Hierarchical Log-linear Models Research Articles

Multi-trial, aggregated, individual participant data mega-analysis of short-term antidepressant versus mood stabilizer monotherapy of bipolar type II major depressive episode.

Few studies have systematically examined the safety and effectiveness of antidepressant versus mood stabilizer monotherapy of bipolar II depression. To date, there are no aggregated or mega-analyses of prospective trials of individual participant-level data (IPD) to inform future treatment guidelines on the relative safety and effectiveness of antidepressant or lithium monotherapy. Data from a series of four independent, similarly designed trials of antidepressant or lithium monotherapy (where longitudinal IPD were available) (n = 393) were aggregated into an IPD dataset (i.e., mega-analysis). Hierarchical log-linear growth models were used to analyze primary outcome of change over time in Hamilton Rating Scale for Depression (HRSD) scores; while secondary outcomes examined Clinical Global Impressions severity (CGI/S) and change (CGI/C) scores, and change over time in Young Mania Rating (YMR) scores. Relative to lithium monotherapy, antidepressant monotherapy demonstrated significantly greater symptom reduction on HRSD scores across time (b = -2.33, t = -6.68, p < 0.0001), significantly greater symptom reduction on the CGI/S across time (b = -0.414, t = -6.32, p < 0.001), and a significant improvement in CGI/C across time (b = -0.47, t = -7.43, p < 0.0001). No differences were observed in change over time for YMR scores between antidepressant and lithium monotherapy (b = 0.06, t = 0.49, p = 0.62). Findings from this IPD mega-analysis of bipolar II depression trials suggest a divergence from current evidence-based guidelines recommending combined mood stabilizer plus antidepressant therapy. The current mega-analysis suggests that antidepressant monotherapy may provide superior short-term effectiveness without clinically meaningful increase in treatment-emergent hypomanic symptoms compared to lithium monotherapy.

Tree-Structured Model with Unbiased Variable Selection and Interaction Detection for Ranking Data

In this article, we propose a tree-structured method for either complete or partial rank data that incorporates covariate information into the analysis. We use conditional independence tests based on hierarchical log-linear models for three-way contingency tables to select split variables and cut points, and apply a simple Bonferroni rule to declare whether a node worths splitting or not. Through simulations, we also demonstrate that the proposed method is unbiased and effective in selecting informative split variables. Our proposed method can be applied across various fields to provide a flexible and robust framework for analyzing rank data and understanding how various factors affect individual judgments on ranking. This can help improve the quality of products or services and assist with informed decision making.

Using of Hierarchical Loglinear Model in Multiway Frequency Tables and an Application on Suicide Cases

The aim of this study was to use Hierarchical Loglinear Model (HLLM) in the analysis of multiway frequency tables and to interpret the main and interaction effects of this model on suicide cases.&#x0D; The data set used in this study was taken from the Turkish Republic State Statistical Institute (TUIK). A total of 6479 cases in 2016 and 2018 years were used in this analysis and the analyzes were made by considering gender, year and age variables.&#x0D; As a result of HDLM analysis, Year, Gender and Age, which are the main effects in suicide cases, and the interactions of Year × Gender and Gender × Age were found significantly (P

Heterogeneity among contingency tables diagnosed by hierarchical log-linear models and their effect on Biplots

The theory of singular value decomposition of matched matrices is used to verify the heterogeneity of rows, columns and between matched two-way tables. An exploratory analysis that can be visualized in biplots and through simulations studies with the hierarchical log-linear model using ordinary residuals and the components of residual deviance. The effect of heterogeneity was studied generating different sample sizes and their behavior was checked by adjusting Poisson’s model. We concluded that the model of ordinary residuals is the one that best reflects the degree of heterogeneity among the matched tables. Finally, na illustrative example is presented in order to guide the researcher to interpret the relationship between the results of the log-linear models with the biplots considering the effects between the sum and difference between the tables.

