Use of posterior predictive assessments to evaluate model fit in multilevel logistic regression

Assessing the fit of a model is an important final step in any statistical analysis, but this is not straightforward when complex discrete response models are used. Cross validation and posterior predictions have been suggested as methods to aid model criticism. In this paper a comparison is made between four methods of model predictive assessment in the context of a three level logistic regression model for clinical mastitis in dairy cattle; cross validation, a prediction using the full posterior predictive distribution and two “mixed” predictive methods that incorporate higher level random effects simulated from the underlying model distribution. Cross validation is considered a gold standard method but is computationally intensive and thus a comparison is made between posterior predictive assessments and cross validation. The analyses revealed that mixed prediction methods produced results close to cross validation whilst the full posterior predictive assessment gave predictions that were over-optimistic (closer to the observed disease rates) compared with cross validation. A mixed prediction method that simulated random effects from both higher levels was best at identifying the outlying level two (farm-year) units of interest. It is concluded that this mixed prediction method, simulating random effects from both higher levels, is straightforward and may be of value in model criticism of multilevel logistic regression, a technique commonly used for animal health data with a hierarchical structure.

Low organisational justice and heavy drinking: a prospective cohort study

Objectives: To investigate whether low perceived organisational injustice predicts heavy drinking among employees. Methods: Data from the prospective occupational cohort study, the 10-Town Study, related to 15 290 Finnish public sector local government employees nested in 2432 work units, were used. Non-drinkers were excluded. Procedural, interactional and total organisational justice, heavy drinking (>=210 g of absolute alcohol per week) and other psychosocial factors were determined by means of questionnaire in 2000-2001 (phase 1) and 2004 (phase 2). Multilevel logistic regression analyses taking into account for the hierarchical structure of the data were conducted and adjustments were made for sex, age, socio-economic position, marital status, baseline heavy drinking, psychological distress and other psychosocial risk factors such as job strain and effort/reward imbalance. Results: After adjustments, participants who reported low procedural justice at phase 1 were about 1.2 times more likely to be heavy drinkers at phase 2 compared with their counterparts with high justice. Low perceived justice in interpersonal treatment and low perceived total organisational justice were associated with an elevated prevalence of heavy drinking only in the socio-demographics adjusted model. Conclusions: This is the first longitudinal study to show that low procedural justice is weakly associated with an increased likelihood of heavy drinking.

Error estimation and adaptive mesh refinement for aerodynamic flows

This lecture course covers the theory of so-called duality-based a posteriori error estimation of DG finite element methods. In particular, we formulate consistent and adjoint consistent DG methods for the numerical approximation of both the compressible Euler and Navier-Stokes equations; in the latter case, the viscous terms are discretized based on employing an interior penalty method. By exploiting a duality argument, adjoint-based a posteriori error indicators will be established. Moreover, application of these computable bounds within automatic adaptive finite element algorithms will be developed. Here, a variety of isotropic and anisotropic adaptive strategies, as well as $hp$-mesh refinement will be investigated.

Cow, farm, and management factors during the dry period that determine the rate of clinical mastitis after calving

The purpose of the research was to investigate cow characteristics, farm facilities, and herd management strategies during the dry period to examine their joint influence on the rate of clinical mastitis after calving. Data were collected over a 2-yr period from 52 commercial dairy farms throughout England and Wales. Cows were separated for analysis into those housed for the dry period (8,710 cow-dry periods) and those at pasture (9,964 cow-dry periods). Multilevel models were used within a Bayesian framework with 2 response variables, the occurrence of a first case of clinical mastitis within the first 30 d of lactation and time to the first case of clinical mastitis during lactation. A variety of cow and herd management factors were identified as being associated with an increased rate of clinical mastitis and these were found to occur throughout the dry period. Significant cow factors were increased parity and at least one somatic cell count ≥200,000 cells/mL in the 90 d before drying off. A number of management factors related to hygiene were significantly associated with an increased rate of clinical mastitis. These included measures linked to the administration of dry-cow treatments and management of the early and late dry-period accommodation and calving areas. Other farm factors associated with a reduced rate of clinical mastitis were vaccination with a leptospirosis vaccine, selection of dry-cow treatments for individual cows within a herd rather than for the herd as a whole, routine body condition scoring of cows at drying off, and a pasture rotation policy of grazing dry cows for a maximum of 2 wk before allowing the pasture to remain nongrazed for a period of 4 wk. Models demonstrated a good ability to predict the farm incidence rate of clinical mastitis in a given year, with model predictions explaining over 85% of the variability in the observed data. The research indicates that specific dry-period management strategies have an important influence on the rate of clinical mastitis during the next lactation.

Cow, farm, and herd management factors in the dry period associated with raised somatic cell counts in early lactation

This study investigated cow characteristics, farm facilities, and herd management strategies during the dry period to examine their joint influence on somatic cell counts (SCC) in early lactation. Data from 52 commercial dairy farms throughout England and Wales were collected over a 2-yr period. For the purpose of analysis, cows were separated into those housed for the dry period (6,419 cow-dry periods) and those at pasture (7,425 cow-dry periods). Bayesian multilevel models were specified with 2 response variables: ln SCC (continuous) and SCC >199,000 cells/mL (binary), both within 30 d of calving. Cow factors associated with an increased SCC after calving were parity, an SCC >199,000 cells/mL in the 60 d before drying off, increasing milk yield 0 to 30 d before drying off, and reduced DIM after calving at the time of SCC estimation. Herd management factors associated with an increased SCC after calving included procedures at drying off, aspects of bedding management, stocking density, and method of pasture grazing. Posterior predictions were used for model assessment, and these indicated that model fit was generally good. The research demonstrated that specific dry-period management strategies have an important influence on SCC in early lactation.