Correction For Selection Bias Research Articles

Selection bias correction for species sensitivity distribution modeling and hazardous concentration estimation.

The species sensitivity distribution (SSD) has been an important development in ecotoxicology, and despite numerous concerns having been raised over many years, it remains the preferred (and often mandated) technique for establishing "safe" concentrations of contaminants in receiving water bodies by jurisdictions around the world. Although universally recognized as a crucial prerequisite for the statistical validity of the procedure, the assumption of random selection of species for SSD modeling is invariably violated. It is shown in the present study that, under very minimal assumptions, nonrandom species selection can result in hazardous concentration estimation errors of a factor of 20 or more. Importantly, if the toxicity data are biased toward the more sensitive species, then the conventional practice of using the lower confidence interval limit for the estimated hazardous concentration may be compensating in the wrong direction.

Performance and Strategy: Simultaneous Equations Analysis of Long-lived Firms

A simultaneous equations model of performance, strategy and size is tested using fieldwork evidence on long-lived firms in Scotland. Estimation is by I3SLS, with correction for sample selection bias. The contributions of this paper are that it: (a) grounds estimation on fieldwork evidence; (b) calibrates performance and competitive strategy; (c) tests and models endogeneity; and (d) computes robust trade-off elasticities between firm size and performance. It shows how this trade-off provides the entrepreneur with two strong incentives: (i) to seek greater efficiency typically by an increase in the human capital of the ‘core’ workforce; (ii) to achieve higher levels of performance by adopting more diverse competitive strategies.

Factor analysis of metabolic syndrome components and predicting type 2 diabetes: Results of 10-year follow-up in a Middle Eastern population.

The relationship among components of metabolic syndrome (MetS) and their association with diabetes is unclear in West Asia. The aim of the present study was to conduct factor analysis of MetS components and the effect these factors have on the incidence of type 2 diabetes (T2D) in a population-based cohort study of the Tehran Lipid and Glucose Study (TLGS). The present study enrolled 1861 men and 2706 women (20-60 years of age), from Tehran (Iran) who were free of diabetes at baseline and followed them for 10 years. A principal component analysis was performed to extract standardized factors from MetS components. Logistic regression was used to detect associations between the extracted factors and the incidence of diabetes. A propensity score was used to correct differential selection bias resulting from loss to follow-up. Factor analysis identified three factors (blood pressure, lipids and glycemia). Waist circumference was shared in three all factors. Blood pressure, lipids and glycemia were related to the incidence of diabetes with odds ratios (95% confidence intervals) of 2.23 (1.31-3.78), 1.89 (1.27-3.67), and 7.54 (4.09-13.91), respectively, in men and 2.13 (1.34-3.40), 2.06 (1.35-3.15), and 13.91 (7.29-26.51), respectively, in women for the third versus the first tertile of these standardized factors. Central adiposity may have a pivotal role in MetS linking other risk factors together. Glycemia had a high impact on the incidence of diabetes, whereas blood pressure and lipid had a similar moderate effect on the incidence of diabetes.

Farm Size and Returns to Cultivation in India: Revisiting an Old Debate

This paper revisits the long-debated question of the relationship between farm size and productivity by studying the relationship between area cultivated and net returns to cultivation in India using a nationally representative data-set. The analysis is carried out separately for the two major agricultural seasons, kharif and rabi, and for both the seasons pooled together. Our findings suggest the existence of an inverse relationship, even when we control for a number of household and farm characteristics and even when we treat factors such as household type (occupation), social group (caste), agro-climatic zone (region) and agricultural season as fixed effects. The result is also robust to correction for selection bias. However, the efficiency of the smallholder as a result of this greater productivity has to be treated with some caution as it ignores the low absolute levels of their returns, which raise questions about the sustainability of their livelihoods. This is further aggravated by the fact that they pay relatively higher unit costs and because of their greater dependence on purchased inputs.

