Sample Selection Correction Research Articles

Women’s formal and informal labour in Brazil: an inequality decomposition (2000-2010)

PurposeThe purpose of this paper is to examine the effects of changes in the relative composition and in the segmentation between formal and informal labour on earnings differentials among women over the last decade in Brazil.Design/methodology/approachThe authors follow Machado and Mata’s method to decompose the changes along the earnings distribution, with correction for sample selection and using microdata from the Demographic Census of 2000 and 2010. Informal labour was divided into informal salaried labour and self-employment, and both groups were compared with the formal labour separately.FindingsThe results indicate that, in both cases, an increase in earnings differentials in the bottom of the earnings distribution due to segmentation, suggesting that the returns to formal labour have grown relatively to informal labour during the period. On the other hand, earnings differentials decrease as one moves up the earnings distribution due to the composition effect, which is stronger on the top of the distribution relatively to the bottom. Furthermore, there are compensating differentials for self-employed women above the 30th quantile, which contributed to reduce the inequality between this group and formal workers.Originality/valueThe paper contributes to a better understanding of the changes taking place in female labour, shedding some light on how they affect different points along the earnings distribution. Furthermore, the adopted approach proposes a new application for the correction of sample bias in the context of quantile regression by employing a logit multinomial, and using the Demographic Census data.

Longitudinal Wealth Data and Multiple Imputation--An Evaluation Study

Statistical analysis in surveys is generally facing missing data. In longitudinal studies for some missing values there might be past or future data points available. The question arises how to successfully transform this advantage into improved imputation strategies. In a simulation study the authors compare six combinations of cross-sectional and longitudinal imputation strategies for German wealth panel data. The authors create simulation data sets by blanking out observed data points: they induce item non response by a missing at random (MAR) and two differential non-response (DNR) mechanisms. We test the performance of multiple imputation using chained equations (MICE), an imputation procedure for panel data known as the row-and-column method and a regression prediction with correction for sample selection. The regression and MICE approaches serve as fallback methods, when only cross-sectional data is available. The row-and-column method performs surprisingly well considering the cross-sectional evaluation criteria. For trend estimates and the measurement of inequality, combining MICE with the row-and-column technique regularly improves the results based on a catalogue of six evaluation criteria including three separate inequality indices. As for wealth mobility, two additional criteria show that a model based approach such as MICE might be the preferable choice. Overall the results show that if the variables, which ought to be imputed, are highly skewed, the row-and-column technique should not be dismissed beforehand.

A double-copula stochastic frontier model with dependent error components and correction for sample selection

In the standard stochastic frontier model with sample selection, the two components of the error term are assumed to be independent, and the joint distribution of the unobservable in the selection equation and the symmetric error term in the stochastic frontier equation is assumed to be bivariate normal. In this paper, we relax these assumptions by using two copula functions to model the dependences between the symmetric and inefficiency terms on the one hand, and between the errors in the sample selection and stochastic frontier equation on the other hand. Several families of copula functions are investigated, and the best model is selected using the Akaike Information Criterion (AIC). The methodology was applied to a sample of 200 rice farmers from Northern Thailand. The main findings are that (1) the double-copula stochastic frontier model outperforms the standard model in terms of AIC, and (2) the standard model underestimates the technical efficiency scores, potentially resulting in wrong conclusions and recommendations.

Characteristics of Demand Structure and Preferences for Wild and Farmed Seafood in Germany: An Application of QUAIDS Modeling with Correction for Sample Selection

AbstractThe production of farmed fish is growing rapidly and presents a sustainable and possibly low-cost alternative to wild fish. Thus, we may expect retail prices of farmed to be lower than prices of wild fish and demand to be less elastic. Otherwise, marketing of farmed fish may generate some extra value that justifies higher prices and may exhibit more elastic demand. To test these hypotheses, we employ monthly household scanner panel data for Germany from 2006 to 2010 for six frozen seafood products that include farmed and wild fish. A QUAIDS model is estimated by a consistent two-step procedure to account for censoring of the dependent variable. We find consumers to be price sensitive, particularly with regard to the high-value seafood species salmon and shrimp. This price elastic market implies that the German seafood industry still has the potential for growing revenues if production increases.

