Bayesian Hurdle Model Research Articles

Seabird abundances projected to decline in response to climate change in Britain and Ireland

Estimating species’ vulnerability to climate change is a key challenge in conservation. Many seabird species are particularly sensitive to the negative effects of warming and are thus threatened by climate change. We projected seabird abundance in 2050 in Britain and Ireland, an internationally important region for seabirds, under climate change scenario RCP8.5. Our projections were based on fitted relationships between abundance and both climatic (air temperature, precipitation) and oceanographic (sea surface temperature, potential energy anomaly) variables, estimated using a Bayesian hurdle model with spatial random effects. The most consistent responses to environmental variables were negative relationships between species presence or abundance and breeding season maximum terrestrial air temperature. Although data and modelling issues meant that future projections could be made for only 15 of the 25 species and with low predictive power for spatial abundance, predictive power was high for spatial presence and moderate for temporal abundance change. Projections suggest that seabirds will generally fare poorly under climate change in Britain and Ireland, with 10 species projected to decline in abundance by 2050. Variable model fit and projection uncertainty limited confidence, which was generally greater for declines than for increases. Fulmar, puffin and Arctic tern were each projected to decline by over 70% with higher confidence. For a few species, colonisation of new areas may reduce projected losses, but this is potentially limited by low availability of suitable habitat. Projected abundance change was more negative for marine specialists than for generalists. Our findings highlight the vulnerability and conservation needs of seabirds in a changing climate.

A Hierarchical Mixed Effect Hurdle Model for Spatiotemporal Count Data and its Application to Identifying Factors Impacting Health Professional Shortages

Summary Count data are common in many fields such as public health. Hurdle models have been developed to model count data when the zero count could be either inflated or deflated. However, when data are repeatedly collected over time and spatially correlated, it is very challenging to model the data appropriately. For example, to study health professional shortage areas, the number of primary care physicians along with other demographic characteristics are collected at the county level in the USA and over different years. Since the data are repeatedly collected over time, counties are nested within the state, and adjacent counties are geographically correlated, the dependence structure of the data is very complex. We develop a Bayesian hurdle model with multilayered random effects to incorporate this complex structure. We use a time-varying random effect for each state to capture the time effect at the state level, and a temporal thin plate spline to capture the spatiotemporal correlation across different counties. We use STAN to obtain samples for inference from the posterior distribution. By using the model proposed, we can identify the important factors which impact health professional shortage areas. Simulation studies also confirm the effectiveness of the model.

Estimating relationships between size and fecundity in the threatened giant garter snake in seminatural and agricultural wetlands

AbstractIncreasing numbers of populations of threatened and endangered plants and animals occur in environments altered by human activity. Life history traits may vary among populations in natural and human‐dominated environments, and therefore, management strategies may need to be tailored to different environments to promote persistence. The primary objectives of this project were to estimate the probability of being pregnant for females of the threatened giant garter snake (Thamnophis gigas) and test for differences in these demographic parameters between populations in seminatural wetlands and wetlands in areas dominated by rice agriculture. We also hypothesized that the probability of breeding and the number of fetuses would increase with increasing body size, a commonly reported relationship for snakes. We analysed data on T. gigas fecundity from 9 years using a Bayesian hurdle model and found no evidence of variation in the probability of being pregnant or in the number of fetuses given pregnancy for T. gigas in the two environments. The probability of being pregnant in a year was higher for females of larger size and ranged from <0.10 for snakes smaller than 631 mm snout‐to‐vent length (SVL) to >0.96 for the largest females. Similarly, given that a female was pregnant, the number of fetuses produced was higher for larger females. Pregnant females of average SVL were expected to produce 14 fetuses, and the largest females were expected to produce 34 fetuses. The inferences from our analyses are needed for subsequent population modeling, which will guide planning and decision‐making for T. gigas.

Farm level risk factors for bovine digital dermatitis in Taranaki, New Zealand: An analysis using a Bayesian hurdle model

As part of a cross-sectional study of bovine digital dermatitis (BDD) in 60,455 cows across 224 herds, in Taranaki, a region of the North Island of New Zealand, from September 2014 to February 2015, questionnaires from 114 farms were analysed to identify the key management practices which affect the probability of a farm being affected by BDD and the prevalence of an individual cow being affected on affected farms. The data from the questionnaires were analysed using a Bayesian hurdle model. Two factors were identified as being associated with farm level infection status: (1) milking parlour type; farms which had rotary platforms were more likely to be recorded as having BDD than those which had herringbone parlours (odds ratio, OR, 3.19; 95% probability interval, PI, 1.31–8.51); and (2) young stock movement; farms whose young stock were reared on farms alongside heifers from other farms had a higher odds of being BDD positive than farms where heifers were kept separate (OR 4.15; 95% PI 1.39–15.27). Two factors were associated with an increase in the prevalence of BDD within affected farms: (1) farms which used outside staff to trim feet had a higher prevalence of BDD (prevalence ratio, PR, 3.13; 95% PI 1.25–7.29) than farms which did not use outside staff; and (2) farms examined in spring (September to November) had a higher prevalence of BDD (PR 2.16; 95% PI 1.05–4.43) than farms examined in summer (December to February).

