Sampling Bias Correction Research Articles

Bunching up the background betters bias in species distribution models

Sets of presence records used to model species’ distributions typically consist of observations collected opportunistically rather than systematically. As a result, sampling probability is geographically uneven, which may confound the model's characterization of the species’ distribution. Modelers frequently address sampling bias by manipulating training data: either subsampling presence data or creating a similar spatial bias in non‐presence background data. We tested a new method, which we call ‘background thickening’, in the latter category. Background thickening entails concentrating background locations around presence locations in proportion to presence location density. We compared background thickening to two established sampling bias correction methods – target group background selection and presence thinning – using simulated data and data from a case study. In the case study, background thickening and presence thinning performed similarly well, both producing better model discrimination than target group background selection, and better model calibration than models without correction. In the simulation, background thickening performed better than presence thinning when the number of simulated presence locations was low, and vice versa. We discuss drawbacks to target group background selection, why background thickening and presence thinning are conservative but robust sampling bias correction methods, and why background thickening is better than presence thinning for small sample sizes. Particularly, background thickening is advantageous for treating sampling bias when data are scarce because it avoids discarding presence records.

Optimizing MaxEnt model in the prediction of species distribution.

Maximum Entropy (MaxEnt) model has been widely used in recent years. However, MaxEnt is highly inclined to produce misleading results if it is not well optimized. We summarized the researches about the model optimization for sampling bias correction, model complexity tuning, presence-absence threshold selection, and model evaluation. Spatial filtering performs best for sampling bias correction, while restricted background method shows the lowest efficacy. Model complexi-ty is mainly determined by three factors: The number of environmental variables, model feature types, and regularization multiplier. Variables filtering is needed when sample size is less than the number of environment variables. The criterion of variables selection should focus on their ecological significance rather than the co-linearity between them. The choice of feature types has relatively limi-ted effects on predictive performance of the model, therefore it is advised to choose simpler models. To control overfitting, it is necessary to conduct species-specific tuning on regularization multiplier, which was usually bigger than the default setting. There are three criteria called objectivity, equality and discriminability for selecting threshold to convert continuous predication (e.g. probability of presence) into binary results. Maximizing the sum of sensitivity and specificity is a sound method for threshold selection. Model evaluation methods could be classified into two main types: Threshold-independent and threshold-dependent. Among the threshold-independent evaluations, information criteria may offer significant advantages over AUC and COR. True Skill Statistics is a better index for threshold-dependent evaluations, because it takes both omission and commission errors into account, and is robust to pseudo-absence assumption and species prevalence.

Small sample bias correction or bias reduction?

Many problems in biomedical and other sciences are subject to biased estimates (maximum likelihood or of similar types). In two seminal papers Cox and Snell (1968) and Firth (1993) deal with first order bias of maximum likelihood estimates. Cox and Snell obtain a correction term that corrects, approximately, first order bias and Firth uses an adjustment to the score function; the solution of the estimating equation obtained by solving the adjusted score function to zero, removes the first order bias of the maximum likelihood estimates approximately. In many applications authors use one of these two procedures for bias correction without being aware that the other exists or whether these two procedures are equivalent. In this paper we investigate the equivalence issue of the two methods through theoretical analysis, simulation study and data analysis. We show that the two methods yield either exactly the same estimates or that the preventive method has some edge over the other.

A matter of timing: how temporal scale selection influences cetacean ecological niche modelling

