nested Latent Class Models

Cited by 210 publications

References 22 publications

Relating Clusters of Adolescent Problems to Adult Criminal Trajectories: a Person-Centered, Prospective Approach

Relating Clusters of Adolescent Problems to Adult Criminal Trajectories: a Person-Centered, Prospective Approach

An Empirical Comparison of Information-Theoretic Selection Criteria for Multivariate Behavior Genetic Models

Estimating latent class model parameters for filter questions with skip patterns

Contact Info

Product

Resources

About