An analysis of trends, frequencies and factors influencing the development of resistance to phosphine in the red flour beetle Tribolium castaneum (Herbst) in Australia

Although resistance to phosphine, the key disinfestant used worldwide in the stored grain environment has been an ongoing industry issue, studies on its trend over large geographic region and over long period of time is very limited. In this study, we critically analysed 20 years' phosphine resistance diagnosis data for the red flour beetle Tribolium castaneum (Herbst) stored in the Australian Grain Insect Resistance Database. Resistance diagnosis on a staggering 6336 samples, along with information on storage types and treatment history was interrogated to establish trends and frequencies of resistance development in this species and factors that may have contributed towards these resistance occurences. Using descriptive statistics, linear and trend analysis and a well established Bayesian hurdle model, we determined that strong resistance in T. castaneum was significantly more prevalent in quarantine intereceptions than in central storages and on farms. The strong resistance incidences had been confined to eastern states of Queensland, New South Wales, Victoria and South Australia, whereas it is yet to be detected in the state of Western Australia. We could not establish any significant correlation between the strong resistance development and any commodity or treatments. After an initial increasing trend in incidences since the first detection of strong resistance in 1997 in this species, the frequency was stabilised during 2001–08; after which there had been an upward trend since 2009 till the last survey in 2013. The conclusions derived from this analyses highlighted the importance of a resistance monitoring program with relevant information being used in Australia as the basis for ongoing and future phosphine resistance management strategies. This research may also proved valuable towards devising similar strategies in overseas countries with phosphine resistance problems.

Monitoring resistance to phosphine in the lesser grain borer, Rhyzopertha dominica, in Australia: A national analysis of trends, storage types and geography in relation to resistance detections

Data on incidence of resistance to phosphine over the last 20 years and factors associated with insect sample collection are stored in the Australian Grain Insect Resistance Database. The database was analysed using descriptive statistics, linear trend analysis and Bayesian hurdle modelling to gain insights into factors contributing to the development of strong resistance in Rhyzopertha dominica. Descriptive statistics indicated that strong resistance was significantly more common in central storages, particularly bunker storages, than on farms. Strong resistance in R. dominica was also associated with wheat, barley and sorghum but there was no significant link to any grain protectant or storage treatment chemical, other than phosphine. Highest frequency of strong resistance was found in northern New South Wales and no detections were made in Western Australia. In eastern Australia, trend analysis indicated that strong resistance detections increased steadily from the first detection in 1997 to about 8% of samples containing resistant insects in 2014. Weak resistance was detected in about 10% of samples in eastern Australia in the early 1990s but this increased rapidly to 40–50% by 1990, at the same time that industry use of phosphine greatly increased, and then to about 80% in 1995. Strong resistance was first detected in this species when weak resistance was diagnosed in close to 80% of population samples. The Bayesian hurdle model identified bunkers, silos and unsealed storages as being associated with development of strong resistance and sheds with a lower frequency. This model also identified an accelerated increase in resistance frequency of strong resistance from 2011 to present. The information gained from this analysis is being used to inform current and future management of resistance to phosphine.

A Bayesian hurdle model for analysis of an insect resistance monitoring database

Motivated by the analysis of the Australian Grain Insect Resistance Database (AGIRD), we develop a Bayesian hurdle modelling approach to assess trends in strong resistance of stored grain insects to phosphine over time. The binary response variable from AGIRD indicating presence or absence of strong resistance is characterized by a majority of absence observations and the hurdle model is a two step approach that is useful when analyzing such a binary response dataset. The proposed hurdle model utilizes Bayesian classification trees to firstly identify covariates and covariate levels pertaining to possible presence or absence of strong resistance. Secondly, generalized additive models (GAMs) with spike and slab priors for variable selection are fitted to the subset of the dataset identified from the Bayesian classification tree indicating possibility of presence of strong resistance. From the GAM we assess trends, biosecurity issues and site specific variables influencing the presence of strong resistance using a variable selection approach. The proposed Bayesian hurdle model is compared to its frequentist counterpart, and also to a naive Bayesian approach which fits a GAM to the entire dataset. The Bayesian hurdle model has the benefit of providing a set of good trees for use in the first step and appears to provide enough flexibility to represent the influence of variables on strong resistance compared to the frequentist model, but also captures the subtle changes in the trend that are missed by the frequentist and naive Bayesian models.