Data vs. Methods: Quasi-Experimental Evaluation of Alternative Sample Selection Corrections for Missing College Entrance Exam Score Data

In 2007, Michigan began requiring all high school students to take the ACT college entrance exam. This natural experiment allows us to evaluate the performance of several parametric and semiparametric sample selection correction models. We apply each model to the censored, prepolicy test score data and compare the predicted values to the uncensored, post-policy distribution. We vary the set of model predictors to imitate the varying levels of data detail to which a researcher may have access. We find that predictive performance is sensitive to predictor choice but not correction model choice. All models perform poorly using student demographics and school- and district-level characteristics as predictors. However, all models perform well when including students’ prior and contemporaneous scores on other tests. Similarly, correction models using group-level data perform better with more finely disaggregated groups, but produce similar predictions under different functional form assumptions. Our findings are not explained by an absence of selection, the assumptions of the parametric models holding, or the data lacking sufficient variation to permit useful semiparametric estimation. We conclude that “data beat methods” in this setting: gains from using less restrictive econometric methods are small relative to gains from seeking richer or more disaggregated data.

School Characteristics and Teacher Turnover: Assessing the Role of Preferences and Opportunities

Job characteristics can affect worker turnover through their effect on utility and through their effect on outside job opportunities. The aim of this study is to identify and estimate the roles of these two channels separately. Our method exploits information on job changes, and relies on an augmented sample selection correction. To illustrate our approach, we use an exhaustive register of Dutch primary school teachers and show a detailed picture of preferences for school characteristics. We also find that the dependence between current and outside job attributes can affect turnover and thus the allocation of teachers across schools.

Booster Seats and Traffic Fatalities Among Children

In an effort to increase booster seat use among children, the National Highway Traffic Safety Administration is encouraging state legislators to promote stricter booster seat laws, yet there is a paucity of information on booster seat efficacy relative to other forms of restraint. Using data from the Fatality Analysis Reporting System for the period 2008-2014 and the sample selection correction proposed by Levitt and Porter (2001), the current study examines the effectiveness of booster seats relative to child safety seats and adult seat belts. For children 6 to 8 years of age, we find that booster seats are more than twice as effective as child safety seats and over 30 percent more effective than standard seat belts at decreasing the likelihood of fatality in a motor vehicle accident. For children 2 to 5 years of age, all three forms of restraint appear equally effective.

Patterns of earnings differentials across three conservative European welfare regimes with alternative education systems

The aim of this paper is to investigate, from a generational perspective, the effect of human capital on individual earnings and earnings differences in Germany, France and Italy, three developed countries in Western Europe with similar conservative welfare regimes but with important differences in their education systems. Income inequalities between and within education levels are explored using a two-stage probit model with quantile regressions in the second stage. More precisely, drawing upon 2005 EU-SILC data, returns on schooling and experience are estimated separately for employees and self-employed full-time workers by means of Mincerian earnings equations with sample selection; the sample selection correction accounts for the potential individual self-selection into the two labour force types. Although some determinants appear to be relatively similar across countries, state-specific differentials are drawn in light of the institutional features of each national context. The study reveals how each dimension of human capital differently affects individuals’ earnings and earnings inequality and, most of all, how their impacts differ along the conditional earnings distribution and across countries. In the comparative perspective, the country's leading position in terms of the highest rewards on education also depends on which earnings distribution (employee vs. self-employed) is analysed.

Longitudinal Wealth Data and Multiple Imputation: An Evaluation Study

Statistical Analysis in surveys is generally facing missing data. In longitudinal studies for some missing values there might be past or future data points available. The question arises how to successfully transform this advantage into improved imputation strategies. In a simulation study the authors compare six combinations of cross‐sectional and longitudinal imputation strategies for German wealth panel data. The authors create simulation data sets by blanking out observed data points: they induce item non response by a missing at random (MAR) and two differential nonresponse (DNR) mechanisms. We test the performance of multiple imputation using chained equations (MICE), an imputation procedure for panel data known as the row‐and‐column method and a regression prediction with correction for sample selection. The regression and MICE approaches serve as fallback methods, when only cross‐sectional data is available. The row‐and‐column method performs surprisingly well considering the cross‐sectional evaluation criteria. For trend estimates and the measurement of inequality, combining MICE with the row‐and‐column technique regularly improves the results based on a catalogue of six evaluation criteria including three separate inequality indices. As for wealth mobility, two additional criteria show that a model based approach such as MICE might be the preferable choice. Overall the results show that if the variables, which ought to be imputed, are highly skewed; the row‐and‐column technique should not be dismissed beforehand.