Bad marriage, broken heart? Age and gender differences in the link between marital quality and cardiovascular risks among older adults.

Working from a life course perspective, we develop hypotheses about age and gender differences in the link between marital quality and cardiovascular risk and test them using data from the first two waves of the National Social Life, Health, and Aging Project. The analytic sample includes 459 married women and 739 married men (aged 57-85 in the first wave) who were interviewed in both waves. We apply Heckman-type corrections for selection bias due to mortality and marriage. Cardiovascular risk is measured as hypertension, rapid heart rate, C-reactive protein, and general cardiovascular events. Results suggest that changes in marital quality and cardiovascular risk are more closely related for older married people than for their younger counterparts and that the link between marital quality and cardiovascular risk is more pronounced among women than among men at older ages. These findings fit with the gendered life course perspective and cumulative disadvantage framework.

Trends in Helicobacter pylori infection among Māori, Pacific, and European Birth cohorts in New Zealand.

The aim of this paper is to estimate the seroprevalence of Helicobacter pylori infection in the New Zealand population by ethnicity and year of birth. A systematic search identified seven studies in New Zealand that reported prevalence of H.pylori infection among 4463 participants. Prevalence data were pooled to estimate the Māori, Pacific, and European seroprevalence of H.pylori in four birth cohorts (1926-40, 1941-55, 1956-70, and 1971-85), by assuming that infection is acquired in childhood and seroprevalence is stable with aging. The best estimates of national seroprevalence were obtained by geographic regional weighting and corrections for selection and measurement bias. Infection rates among all ethnic groups declined in more recent birth cohorts. Prevalence was highest among Pacific peoples (ranging from 39-83%) followed by Māori (18-57%) and then European (7-35%). The absolute ethnic differences in seroprevalence decreased in subsequent cohorts, but the relative ethnic differences increased. There is scope to much further reduce Māori and especially Pacific people's risk of H.pylori infection. Solutions to reduce H.pylori prevalence and its sequelae should focus on people at greatest risk of the infection. Further evaluation of strategies to address H.pylori infection is warranted. Interventions to be evaluated could include household crowding reduction and eradication therapy for asymptomatic infected persons to reduce their risk of noncardia stomach cancer.

Approaching the female labor supply from the unpaid work and non-work functions

Purpose – The purpose of this paper is to study the allocation of time to paid work, unpaid work and non-work by women in Britain in 1998-1999. To infer the labor supply from the other time-use expressions. Design/methodology/approach – The paper uses weekly diary data to estimate the unpaid work and non-work functions. It infers the (residual) paid work expression. As the latter is recovered from uncensored regressions, it makes direct use of the complete set of observations. Hence, it contains more information than the conventional labor supply functions that are estimated from the data obtained from paid work participants via the Tobit and Heckit or selection-bias correction (SBC) techniques. Findings – The women surveyed generally allocated 69 percent of their time toward non-work, 18 percent to unpaid work, and 13 percent to paid work. The non-work function is dominated by the autonomous component, and all three functions depend on subjects’ age, wage, non-labor income, household composition, the unpaid work contributions of adult relatives and region of residence. The unpaid work and non-work functions are more consistent with the SBC rather than the Tobit version of the labor supply. Moreover, the Tobit predicts unrealistic paid work allocations for women engaging in very little non-work. Research limitations/implications – The unpaid work and non-work functions are regressed separately, as often the case in the literature. Their consideration within a seemingly unrelated regression framework necessitates a reduction in the number of observations to match those considered in the Tobit and SBC versions of the labor supply. Nuances may arise when the time reported in the diaries does not add up to 24 daily hours for all respondents. Knowledge of the recovered regional, age, household member and other effects on women time allocation might had come handy to economic development authorities who sought to attract women out of the household into market production, and from part-time to full-time employment in the context of the 2000-2010 Lisbon Strategy. Similar lessons may be valid today. Originality/value – The data set derives from a survey that has not been used before. It relies on week-long diaries in order to avoid the occurrence of many zeros in a good number of activities (which is the norm in short diaries), and to ensure the study of a censored time-use function through its uncensored complements. The findings are compared to those of a weekly diary survey conducted in 1987 that solicited similar information. Hence, the study fills a gap in time-use analysis. Identifying the factors which influence the number of hours that women engage in work (both paid and unpaid) and non-work is useful for economic policy purposes. The study exposes a limitation in the conventional estimation of the labor supply which, in turn, casts doubt on the reliability of empirical results for policy making.