Modelling in the marine environment faces unique challenges that place greater emphasis on model accuracy. The spatio-temporal variability of this environment presents challenges when trying to develop useful habitat models. We tested how different temporal scales influence model predictions for cetaceans with different ecological requirements. We used 7 years of (opportunistic) whale watching data (>16000 cetacean sightings) collected in the Azores archipelago under the MONICET platform. We modelled the distribution of 10 cetacean species with a sampling bias correction. Distribution modelling was performed at 2 spatial scales (2 and 4 km) and 2 temporal resolutions (8 d vs. monthly averages). We used a MAXENT analysis with 3 different validation procedures. Generally, the 8 d means produced better results. In some cases (e.g. baleen whales), predictions using monthly means were no better than null models. Finer temporal grains provided essential insights, especially for species influenced by dynamic variables (e.g. sea surface temperature). For species more influenced by static variables (e.g. bathymetry), differences between temporal scales were smaller. The selection of the right temporal scale can be essential when modelling the niches of cetaceans. Datasets with high temporal resolution (e.g. whale watching data) can provide an excellent basis for these analyses, allowing use of finer temporal grains. Our models showed good predictive performance; however, limitations related to the spatial coverage were found. Merging datasets with different temporal and spatial resolutions could help to improve niche estimates. Models with better predictive capacity and transferability are needed to implement more efficient protection and conservation measures.

Nonparametric change point estimation for survival distributions with a partially constant hazard rate.

We present a new method for estimating a change point in the hazard function of a survival distribution assuming a constant hazard rate after the change point and a decreasing hazard rate before the change point. Our method is based on fitting a stump regression to p values for testing hazard rates in small time intervals. We present three real data examples describing survival patterns of severely ill patients, whose excess mortality rates are known to persist far beyond hospital discharge. For designing survival studies in these patients and for the definition of hospital performance metrics (e.g.mortality), it is essential to define adequate and objective end points. The reliable estimation of a change point will help researchers to identify such end points. By precisely knowing this change point, clinicians can distinguish between the acute phase with high hazard (time elapsed after admission and before the change point was reached), and the chronic phase (time elapsed after the change point) in which hazard is fairly constant. We show in an extensive simulation study that maximum likelihood estimation is not robust in this setting, and we evaluate our new estimation strategy including bootstrap confidence intervals and finite sample bias correction.

Bias correction for time series factor models

ABSTRACTIn this paper we work with multivariate time series that follow a Factor Model. In particular, we consider the setting where factors are dominated by highly persistent AutoRegressive (AR) processes and samples that are rather small. Therefore, the factors' AR models are estimated using small sample bias correction techniques. A Monte Carlo study reveals that bias-correcting the AR coefficients of the factors allows to obtain better results in terms of prediction interval coverage. As expected, the simulation shows that bias-correction is more successful for smaller samples. We present the results assuming the AR order and number of factors are known as well as unknown. We also study the advantages of this technique for a set of Industrial Production Indexes of several European countries.

Wrong, but useful: regional species distribution models may not be improved by range-wide data under biased sampling.

Species distribution modeling (SDM) is an essential method in ecology and conservation. SDMs are often calibrated within one country's borders, typically along a limited environmental gradient with biased and incomplete data, making the quality of these models questionable. In this study, we evaluated how adequate are national presence‐only data for calibrating regional SDMs. We trained SDMs for Egyptian bat species at two different scales: only within Egypt and at a species‐specific global extent. We used two modeling algorithms: Maxent and elastic net, both under the point‐process modeling framework. For each modeling algorithm, we measured the congruence of the predictions of global and regional models for Egypt, assuming that the lower the congruence, the lower the appropriateness of the Egyptian dataset to describe the species' niche. We inspected the effect of incorporating predictions from global models as additional predictor (“prior”) to regional models, and quantified the improvement in terms of AUC and the congruence between regional models run with and without priors. Moreover, we analyzed predictive performance improvements after correction for sampling bias at both scales. On average, predictions from global and regional models in Egypt only weakly concur. Collectively, the use of priors did not lead to much improvement: similar AUC and high congruence between regional models calibrated with and without priors. Correction for sampling bias led to higher model performance, whatever prior used, making the use of priors less pronounced. Under biased and incomplete sampling, the use of global bats data did not improve regional model performance. Without enough bias‐free regional data, we cannot objectively identify the actual improvement of regional models after incorporating information from the global niche. However, we still believe in great potential for global model predictions to guide future surveys and improve regional sampling in data‐poor regions.

