Semiparametric Additive Regression Model Research Articles

Variational inference: uncertainty quantification in additive models

Markov chain Monte Carlo (MCMC)-based simulation approaches are by far the most common method in Bayesian inference to access the posterior distribution. Recently, motivated by successes in machine learning, variational inference (VI) has gained in interest in statistics since it promises a computationally efficient alternative to MCMC enabling approximate access to the posterior. Classical approaches such as mean-field VI (MFVI), however, are based on the strong mean-field assumption for the approximate posterior where parameters or parameter blocks are assumed to be mutually independent. As a consequence, parameter uncertainties are often underestimated and alternatives such as semi-implicit VI (SIVI) have been suggested to avoid the mean-field assumption and to improve uncertainty estimates. SIVI uses a hierarchical construction of the variational parameters to restore parameter dependencies and relies on a highly flexible implicit mixing distribution whose probability density function is not analytic but samples can be taken via a stochastic procedure. With this paper, we investigate how different forms of VI perform in semiparametric additive regression models as one of the most important fields of application of Bayesian inference in statistics. A particular focus is on the ability of the rivalling approaches to quantify uncertainty, especially with correlated covariates that are likely to aggravate the difficulties of simplifying VI assumptions. Moreover, we propose a method, where we combine both advantages of MFVI and SIVI and compare its performance. The different VI approaches are studied in comparison with MCMC in simulations and an application to tree height models of douglas fir based on a large-scale forestry data set.

On comparison of net survival curves

BackgroundRelative survival analysis is a subfield of survival analysis where competing risks data are observed, but the causes of death are unknown. A first step in the analysis of such data is usually the estimation of a net survival curve, possibly followed by regression modelling. Recently, a log-rank type test for comparison of net survival curves has been introduced and the goal of this paper is to explore its properties and put this methodological advance into the context of the field.MethodsWe build on the association between the log-rank test and the univariate or stratified Cox model and show the analogy in the relative survival setting. We study the properties of the methods using both the theoretical arguments as well as simulations. We provide an R function to enable practical usage of the log-rank type test.ResultsBoth the log-rank type test and its model alternatives perform satisfactory under the null, even if the correlation between their p-values is rather low, implying that both approaches cannot be used simultaneously. The stratified version has a higher power in case of non-homogeneous hazards, but also carries a different interpretation.ConclusionsThe log-rank type test and its stratified version can be interpreted in the same way as the results of an analogous semi-parametric additive regression model despite the fact that no direct theoretical link can be established between the test statistics.

Modeling destructive earthquake casualties based on a comparative study for Turkey

The statistical modeling of destructive earthquakes is an indispensable tool for extracting information for prevention and risk reduction casualties after destructive earthquakes in a seismic region. The linear regression (LR) model can reveal the relation between casualty rate and related covariates based on earthquake catalog. However, if some covariates affect the casualty rate parametrically and some of them nonparametrically, the LR model may entail serious bias and loss of power when estimating or making inference about the effect of parameters. We suggest that semi-parametric beta regression (SBR), semi-parametric additive regression (SAR), and beta regression (BR) models could provide a more suitable description than the LR model to analyze the observed casualties after destructive earthquakes. We support this argument using destructive earthquakes occurred in Turkey between 1900 and 2012 having surface wave magnitudes five or more. The LR, SAR, BR, and SBR models are compared within the context of this data. The data strongly support that the SBR and SAR models can lead to more precise results than the BR and LR models. Furthermore, the SBR is the best model for the earthquake data since the beta distribution provides a flexible model that can be used to analyze the data involving proportions or rates. The results from this model suggest that the casualty rate depends on energy, damaged buildings, and the number of aftershocks of a destructive earthquake.