Seroprevalencia de VIH en población mexicana de entre 15 y 49 años: resultados de la Ensanut 2012

To estimate the HIV seroprevalence among Mexicans aged 15 to 49 years old and living in households, and to describe the profile of serorreactive individuals. Cross-sectional study with a national probabilistic sample of individuals aged 15 to 49 years with behavioral data from direct interview (face-to-face) at households and HIV screening using capillary blood collected from the same individuals. A seroprevalence of 0.15% (95%CI 0.09-0.21) was estimated for Mexicans aged 15 to 49; seroprevalence among women was 0.07% (95%CI 0.03-0.11) and 0.24% (95%CI 0.11-0.36) for men. HIV serorreactive population is composed of younger men, with a higher socioeconomic level compared to the general population, and with a higher insurance coverage-social protection on health in general and social security in particular. Only 50% of the serorreactive individuals may be aware of their status as living with HIV. The estimated HIV seroprevalence in the NHNS 2012 suggests a stable pattern since 2000. The estimated prevalence among individuals 15 to 49 years was adjusted both for selection bias correction and to include MSM estimations (under the assumption that MSM is a population hard to reach in a household survey), resulting in a total seroprevalence of 0.23% and an estimated number of people with HIV of 140,000.

Non-response bias in student evaluations of teaching

For as long as colleges and universities have been conducting student evaluations of teaching (SET), faculty have questioned the validity of the information collected. Substantial effort has been expended to improve the SET process, and researchers have communicated a consistent understanding of why students evaluate teachers as they do. Most of these conclusions have been based on an analysis of SET data gathered at the end of the semester by a sample of students who may not represent all students enrolled in the class. This clearly creates the potential for a sample selection bias that puts into question much of what we have learned about why students evaluate their instructors as they do.In this paper, we gather a unique data set from individual students who provide either online SET forms or in-class SET forms. We use these responses to address the concerns regarding the validity of SET data which results from the low response rates often associated with online SET evaluations. The Heckman (1979) sample selection bias correction estimator is used to quantify the bias resulting from the low online response rate and finds this bias to be negligibly small.

Effects of sample selection bias on the accuracy of population structure and ancestry inference.

Population stratification is an important task in genetic analyses. It provides information about the ancestry of individuals and can be an important confounder in genome-wide association studies. Public genotyping projects have made a large number of datasets available for study. However, practical constraints dictate that of a geographical/ethnic population, only a small number of individuals are genotyped. The resulting data are a sample from the entire population. If the distribution of sample sizes is not representative of the populations being sampled, the accuracy of population stratification analyses of the data could be affected. We attempt to understand the effect of biased sampling on the accuracy of population structure analysis and individual ancestry recovery. We examined two commonly used methods for analyses of such datasets, ADMIXTURE and EIGENSOFT, and found that the accuracy of recovery of population structure is affected to a large extent by the sample used for analysis and how representative it is of the underlying populations. Using simulated data and real genotype data from cattle, we show that sample selection bias can affect the results of population structure analyses. We develop a mathematical framework for sample selection bias in models for population structure and also proposed a correction for sample selection bias using auxiliary information about the sample. We demonstrate that such a correction is effective in practice using simulated and real data.

