Predictor-response Relationships Research Articles

Transforming healthcare performance monitoring - A cutting-edge approach with generalized additive profiles: GAMs for healthcare quality monitoring.

Recent findings indicate a growing trend in data analysis within healthcare using statistical process control. However, the diversity of variables involved necessitates the expansion of new process control methodologies. This study examined control chart applications in cardiology by using generalized additive models (GAMs) to construct profiles while involving multiple healthcare variables (08). Two distinct statistics: deviation (D), and Hotelling (T2) were employed for constructing control charts: a commonly used single-variable statistic for nonparametric profiles and an innovative multivariate statistic that assesses the contribution of each element to process changes. These statistics were tested for monitoring ischemic and hemorrhagic strokes in 1-year acute stroke (369) patients at the Faisalabad Institute of Cardiology. Demographic parameters (age, gender), vascular risk factors (diabetes, family history, sleep), socioeconomic variables (smoking, location), and blood pressure are included in the model. The research includes the computation of zero-state average run length (ARL) for assessing the performance of control charts. The characteristics of the proposed profile were analyzed, such as the T2 control chart, performing better than the D chart for medium-to-large shifts (δ ≥ 0.50). On the other hand, for small δ = 0.25, the D control chart produces smaller ARL values but more significant standard deviations. While both statistics contribute to profile monitoring, T2 is more effective at identifying and tracing medium and large shifts. In conclusion, such handy tools may aid healthcare performance monitoring, especially for complicated predictor-response relationships. Monitored profiles demonstrated that GAMs are useful for healthcare analysis and process monitoring.

Random Survival Forests With Competing Events: ASubdistribution-Based Imputation Approach.

Random survival forests (RSF) can be applied to many time-to-event research questions and are particularly useful in situations where the relationship between the independent variables and the event of interest is rather complex. However, in many clinical settings, the occurrence of the event of interest is affected by competing events, which means that a patient can experience an outcome other than the event of interest. Neglecting the competing event (i.e., regarding competing events as censoring) will typically result in biased estimates of the cumulative incidence function (CIF). A popular approach for competing events is Fineand Gray's subdistribution hazard model, which directly estimates the CIF by fitting a single-event model defined on a subdistribution timescale. Here, we integrate concepts from the subdistribution hazard modeling approach into the RSF. We develop several imputation strategies that use weights as in a discrete-time subdistribution hazard model to impute censoring times in cases where a competing event is observed. Our simulations show that the CIF is well estimated if the imputation already takes place outside the forest on the overall dataset. Especially in settings with a low rate of the event of interest or a high censoring rate, competing events must not be neglected, that is, treated as censoring. When applied to a real-world epidemiological dataset on chronic kidney disease, the imputation approach resulted in highly plausible predictor-response relationships and CIF estimates of renalevents.

Parametric Quantile Beta Regression Model

SummaryIn this paper, we develop a fully parametric quantile regression model based on the generalised three‐parameter beta (GB3) distribution. Beta regression models are primarily used to model rates and proportions. However, these models are usually specified in terms of a conditional mean. Therefore, they may be inadequate if the observed response variable follows an asymmetrical distribution. In addition, beta regression models do not consider the effect of the covariates across the spectrum of the dependent variable, which is possible through the conditional quantile approach. In order to introduce the proposed GB3 regression model, we first reparameterise the GB3 distribution by inserting a quantile parameter, and then we develop the new proposed quantile model. We also propose a simple interpretation of the predictor–response relationship in terms of percentage increases/decreases of the quantile. A Monte Carlo study is carried out for evaluating the performance of the maximum likelihood estimates and the choice of the link functions. Finally, a real COVID‐19 dataset from Chile is analysed and discussed to illustrate the proposed approach.