Evaluation of Blood Pressure Control Medicines Using Health and Medical Checkup Data in Japan: Alternative Methods for Randomized Controlled Trials

Background: Many people take medicines to control high blood pressure (BP), or hypertension. Randomized clinical trials (RCT) are usually used for the evaluation of effects of medicines. However, RCT have some serious problems. Data and Methods: We evaluated the effects of BP medicines in Japan using a dataset containing 113,979 cases. We employed four statistical methods in the analysis. First, we simply compared the systolic blood pressure (SBP) of individuals with and without BP medicines. We then used a regression model with a dummy variable, representing taking medicines or not. We replaced the dummy variable by its expected value, and estimated the regression model again. Finally, we selected individuals who had both taken and not taken medicines at different times. The effect of sample selection was also considered in the estimation. Results: For the simple comparison, SBP with BP medicines was 11 mmHg higher than without medicines. In the next regression analysis, SBP with BP medicines was still 5 mmHg higher. When the dummy variable was replaced by its expected value, SBP with medicines decreased by 7 mmHg. For individuals taking medicines at some times and not at others, SBP decreased by 9 and 8 mmHg in models with and without a sample bias correction, respectively. Conclusion: The methods eliminated some problems of RCT and might be attractive. However, we obtained contradictory conclusions depending on the statistical methods employed, despite using the identical dataset. Statistical methods must be selected carefully to obtain a reliable evaluation. Limitations: The dataset was observatory, and the sample period was only 3 years.

Modeling relative habitat suitability of southern Florida for invasive Burmese pythons (Python molurus bivittatus)

Invasive Burmese pythons are altering the ecology of southern Florida and their distribution is expanding northward. Understanding their habitat use is an important step in understanding the pathways of the invasion. This study identifies key landscape variables in predicting relative habitat suitability for pythons at the present stage of invasion through presence-only ecological niche modeling using geographical sampling bias correction. We used 2014 presence-only observations from the EDDMapS database and three landscape variables to model habitat suitability: fine-scale land cover, home range-level land cover, and distance to open freshwater or wetland. Ten geographical sampling bias correction scenarios based on road presence and sampling effort were evaluated to improve the efficacy of modeling. The best performing models treated road presence as a binary factor rather than a continuous decrease in sampling effort with distance from roads. Home range-level cover contributed the most to the final prediction, followed by proximity to water and fine-scale land cover. Estuarine habitat and freshwater wetlands were the most important variables to contribute to python habitat suitability at both the home range-level and fine-scale. Suitability was highest within 30 m of open freshwater and wetlands. This study provides quantifiable, predictive relationships between habitat types and python presence at the current stage of invasion. This knowledge can elucidate future targeted studies of python habitat use and behavior and help inform management efforts. Furthermore, it illustrates how estimates of relative habitat suitability derived from MaxEnt can be improved by both multi-scale perspectives on habitat and consideration of a variety of bias correction scenarios for selecting background points.

Improved species‐occurrence predictions in data‐poor regions: using large‐scale data and bias correction with down‐weighted Poisson regression and Maxent

Species distribution modelling (SDM) has become an essential method in ecology and conservation. In the absence of survey data, the majority of SDMs are calibrated with opportunistic presence‐only data, incurring substantial sampling bias. We address the challenge of correcting for sampling bias in the data‐sparse situations. We modelled the relative intensity of bat records in their entire range using three modelling algorithms under the point‐process modelling framework (GLMs with subset selection, GLMs fitted with an elastic‐net penalty, and Maxent). To correct for sampling bias, we applied model‐based bias correction by incorporating spatial information on site accessibility or sampling efforts. We evaluated the effect of bias correction on the models’ predictive performance (AUC and TSS), calculated on spatial‐block cross‐validation and a holdout data set. When evaluated with independent, but also sampling‐biased test data, correction for sampling bias led to improved predictions. The predictive performance of the three modelling algorithms was very similar. Elastic‐net models have intermediate performance, with slight advantage for GLMs on cross‐validation and Maxent on hold‐out evaluation. Model‐based bias correction is very useful in data‐sparse situations, where detailed data are not available to apply other bias correction methods. However, bias correction success depends on how well the selected bias variables describe the sources of bias. In this study, accessibility covariates described bias in our data better than the effort covariate, and their use led to larger changes in predictive performance. Objectively evaluating bias correction requires bias‐free presence–absence test data, and without them the real improvement for describing a species’ environmental niche cannot be assessed.