Credit Risk Modeling of Residential Mortgage Lending in Russia

This paper analyzes the problems of credit risk modeling of residential mortgage lending in Russia. Using unique mortgage loan and macro data from a regional branch of the Agency of Home Mortgage Lending (2008-2012), we find that borrower and mortgage loan characteristics affect the loan performance and play an important role in predicting default as well as a macroeconomic situation. On the residential mortgage market, borrowers with undeclared income have the lowest probability of default, mainly explained by the difference in declared and real income. Obtained results are robust under parametric and semiparametric specifications with correction for selectivity bias.

The shadow of death model revisited with an application to French firms

This article provides an empirical model of the shadow of death in which the exit probability of a firm depends on the firm’s productive performance and the firm’s level of sunk costs, which are viewed as barriers to exit. The shadow of death effect is treated by assuming a relationship between the propensity to exit and both the contemporaneous and lagged values of efficiency and sunk costs. To estimate the unobserved productive efficiency, we use the Ackerberg et al. (2006) estimator extended by the addition of a correction for selection bias. We use an unbalanced sample of approximately 100 000 French firms over the period 1997 to 2002. Our results indicate that the probability of exit is negatively affected by unobserved individual efficiency and the level of sunk costs. The shadow of death effect applies mainly in manufacturing, where both productive efficiency and sunk costs decrease during several years before exit. In service sectors, the exit process seems to occur more suddenly.

Inter-industry job mobility in the knowledge economy in Finland

Purpose – The purpose of this paper is to investigate inter-industry labor mobility, paying special attention to workers who move into high-tech (HT) sectors or knowledge-intensive business services (KIBSs). This study inquires whether skilled workers are mobile and whether the characteristics of mobile workers support the effective transfer of knowledge across industries. Design/methodology/approach – Census data representing 7 percent of Finnish residents were used. The micro-econometric estimation method with correction of sample selection bias was applied. Findings – The results show that young workers are the most mobile, whereas mobility decreased for those with previous work experience, higher education and higher income level. These findings indicate that the highly skilled workers are not necessarily the most mobile, a trend that may weaken the effectiveness of knowledge spillover. However, on average, highly educated workers move into KIBS sectors more often than to other sectors, and HT sectors attract workers who have higher incomes, which may indicate that their skills are highly valued. As a result, knowledge spillovers may emerge. The knowledge spillovers resulting from job mobility are concentrated in large growth centers that have universities. Originality/value – This study provides a new and topical viewpoint to the mobility literature by focussing on skilled workers and their movement toward the HT and KIBS sectors.

Genotype-by-environment interaction of growth traits in rainbow trout (Oncorhynchus mykiss): A continental scale study1

Rainbow trout is a globally important fish species for aquaculture. However, fish for most farms worldwide are produced by only a few breeding companies. Selection based solely on fish performance recorded at a nucleus may lead to lower-than-expected genetic gains in other production environments when genotype-by-environment (G × E) interaction exists. The aim was to quantify the magnitude of G × E interaction of growth traits (tagging weight; BWT, harvest weight; BWH, and growth rate; TGC) measured across 4 environments, located in 3 different continents, by estimating genetic correlations between environments. A total of 100 families, of at least 25 in size, were produced from the mating 58 sires and 100 dams. In total, 13,806 offspring were reared at the nucleus (selection environment) in Washington State (NUC) and in 3 other environments: a recirculating aquaculture system in Freshwater Institute (FI), West Virginia; a high-altitude farm in Peru (PE), and a cold-water farm in Germany (GER). To account for selection bias due to selective mortality, a multitrait multienvironment animal mixed model was applied to analyze the performance data in different environments as different traits. Genetic correlation (rg) of a trait measured in different environments and rg of different traits measured in different environments were estimated. The results show that heterogeneity of additive genetic variances was mainly found for BWH measured in FI and PE. Additive genetic coefficient of variation for BWH in NUC, FI, PE, and GER were 7.63, 8.36, 8.64, and 9.75, respectively. Genetic correlations between the same trait in different environments were low, indicating strong reranking (BWT: rg = 0.15 to 0.37, BWH: rg = 0.19 to 0.48, TGC: rg = 0.31 to 0.36) across environments. The rg between BWT in NUC and BWH in both FI (0.31) and GER (0.36) were positive, which was also found between BWT in NUC and TGC in both FI (0.10) and GER (0.20). However, rg were negative between BWT in NUC and both BWH (-0.06) and TGC (-0.20) in PE. Correction for selection bias resulted in higher additive genetic variances. In conclusion, strong G × E interaction was found for BWT, BWH, and TGC. Accounting for G × E interaction in the breeding program, either by using sib information from testing stations or environment-specific breeding programs, would increase genetic gains for environments that differ significantly from NUC.

