Semiparametric Logistic Regression Model Research Articles

Semi Parametric Logistic Regression Model with the Outputs Representing Trapezoidal Intuitionistic Fuzzy Number

In this paper, the fuzzy logic and the trapezoidal fuzzy intuitionistic number were presented, as well as some properties of the trapezoidal fuzzy intuitionistic number and semi- parametric logistic regression model when using the trapezoidal fuzzy intuitionistic number. The output variable represents the dependent variable sometimes cannot be determined in only two cases (response, non-response)or (success, failure) and more than two responses, especially in medical studies; therefore so, use a semi parametric logistic regression model with the output variable (dependent variable) representing a trapezoidal fuzzy intuitionistic number.
 the model was estimated on simulation data when sample sizes 25,50 and 100, as the parametric part was estimated by two methods of estimation, are fuzzy ordinary least squares estimators FOLSE method and suggested fuzzy weighted least squares estimators SFWLSE , while the non-parametric part is estimated by Nadaraya Watson estimation and Nearest Neighbor estimator. The results were the fuzzy ordinary least squares estimators method was better than the suggested fuzzy weighted least squares estimators while, in the non-parametric portion, the Nadaraya Watson estimators had better than Nearest Neighbor estimators to estimate the model

Bayesian Variable Selection For Semiparametric Logistic Regression

Lasso variable selection is an attractive approach to improve the prediction accuracy. Bayesian lasso approach is suggested to estimate and select the important variables for single index logistic regression model. Laplace distribution is set as prior to the coefficients vector and prior to the unknown link function (Gaussian process). A hierarchical Bayesian lasso semiparametric logistic regression model is constructed and MCMC algorithm is adopted for posterior inference. To evaluate the performance of the proposed method BSLLR is through comparing it to three existing methods BLR, BPR and BBQR. Simulation examples and numerical data are to be considered. The results indicate that the proposed method get the smallest bias, SD, MSE and MAE in simulation and real data. The proposed method BSLLR performs better than other methods.

Bayesian Semi-Parametric Logistic Regression Model with Application to Credit Scoring Data

Bayesian Semi-Parametric Logistic Regression Model with Application to Credit Scoring Data

An empirical comparison of classification algorithms for mortgage default prediction: evidence from a distressed mortgage market

This paper evaluates the performance of a number of modelling approaches for future mortgage default status. Boosted regression trees, random forests, penalised linear and semi-parametric logistic regression models are applied to four portfolios of over 300,000 Irish owner-occupier mortgages. The main findings are that the selected approaches have varying degrees of predictive power and that boosted regression trees significantly outperform logistic regression. This suggests that boosted regression trees can be a useful addition to the current toolkit for mortgage credit risk assessment by banks and regulators.

A test for the linearity of the nonparametric part of a semiparametric logistic regression model

A semiparametric logistic regression model is proposed in which its nonparametric component is approximated with fixed-knot cubic B-splines. To assess the linearity of the nonparametric component, we construct a penalized likelihood ratio test statistic. When the number of knots is fixed, the null distribution of the test statistic is shown to be asymptotically the distribution of a linear combination of independent chi-squared random variables, each with one degree of freedom. We set the asymptotic null expectation of this test statistic equal to a value to determine the smoothing parameter value. Monte Carlo experiments are conducted to investigate the performance of the proposed test. Its practical use is illustrated with a real-life example.

A comparison of forest fire indices for predicting fire risk in contrasting climates in China

The relationships between fire danger indices and fire risk have been extensively studied in many regions of the world. This work uses partial effect analysis in semiparametric logistic regression models to assess the nonlinear relationships among location, day, altitude, fire danger indices, normalized difference vegetation index (NDVI), and fire ignition from 1996 to 2008 in four different climatic regions in China. The four regions are North China (NR), Northeast China (NE), Southeast China (SE), and Southwest China (SW). The three main results are as follows: First, different fire danger indices are selected as significant variables dependent on the region. The inter-regional difference could be partially explained by difference in local weather and vegetation conditions. Second, spatial location exerts highly significant effects in all four regions. NDVI values are selected as explained variable for NR, NE, and SE on fire ignitions. On a daily scale, altitude influences fire ignition for NR, SE, and SW. Third, the robustness of the probability models used in NE, SE, and SW is better than that in NR on a daily scale. The semiparametric logistic regression model used in this study is useful for assessing the ability of fire danger indices to estimate probabilities of fire ignition on a daily scale. This study encourages further research on assessing the predictive ability of fire danger indices developed at other temporal and spatial scales in China.