Selection and the age – productivity profile. Evidence from chess players

We use data on professional chess tournaments to study how endogenous selection affects the relationship between age and mental productivity in a brain-intensive profession. We show that less talented players are more likely to drop out, and that the age-productivity gradient is heterogeneous by ability, making fixed effects estimators inconsistent. Since we do not observe the players who dropped out of chess before the beginning of our sampling period, we cannot exploit the standard Heckman sample selection correction procedure. Therefore, we correct for selection by using an imputation method that repopulates the sample by applying to older cohorts the self-selection patterns observed in younger cohorts. We estimate the age-productivity profile on the repopulated sample using median regressions, and find that median productivity increases by close to 5 percent from initial age (15) to peak age (21.6), and declines substantially after the peak. At age 50, it is about 10 percent lower than at age 15. We compare profiles in the unadjusted and in the repopulated sample and show that failure to adequately address endogenous selection in the former leads to substantially over-estimating productivity at any age relative to initial age.

Please Call Again: Correcting Non-Response Bias in Treatment Effect Models

We propose a novel selectivity correction procedure to deal with survey attrition, at the crossroads of the Heckit and of the bounding approach of Lee (2009). As a substitute for the instrument needed in sample selectivity correction models, we use information on the number of attempts that were made to obtain response to the survey from each individual who responded. We obtain set identification, but if the number of attempts to reach each individual is high enough, we can come closer to point identification. We apply our sample selection correction in the context of a job-search experiment with low and unbalanced response rates.

Gender wage gap: A semi-parametric approach with sample selection correction

Sizeable gender differences in employment rates are observed in many countries. Sample selection into the workforce might therefore be a relevant issue when estimating gender wage gaps. We propose a semi-parametric estimator of densities in the presence of covariates which incorporates sample selection. We describe a simulation algorithm to implement counterfactual comparisons of densities. The proposed methodology is used to investigate the gender wage gap in Italy. We find that, when sample selection is taken into account, the gender wage gap widens, especially at the bottom of the wage distribution.

Assessing the Returns to Education in Georgia

The economic returns to education in transition countries have been extensively evaluated in the literature. The present study contributes to this literature by estimating the returns to education in Georgia during the last transition period 2000-04. We find very low returns to education in Georgia and little evidence of an increasing trend in the returns. This picture contrasts with somewhat higher rates of return to education in the mid-1990s in Georgia and the recent estimates from other transition countries. A further analysis of the shifts in the supply and demand for education sheds light on possible causes. In particular, on the supply side, the decline in the quality of education in the 1990s has negated the improvements in the provision of skills needed by market economies during this period. On the demand side, the expansion of the Georgian economy has taken place in the direction of fields such as public administration and education that employ a highly educated workforce but do not remunerate well. Yet it would be a mistake to conclude that education is not a valuable asset in Georgia. The role of education is largely manifested in its impact on the employability of individuals, an issue that has been overlooked in the transition literature. Once this impact is taken into account, education is shown to play an increasingly important role in influencing the earnings of the working population in Georgia. The paper uses the ordinary least squares approach, instrumental variables approach, and sample selection correction, taking into account conditional and unconditional marginal effects of education on earnings.

The Accuracy of Predicted Wages of the Non‐Employed and Implications for Policy Simulations from Structural Labour Supply Models*

We examine the accuracy of predicted wages for the non‐employed. We argue that unemployment, marginal attachment, and not in the labour force are three distinct states. Using panel data from Australia, we test the accuracy of predicted wages for these three groups of non‐employed using sample selection models. Focusing on those individuals who subsequently enter employment, we find that predictions which incorporate the estimated sample selection correction perform poorly, particularly for the marginally attached and the not in the labour force. These results have important implications for policy simulations from structural labour supply models.

Gender Wage Gap: A Semi-Parametric Approach with Sample Selection Correction

Sizeable gender differences in employment rates are observed in many countries. Sample selection into the workforce might therefore be a relevant issue when estimating gender wage gaps. This paper proposes a new semi-parametric estimator of densities in the presence of covariates which incorporates sample selection. We describe a simulation algorithm to implement counterfactual comparisons of densities. The proposed methodology is used to investigate the gender wage gap in Italy. It is found that when sample selection is taken into account gender wage gap widens, especially at the bottom of the wage distribution. Explanations are offered for this empirical finding.