Estimating the Effect of Crime Risk on Property Values and Time on Market: Evidence from Megan's Law in Virginia

We examine neighborhood externalities that arise from the perceived risk associated with the proximity of a registered sex offender's residence. We find large negative externality effects on a property's price and liquidity, employing empirical techniques that include a fixed‐effects OLS model, a correction for sample selection bias and censoring using a Heckman treatment, and a three‐stage least‐squares model to account for simultaneity bias in the joint determination of a home's sale price and liquidity. Additionally, we find amplified effects for homes with more bedrooms (a proxy for children) and if the nearby offender is designated by the state as “violent.”

The impact of government income transfers on the Brazilian job market

This paper intends to analyze the impact of transferences over the formal and informal labor supply from the family heads. For the analyze of the effects in scope of the entrance decisions in job market it was estimated a multinomial logit, while in the scope of the working time it was used a variation of the method of Durbin and McFadden (1984) for selection bias correction. It has verified that transferences have positive effect over the probabilities that the individual doesn't work and does informally work. However, this last one seems to be related to a substitution effect, once it has been observed a parallel negative effect over the probability to work on the formal sector. It has been obtained yet that the benefits negatively impacts on the offered hours by the family heads whether in the formal or non-formal sector, effect that has happened to be verified on the hours offering in formal sector of all individuals. On the other hand, as we consider the hours of working of those who are engaged in informal sector, it has obtained that income transferences perform not as a discourage issue but as a magnifier factor of worked hours.

Aid and donor trust in recipient country systems

The 2005 Paris Declaration on Aid Effectiveness sets targets for increased use by donors of recipient country systems for managing aid. The target is premised on a view that country systems are strengthened when donors trust recipients to manage aid funds, but undermined when donors manage aid through their own separate parallel systems. This study provides an analytical framework for understanding donors' decisions to trust or bypass country systems. Empirical tests are conducted using data from three OECD-DAC surveys designed to monitor progress toward Paris Declaration goals. Tests show that use of recipient country systems is positively related to (1) the donor's reputational stake in the country's development, as proxied by the donor's share of aid provided to the recipient; (2) the trustworthiness or quality of those systems, as measured by cross-country corruption indicators; and (3) donors' risk tolerance, as proxied by public support for aid provision in donor countries. Findings are robust to corrections for potential sample selection, omitted variables or endogeneity bias.

Roles of extension and ethno-religious networks in acceptance of resource-conserving agriculture among Ethiopian farmers

This article analyses roles of social and extension networks in adoption of resource-conserving practices among Ethiopian farmers. We gathered data from 297 randomly sampled households on their agricultural practices, social networks, access to the extension, and geographical location. After examining general determinants of practising resource-conserving agriculture, we employ a two-stage regression with full-maximum likelihood correction for selection bias to establish the roles of general social networks and external professionals in acceptance of conservation techniques. In accordance with previous research, probit regression in the first stage shows that the access to extension increases with farmers’ wealth and the size of their personal networks, and decreases with the distance of their households from village centres. However, after accounting for this unequal access to extension, the second-stage linear regression shows that regardless of education, wealth or geographical location, those whose religion and ethnicity match with their agent, report learning more about conservation from extension sources. Furthermore, farmers who are socially well connected within the community tend to be less receptive to agents’ recommendations regarding resource conservation. Dissemination policy of conservation agriculture should consider the ethnic and religious affinity between farmers and their extension agents. It also needs to pay more attention to socially and geographically isolated individuals.