A random forest approach to improve estimates of tributary nutrient loading

Estimating constituent loads from discrete water quality samples coupled with stream discharge measurements is critical for management of freshwater resources. Nutrient loads calculated based on discharge-concentration relationships form the basis of government nutrient load targets and scientific studies of the response of receiving waters to external loads. In this study, a new model is developed using random forests and applied to estimate concentrations and loads of total phosphorus, dissolved phosphorus, total nitrogen, and chloride, using data from 17 tributaries to Lake Champlain monitored from 1992 to 2021. I benchmark this model against one of the most widespread models currently used to estimate nutrient loads, Weighted Regressions on Time, Discharge, and Season (WRTDS). The random forest model outperformed both the base WRTDS model and an extension of the WRTDS model using Kalman filtering in the great majority of cases, likely due to the inclusion of rate-of-change in discharge and antecedent discharge over different leading windows as predictors, and to the flexibility of the random forest to model predictor-response relationships. The random forest also had useful visualization capabilities which provided important process insights. WRTDS remains a useful model for many applications, but this study represents a promising new approach for load estimation which can be applied easily to existing datasets, and which is easy to customize for different applications.

Comparative efficacy of three Bayesian variable selection methods in the context of weight loss in obese women.

The use of high-dimensional data has expanded in many fields, including in clinical research, thus making variable selection methods increasingly important compared to traditional statistical approaches. The work aims to compare the performance of three supervised Bayesian variable selection methods to detect the most important predictors among a high-dimensional set of variables and to provide useful and practical guidelines of their use. We assessed the variable selection ability of: (1) Bayesian Kernel Machine Regression (BKMR), (2) Bayesian Semiparametric Regression (BSR), and (3) Bayesian Least Absolute Shrinkage and Selection Operator (BLASSO) regression on simulated data of different dimensions and under three scenarios with disparate predictor-response relationships and correlations among predictors. This is the first study describing when one model should be preferred over the others and when methods achieve comparable results. BKMR outperformed all other models with small synthetic datasets. BSR was strongly dependent on the choice of its own intrinsic parameter, but its performance was comparable to BKMR with large datasets. BLASSO should be preferred only when it is reasonable to hypothesise the absence of synergies between predictors and the presence of monotonous predictor-outcome relationships. Finally, we applied the models to a real case study and assessed the relationships among anthropometric, biochemical, metabolic, cardiovascular, and inflammatory variables with weight loss in 755 hospitalised obese women from the Follow Up OBese patients at AUXOlogico institute (FUOBAUXO) cohort.

Identifying the determinants of crop yields in China since 1952 and its policy implications

Food security is one of the main challenges facing humanity, and increasing crop yields is critical to meet the growing global food demand. Grain production in China has remarkably improved since its founding in 1949, but the growth remains uneven across regions. The goal of this research was to assess the determinants of the yields of three major food crops (maize, rice, and wheat) in China and the mechanism of their responses to increased crop yields. Boosted regression tree models were created to capture the linked complex predictor-response relationships between crop yields and individual explanatory variables using prefecture-level agricultural statistics in China coupled with geographic data during the period from 1952 to 2017. The results showed that technological inputs (e.g., fertilizers, electricity consumption, and power of agricultural machinery) played a key role in the increase of crop yields, which explained 47%, 39%, and 62% of the variances in the yields of maize, rice, and wheat during the period 1952–2017, respectively. However, the contribution of technology became weaker over time, while the contribution of agro-environmental conditions and structural characteristics of administrative regions became stronger. Partial dependence plots indicated that encouraging higher technological inputs, accelerating large-scale grain production, shortening the urban-rural income gap, and improving the education of farmers were conductive to increasing crop yields. Overall, our results suggest that grain production policies aimed at increasing crop yields should better reflect the spatial heterogeneity of yield gaps; low-yield-gap regions should improve the utilization efficiency of water and fertilizer and formulate measures to inhibit non-grain production on cultivated land, while high-yield-gap regions should focus on improving technological inputs and promoting agricultural restructuring. This study provides deep insights into the processes of how individual explanatory variables affect crop yields, which is essential to develop differentiated grain production policies and provide valuable references for shortening yield gaps in high-yield-gap regions.