Common Dolphin (Delphinus delphis) Bycatch in New Zealand Commercial Trawl Fisheries

Marine mammals are regularly reported as bycatch in commercial and artisanal fisheries, but data are often insufficient to allow assessment of these incidental mortalities. Observer coverage of the mackerel trawl fishery in New Zealand waters between 1995 and 2011 allowed evaluation of common dolphin Delphinus delphis bycatch on the North Island west coast, where this species is the most frequently caught cetacean. Observer data were used to develop a statistical model to estimate total captures and explore covariates related to captures. A two-stage Bayesian hurdle model was used, with a logistic generalised linear model predicting whether any common dolphin captures occurred on a given tow of the net, and a zero-truncated Poisson distribution to estimate the number of dolphin captures, given that there was a capture event. Over the 16-year study period, there were 119 common dolphin captures reported on 4299 observed tows. Capture events frequently involved more than one individual, with a maximum of nine common dolphin observed caught in a single tow. There was a peak of 141 estimated common dolphin captures (95% c.i.: 56 to 276; 6.27 captures per 100 tows) in 2002–03, following the marked expansion in annual effort in this fishery to over 2000 tows. Subsequently, the number of captures fluctuated although fishing effort remained relatively high. Of the observed capture events, 60% were during trawls where the top of the net (headline) was <40 m below the surface, and the model determined that this covariate best explained common dolphin captures. Increasing headline depth by 21 m would halve the probability of a dolphin capture event on a tow. While lack of abundance data prevents assessment of the impact of these mortalities on the local common dolphin population, a clear recommendation from this study is the increasing of headline depth to reduce common dolphin captures.

Agricultural, socioeconomic and environmental variables as risks for human verotoxigenic Escherichia coli(VTEC) infection in Finland

BackgroundVerotoxigenic E. coli (VTEC) is the cause of severe gastrointestinal infection especially among infants. Between 10 and 20 cases are reported annually to the National Infectious Disease Register (NIDR) in Finland. The aim of this study was to identify explanatory variables for VTEC infections reported to the NIDR in Finland between 1997 and 2006. We applied a hurdle model, applicable for a dataset with an excess of zeros.MethodsWe enrolled 131 domestically acquired primary cases of VTEC between 1997 and 2006 from routine surveillance data. The isolated strains were characterized by virulence type, serogroup, phage type and pulsed-field gel electrophoresis. By applying a two-part Bayesian hurdle model to infectious disease surveillance data, we were able to create a model in which the covariates were associated with the probability for occurrence of the cases in the logistic regression part and the magnitude of covariate changes in the Poisson regression part if cases do occur. The model also included spatial correlations between neighbouring municipalities.ResultsThe average annual incidence rate was 4.8 cases per million inhabitants based on the cases as reported to the NIDR. Of the 131 cases, 74 VTEC O157 and 58 non-O157 strains were isolated (one person had dual infections). The number of bulls per human population and the proportion of the population with a higher education were associated with an increased occurrence and incidence of human VTEC infections in 70 (17%) of 416 of Finnish municipalities. In addition, the proportion of fresh water per area, the proportion of cultivated land per area and the proportion of low income households with children were associated with increased incidence of VTEC infections.ConclusionsWith hurdle models we were able to distinguish between risk factors for the occurrence of the disease and the incidence of the disease for data characterised by an excess of zeros. The density of bulls and the proportion of the population with higher education were significant both for occurrence and incidence, while the proportion of fresh water, cultivated land, and the proportion of low income households with children were significant for the incidence of the disease.

HIV testing and retesting for men and women in Switzerland.

This study was conducted to describe voluntary HIV testing in the general population in Switzerland and to estimate yearly HIV test incidence. In 1994, a representative telephone survey of individuals aged 17 to 45 years obtained self-reported information on HIV testing. In addition to describing cumulative HIV test incidence, yearly HIV test incidence over time was estimated by a Bayesian hurdle model allowing for the plausible scenario of test consumption differing between first test and subsequent retests. Overall, 33% of the Swiss population (age 17 to 45 years) has been tested at some time for HIV on a voluntary basis (30% men, 36% women). For the time period 1990-1994, the result showed for 35-year-old individuals with supposedly low risk behavior, that 1) annual test incidence (first test or retest) showed a greater increase for men (4.2 to 5.9%) than for women (5.0 to 6.0%); 2) annual first test incidence increased moderately and differed for men and women (2.9 to 3.4% for men, 4.6 to 5.2% for women), and 3) annual retest incidence was twice as high for men (17.6%) as for women (8.6%). In conclusion, a substantial part of the Swiss population has been tested at some stage for HIV on a voluntary basis, and differences exist for testing and retesting between men and women.