Multinomial Probit Regression Model Research Articles

Gender role differentiation in household fuel transition decision-making: Implications for education and training in Uganda

IntroductionThe role of gender and gender role differentiation has been of long standing interest and has remained a concern regarding the access and use of energy fuels for cooking in households. Although there seems to be a thin line between gender. However, studies on gender role differentiation in household fuel transition have framed gender as the biological construction of male and female rather than social roles.MethodsThis study used A multinomial probit regression model (MNP) to analyze the effect of gender role differentiation on household transition decisions from high to low-polluting fuels and their implications on education and training in Uganda. The study used the National Household Survey data collected by Uganda Bureau of Statistics.Findings and discussionThe findings revealed that the gender role differentiation significantly affected household fuel transition decisions. The study concludes by highlighting the implications of such gender role differentiation on education and training in Uganda.

Factors Influencing the Choice of Storage Technologies by Smallholder Potato Farmers in Eastern and Southwestern Uganda

Potato is a key food and cash crop in Uganda, mainly produced by smallholder farmers in the eastern and southwestern highlands of the country. This study assessed different factors influencing the choice of storage technologies by Ugandan potato farmers. Data were collected from 240 potato farmers using structured questionnaires in key potato producing districts in eastern and southwestern Uganda. Data were analysed using descriptive statistics and the multinomial probit regression model. Results indicate that potato farmers have limited access to credit and adequate extension services. Furthermore, most of the potato production is sold immediately after harvest. Although significant quantities of potato are stored as food for the household and seed for the next season, very few farmers store ware potato for later sale at a higher price. The farmer households generally use light storage technologies designed for seed storage, while dark stores required for proper ware potato storage are rarely used. Results for factors influencing the choice of storage technologies were mixed, and the extent and direction of influence varied with technology. The predominant factors that positively influenced the choice of dark storage technologies or a combination of different storage technologies included monthly income from sources other than potato sales, access to storage management advice and access to credit. This study recommends enhancing farmers’ access to adequate extension services and credit to promote good ware potato storage conditions.

Cooking fuel choices of households in urban areas in Uganda: a multinomial probit regression analysis

PurposeThis study explores the economic and sociodemographic factors that influence households' decisions on the type of fuel used for cooking in urban areas in Uganda.Design/methodology/approachIn total, two cross-section data surveyed by the Uganda Bureau of Statistics (UBOS) in 2012/13 and 2016/17 were used to analyze consumption of energy for cooking purposes in urban areas of Uganda. This paper employed a multinomial probit regression model and the corresponding marginal effects to analyze cooking fuel choices, which are biomass, electricity and gas and kerosene combined.FindingsThe results showed that household expenditure was statistically significant for the choice of cooking fuel chosen. Furthermore, kitchen type, dwelling type and apartment tenure type are found to be significantly influence the choice of household cooking fuel decisions.Originality/valueThis study takes into consideration the combined influence of the kitchen type, dwelling and tenure type as explanatory variables for the choice of cooking fuel for households in urban areas in Uganda. These factors have not been considered in previous studies done in Uganda, especially within the context of urban households when making choices for cooking fuel.

Family help received by Mexican older adults across socioeconomic strata: changes over a critical decade.

ABSTRACTObjective.To identify how patterns of family economic support help alleviate the cumulative effects of inequality, with focus on the financial support that children give their elderly parents.Methods.This paper uses data from two cross-sections, 2001 and 2012, of the Mexican Health and Aging Study for the 50 years and older population. Analysis includes descriptive statistics to estimate differences in economic support based on family and individual characteristics; and a multinomial probit regression model, in each cross-section, to analyze the amount of money received for economic help and the associated characteristics.Results.Economic help received was significantly reduced, both in proportion, from 20% to 10% between 2001 and 2012, and in the amount received, with differences by income quintile. In 2001, 14.9% of those in the lowest quintile (Q1) would move to Q4–Q5 with children’s help; in 2012, this was 9.1%. The adjusted probability of receiving any amount of money from children decreased from 0.511 in 2001 to 0.340 in 2012.Conclusions.In Mexico, economic inequality in the 50 years and older population remains a constant. Economic help received from children varied by income quintile and plays an important role for those in the lowest income groups. More research is needed to understand the patterns of intergenerational exchanges as these cohorts of older adults continue to age and as future cohorts are entering old age with more pronounced changes than the current cohorts experienced over this critical decade.

Investigating unmet health care needs under the National Health Insurance program in Taiwan: A latent class analysis.