Exploring the determinants and distribution patterns of soil mattic horizon thickness in a typical alpine environment using boosted regression trees

Mattic horizon, grassroots intertwined surface horizon for soils in alpine meadow, is of great importance for alpine ecosystems with its thickness serving as a critical indicator for the extent of alpine meadow degradation. However, the distribution patterns and key environmental factors controlling mattic horizon thickness (MHT) remain unknown. We evaluated the potential of a boosted regression trees (BRT) based variable selection approach to identify the determinants of MHT, to reveal the predictor-response relationship, and to map the spatial distributions of MHT. Results showed that the selected six key environmental variables can reliably reflect the pedogenic processes of mattic horizon in such alpine environment. Among which, the most important environmental variables controlling MHT were modified soil-adjusted vegetation index 2 (relative importance of 31.9%), elevation (22.7%), followed by slope (13.2%), topographic wetness index (11.4%), relative slope position (10.4%), and mean annual precipitation (10.4%). Additionally, the BRT model using those six key variables for mapping MHT performed an acceptable prediction with R2 value of 0.447 (standard deviation of 0.019). The variable selection procedure dramatically reduced the number of environmental variables by 77% (from 26 to 6), which could not only improve computational efficiency but also significantly improve model performance with a 30.3% increase in R2. Our results demonstrate that the proposed BRT algorithm-based variable selection approach can reliably identify the determinants of MHT both statistically and pedologically. It can be used as a promising approach for revealing the predictor-response relationship, mapping the distribution patterns of MHT, and may provide a solution to map soil depth over large areas.

Maximizing donors' gifts: A comparison of actual and expected solid organ yield among VCA donors.

Vascularized Composite Allograft (VCA) transplantation provides life-changing transplants, but VCA adds complexity to the donation process and timing, possibly impeding solid organ donation. Expanding upon descriptive analyses, this study examines risk-adjusted predictions versus the observed number of organs donated by VCA donors. Our cohort included VCA donors in the United States during January 1, 2008-December 31, 2017 (n=51), using OPTN Deceased Donor Registration Form data and the Scientific Registry of Transplant Recipients (SRTR) donor yield models to calculate observed-to-expected (O:E) yield ratios. Almost all VCA donors' livers (48/51; 94.1%) and kidneys (92/102; 90.2%) were transplanted, with fewer hearts (28/51; 54.9%), lungs (46/102; 45.1%), pancreata (15/51; 29.4%), and intestines (3/51; 5.9%) transplanted. O:E ratios for overall organ yield were slightly greater than expected for VCA donors (1.10; 95% CI: 1.02-1.17). Liver (1.17: 1.08-1.27) and lung yields (1.38: 1.07-1.68) were both greater than expected, while kidney, heart, and pancreas yields were similar to expected. Across VCA types, bilateral upper limb and abdominal wall donors had better-than-expected yields while uterus, face, and unilateral upper limb donors all had similar-to-expected yields. Solid organ yield among VCA donors was as good or better than predicted, suggesting that VCA donation does not compromise recovery and transplantation of lifesaving organs.

Mapping the geogenic radon potential for Germany by machine learning

The radioactive gas radon (Rn) is considered as an indoor air pollutant due to its detrimental effects on human health. In fact, exposure to Rn belongs to the most important causes for lung cancer after tobacco smoking. The dominant source of indoor Rn is the ground beneath the house. The geogenic Rn potential (GRP) - a function of soil gas Rn concentration and soil gas permeability - quantifies what “earth delivers in terms of Rn” and represents a hazard indicator for elevated indoor Rn concentration. In this study, we aim at developing an improved spatial continuous GRP map based on 4448 field measurements of GRP distributed across Germany. We fitted three different machine learning algorithms, multivariate adaptive regression splines, random forest and support vector machines utilizing 36 candidate predictors. Predictor selection, hyperparameter tuning and performance assessment were conducted using a spatial cross-validation where the data was iteratively left out by spatial blocks of 40 km*40 km. This procedure counteracts the effect of spatial auto-correlation in predictor and response data and minimizes dependence of training and test data. The spatial cross-validated performance statistics revealed that random forest provided the most accurate predictions. The predictors selected as informative reflect geology, climate (temperature, precipitation and soil moisture), soil hydraulic, soil physical (field capacity, coarse fraction) and soil chemical properties (potassium and nitrogen concentration). Model interpretation techniques such as predictor importance as well as partial and spatial dependence plots confirmed the hypothesized dominant effect of geology on GRP, but also revealed significant contributions of the other predictors. Partial and spatial dependence plots gave further valuable insight into the quantitative predictor-response relationship and its spatial distribution. A comparison with a previous version of the German GRP map using 1359 independent test data indicates a significantly better performance of the random forest based map.