A multisource approach to health care use: concordance between register and self-reported physician visits in the foreign-born population in Finland

BackgroundReliable information on the use of health services is important for health care planning, monitoring and policy. It is critical to assess the validity of the sources used for this purpose, including register and survey-based data. Studies on foreign-born populations’ health care use have usually implemented either survey or register data. The concordance of such data among groups of different cultural background remains largely unknown. In this study, we presented an approach to examine routinely how survey and register-related characteristics may explain disagreement found between the two information sources.MethodsWe linked register- and survey-based data pertaining to the Finnish Register of Primary Health Care general physician visits and the Survey on Well-Being among Foreign Born Population (FinMonik, 2018–2019), a nationally representative survey. The sample comprised n = 5,800 informants for whom registered general physician visits were tracked in the 12-month period preceding their participation in the survey. Cohen’s kappa was used as measure of multisource concordance, hierarchical loglinear models for the association between single predictors and multisource discrepancy, and a logistic regression model for examining source-related predictors of source discrepancy. Survey weights were used in all sample analyses.ResultsSource concordance was poor. When dichotomizing general physician visits (zero vs one or more), 35% of informants had reported one or more visits while none were found from register. Both register- and informant-related predictors were associated to this discrepancy (i.e. catchment area, private health care use, inability to work, region of origin and reason for migration).ConclusionsWe found high discrepancy between the reported and the registered physician visits among the foreign-born population in Finland, with a particularly high number of reported physician visits when none were found in the register. There was a strong association between the specific catchment area and mismatch, indicating that both register under-coverage and survey over-report are plausible and may coexist behind the discrepancy. However, associations of informant’s characteristics and mismatch were less pronounced. Implications on the validity of medical information sources are discussed.

Internationalization–Innovation Strategy in New Ventures: A Log-linear Analysis

Despite the well-established empirical evidence of a positive mutual effect of internationalization and innovation, the extant literature provides a limited understanding of factors that could strengthen the adoption of a dual strategy in new ventures. This article addresses this critical gap through the examination of a large dataset on entrepreneurial activity from the Global Entrepreneurship Monitor, which covers 50 countries in both developed and developing contexts. This article advances an estimation of a hierarchical log-linear model designed to uncover the mutual influence of micro (i.e., entrepreneurs and firm particularities) and macro (i.e., industry and environment features) factors and to identify which variables are of great importance with respect to the innovation–internationalization strategy in new ventures. The results suggest that (a) combinations of micro and macro factors explain the adoption of the internationalization–innovation strategy in new ventures; (b) gender barriers to adoption appear in developed countries; (c) opportunity recognition is relevant to adoption in any country; and (d) business skills are significant to adoption only in low-technology sectors. This article adds to the extant literature by providing a comprehensive understanding of how micro- and macro-level factors concurrently affect innovation and internationalization strategies for new ventures.

Data Information in Contingency Tables: A Fallacy of Hierarchical Loglinear Models

Information identities derived from entropy and relative entropy can be useful in statistical inference. For discrete data analyses, a recent study by the authors showed that the fundamental likelihood structure with categorical variables can be expressed in different yet equivalent information decompositions in terms of relative entropy. This clarifies an essential difference between the classical analysis of variance and the analysis of discrete data, revealing a fallacy in the analysis of hierarchical loglinear models. The discussion here is focused on the likelihood information of a three-way contingency table, without loss of generality. A classical three-way categorical data example is examined to illustrate the findings.

Linear Information Models: An Introduction

Relative entropy identities yield basic decompositions of cat egorical data log-likelihoods. These naturally lead to the development of information models in contrast to the hierarchical log-linear models. A recent study by the authors clarified the principal difference in the data likelihood analysis between the two model types. The proposed scheme of log-likelihood decomposition introduces a prototype of linear information models, with which a basic scheme of model selection can be formulated accordingly. Empirical studies with high-way contingency tables are exem plified to illustrate the natural selections of information models in contrast to hierarchical log-linear models.

An Application of Bayesian Model Averaging Approach to Traffic Accidents Data Over Hierarchical Log-Linear Models

In this article, a Bayesian model averaging approach for hier archical log-linear models is considered. Posterior model probabilities are approximately calculated for hierarchical log-linear models. Dimension of interested model space is reduced by using Occam’s window and Occam’s razor approaches. 2002 road traffic accident data of Turkey is analyzed by using the considered approach

Loglinear Modeling of Academic Performance Data

The focus of this article was to fit a hierarchical loglinear model to data on academic performance. Data on gender, university attended for B.Sc., B.Sc. and M.Sc. grades of 116 M.Sc. graduates were collected from Department of Statistics, University of Ilorin, Ilorin, Nigeria. Model estimation was carried out by iterative proportional fitting method. Likelihood ratio statistic was utilized for goodness of fit test. The final model generating class contained University, Gender, and B.Sc.*M.Sc., and in harmony with the principle of hierarchy, also contained B.Sc. and M.Sc. grades. Significant interaction was found between B.Sc. and M.Sc. grades only. All other 2-factor and all 3-factor interactions were found not to be significant. Thus, M.Sc. grade was neither associated with gender and university nor with their interaction. The likelihood ratio statistic with p-value of 0.722 suggested model adequacy. The study concluded that only B.Sc. grade was associated with M.Sc. grade obtained by students on graduation. The need to extend study to other departments in the University was recommended.  Â