不同区域森林火灾对生态因子的响应及其概率模型

火灾是影响森林生态系统过程的重要干扰之一,其对森林生态系统内各生态因子的响应各不相同。由于植被状况及生态环境的不同,森林火灾的时空分布特征在中国不同植被气候类型内表现不同,根据植被气候类型分类系统,将中国主要森林火灾地区划分为4个区域:东北(冷温带松林)、华北(落叶阔叶林)、东南(常绿阔叶林)和西南(热带雨林),应用遥感监测数据和地面环境数据,以时空变量、生态因子(植被生长变化指数、湿度等)为可选自变量,应用半参数化Logistic回归模型,就森林火险对不同生态影响因子的响应规律进行了分析,建立了基于生态因子的着火概率模型和大火蔓延概率模型,通过模拟及实际数据散点图、火险概率图,评估了模型应用价值。结果表明,土壤湿度及植被含水量在落叶阔叶林、常绿阔叶林、热带雨林地区对着火概率影响显著。在4个植被气候区内,土壤及凋落物湿度对大火蔓延的作用较小。在冷温带松林、落叶阔叶林、常绿阔叶林地区,植被生长的年内变化对火灾发生的影响显著,在常绿阔叶林地区,年内植被生长变化对大火蔓延的作用较小。森林火险概率与各生态因子的相关关系主要呈现出非线性。不同植被气候区内,火险概率受不同生态因子组合的影响,这与不同区域的植被状况及生态环境不同有关。在不同植被气候类型,应用时空变量、生态因子建立半参数化logistic回归模型,进行着火概率和大火蔓延概率的模拟具有可行性和实际应用能力。为进一步分析森林生态系统与火灾之间的动态关系、展开生态系统火灾干扰研究提供了理论基础。;Forest fires are considered an important disturbance factor for forest ecosystems. Forest fires are influenced by ecological factors. There are different relationships between diverse ecological factors and forest fires in disparate vegetation regions in China. The objectives of this study were two-fold: i) to assess relationships between ecological factors (NDVI and weather elements) and probability of occurrence of fires (i.e. fire risk), and, ii) to establish fire probability models in four contrasting vegetation regions across China.<br> This study covers the monsoonal region where plenty of rainfall is received but significant seasonal variation within dry and wet seasons are characterised by air mass transitions between inland air and oceanic air. The study area was divided into four sub-regions according to their distinct biomes: i) north China plain with a dominant vegetation of deciduous broad-leaved trees; ii) north-east of China dominated by cool temperate coniferous forest; iii) southeast of China dominated by mixed evergreen broadleaf and deciduous broad-leaved forest, and iv) southwest of China dominated by tropical rain forest. Fire data were extracted from the Along Track Scanning Radiometer of the European Space Agency. Daily values of weather elements from 245 stations covering majority of the four climatic regions above were obtained from China Meteorological Administration. Normalized Difference Vegetation Index (NDVI) was applied as a measure of vegetation status. We linked vegetation with location, time, altitude, weather elements,, and fire characteristics during 1998-2007 in the four regions above using semi-parametric logistic (SPL) regression models. Non-linear relationships between different ecological factors and fire risk (i.e. probability of fire ignition and occurrence of large fire events) were assessed by semi-parametric logistic regression models.<br> We analyzed characteristics of forest fire activity in the four contrasting vegetation regions across China. NDVI was found to be a significant indicator of probability of fire ignition in all the four forest regions except tropical rain forest. On the other hand, NDVI had a significant effect on probability of occurrence of large fire events in all the three forest regions except for evergreen broadleaf region. There was a significant correlation between probability of fire ignition and altitude. Location was found to have a highly significant effect on fire risk in all the four regions, except for the cool temperate coniferous forest. Mean monthly values of ecological factors had more effect on probability of fire ignition than monthly maximum values of those ecological factors. On the contrary, maximum monthly values of ecological factors had more effect on probability of large fire events than mean monthly values of the ecological factors. For each of the four regions of China, a different model of probability of fire ignition and occurrence of large fire events was established. Graphs of observed versus estimated probabilities of large fire events and fire danger maps were produced from the probability models to assess robustness of these models. The results of these final models with selected ecological factors as explanatory variables seem reasonable. The analysis presents some of the dynamics of ecosystem-fire interactions and their value for management systems.