A Random Forest Approach for Bounded Outcome Variables

Random forests have become an established tool for classification and regression, in particular in high-dimensional settings and in the presence of nonadditive predictor-response relationships. For bounded outcome variables restricted to the unit interval, however, classical modeling approaches based on mean squared error loss may severely suffer as they do not account for heteroscedasticity in the data. To address this issue, we propose a random forest approach for relating a beta distributed outcome to a set of explanatory variables. Our approach explicitly makes use of the likelihood function of the beta distribution for the selection of splits during the tree-building procedure. In each iteration of the tree-building algorithm it chooses one explanatory variable in combination with a split point that maximizes the log-likelihood function of the beta distribution with the parameter estimates derived from the nodes of the currently built tree. Results of several simulation studies and an application using data from the U.S.A. National Lakes Assessment Survey demonstrate the properties and usefulness of the method, in particular when compared to random forest approaches based on mean squared error loss and parametric regression models. Supplementary materials for this article are available online.

Leafing Patterns and Drivers across Seasonally Dry Tropical Communities

Investigating the timing of key phenological events across environments with variable seasonality is crucial to understand the drivers of ecosystem dynamics. Leaf production in the tropics is mainly constrained by water and light availability. Identifying the factors regulating leaf phenology patterns allows efficiently forecasting of climate change impacts. We conducted a novel phenological monitoring study across four Neotropical vegetation sites using leaf phenology time series obtained from digital repeated photographs (phenocameras). Seasonality differed among sites, from very seasonally dry climate in the caatinga dry scrubland with an eight-month long dry season to the less restrictive Cerrado vegetation with a six-month dry season. To unravel the main drivers of leaf phenology and understand how they influence seasonal dynamics (represented by the green color channel (Gcc) vegetation index), we applied Generalized Additive Mixed Models (GAMMs) to estimate the growing seasons, using water deficit and day length as covariates. Our results indicated that plant-water relationships are more important in the caatinga, while light (measured as day-length) was more relevant in explaining leafing patterns in Cerrado communities. Leafing behaviors and predictor-response relationships (distinct smooth functions) were more variable at the less seasonal Cerrado sites, suggesting that different life-forms (grasses, herbs, shrubs, and trees) are capable of overcoming drought through specific phenological strategies and associated functional traits, such as deep root systems in trees.

Effectiveness of filter trading as an intraday trading rule

PurposeFilter trading is a technical trading rule that has been used extensively to test the efficient market hypothesis in the context of long-term trading. In this paper, the authors adopt the rule to analyze intraday trading, in which an open position is not left overnight. This paper aims to explore the relationship between intraday filter trading profitability and intraday realized volatilities. The bivariate thin plate spline (TPS) model is chosen to fit the predictor-response surface for high frequency data from the Hang Seng index futures (HSIF) market. The hypotheses follow the adaptive market hypothesis, arguing that intraday filter trading differs in profitability under different market conditions as measured by realized volatility, and furthermore, the optimal filter size for trading on each day is related to the realized volatility. The empirical results furnish new evidence that range-based realized volatilities (RaV) are more efficient in identifying trading profit than return-based volatilities (ReV). These results shed light on the efficiency of intraday high frequency trading in the HSIF market. Some trading suggestions are given based on the findings.Design/methodology/approachAmong all the factors that affect the profit of filter trading, intraday realized volatility stands out as an important predictor. The authors explore several intraday volatilities measures using range-based or return-based methods of estimation. The authors then study how the filter trading profit will depend on realized volatility and how the optimal filter size is related to the realized volatility. The bivariate TPS model is used to model the predictor-response relationship.FindingsThe empirical results show that range-based realized volatility has a higher predictive power on filter rule trading profit than the return-based realized volatility.Originality/valueFirst, the authors contribute to the literature by investigating the profitability of the filter trading rule on high frequency tick-by-tick data of HSIF market. Second, the authors test the assumption that the magnitude of the intraday momentum trading profit depends on the realized volatilities and aims to identify a relationship between them. Furthermore, the authors consider several intraday realized volatilities and find the RaV have the higher prediction power than ReV. Finally, the authors find some relationship between the optimal filter size and the realized volatilities. Based on the observations, the authors also give some trading suggestions to the intraday filter traders.