Projecting future changes in element concentrations of approximately 100 untreated discharges from legacy mines in Japan by a hierarchical log-linear model

Understanding future changes in the concentrations of elements such as Cd in mine drainages, which can cause severe environmental impacts, is crucial to strategically optimize the treatment and management of such drainages. In this study, on the basis of 17-year data (2003–2019) for 99 untreated drainages from legacy mines in Japan, we developed a Bayesian hierarchical log-linear model that can capture temporal changes in the concentrations of seven elements including six metals (Cd, Pb, As, Cu, Zn, Fe, and Mn) in individual mine drainages. We also projected future changes to understand the prospective trends nationwide. The modeling results showed that, during 2003–2019, although overall decreasing trends were observed for most elements across all the drainages evaluated, decreases in the concentrations of these elements were not evident in many mine drainages (5%–28% of drainages for individual elements); in addition, any rise in the number of mine drainages with element concentrations below nationwide drainage standards over the next 100 years will likely be limited (e.g., approximately 10 drainages for Zn and Fe at median estimates). These results have significant implications for future strategies to manage mine drainages: it is probably too optimistic to assume that the element concentrations of mine drainages will always decrease, or that these drainages will satisfy drainage standards (permits) in the not so distant future.

Spatial-Temporal Modelling of Disease Risk Accounting for PM2.5 Exposure in the Province of Pavia: An Area of the Po Valley

Spatio-temporal Bayesian disease mapping is the branch of spatial epidemiology interested in providing valuable risk estimates in certain geographical regions using administrative areas as statistical units. The aim of the present paper is to describe spatio-temporal distribution of cardiovascular mortality in the Province of Pavia in 2010 through 2015 and assess its association with environmental pollution exposure. To produce reliable risk estimates, eight different models (hierarchical log-linear model) have been assessed: temporal parametric trend components were included together with some random effects that allowed the accounting of spatial structure of the region. The Bayesian approach allowed the borrowing information effect, including simpler model results in the more complex setting. To compare these models, Watanabe–Akaike Information Criteria (WAIC) and Leave One Out Information Criteria (LOOIC) were applied. In the modelling phase, the relationship between the disease risk and pollutants exposure (PM2.5) accounting for the urbanisation level of each geographical unit showed a strong significant effect of the pollutant exposure (OR = 1.075 and posterior probability, or PP, >0.999, equivalent to p < 0.001). A high-risk cluster of Cardiovascular mortality in the Lomellina subareas in the studied window was identified.

Challenges in monitoring mobile populations: Applying bayesian multi‐site mark–recapture abundance estimation to the monitoring of a highly mobile coastal population of bottlenose dolphins

Abstract Monitoring the abundance of mobile and wide‐ranging cetacean populations for conservation management is challenging, especially when the management is focused on static protected areas. Where abundance estimates are derived from mark–recapture data, such as photo‐identification of naturally marked individuals, unpredictable movements of animals in and out of the survey area can reduce ‘capture’ probabilities and affect the precision and accuracy of resulting estimates. A Bayesian hierarchical log–linear model was applied to photo‐identification data collected in summer 2014 to derive a multi‐site abundance estimate for a population of bottlenose dolphins, Tursiops truncatus, ranging widely throughout the coastal waters of western Ireland. In addition, the effects of varying levels of sampling effort on the minimum detectable decrease in population size were examined. The abundance (median) of dolphins was estimated as 189 (coefficient of variation (standard deviation/mean), 0.11; 95% highest‐posterior density interval, 162–232). Over 50% of the well‐marked dolphins encountered throughout the study were sighted in more than one distinct coastal area, thus displaying high mobility. In addition, it was found that it would require biennial surveys to detect a 25% decline in abundance within the six‐year reporting period of the EU’s Habitats Directive. Given that the Special Area of Conservation designated for these dolphins consists of two separate areas covering a substantial portion of the west coast of Ireland, the multisite approach is appropriate for monitoring this population. It produces a more precise estimate and is well suited for sparse recapture data collected opportunistically at multiple sites, when the lack of resources prevents large‐scale surveys or when concentrating surveys on smaller localized areas fails to capture the broad range and unpredictable occurrence of the animals. The Bayesian multi‐site approach could be applied to the management of other wide‐ranging marine or terrestrial taxa.