In Taiwan, although the implementation of the National Health Insurance (NHI) program reduced financial barriers and enhanced accessibility for individuals to utilize health care services, an unequal distribution of medical care resources still exists. This paper is focusing on analyzing factors that are associated with unmet health care needs among the middle-aged and elderly under the NHI in Taiwan. Statistical analysis from the 2007 Survey of Health and Living Status of the Elderly in Taiwan. We firstly adopted latent class analysis to classify individuals' observable reasons for feeling unwell but not seeing a doctor within the last 3months into three latent perceived barriers classes. We further used a multinomial probit regression model to analyze factors that are associated with each perceived barrier class to the access of health care service. Results indicate relative to the "relatively no barriers" class, individuals with a high level of educational attainment tend to more likely to be in the "accommodation barriers" class, and individuals live in the most developed areas with the densest medical facilities tend to less likely to be in the "accessibility barriers" class. We identified possible risk factors for each perceived barrier, which could provide important insights for health authorities and medical providers when targeting policies and interventions to efficiently assist people in need.

Sparse bayesian kernel multinomial probit regression model for high-dimensional data classification

In this paper we introduce a sparse Bayesian kernel multinomial probit regression model for multi-class cancer classification. The relationship between the cancer types and gene expression measurements is explained by an unknown function which belongs to an abstract functional space like the reproducing kernel Hilbert space. We assign a sparse prior for regression parameters and perform variable selection by indexing the covariates of the model with a binary vector. The correlation prior for the binary vector assigned in this paper is able to distinguish models with the same size. The proposed method is successfully tested on one simulated data set and two publicly available real life data sets.

Strategies to Manage Risk and their Role in Impacting Economic Performance among Farm Households

Data from the 2015 ARMS and a multinomial probit regression model were used in an attempt to discern the impact of socio-economic factors on the likelihood of a farm household falling in a favorable income-wealth category delineated by above-median income and wealth levels. A primary determinant considered was the number of risk management strategies utilized by the farm household. Findings indicate that farmers who layer up the adoption of risk management strategies are most likely to secure an economic performance status characterized by ‘low-income, high-wealth’. Aging farm operators and those with college education and who operate larger sized farms are found more likely to be in the top economic performance category of ‘high-income, high-wealth’.

Households’ Choices of Fuelwood Sources: Implications for Agroforestry Interventions in the Southern Highlands of Tanzania

Fuelwood is the main source of energy for various household uses in many developing countries; and management of tree resources where it is obtained can be best undertaken when household choice patterns are understood. In this paper, households’ decision to obtain fuelwood for domestic consumption as influenced by household and fuelwood source characteristics is analyzed using a multinomial probit regression model. Data is obtained from 254 randomly selected household drawn from Mbarali district, south-western Tanzania. Results indicate that households are heavily dependent on natural forests for household energy provision and that the choice to obtain fuelwood from the forest, farm or market depends, among other factors on the availability of preferred fuelwood tree species at these sources. Acacia tortilis, Brachystegia spp. and Faidherbia albida are the significantly preferred tree species and hence affect the decision of where to obtain fuelwood. This revelation highlights the pressure applied to surrounding forests as well as to the aforementioned tree species which require immediate management interventions. The gender of the household head and whether the household lives in peri-urban or rural areas also influence choice of fuelwood source. Promotion of tree planting and on-farm management of tree species similar to preferred species found in natural forests is recommended. To the extent that consumer preferences are likely to change over time, further research using panel datasets is necessary to unravel inter-temporal preferences for fuelwood sources.

Critical values of mechanical stress and strain during traumatic cervical spinal cord injury: Clinical study with the use of Finite Element Modelling

Study design: Prospective case-control study. Objective: Aim of the study was to assess critical values of mechanical stress and strain in the cervical spinal cord based on the Finite Element Method (FEM) simulations of specified clinical cases. Summary of background data: The knowledge about the values and distribution of tension and deformation, which are noted at the moment of traumatic spinal cord injury (tSCI) may enable determination of the range of primary and secondary injury. Methods: Total of 28 patients after cervical spine (C-spine) injury were enrolled, 14 with neurological symptoms of tSCI (study group, SG) and 14 neurologically intact (control group, CG). Both groups were age and sex matched. A three-dimensional (3D) numerical model of the cervical spinal cord containing dura and pia matter together with denticulate ligament was created. The variable boundary conditions were established individually for each case and allowed to reconstruct the moment of the injury in computer environment. Factors differentiating between SG and CG were tested with multiple logistic regression model. The predictors of ASIA scale outcomes were evaluated in the ordinal multinomial probit regression model. Results: There were no correlations between age, sex and the level of injury and the values of stress and strain. The results in longitudinal axis (z), in stress (OR-6.3; 95%CI 3.94-8.78; p < .033) and strain (OR-7.8; 95%CI 3.03-10.19; p < .046) were the risk factors of neurological deficits after tSCI. The cut off value for stress was 8.1 kPa (sensitivity-85.7%; specificity-78.6%; AUC-0.819, p < .001), and for strain 0.0117 (sensitivity-92.9%; specificity-72.5%; AUC-0.645, p < .001). Results in the longitudinal axis (z), in stress and strain correspond with grading in ASIA scale. One grade change in ASIA scale correlates with the decrease in z axis by 4.01 kPa and 0.012 in stress and strain respectively. Conclusions: The severity of damage of osseous and ligament structures of the spine, significantly influences the range of the mechanical stress applied to the spinal cord. Neural tissue of the spinal cord is the most resistant to the mechanical stimulus acting in sagittal direction, distraction appears to be the most destructive component of the injury phenomenon.