DECO: decompose heterogeneous population cohorts for patient stratification and discovery of sample biomarkers using omic data profiling.

MotivationPatient and sample diversity is one of the main challenges when dealing with clinical cohorts in biomedical genomics studies. During last decade, several methods have been developed to identify biomarkers assigned to specific individuals or subtypes of samples. However, current methods still fail to discover markers in complex scenarios where heterogeneity or hidden phenotypical factors are present. Here, we propose a method to analyze and understand heterogeneous data avoiding classical normalization approaches of reducing or removing variation.Results DEcomposing heterogeneous Cohorts using Omic data profiling (DECO) is a method to find significant association among biological features (biomarkers) and samples (individuals) analyzing large-scale omic data. The method identifies and categorizes biomarkers of specific phenotypic conditions based on a recurrent differential analysis integrated with a non-symmetrical correspondence analysis. DECO integrates both omic data dispersion and predictor–response relationship from non-symmetrical correspondence analysis in a unique statistic (called h-statistic), allowing the identification of closely related sample categories within complex cohorts. The performance is demonstrated using simulated data and five experimental transcriptomic datasets, and comparing to seven other methods. We show DECO greatly enhances the discovery and subtle identification of biomarkers, making it especially suited for deep and accurate patient stratification.Availability and implementationDECO is freely available as an R package (including a practical vignette) at Bioconductor repository (http://bioconductor.org/packages/deco/).Supplementary information Supplementary data are available at Bioinformatics online.

Modeling biomarker ratios with gamma distributed components

We propose a regression model termed “extended GB2 model”, which is designed to analyze ratios of biomarkers in epidemiological and medical research. Typical examples of biomarker ratios are given by the LDL/HDL cholesterol ratio in cardiovascular research and the amyloid-$\beta$ 42/40 ratio in dementia research. Unlike regression modeling with a log-transformed response, which is often used to describe ratio outcomes in observational studies, the extended GB2 model directly links the expectation of the untransformed biomarker ratio to a set of covariates. This strategy allows for a simple interpretation of the predictor-response relationships in terms of multiplicative increases/decreases of the expected outcome, similar to Poisson and Cox regression. In the theoretical part of the paper, we derive the log-likelihood of the proposed model, analyze its properties, and provide details on confidence intervals and hypothesis testing. We will also present the results of a simulation study demonstrating the robustness of the proposed modeling approach against model misspecification. The usefulness of the method is demonstrated by two applications on the aforementioned LDL/HDL cholesterol and amyloid-$\beta$ 42/40 ratios. For this, we analyze data from a cohort study on kidney disease and from a large observational database on neurodegenerative diseases.