Animal-based measures on fattening heavy pigs at the slaughterhouse and the association with animal welfare at the farm level: a preliminary study

Monitoring animal welfare (AW) in pig farms requires both proper indicators and a feasible approach. Animal-based measures (ABMs) are well-established AW indicators. Furthermore, AW screening at the slaughterhouses could be useful for identifying problems on farm. The aim of this study was to evaluate ABMs at the slaughterhouse and, when possible, to compare these ABMs with those collected on the farm. The study was carried out in northern Italy in a commercial abattoir and in a sample of farms. Animal-based measures were recorded on pigs from 62 batches of 54 farms, during ante-mortem (n=10 085 pigs) and post-mortem (n=7952 pigs) inspections. Sixteen of 54 farms were selected to compare ABMs collected at the slaughterhouse with ABMs collected on the farm. Overall, 2295 pigs (mean pigs examined per farm 119±45) were inspected at the slaughterhouse (group S) and 420 pigs (mean pigs per farm 26±5) on the farm (group F). Non-animal-based measures were also collected at the 16 farms. Differences between groups S and F, at the animal level, were assessed by a two-tailed paired Wilcoxon–Mann–Whitney test. Differences at the site of observation level (farm and slaughterhouse) were assessed by Fisher’s exact test using a hierarchical log-linear modelling for contingency tables. The most frequent ABMs at the slaughterhouse were manure on the body (47.7%), followed by dermatitis (28.0%), white spot (25.4%) and bursitis (24.7%). Recording ABMs at the slaughterhouse and on the farm usually yielded similar results; however, there were some exceptions. In particular, significant differences were found for non-uniformity of size (P<0.05) and dermatitis (P<0.001), which were higher at the slaughterhouse than on the farm. Results of log-linear modelling underlined the effect of the farm of origin on the percentage of pigs with bursitis, manure on the body and ear injuries that were observed at the slaughterhouse. In group S, significant associations between manure on the body and insufficient presence of clean and dry areas in the corresponding farm were found (P<0.05). Although these results should be interpreted with care due to the limited sample of farms, the slaughterhouse could be a feasible site of observation of ABMs, which could then be integrated in monitoring of AW on farm. Considering the number of slaughtered batches per farm, it would be possible to repeat assessments several times throughout the year for each farm, which could help provide an index for the continuous monitoring of AW.

Approximating faces of marginal polytopes in discrete hierarchical models

The existence of the maximum likelihood estimate in a hierarchical log-linear model is crucial to the reliability of inference for this model. Determining whether the estimate exists is equivalent to finding whether the sufficient statistics vector $t$ belongs to the boundary of the marginal polytope of the model. The dimension of the smallest face $\mathbf{F}_{t}$ containing $t$ determines the dimension of the reduced model which should be considered for correct inference. For higher-dimensional problems, it is not possible to compute $\mathbf{F}_{t}$ exactly. Massam and Wang (2015) found an outer approximation to $\mathbf{F}_{t}$ using a collection of submodels of the original model. This paper refines the methodology to find an outer approximation and devises a new methodology to find an inner approximation. The inner approximation is given not in terms of a face of the marginal polytope, but in terms of a subset of the vertices of $\mathbf{F}_{t}$. Knowing $\mathbf{F}_{t}$ exactly indicates which cell probabilities have maximum likelihood estimates equal to $0$. When $\mathbf{F}_{t}$ cannot be obtained exactly, we can use, first, the outer approximation $\mathbf{F}_{2}$ to reduce the dimension of the problem and then the inner approximation $\mathbf{F}_{1}$ to obtain correct estimates of cell probabilities corresponding to elements of $\mathbf{F}_{1}$ and improve the estimates of the remaining probabilities corresponding to elements in $\mathbf{F}_{2}\setminus\mathbf{F}_{1}$. Using both real-world and simulated data, we illustrate our results, and show that our methodology scales to high dimensions.