Forest cover and level of protection influence the island-wide distribution of an apex carnivore and umbrella species, the Sri Lankan leopard (Panthera pardus kotiya)

Apex predators fulfil potentially vital ecological roles. Typically wide-ranging and charismatic, they can also be useful surrogates for biodiversity preservation, making their targeted conservation imperative. The Sri Lankan leopard (Panthera pardus kotiya), an endangered, endemic sub-species, is the island’s apex predator. Of potential keystone importance, this carnivore also fulfills “umbrella” and “flagship” criterion and is of high ecological and existence value. Apex predator conservation requires identifying factors underlying distribution, so we used multi-scale maximum entropy modelling with sampling bias correction to investigate a broad suite of relevant ecological, climatic and anthropogenic factors in order to identify potentially suitable leopard habitat. Presence locations were determined from 15 years of surveys, observations and verified reports. The best bias correction procedure and scale were uncertain, so we employed a novel method of using information from all models across analyses to determine top models and identify influential variables. Leopard presence was most strongly linked to the landscape proportion encompassed by Protected Areas strictly limiting human presence, with more porous Protected Areas less influential. All three forest composition and configuration metrics investigated (area weighted mean patch size, patch density and forest connectivity) were influential, with increased patch size and higher connectivity predicting better habitat suitability for leopards. Habitat suitability was also better where cropland extent and urban patch size were small. In summary, ground-level protection and natural forest extent and connectivity are of profound importance to Sri Lankan leopard distribution and are key factors in ensuring the ecological integrity of the island’s faunal assemblages.

Methods for Analyzing Binary Repeated Measures: The Small Sample Case

Binary repeated measurements occur often in a variety of fields. Particularly in medicine, small samples are used in the early phases (phase I and II) of clinical trials, in bio equivalence studies and in crossover trials where human participation is multitudinous. Hence, it is vital to develop a precise method to analyze binary Repeated Measures Data (RMD) with small sample size which is related to humans and even to animals. As a result, this simulation study was carried out in SAS to examine the performance of the two methods used with general sample sizes; the Generalized Estimating Equations (GEE) method and Generalized Linear Mixed Models (GLMM) towards analysis of binary RMD with small sample size, after adjusting the bias that occurs in small samples. Being motivated by the study of literature, large scale simulations are carried out for each method with the facilitation of PROC GENMOD and PROC GLIMMIX procedures respectively, along with varying options of small sample bias correction methods available in SAS, the Sandwich Variance Estimation (SVE) technique and its variants. Each method with all possible SVE techniques available in SAS were compared and contrasted with respect to the properties; Type I error, power, unbiasedness, consistency, sufficiency, convergence, speed of computation and efficiency. The results obtained from the simulation study depicted that for binary RMD which adhere to AR(1) process, with no missing values and with no covariates, GLMM with SVE techniques FIROEEQ and ROOT perform equally and exceptionally well for a small sample size binary repeated case with respect to all the properties of parameter estimates considered except for sufficiency. However the GEE method with the naive option while being marginal with respect to type I error, performs well in analyzing very small sample sizes and satisfies all the properties including sufficiency.

The Role of Oil Prices in Exchange Rate Movements: The CIS Oil Exporters

Undoubtedly, oil prices play a crucial role in the macroeconomic performances of oil-exporting developing countries. In this regard, the exchange rate is one of the key macroeconomic indicators worthy of investigation. Existing literature shows that world oil prices play an important role in the appreciation of the exchange rates of oil-exporting developing countries. However, only a few studies have examined this issue by considering all three oil-exporting countries of the Commonwealth Independent States, namely Azerbaijan, Kazakhstan and Russia, together. In order to fill this gap and given the increasing importance of these economies in the world’s energy markets, this paper examines the role of oil prices in the movement of real effective exchange rates of the above-mentioned CIS countries. We applied the autoregressive distributed lag bounds testing method with a small sample bias correction to the data of these countries over the 2004Q1–2013Q4 period. The estimation results indicate that oil prices are certainly a main driver behind real effective exchange rate appreciation in the selected economies. Moreover, estimations show that productivity, to some extent, can also lead to the appreciation. The policy implication of this research is that an appreciation of the real exchange rate is harmful for the exports of non-oil goods and services in these countries. Since oil prices lead to the appreciation mainly through higher domestic prices, which is a result of tremendous public spending, decision-makers should reconsider the prevailing fiscal policy to make it much more counter-cyclical.

