On Model Selection Criterion in Capture –Recapture Experiments with Sparse Data

Full Text PDF PDF
Author(s) Danjuma Jibasen | Ezra Gayawan
Pages 750-753
Volume 3
Issue 12
Date December, 2014
Keywords Akaike Information Criterion, Likelihood ratio test, hypothesis testing, confidence interval estimation, sparse data

Model selection involving sparse data is always difficult, this is because with sparse data, quite different models can appear to fit adequately with highly diverse point, it is also almost impossible to test the underlying assumptions and select the “best” model”. Some ecological as well as epidemiological experiments result in sparse data. In this paper we propose a modified Akaike information criterion (call it AICJ) for selecting models in capture-recapture experiments resulting in sparse data. The proposed criterion was compared with the Akaike Information Criterion (AIC) and likelihood ratio test G2. It was found that AICJ performed well in comparison to AIC and G2. We therefore recommend that instead of concluding that model selection criterion may give misleading results for sparse data, or even that selection criterion does not exist, AICJ can be used. The proposed criterion is based on the link between interval estimation, hypothesis testing and model selection.

< Back to December Issue