Sparse Bayesian multinomial probit regression model with correlation prior for high-dimensional data classification

Selecting a small number of relevant genes for cancer classification has received a great deal of attention in microarray data analysis. In this paper, a sparse Bayesian multinomial probit regression model with correlation prior is proposed. Based on simulated and real datasets, we demonstrate that the proposed method performs better than five other competing methods in terms of variable selection and classification.

Sparse Bayesian variable selection in multinomial probit regression model with application to high-dimensional data classification

ABSTRACTHere we consider a multinomial probit regression model where the number of variables substantially exceeds the sample size and only a subset of the available variables is associated with the response. Thus selecting a small number of relevant variables for classification has received a great deal of attention. Generally when the number of variables is substantial, sparsity-enforcing priors for the regression coefficients are called for on grounds of predictive generalization and computational ease. In this paper, we propose a sparse Bayesian variable selection method in multinomial probit regression model for multi-class classification. The performance of our proposed method is demonstrated with one simulated data and three well-known gene expression profiling data: breast cancer data, leukemia data, and small round blue-cell tumors. The results show that compared with other methods, our method is able to select the relevant variables and can obtain competitive classification accuracy with a small subset of relevant genes.

Risk Score Algorithm for Treatment of Persistent Apical Periodontitis

Persistent apical periodontitis related to a nonvital tooth that does not resolve following root canal treatment may be compatible with health and may not require further intervention. This research aimed to develop a Deterioration Risk Score (DRS) to differentiate lesions requiring further intervention from lesions likely to be compatible with health. In this cross-sectional study, patient records (2003-2008) were screened for root-filled teeth with periapical radiolucency visible on periapical radiographs taken at treatment and at recruitment at least 4 yr later. The final sample consisted of 228 lesions in 182 patients. Potential demographic and treatment risk factors were screened against 3 categorical outcomes (improved/unchanged/deteriorated), and a multivariate independent multinomial probit regression model was built. A 5-level DRS was constructed by summing values of adjusted regression coefficients in the model, based on predicted probabilities of deterioration. Most lesions (127, 55.7%) had improved over time, while 32 (14.0%) remained unchanged, and 69 (30.3%) had deteriorated. Significant predictors of deterioration were as follows: time since treatment (relative risk [RR]: 1.11, 95% confidence interval [CI]: 1.01-1.22, p = .030, rounded beta value = 1, for every year increase after 4 yr), current pain (RR: 3.79, 95% CI: 1.48-9.70, p = .005, rounded beta value = 13), sinus tract present (RR: 4.13, 95% CI: 1.11-15.29, p = .034, rounded beta value = 14), and lesion size (RR: 7.20, 95% CI: 3.70-14.02, p < .001, rounded beta value = 20). Persistent apical periodontitis with DRS <15 represented very low risk; 15-20, low risk; 21-30, moderate risk; 31-40, high risk; and >40, very high risk. DRS could help the clinician identify persistent apical periodontitis at low risk for deterioration, and it would not require intervention. When validated, this tool could reduce the risk of overtreatment and contribute toward targeted care and better efficiency in the timely management of disease.

PP2 - Multinational Consistency of a Discrete Choice Model in Quantifying Health States for the Extended 5-Level EQ-5D

Objectives: To investigate the feasibility of choice experiments for EQ-5D-5L states using computer-based data collection, and to examine the consistency of the estimated parameters values derived after modeling the stated preference data across countries in a multinational study. Methods: Similar choice experiments were executed in Canada, England, The Netherlands, and United States (US). Interactive software was developed to standardize the format of the choice tasks across countries, except for England where face-to-face interviewers were used. The choice task required respondents to choose between two sub-optimal health states. A Bayesian design was used to generate 200 pairs of states that were randomly grouped into 20 blocks. Each respondent completed one block consisting of 10 pairs. A main-effect alternative-specific multinomial probit regression model was used to estimate regression coefficients and to derive values for each health state that capture the relative differences in levels between states. Results: In total there were 1775 respondents, at least 400 respondents from each country, who completed 17750 paired comparisons, resulting into 35500 assessed health states. The mean time to perform one choice task was between 29.2 (US) and 45.2 (England) seconds. All regression coefficients were statistically significant, except level 2 for Usual Activities in The Netherlands (p= 0.51). Three regression coefficients with illogical ordering were observed (The Netherlands: level 3 Pain/ Discomfort, England: Level 3 Usual Activities & Pain/Discomfort). Predictions for the complete set of 3125 EQ-5D-5L states were similar for the four countries. Intra class correlation coefficients between the countries were high: from 0.89 (England vs. US) through 0.99 (Canada vs. US). Conclusions: This proof of concept study indicates that computer-based choice tasks for the EQ-5D-5L in the general population are feasible and parameter of the choice tasks estimates are generally consistent and logical, and the estimated values are largely consistent between the 4 countries.