Probability Models of Fire Risk Based on Forest Fire Indices in Contrasting Climates over China

Abstract: Fire weather indices have been widely applied to predict fire risk in many regions of the world. The objectives of this study were to establish fire risk probability models based on fire indices over different climatic regions in China. We linked the indices adopted in Canadian, US, and Australia with location, time, altitude, vegetation and fire characteristics during 1998–2007 in four regions using semi— parametric logistic (SPL) regression models. Different combinations of fire risk indices were selected as explanatory variables for specific regional probability model. SPL regression models of probability of fire ignition and large fire events were established to describe the non—linear relationship between fire risk indices and fire risk probabilities in the four regions. Graphs of observed versus estimated probabilities, fire risk maps, graphs of numbers of large fire events were produced from the probability models to assess the skill of these models. Fire ignition in all regions showed a ...

Efficient semiparametric mean‐association estimation for longitudinal binary responses

Semiparametric methods for longitudinal data with association within subjects have recently received considerable attention. However, existing methods for semiparametric longitudinal binary regression modeling (i) mainly concern mean structures with association parameters treated as nuisance; (ii) generally require a correct specification of the covariance structure for misspecified covariance structure may lead to inefficient mean parameter estimates; and (iii) usually run into computation and estimation problems when the time points are irregularly and possibly subject specific. In this article, we propose a semiparametric logistic regression model, which simultaneously takes into account both the mean and response-association structures (via conditional log-odds ratio) for multivariate longitudinal binary outcomes. Our main interest lies in efficient estimation of both the marginal and association parameters. The estimators of the parameters are obtained via the profile kernel approach. We evaluate the proposed methodology through simulation studies and apply it to a real dataset. Both theoretical and empirical results demonstrate that the proposed method yields highly efficient estimators and performs satisfactorily.

Correcting for Population Stratification in Genomewide Association Studies

Genomewide association studies have become the primary tool for discovering the genetic basis of complex human diseases. Such studies are susceptible to the confounding effects of population stratification, in that the combination of allele-frequency heterogeneity with disease-risk heterogeneity among different ancestral subpopulations can induce spurious associations between genetic variants and disease. This article provides a statistically rigorous and computationally feasible solution to this challenging problem of unmeasured confounders. We show that the odds ratio of disease with a genetic variant is identifiable if and only if the genotype is independent of the unknown population substructure conditional on a set of observed ancestry-informative markers in the disease-free population. Under this condition, the odds ratio of interest can be estimated by fitting a semiparametric logistic regression model with an arbitrary function of a propensity score relating the genotype probability to ancestry-informative markers. Approximating the unknown function of the propensity score by B-splines, we derive a consistent and asymptotically normal estimator for the odds ratio of interest with a consistent variance estimator. Simulation studies demonstrate that the proposed inference procedures perform well in realistic settings. An application to the well-known Wellcome Trust Case-Control Study is presented. Supplemental materials are available online, including a comment by Tchetgen Tchetgen and rejoinder.

A Semiparametric Estimation of Mean Functionals With Nonignorable Missing Data

Parameter estimation with nonignorable missing data is a challenging problem in statistics. The fully parametric approach for joint modeling of the response model and the population model can produce results that are quite sensitive to the failure of the assumed model. We propose a more robust modeling approach by considering the model for the nonresponding part as an exponential tilting of the model for the responding part. The exponential tilting model can be justified under the assumption that the response probability can be expressed as a semiparametric logistic regression model.In this paper, based on the exponential tilting model, we propose a semiparametric estimation method of mean functionals with nonignorable missing data. A semiparametric logistic regression model is assumed for the response probability and a nonparametric regression approach for missing data discussed in Cheng (1994) is used in the estimator. By adopting nonparametric components for the model, the estimation method can be made robust. Variance estimation is also discussed and results from a simulation study are presented. The proposed method is applied to real income data from the Korean Labor and Income Panel Survey.