A spatial ensemble approach for broad-area mapping of land surface properties

Understanding rapid global change requires land cover maps with broad spatial extent, but also fine spatial and temporal resolution. Developing such maps presents a unique challenge, as variability in relationships between spectral characteristics (i.e., predictors) and a response variable is likely to increase with the size of the region across which a model is built and applied. Although most mapping approaches apply the same predictor-response relationships globally across the entire modeling region, learned relationships from one local area may be invalid for another when predicting across broad extents. Here, we adapted a spatial ensemble approach borrowed from species distribution modeling to land cover mapping, and evaluated whether the approach could faithfully represent spatial variation in relationships between land cover and spectral data. The spatiotemporal exploratory model (STEM) uses an ensemble of regression trees defined within spatially overlapping support sets, producing a broad-extent map that reflects variability at the spatial scale of each constituent support set. As test cases for reference maps, we used 30-m-resolution forest canopy and impervious surface cover layers from the 2001 U.S. National Land Cover Database (NLCD) for the states of Washington, Oregon, and California. When testing strategies for support set size and sampling intensity, we found that predictor-response relationships were strongest when individual components of the spatial ensemble were small and when sampling intensity was high. Compared to aspatial bagged decision tree and random forest models, we found that the STEM approach successfully captured variation in our source maps, both globally and at scales smaller than the modeling region. Leveraging the spatial structure of a STEM, we also mapped per-pixel spatial variation in prediction confidence and the importance of different predictor variables. After testing appropriate spatial ensemble and sampling strategies, we extended the predictor-response relationships gleaned from the 2001 source maps into a yearly time series based on temporally-smoothed spectral data from the LandTrendr algorithm. The end products were yearly forest canopy and impervious surface cover time series representing 1990–2012. Formal evaluation showed that our temporally extended maps also closely resembled NLCD maps from 2011. The aim of this research was to cultivate the implicit relationships between spectral data and a given map, not improve them, but as the need for time series maps produced at both broad extents and fine resolutions increases, our results demonstrate that an ensemble of locally defined estimators is potentially more appropriate than conventional ensemble models for land cover mapping across broad extents.

Performance characterisation for risk assessment of striking ship impacts based on struck ship damaged volume

Ship collision accidents are rare events but pose huge threat to human lives, assets, and the environment. Many researchers have sought for effective models that compute ship stochastic response during collisions by considering the variability of ship collision scenario parameters. However, the existing models were limited by the capability of the collision computational models and did not completely capture collision scenario, and material and geometric uncertainties. In this paper, a novel framework to performance characterisation of ships in collision involving a variety of striking ships is developed, by characterising the structural consequences with efficient response models. A double-hull oil carrier is chosen as the struck ship to demonstrate the applicability of the proposed framework. Response surface techniques are employed to generate the most probable input design sets which are used to sample an automated finite element tool to compute the chosen structural consequences. The resulting predictor-response relationships are fitted with suitable surrogate models to probabilistically characterise the struck ship damage under collisions. As demonstrated in this paper, such models are extremely useful to reduce the computational complexity in obtaining probabilistic design measures for ship structures. The proposed probabilistic approach is also combined with available collision frequency models from literature to demonstrate the risk tolerance computations.

A State-Based Sensitivity Analysis for Distinguishing the Global Importance of Predictor Variables in Artificial Neural Networks

Artificial neural networks (ANNs) are powerful empirical approaches used to model databases with a high degree of accuracy. Despite their recognition as universal approximators, many practitioners are skeptical about adopting their routine usage due to lack of model transparency. To improve the clarity of model prediction and correct the apparent lack of comprehension, researchers have utilized a variety of methodologies to extract the underlying variable relationships within ANNs, such as sensitivity analysis (SA). The theoretical basis of local SA (that predictors are independent and inputs other than variable of interest remain “fixed” at predefined values) is challenged in global SA, where, in addition to altering the attribute of interest, the remaining predictors are varied concurrently across their respective ranges. Here, a regression-based global methodology, state-based sensitivity analysis (SBSA), is proposed for measuring the importance of predictor variables upon a modeled response within ANNs. SBSA was applied to network models of a synthetic database having a defined structure and exhibiting multicollinearity. SBSA achieved the most accurate portrayal of predictor-response relationships (compared to local SA and Connected Weights Analysis), closely approximating the actual variability of the modeled system. From this, it is anticipated that skepticisms concerning the delineation of predictor influences and their uncertainty domains upon a modeled output within ANNs will be curtailed.

