The African dwarf crocodile (Osteolaemus tetraspis) is endemic to closed-canopy forests of Central and West Africa and is the smallest of the world’s true crocodiles. The species is difficult to study in the wild and therefore poorly known, but likely plays an important ecological role as a top aquatic predator in cool water forest systems. The dwarf crocodile is also a major food and economic resource to local people and, as a result, is threatened with overhunting for the bushmeat trade. The image depicts a collection of young dwarf crocodiles, possibly representing three cohorts, measured in a capture-recapture study in Loango National Park, Gabon.
The article linked to the image is On thinning of chains in MCMC by William Link and Mitchell Eaton. In the article, the authors caution against the routine practice of thinning chains in Markov chain Monte Carlo (MCMC) simulations. Many analysts, recognizing that MCMC precision decreases as the autocorrelation of the chains increases, routinely thin (sub-sample) their chains. Thinning reduces autocorrelation, but the associated gains in precision are more than offset by the reduction in chain length. Thinning of chains is therefore wasteful, though occasionally justified under circumstances discussed in the article.
To illustrate, the authors refer to a recent application (Eaton and Link 2011, Ecological Applications) in which they applied Bayesian multimodel inference to evaluate two growth models used to estimate individual dwarf crocodile age from capture-recapture data. They demonstrate analytically that thinning their model-selection chains would have decreased autocorrelation but would also decrease the precision with which posterior model probabilities were approximated.
The young dwarf crocodiles were photographed by Mitchell Eaton in 2004.