Apoyo público e innovación a nivel de firmas

This article intends to contribute to the analyses of the effects of public policies on innovation at the firm level, based on the case of a small developing economy (Uruguay). It studies the relation between public support and the probability of innovating at the market level, introducing a lag between support and innovation. The econometric analysis uses binomial Probit models with Heckman sampling bias correction. The article concludes that public support matters for innovation at the market level, whether by stimulating innovation or by facilitating it in already innovative firms. Moreover, the innovation activities of Uruguayan firms receiving public support do not seem to require a long time to develop. Finally, it is suggested that the relation between firm internationalization and the probability of innovating be further studied.

Performance tradeoffs in target‐group bias correction for species distribution models

Species distribution models (SDMs) are often calibrated using presence‐only datasets plagued with environmental sampling bias, which leads to a decrease of model accuracy. In order to compensate for this bias, it has been suggested that background data (or pseudoabsences) should represent the area that has been sampled. However, spatially‐explicit knowledge of sampling effort is rarely available. In multi‐species studies, sampling effort has been inferred following the target‐group (TG) approach, where aggregated occurrence of TG species informs the selection of background data. However, little is known about the species‐ specific response to this type of bias correction.The present study aims at evaluating the impacts of sampling bias and bias correction on SDM performance. To this end, we designed a realistic system of sampling bias and virtual species based on 92 terrestrial mammal species occurring in the Mediterranean basin. We manipulated presence and background data selection to calibrate four SDM types. Unbiased (unbiased presence data) and biased (biased presence data) SDMs were calibrated using randomly distributed background data. We used real and TG‐estimated sampling efforts in background selection to correct for sampling bias in presence data.Overall, environmental sampling bias had a deleterious effect on SDM performance. In addition, bias correction improved model accuracy, and especially when based on spatially‐explicit knowledge of sampling effort. However, our results highlight important species‐specific variations in susceptibility to sampling bias, which were largely explained by range size: widely‐distributed species were most vulnerable to sampling bias and bias correction was even detrimental for narrow‐ranging species. Furthermore, spatial discrepancies in SDM predictions suggest that bias correction effectively replaces an underestimation bias with an overestimation bias, particularly in areas of low sampling intensity. Thus, our results call for a better estimation of sampling effort in multispecies system, and cautions the uninformed and automatic application of TG bias correction.

Adaptive Correction of Sampling Bias in Dynamic Call Graphs

This article introduces a practical low-overhead adaptive technique of correcting sampling bias in profiling dynamic call graphs. Timer-based sampling keeps the overhead low but sampling bias lowers the accuracy when either observable call events or sampling actions are not equally spaced in time. To mitigate sampling bias, our adaptive correction technique weights each sample by monitoring time-varying spacing of call events and sampling actions. We implemented and evaluated our adaptive correction technique in Jikes RVM, a high-performance virtual machine. In our empirical evaluation, our technique significantly improved the sampling accuracy without measurable overhead and resulted in effective feedback directed inlining.

Inference for impulse response coefficients from multivariate fractionally integrated processes

ABSTRACTThis article considers a multivariate system of fractionally integrated time series and investigates the most appropriate way for estimating Impulse Response (IR) coefficients and their associated confidence intervals. The article extends the univariate analysis recently provided by Baillie and Kapetanios (2013), and uses a semiparametric, time domain estimator, based on a vector autoregression (VAR) approximation. Results are also derived for the orthogonalized estimated IRs which are generally more practically relevant. Simulation evidence strongly indicates the desirability of applying the Kilian small sample bias correction, which is found to improve the coverage accuracy of confidence intervals for IRs. The most appropriate order of the VAR turns out to be relevant for the lag length of the IR being estimated.