Using custom scientific workflow software and GIS to inform protected area climate adaptation planning in the Greater Yellowstone Ecosystem

Anticipating the ecological effects of climate change to inform natural resource climate adaptation planning represents one of the primary challenges of contemporary conservation science. Species distribution models have become a widely used tool to generate first-pass estimates of climate change impacts to species probabilities of occurrence. There are a number of technical challenges to constructing species distribution models that can be alleviated by the use of scientific workflow software. These challenges include data integration, visualization of modeled predictor–response relationships, and ensuring that models are reproducible and transferable in an adaptive natural resource management framework. We used freely available software called VisTrails Software for Assisted Habitat Modeling (VisTrails:SAHM) along with a novel ecohydrological predictor dataset and the latest Coupled Model Intercomparison Project 5 future climate projections to construct species distribution models for eight forest and shrub species in the Greater Yellowstone Ecosystem in the Northern Rocky Mountains USA. The species considered included multiple species of sagebrush and juniper, Pinus flexilis, Pinus contorta, Pseudotsuga menziesii, Populus tremuloides, Abies lasciocarpa, Picea engelmannii, and Pinus albicaulis. Current and future species probabilities of occurrence were mapped in a GIS by land ownership category to assess the feasibility of undertaking present and future management action. Results suggested that decreasing spring snowpack and increasing late-season soil moisture deficit will lead to deteriorating habitat area for mountain forest species and expansion of habitat area for sagebrush and juniper communities. Results were consistent across nine global climate models and two representative concentration pathway scenarios. For most forest species their projected future distributions moved up in elevation from general federal to federally restricted lands where active management is currently prohibited by agency policy. Though not yet fully mature, custom scientific workflow software shows considerable promise to ease many of the technical challenges inherent in modeling the potential ecological impacts of climate change to support climate adaptation planning.

Pattern-Aided Regression Modeling and Prediction Model Analysis

This paper first introduces pattern aided regression (PXR) models, a new type of regression models designed to represent accurate and interpretable prediction models. This was motivated by two observations: (1) Regression modeling applications often involve complex diverse predictor-response relationships , which occur when the optimal regression models (of given regression model type) fitting two or more distinct logical groups of data are highly different. (2) State-of-the-art regression methods are often unable to adequately model such relationships. This paper defines PXR models using several patterns and local regression models, which respectively serve as logical and behavioral characterizations of distinct predictor-response relationships. The paper also introduces a contrast pattern aided regression (CPXR) method, to build accurate PXR models. In experiments, the PXR models built by CPXR are very accurate in general, often outperforming state-of-the-art regression methods by big margins. Usually using (a) around seven simple patterns and (b) linear local regression models, those PXR models are easy to interpret; in fact, their complexity is just a bit higher than that of (piecewise) linear regression models and is significantly lower than that of traditional ensemble based regression models. CPXR is especially effective for high-dimensional data. The paper also discusses how to use CPXR methodology for analyzing prediction models and correcting their prediction errors.

The ENSO–Precipitation Teleconnection and Its Modulation by the Interdecadal Pacific Oscillation

Abstract This study evaluates the role of the interdecadal Pacific oscillation (IPO) in modulating the El Niño–Southern Oscillation (ENSO)–precipitation relationship. The standard IPO index is described together with several alternatives that were derived using a low-frequency ENSO filter, demonstrating that an equivalent IPO index can be obtained as a low-frequency version of ENSO. Several statistical artifacts that arise from using a combination of raw and smoothed ENSO indices in modeling the ENSO–precipitation teleconnection are then described. These artifacts include the potentially spurious identification of low-frequency variability in a response variable resulting from the use of smoothed predictors and the potentially spurious modulation of a predictor–response relationship by the low-frequency version of the predictor under model misspecification. The role of the IPO index in modulating the ENSO–precipitation relationship is evaluated using a global gridded precipitation dataset, based on three alternative statistical models: stratified, linear, and piecewise linear. In general, the information brought by the IPO index, beyond that already contained in the Niño-3.4 index, is limited and not statistically significant. An exception is in northeastern Australia using annual precipitation data, and only for the linear model. Stratification by the IPO index induces a nonlinear ENSO–precipitation relationship, suggesting that the apparent modulation by the IPO is likely to be spurious and attributable to the combination of sample stratification and model misspecification. Caution is therefore required when using smoothed climate indices to model or explain low-frequency variability in precipitation.