The penalized profile sampler

The penalized profile sampler for semiparametric inference is an extension of the profile sampler method [B.L. Lee, M.R. Kosorok, J.P. Fine, The profile sampler, Journal of the American Statistical Association 100 (2005) 960–969] obtained by profiling a penalized log-likelihood. The idea is to base inference on the posterior distribution obtained by multiplying a profiled penalized log-likelihood by a prior for the parametric component, where the profiling and penalization are applied to the nuisance parameter. Because the prior is not applied to the full likelihood, the method is not strictly Bayesian. A benefit of this approximately Bayesian method is that it circumvents the need to put a prior on the possibly infinite-dimensional nuisance components of the model. We investigate the first and second order frequentist performance of the penalized profile sampler, and demonstrate that the accuracy of the procedure can be adjusted by the size of the assigned smoothing parameter. The theoretical validity of the procedure is illustrated for two examples: a partly linear model with normal error for current status data and a semiparametric logistic regression model. Simulation studies are used to verify the theoretical results.

The penalized profile sampler

The penalized profile sampler for semiparametric inference is an extension of the profile sampler method (Lee, Kosorok and Fine, 2005) obtained by profiling a penalized log-likelihood. The idea is to base inference on the posterior distribution obtained by multiplying a profiled penalized log-likelihood by a prior for the parametric component, where the profiling and penalization are applied to the nuisance parameter. Because the prior is not applied to the full likelihood, the method is not strictly Bayesian. A benefit of this approximately Bayesian method is that it circumvents the need to put a prior on the possibly infinite-dimensional nuisance components of the model. We investigate the first and second order frequentist performance of the penalized profile sampler, and demonstrate that the accuracy of the procedure can be adjusted by the size of the assigned smoothing parameter. The theoretical validity of the procedure is illustrated for two examples: a partly linear model with normal error for current status data and a semiparametric logistic regression model. As far as we are aware, there are no other methods of inference in this context known to have second order frequentist validity.

Detection of Simulated Nodules on Clinical Radiographs: Dose Reduction at Digital Posteroanterior Chest Radiography

To determine to what extent dose reduction results in decreased detection of simulated nodules on patient digital posteroanterior (PA) chest radiographs. Raw data from 20 clinical digital PA chest images that were reported as having normal findings and that were obtained with a slot-scan charge-coupled device system were used. For research protocol that concerns data with patient identities concealed, institutional review board approval is not required. One hundred twenty nodules varying in size and signal intensity were digitally simulated and added to the chest images. Hard copies were printed to represent a 100% dose and, by adding noise, to represent simulated patient doses of 50%, 25%, and 12%. Four radiologists reviewed images. Each lesion was registered as "detected" or "not detected." A semiparametric logistic regression model was used for statistical analysis. The decrease in radiation dose from 100% to 50%, 25%, or 12% had no effect on lesion detection in the lungs. The decrease in radiation dose had an effect on lesion detection in the mediastinum, as probabilities deteriorated from the 100% dose to the 50%, 25%, and 12% dose with each step. Probabilities of smaller detection rates when compared with that of the reference category (100% dose) were 0.97 (95% confidence interval [CI]: -0.86, 0.012) for the 50% dose, 1 (CI: -0.59, -0.61) for the 25% dose, and 1 (CI: -2.41, -1.22) for the 12% dose. CIs for the effects were on the log(odds). Detection probability decreased with smaller and lower signal intensity lesions. At clinical digital radiography, dose reduction resulted in decreased observer detection of simulated nodules in the mediastinum but not in the lungs.

Sub‐alpine amphibian distributions related to species palatability to non‐native salmonids in the Klamath mountains of northern California

