DIC in variable selection

Linde, Angelika van der

doi:10.1111/j.1467-9574.2005.00278.x

Cited by 112 publications

(25 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…As a sensitivity analysis, we estimated changes in annual mortality when the target covariate assumed values representing the 10% and 90% quantiles of observed values while we held others at their mean values. We based predictions of the effects of varying levels of covariates on the model with the lowest deviance information criterion (DIC; van der Linde , Abadi et al , Spiegelhalter et al ).…”

Section: Methodsmentioning

confidence: 99%

Effects of wolf pack size and winter conditions on elk mortality

et al. 2019

View full text Add to dashboard Cite

Elk (Cervus canadensis) are high‐profile game animals for many states in the western United States, yet over the past several decades some populations have experienced a persistent and broad‐scale decline in recruitment. Over this same period, gray wolves (Canis lupus) have become an integral component of many western landscapes and agencies are increasingly challenged to maximize hunting opportunities of ungulates via predator management while simultaneously ensuring wolf conservation. To better understand the implications of predator management on elk populations, we monitored survival of 1,244 adult female elk and 806 6‐month‐old calves from 29 populations distributed throughout Idaho, USA, from 2004 to 2016. We developed predictive models of mortality that related mortality risk to wolf pack size, winter conditions, and individual‐level characteristics. Annual mortality rates (excluding harvest) for adult females and calves were 0.09 and 0.40, respectively. Calf mortality was predicted best with a model that included additive effects of chest girth at time of capture, mean size of surrounding wolf packs, and snow depth. Adult female mortality was predicted best with a model that included female age, mean size of surrounding wolf packs, and snow depth. Based on a sensitivity analysis, chest girth had the largest effect on risk of mortality for calves followed by pack size and snow depth. Other than the effect of senescence in the oldest (>15 yr) individuals, pack size and snow depth had the largest effect on risk of mortality for adult females. We estimated cause‐specific mortality and predation was the dominant cause of known‐fate mortalities for adult females (35% mountain lion [Puma concolor] and 32% wolf) and calves (45% mountain lion and 28% wolf), whereas malnutrition accounted for 9% and 10% of adult female and calf mortalities, respectively. Wolves preferentially selected smaller calves and older adult females, whereas mountain lions showed little preference for calf size or age class of adult females. Our study indicates managers can increase elk survival by reducing wolf pack sizes on surrounding winter ranges, especially in areas where, or during years when, snow is deep. Additionally, managers interested in improving over‐winter calf survival can implement actions to increase the size of calves entering winter by increasing the nutritional quality of summer and early fall forage resources. Although our study was prompted by management questions related to wolves, mountain lions killed more elk than wolves and differences in selection of individual elk indicate mountain lions may have comparably more of an effect on elk population dynamics. Although we were unable to relate changes in mountain lion populations to elk survival in our study, future research should seek a better understanding of multi‐predator systems, including how management of one predator affect others and ultimately how these interactions affect elk survival. © 2019 The Wildlife Society

show abstract

Section: Methodsmentioning

confidence: 99%

Effects of wolf pack size and winter conditions on elk mortality

et al. 2019

View full text Add to dashboard Cite

show abstract

“…Estimates for σ w 2 , σ v 2 , α t , predicted recruitment, and β terms were recorded along with their respective 95% credible intervals. To evaluate whether any external variable explained a significantly greater amount of variation after accounting for dynamic changes in α, each of the four environmental variables were added singly to the KF-RW model, and model selection was carried out by comparing model deviance information criterion (DIC) scores among models, where lower DIC scores indicate more-preferred models, and all models within 2-3 DIC units of the lowest DIC score are considered useful in explaining variation in rainbow smelt recruitment (Spiegelhalter et al, 2002;van der Linde, 2005). To compare the fits of the KF-RW models and the most useful fixed-effect models using DIC, the three top-ranked fixed-effect models (see Results) were fitted using the same procedure.…”

Section: Ricker Modelmentioning

confidence: 99%

Non-stationary recruitment dynamics of rainbow smelt: The influence of environmental variables and variation in size structure and length-at-maturation

Feiner

Bunnell

Höök

et al. 2015

Journal of Great Lakes Research

View full text Add to dashboard Cite

“…The pointwise log-likelihood can be used, among others, to calculate the Watanabe-Akaike information criterion (WAIC) proposed by Watanabe (2010) and leave-one-out cross-validation (LOO; Gelfand, Dey, and Chang 1992;Vehtari, Gelman, and Gabry 2015; see also Ionides 2008) both allowing to compare different models applied to the same data (lower WAICs and LOOs indicate better model fit). The WAIC can be viewed as an improvement of the popular deviance information criterion (DIC), which has been criticized by several authors (Vehtari et al 2015;Plummer 2008;Van der Linde 2005; see also the discussion at the end of the original DIC paper by Spiegelhalter, Best, Carlin, and Van der Linde 2002) in part because of problems arising from fact that the DIC is only a point estimate. In brms, WAIC and LOO are implemented using the loo package (Vehtari, Gelman, and Gabry 2017) also following the recommendations of Vehtari et al (2015).…”

Section: Parameter Estimationmentioning

confidence: 99%

brms: An R Package for Bayesian Multilevel Models Using Stan

Bürkner¹

2017

J. Stat. Soft.

7,013

5,501

View full text Add to dashboard Cite

The brms package implements Bayesian multilevel models in R using the probabilistic programming language Stan. A wide range of distributions and link functions are supported, allowing users to fit -among others -linear, robust linear, binomial, Poisson, survival, ordinal, zero-inflated, hurdle, and even non-linear models all in a multilevel context. Further modeling options include autocorrelation of the response variable, user defined covariance structures, censored data, as well as meta-analytic standard errors. Prior specifications are flexible and explicitly encourage users to apply prior distributions that actually reflect their beliefs. In addition, model fit can easily be assessed and compared with the Watanabe-Akaike information criterion and leave-one-out cross-validation.

show abstract

DIC in variable selection

Cited by 112 publications

References 9 publications

Effects of wolf pack size and winter conditions on elk mortality

Effects of wolf pack size and winter conditions on elk mortality

Non-stationary recruitment dynamics of rainbow smelt: The influence of environmental variables and variation in size structure and length-at-maturation

brms: An R Package for Bayesian Multilevel Models Using Stan

Contact Info

Product

Resources

About