Using categorical data analyses in determination of dust-related occupational diseases in mining

Dust-related occupational diseases are common in the mining sector. It is important to identify employees who have high potential for these diseases and to investigate the factors affecting disease formation. For this reason, dust and dust-related occupational diseases should be carefully investigated in mining operations. In this study, dust-related occupational diseases in an open-pit lignite mine were investigated. Firstly, dust measurements were performed and then a health check of all employees was implemented. The obtained data set was categorized by taking into account the occupation, age, experience and level of dust exposure of the employees. While the logistic regression analysis was performed to determine the probability of the diseases, a hierarchical loglinear model was established to investigate the factors in the occurrence of these diseases for those employees with the diseases. Therefore, the most important factors for the development of the diseases were determined by the hierarchical loglinear model.

Change-Point Detection in Multivariate Categorical Processes With Variable Selection

Statistical process control (SPC) has been widely used to control and improve the quality of products in manufacturing processes. Currently, a limited number of schemes is available for change-point detection in multivariate categorical processes (MCPs), where each quality characteristic of products is measured by several attribute levels. Furthermore, existing few methods neglect the dependence structure among quality characteristics so as to provide low efficiency in detecting change points. This paper develops one change-point detection scheme based on the hierarchical log-linear model, which integrates a variable selection procedure to focus only on significant interaction effects, or equivalently dependent relationships among the factors. Therefore, the scheme can achieve higher efficiency than its competing method. The simulation study demonstrates the superiority of the proposed scheme and a real application shows the implementation of the scheme.

Transformed statistics for tests of conditional independence in [formula omitted] contingency tables

Tests of the hypothesis of conditional independence in J×K×L contingency tables are considered. An expression to approximate the null distribution of the test statistics is derived. Using this expression, transformed statistics are obtained which converge to a chi-square limiting distribution faster than the original statistics do. Simulations are used to compare the transformed statistics with the original ones, and transformed statistics are proposed based on a Bartlett-type adjustment. Through this work and earlier ones, we cover testing hierarchical loglinear models in three-way tables except for a model having three interaction terms.

Use of Loglinear Models to Investigate the Patterns of Association among Fitness Test Performances

Background: Physical fitness tests are often administered as a battery, allowing for several different fitness scores. Furthermore, fitness scores are generally considered outcome variables, with no one test causing the performance of another. Therefore, using loglinear models to examine the relationship between fitness test performances is appropriate. Methods: This study used data from a campus-based fitness study of college students. Four physical fitness measures were used in the analysis. Cardiorespiratory fitness (CRF) was assessed by a multi-stage running test (VO2max, ml/kg/min). Lower-body power was assessed by a countermovement vertical jump (VJ) test (inches). Muscular endurance was assessed by a maximal push-up (PU) test (repetitions). Muscular strength was assessed by a 1RM bench press (BP) test. Participants were categorized into high or low fitness groups using each sex-specific fitness score median. Results: Results of the initial hierarchical loglinear models indicated no significant four-way or three-way interactions. A simplified loglinear model with all two-factor interactions provided adequate fit to the data. The odds of high BP performance were much greater with high PU (OR=3.2) and VJ (OR=2.3) performance. The odds of high CRF performance were also greater with high PU (OR=2.4) and VJ (OR=2.7) performance. Finally, odds of high PU performance were much greater with high VJ (OR=3.6) performance. CRF and BP were statistically independent. Conclusion: Results from this study indicate that the performance on a fitness test can be related to at least one other fitness test performance.

Accounting for Surveyor Effort in Large-Scale Monitoring Programs

Abstract Accounting for errors in wildlife surveys is necessary for reliable status assessments and quantification of uncertainty in estimates of population size. We apply a hierarchical log-linear Poisson regression model that accounts for multiple sources of variability in count data collected for the Integrated Waterbird Management and Monitoring Program during 2010–2014. In some large-scale monitoring programs (e.g., Christmas Bird Count) there are diminishing returns in numbers counted as survey effort increases; therefore, we also explore the need to account for variable survey duration as a proxy for effort. In general, we found a high degree of concordance between counts and effort-adjusted estimates of relative abundance from the Integrated Waterbird Management and Monitoring Program (x̄difference = 0.02%; 0.25% SD). We suggest that the model-based adjustments were small because there is only a weak asymptotic relationship with effort and count. Whereas effort adjustments are reasonable and effective when applied to count data from plots of standardized area, such adjustments may not be necessary when the area of sample units is not standardized and surveyor effort increases with number of birds present. That is, large units require more effort only when there are many birds present. The general framework we implemented to evaluate effects of varying survey effort applies to a wide variety of wildlife monitoring efforts.