SMOOTHING SPLINE IN SEMIPARAMETRIC ADDITIVE REGRESSION MODEL WITH BAYESIAN APPROACH

Semiparametric additive regression model is a combination of parametric and nonparametric regression models. The parametric components are not linear but following a polynomial pattern, while the nonparametric components are unknown pattern and assumed to be contained in the Sobolev space. The nonparametric components can be approximated by smoothing spline functions. In the development of smoothing spline, the classical statistical approach cannot be applied for solving the inference problem such as constructing confidence intervals for the regression curve. To construct confidence interval of smoothing spline curve in the semiparametric additive regression model, we propose to use Bayesian approach, by assuming improper Gaussian distribution for prior distribution in nonparametric components and multivariate normal distribution for parametric components. In this study, we obtain parameter estimators for parametric component and smoothing spline estimators for the nonparametric component in semiparametric additive regression model. Moreover, we also develop a smoothing parameters selection method simultaneously using Generalized Maximum Likelihood (GML) and confidence intervals for the parameters of the parametric component and the smoothing spline functions of the nonparametric component using Bayesian approach. By computing each posterior mean and posterior variance of parametric component parameters and smoothing spline functions, confidence intervals can be constructed for the parametric component parameters and confidence interval smoothing spline functions for nonparametric components in semiparametric additive regression models. We create R-code to implement estimation model and inference procedure. Our simulation studies reveal estimation and inference method perform reasonably well.

Estimation of the relationship between body mass index and EQ-5D health utilities in individuals with type 2 diabetes: Evidence from the population-based KORA studies

ObjectivesObesity is known to be an important risk factor for type 2 diabetes and its related comorbid conditions; however, its specific impact on generic health-related quality of life (HRQL) is less clear. The objective of this study was to estimate the association between body mass index (BMI) and HRQL in individuals with type 2 diabetes. MethodsThe EQ-5D quality of life questionnaire was administered in a follow-up of 10,385 participants aged 33–94 of the population-based German MONICA/KORA surveys. 1033 participants with type 2 diabetes were identified by self-report combined with validated physician diagnoses. Semiparametric additive regression models were used to estimate the effect of BMI on EQ-5D health utilities adjusted for age, sex, education and comorbidities. ResultsBMI was significantly associated with EQ-5D health utilities even after adjustment for macro- and microvascular complications. The functional relationship between BMI and utilities was nonlinear, reflecting optimal health around 26kg/m2 and significantly decreasing health utilities with increasing levels of overweight and obesity (−0.09 points between BMI values 26 and 40). Among the diabetic complications, the history of a stroke (−0.13) and neuropathy (−0.10) were the strongest predictors of reduced health utility scores. ConclusionsBMI is strongly associated with health utilities in persons with type 2 diabetes. This suggests that lifestyle measures to reduce obesity can markedly improve patients' health-related quality of life and that the negative effect of potential weight gain should be taken into account when determining patient preferences for different type 2 diabetes treatment options.

Estimating the error distribution function in semiparametric additive regression models

We consider semiparametric additive regression models with a linear parametric part and a nonparametric part, both involving multivariate covariates. For the nonparametric part we assume two models. In the first, the regression function is unspecified and smooth; in the second, the regression function is additive with smooth components. Depending on the model, the regression curve is estimated by suitable least squares methods. The resulting residual-based empirical distribution function is shown to differ from the error-based empirical distribution function by an additive expression, up to a uniformly negligible remainder term. This result implies a functional central limit theorem for the residual-based empirical distribution function. It is used to test for normal errors.

On the construction of efficient estimators in semiparametric models

This paper deals with the construction of efficient estimators in semiparametric models without the sample splitting technique. Schick (1987) gave sufficient conditions using the leave-one-out technique for a construction without sample splitting. His conditions are stronger and more cumbersome to verify than the necessary and sufficient conditions for the existence of efficient estimators which suffice for the construction based on sample splitting. In this paper we use a conditioning argument to weaken Schick′s conditions. We shall then show that in a large class of semiparametric models and for properly chosen estimators of the score function the resulting weaker conditions reduce to the minimal conditions for the construction with sample splitting. In other words, in these models efficient estimators can be constructed without sample splitting under the same conditions as those used for the construction with sample splitting. We demonstrate our results by constructing an efficient estimator using these ideas in a semiparametric additive regression model.

Root-n-consistent and efficient estimation in semiparametric additive regression models

In this paper we consider the semiparametric additive regression model whose regression function is the sum of a linear parametric component and several smooth nonparametric components. We construct root-n-consistent and then efficient estimators of the finite dimensional parameter.