Factores determinantes del salario del sector privado en el Ecuador para el año 2014: un caso de estudio en la ciudad de Guayaquil

In Ecuador, wage compensation for private sector workers has become a difficult problem to deal with and conciliate throughout history. The current Ecuadorean Constitution approved in Montecristi in 2008 strongly protects labour rights; however, many companies do not obey the law. This study investigates the variables that are significant to address private-sector wages in Guayaquil city. An analysis of the results in the previous literature and a review by Mincer (1974) suggest these variables: years of work experience, education, gender of the employee, position, seniority in the company, and marital status. Interactive and quadratic variables were also included. The inference methodology selected was ordinary least squares (OLS) with robust standard errors of White, and taking into account the sample bias correction proposed by Heckman (1979). The results show that the significant variables concerning human capital depend on the type of company they are working for and its interaction with other variables. Thus, empirical evidence was found to support that small companies are not paying their labour force in terms of productivity, specifically to new employees.

Mapping Species Distributions with MAXENT Using a Geographically Biased Sample of Presence Data: A Performance Assessment of Methods for Correcting Sampling Bias

MAXENT is now a common species distribution modeling (SDM) tool used by conservation practitioners for predicting the distribution of a species from a set of records and environmental predictors. However, datasets of species occurrence used to train the model are often biased in the geographical space because of unequal sampling effort across the study area. This bias may be a source of strong inaccuracy in the resulting model and could lead to incorrect predictions. Although a number of sampling bias correction methods have been proposed, there is no consensual guideline to account for it. We compared here the performance of five methods of bias correction on three datasets of species occurrence: one “virtual” derived from a land cover map, and two actual datasets for a turtle (Chrysemys picta) and a salamander (Plethodon cylindraceus). We subjected these datasets to four types of sampling biases corresponding to potential types of empirical biases. We applied five correction methods to the biased samples and compared the outputs of distribution models to unbiased datasets to assess the overall correction performance of each method. The results revealed that the ability of methods to correct the initial sampling bias varied greatly depending on bias type, bias intensity and species. However, the simple systematic sampling of records consistently ranked among the best performing across the range of conditions tested, whereas other methods performed more poorly in most cases. The strong effect of initial conditions on correction performance highlights the need for further research to develop a step-by-step guideline to account for sampling bias. However, this method seems to be the most efficient in correcting sampling bias and should be advised in most cases.

Nondetection sampling bias in marked presence‐only data

Species distribution models (SDM) are tools used to determine environmental features that influence the geographic distribution of species' abundance and have been used to analyze presence-only records. Analysis of presence-only records may require correction for nondetection sampling bias to yield reliable conclusions. In addition, individuals of some species of animals may be highly aggregated and standard SDMs ignore environmental features that may influence aggregation behavior.We contend that nondetection sampling bias can be treated as missing data. Statistical theory and corrective methods are well developed for missing data, but have been ignored in the literature on SDMs. We developed a marked inhomogeneous Poisson point process model that accounted for nondetection and aggregation behavior in animals and tested our methods on simulated data.Correcting for nondetection sampling bias requires estimates of the probability of detection which must be obtained from auxiliary data, as presence-only data do not contain information about the detection mechanism. Weighted likelihood methods can be used to correct for nondetection if estimates of the probability of detection are available. We used an inhomogeneous Poisson point process model to model group abundance, a zero-truncated generalized linear model to model group size, and combined these two models to describe the distribution of abundance. Our methods performed well on simulated data when nondetection was accounted for and poorly when detection was ignored.We recommend researchers consider the effects of nondetection sampling bias when modeling species distributions using presence-only data. If information about the detection process is available, we recommend researchers explore the effects of nondetection and, when warranted, correct the bias using our methods. We developed our methods to analyze opportunistic presence-only records of whooping cranes (Grus americana), but expect that our methods will be useful to ecologists analyzing opportunistic presence-only records of other species of animals.