Gender Wage Gap: A Semi-Parametric Approach with Sample Selection Correction

Sizeable gender differences in employment rates are observed in many countries. Sample selection into the workforce might therefore be a relevant issue when estimating gender wage gaps. This paper proposes a new semi-parametric estimator of densities in the presence of covariates which incorporates sample selection. We describe a simulation algorithm to implement counterfactual comparisons of densities. The proposed methodology is used to investigate the gender wage gap in Italy. It is found that when sample selection is taken into account gender wage gap widens, especially at the bottom of the wage distribution. Explanations are offered for this empirical finding.

A stochastic frontier model with correction for sample selection

Heckman’s (Ann Econ Soc Meas 4(5), 475–492, 1976; Econometrica 47, 153–161, 1979) sample selection model has been employed in three decades of applications of linear regression studies. This paper builds on this framework to obtain a sample selection correction for the stochastic frontier model. We first show a surprisingly simple way to estimate the familiar normal-half normal stochastic frontier model using maximum simulated likelihood. We then extend the technique to a stochastic frontier model with sample selection. In an application that seems superficially obvious, the method is used to revisit the World Health Organization data (WHO in The World Health Report, WHO, Geneva 2000; Tandon et al. in Measuring the overall health system performance for 191 countries, World Health Organization, 2000) where the sample partitioning is based on OECD membership. The original study pooled all 191 countries. The OECD members appear to be discretely different from the rest of the sample. We examine the difference in a sample selection framework.

Social Influence Given (Partially) Deliberate Matching: Career Imprints in the Creation of Academic Entrepreneurs

Actors often match with associates on a small set of dimensions that matter most for the particular relationship at hand. In so doing, they are exposed to unanticipated social influences because counterparts have more interests, attitudes, and preferences than would-be associates considered when they first chose to pair. This implies that some apparent social influences (those tied to the rationales for forming the relationship) are endogenous to the matching process, while others (those that are incidental to the formation of the relationship) may be conditionally exogenous, thus enabling causal estimation of peer effects. We illustrate this idea in a new data set tracking the training and professional activities of academic biomedical scientists. In qualitative and quantitative analyzes, we show that scientists match to their postdoctoral mentors based on two dominant factors, geography and scientific focus. They then adopt their advisers' orientations toward commercial science as evidenced by the transmission of patenting behavior, but they do not match on this dimension. We demonstrate this in two-stage models that adjust for the endogeneity of the matching process, using a modification of propensity score estimation and a sample selection correction with valid exclusion restrictions. Furthermore, we draw on qualitative accounts of the matching process recorded in oral histories of the career choices of the scientists in our data. All three methods - qualitative description, propensity score estimators, and those that tackle selection on unobservable factors - are potential approaches to establishing evidence of social influence in partially endogenous networks, and they may be especially persuasive in combination.

The Accuracy of Predicted Wages of the Non-Employed And Implications for Policy Simulations From Structural Labour Supply Models

We examine the accuracy of predicted wages for the non-employed. We argue that unemployment, marginal attachment, and not in the labour force are three distinct states. Using panel data from Australia, we test the accuracy of predicted wages for these three groups of non-employed using sample selection models. Focusing on those individuals who subsequently enter employment, we find that predictions which incorporate the estimated sample selection correction perform poorly, particularly for the marginally attached and the not in the labour force. These results have important implications for policy simulations from structural labour supply models.

A Stochastic Frontier Model with Correction for Sample Selection

Heckman s (1979) sample selection model has been employed in three decades of applications of linear regression studies. The formal extension of the method to nonlinear models, however, is of more recentvintage. A generic solution for nonlinear models is proposed in Terza (1998). We have developed simulation based approach in Greene (2006). This paper builds on this framework to obtain a sample selection correction for the stochastic frontier model. We first show a surprisingly simple way to estimate the familiar normal-half normal stochastic frontier model (which has a closed form log likelihood) using maximum simulated likelihood. The next step is to extend the technique to a stochastic frontier modelwith sample selection. Here, the log likelihood does not exist in closed form, and has not previously been analyzed. We develop a simulation based estimation method for the stochastic frontier model. In an application that seems superficially obvious, the method is used to revisit the World Health Organization data [WHO (2000), Tandon et al. (2000)] where the sample partitioning is based on OECD membership. The original study pooled all 191 countries. The OECD members appear to be discretely different from therest of the sample. We examine the difference in a sample selection framework.