ABSTRACTThe goal of this study was to examine how introduced trout influence the distributions and abundances of a sub‐alpine amphibian assemblage whose members display a variety of different life‐history and defence strategies. Our study was conducted in the sub‐alpine lentic habitats of three wilderness areas that form the core of the Klamath‐Siskiyou Bioregion of northern California, a biodiversity ‘hotspot’ that supports the highest diversity of sub‐alpine, lentic‐breeding amphibians in the western USA. These wilderness areas contain no native fishes, but all have been populated with non‐native trout for recreational fishing. Five of the eight amphibian species that occur in this region were sufficiently common to use in our study; these included one that breeds in both temporary and permanent waters and is palatable to fish (Pacific treefrog, Pseudacris regilla), two that breed primarily in permanent waters and are unpalatable to fish (western toad, Bufo boreas, and rough‐skinned newt, Taricha granulosa), and two that breed primarily in permanent waters and are palatable to fish (Cascades frog, Rana cascadae, and long‐toed salamander, Ambystoma macrodactylum). Based on life histories and predator defence strategies (i.e. palatable or not), we predicted that the three palatable species would likely be negatively correlated with introduced trout, but with P. regilla less impacted because of its use of both temporary and permanent waters. We predicted that B. boreas and T. granulosa would not be significantly correlated with introduced trout due to the lack of any predator/prey interactions between them. We surveyed 728 pond, lake, or wet meadow sites during the summers of 1999–2002, using timed gill‐net sets to measure trout occurrence and relative density, and visual encounter surveys to determine amphibian presence and abundance. We used semiparametric logistic regression models to quantify the effect of trout presence/absence and density on the probability of finding amphibian species in a water body while accounting for variation within and among the various lentic habitats sampled. The distributions of P. regilla, A. macrodactylum and R. cascadae were strongly negatively correlated with trout presence across all three wilderness areas. Ambystoma macrodactylum was 44 times more likely to be found in lakes without fish than in lakes with fish. Rana cascadae and P. regilla were 3.7 and 3.0 times more likely, respectively, to be found in fishless than fish‐containing waters. In contrast, the two unpalatable species were either uncorrelated (T. granulosa) or positively correlated (B. boreas) with fish presence. We found that the relative density of fish (catch per unit effort) was negatively correlated with the combined abundances of the three palatable amphibians, and also with both the length and the condition of the fish themselves. Our results are consistent with a compelling body of evidence that introduced fishes greatly alter the aquatic community structure of mountain lakes, ponds, and wet meadows.

Comparison of Eight Different Digital Chest Radiography Systems: Variation in Detection of Simulated Chest Disease

In a short period, a variety of technically different digital radiography chest systems have become available for clinical use. Our purpose was to assess the diagnostic performance of eight different digital radiography chest systems for detection of simulated chest disease under clinical conditions. Assessed were four different flat-panel detector systems, two different charge-coupled device systems, one selenium-coated drum, and one storage phosphor system. For each system, 10 chest images of an anthropomorphic chest phantom were obtained. Each image contained one to 12 simulated chest lesions. Eight radiologists performed soft-copy interpretations. Entrance dose was measured and effective dose calculated. A semi-parametric logistic regression model was used for statistical analysis. Statistically significant differences were found in the diagnostic performance of the eight digital chest systems (p = 0.01). Best performance was observed with the charge-coupled device system with slot-scan technology, yielding a sensitivity of 46% (132 of 288) lesions detected. The performance of three flat-panel detectors and the selenium-drum system was not significantly different from the slot-scan charge-coupled device system. Fewer lesions were detected with the storage phosphor system than with all other digital technologies, with a sensitivity of 34% (99 of 288) lesions detected, slot-scan charge-coupled device system versus storage phosphor system, p < 0.001. The effective dose varied among the digital systems. We found differences in diagnostic performance among the eight different digital chest systems. Differences in detection rates are predominantly explained by detector design.

ANTILYMPHOCYTE INDUCTION THERAPY IN CADAVER RENAL TRANSPLANTATION

Antilymphocyte induction therapy in cadaver renal transplantation is controversial. The effectiveness of antilymphocyte therapy in the current era of cyclosporine and tacrolimus use has been questioned. The United Network for Organ Sharing data set for the Center-Specific Outcomes Analysis, which has been verified by the transplant centers, was used for this study. At the time information in the database was confirmed, all transplant centers were queried on their use of an antilymphocyte preparation at the time of transplantation, and whether it was used within 24 hr of transplant surgery, the duration of the specific reagent. This allowed us to analyze 24,191 cadaver transplant procedures performed between the October 1, 1987, and the January 31, 1991. Using Cox regression analysis, as well as semiparametric logistic regression models, we demonstrated improved allograft outcomes in patients who received either Minnesota antilymphocyte globulin for 5 days or more or OKT3 for 7 days or more. The relative risk was 0.82 for Minnesota antilymphocyte globulin and 0.86 for OKT3 (for both, P<0.001). Semiparametric models were then used to compare the effectiveness of the antilymphocyte preparation in both a patient with at least a three-antigen mismatch and patients who had a zero-antigen mismatch. The improvement in graft survival was seen in both match grades. These data demonstrate the improved outcomes with the use of antilymphocyte preparations during the early posttransplant period in the modern cyclosporine era.