Spatial Identification of Coal Power Plant Emission Effects on Education Access

Adverse educational outcomes have been repeatedly linked to recurring school absenteeism. Asthma is the most prevalent pediatric chronic illness and is the leading cause of health-related school absenteeism. This study seeks to understand the impacts of coal power plant emissions on asthma aggravation and resulting school absenteeism. Data from a unique, nationally administered pediatric asthma survey are combined with power plant emissions information to estimate the emissions' impacts on school absenteeism by asthmatic children. Exogenous wind pattern information and structural corrections for selection bias are used to identify households affected by emissions. Empirical results show a robustly positive relationship between power plant emissions and asthma-induced school absences. Moreover, we show that without appropriate identification and controls for self-selection, effects of coal power plant emissions on school absenteeism are underestimated.

Using Propensity Scores to Reduce Case-control Selection Bias

To the Editor: When sampling frames differ for cases and controls in a population based study, there are several analytical options to increase study validity. One is to exclude cases not found in the sampling frame for controls. Another, when the case sampling frame is nearly complete, is to link cases to the sampling frame for controls in order to obtain the information needed to model coverage propensity scores.1 Propensity scores have been used to address selection bias due to non-response and treatment-selection,2, 3 but we have found no studies that used propensity scores to address selection bias due to inadequate sampling frame coverage. We describe such an application here. Our analyses use data from the Wisconsin Women’s Health Study. Cases were women with incident invasive breast cancer reported to Wisconsin’s mandatory cancer registry. Controls were identified using a master file of licensed drivers who consented to have their private information released by the Wisconsin Department of Transportation. An estimated 68% of Wisconsin women with valid driver’s licenses provided this consent in 2008.4 Case eligibility was limited to women with a Wisconsin driver’s license (verified by self-report). We calculated odds ratio (OR) estimates of breast cancer risk using 3 established methods to correct selection bias, and compared these with published estimates.5, 6 The 3 methods were (1) excluding cases not in the control sampling frame, (2) applying propensity scores as weights, and (3) using quintiles of propensity scores to stratify analyses. We calculated standard errors of OR estimates to compare their precision. Additional details, including study inclusion criteria, are described elsewhere.4 The 2,988 breast cancer cases were linked to the driver’s license file to determine whether cases matched a record from the master-file of drivers (sampling frame for controls).4, 7 Propensity scores for inclusion in the sampling frame were calculated using logistic regression of 24 potential confounding factors ascertained through interviews with participants. The Table shows results of applying the 3 correction methods to a re-analysis of antidepressant medications and breast cancer risk.6 Of the 3 methods, stratifying by quintile of propensity score produced odds ratio3 with the best precision. Weighting, stratification and exclusion produced odds ratio3 with a similar degree of bias correction. However, the odds ratio3 changed only slightly after applying any of the correction methods. Similar results were found in another re-analysis of breast cancer risk, that one in relation to reproductive factors.4 Historically, the standard procedure in epidemiologic case-control studies has been to exclude cases that could not have been approached to serve as controls.8 This procedure assumes that factors that predict inclusion in the control sampling frame are similar for cases and potential controls. The two propensity-score-based methods implemented here also require this assumption and in addition, require correct model specification for propensity score estimation. The advantage of the propensity score methods is that they preserve the study base by allowing all cases to be used. Selection bias caused by incomplete sampling-frame coverage is frequently overlooked. In this study we linked the case and control sampling frames to demonstrate the use of propensity scores to adjust for selection bias due to incomplete study-base coverage of the sampling frames.