A Typology of Socioeconomic Disadvantage among Farm Operator Households: Assessment of Contributing Factors Using Evidence from a National Survey

Data from the 2004 Agricultural Resource Management Survey are used to construct a typology of socioeconomic disadvantage. Factors that are associated with the likelihood of farm households falling in a socio-economic disadvantaged category are estimated using a multinomial probit regression model. The findings, among others, highlight the importance of factors such as farm operators being married, and being older and more educated in reducing the likelihood of socioeconomic disadvantage. Yet other factors with the potential to lowering the likelihood of socioeconomic disadvantage are those related to increases in off-farm earnings and farm size, and farm program participation.

Factors Associated with Past Research Participation Among Low-Income Persons Living with HIV

We described influences on past research participation among low-income persons living with HIV (PLWH) and examined whether such influences differed by study type. We analyzed a convenience sample of individuals from a large, urban clinic specializing in treating low-income PLWH. Using a computer-assisted survey, we elicited perceptions of research and participating in research, barriers, benefits, "trigger" influences, and self-efficacy in participating in research. Of 193 participants, we excluded 14 who did not identify any type of study participation, and 17 who identified "other" as study type, resulting in 162 cases for analysis. We compared results among four groups (i.e., 6 comparisons): past medical participants (n=36, 22%), past behavioral participants (n=49, 30%), individuals with no past research participation (n=52, 32%), and persons who had participated in both medical and behavioral studies (n=25, 15%). Data were analyzed using chi-square tests for categorical variables and ANOVA for continuous variables. We employed a multinomial probit (MNP) model to examine the association of multiple factors with the outcome. Confidence in ability to keep appointments, and worry about being a 'guinea pig' showed statistical differences in bivariate analyses. The MNP regression analysis showed differences between and across all 6 comparison groups. Fewer differences were seen across groupings of medical participants, behavioral participants, and those with no past research experience, than in comparisons with the medical-behavioral group. In the MNP regression model 'age' and level of certainty regarding 'keeping yourself from being a guinea pig' showed significant differences between past medical participants and past behavioral participants.

Multi-class cancer classification using multinomial probit regression with Bayesian gene selection

We consider the problems of multi-class cancer classification from gene expression data. After discussing the multinomial probit regression model with Bayesian gene selection, we propose two Bayesian gene selection schemes: one employs different strongest genes for different probit regressions; the other employs the same strongest genes for all regressions. Some fast implementation issues for Bayesian gene selection are discussed, including preselection of the strongest genes and recursive computation of the estimation errors using QR decomposition. The proposed gene selection techniques are applied to analyse real breast cancer data, small round blue-cell tumours, the national cancer institute's anti-cancer drug-screen data and acute leukaemia data. Compared with existing multi-class cancer classifications, our proposed methods can find which genes are the most important genes affecting which kind of cancer. Also, the strongest genes selected using our methods are consistent with the biological significance. The recognition accuracies are very high using our proposed methods.

Gene Prediction Using Multinomial Probit Regression with Bayesian Gene Selection

A critical issue for the construction of genetic regulatory networks is the identification of network topology from data. In the context of deterministic and probabilistic Boolean networks, as well as their extension to multilevel quantization, this issue is related to the more general problem of expression prediction in which we want to find small subsets of genes to be used as predictors of target genes. Given some maximum number of predictors to be used, a full search of all possible predictor sets is combinatorially prohibitive except for small predictors sets, and even then, may require supercomputing. Hence, suboptimal approaches to finding predictor sets and network topologies are desirable. This paper considers Bayesian variable selection for prediction using a multinomial probit regression model with data augmentation to turn the multinomial problem into a sequence of smoothing problems. There are multiple regression equations and we want to select the same strongest genes for all regression equations to constitute a target predictor set or, in the context of a genetic network, the dependency set for the target. The probit regressor is approximated as a linear combination of the genes and a Gibbs sampler is employed to find the strongest genes. Numerical techniques to speed up the computation are discussed. After finding the strongest genes, we predict the target gene based on the strongest genes, with the coefficient of determination being used to measure predictor accuracy. Using malignant melanoma microarray data, we compare two predictor models, the estimated probit regressors themselves and the optimal full-logic predictor based on the selected strongest genes, and we compare these to optimal prediction without feature selection.