Address Sample Selection Research Articles

Uplift modeling and its implications for appointment date prediction in attended home delivery

Successful attended home delivery (AHD) is the most important aspect of e-commerce order fulfillment. Prior literature focuses on incentive scheme development for customers' choices of delivery windows and predictive analytics for delivery results, but it is not clear whether the effect of AHD on the appointment date set by customers increases the success rate of AHD. Therefore, we developed an uplift modeling method, PSM-NDML, as a relevant prescriptive analytic tool for AHD on an appointment date, which aims to estimate the causal effect of the by-appointment delivery on the delivery result. PSM-NDML integrates propensity score matching and double machine learning, effectively addressing sample selection bias, low predictive performance, and poor interpretability. Applied to a real-world product delivery dataset of a Chinese logistics company, PSM-NDML achieves superior performance relative to ten other state-of-the-art uplift models in terms of cumulative gain and the Qini coefficient. The predicted responses gained from PSM-NDML are also visually interpreted at the global and local levels, which reveals various managerial insights. In practice, the findings expand managers' understanding of the heterogeneous effects of AHD on appointment dates and provide decision support for logistics companies in the development of home delivery plans.

Making the most of AI and machine learning in organizations and strategy research: Supervised machine learning, causal inference, and matching models

AbstractResearch SummaryWe spotlight the use of machine learning in two‐stage matching models to deal with sample selection bias. Recent advances in machine learning have unlocked new empirical possibilities for inductive theorizing. In contrast, the opportunities to use machine learning in regression studies involving large‐scale data with many covariates and a causal claim are still less well understood. Our core contribution is to guide researchers in the use of machine learning approaches to choosing matching variables for enhanced causal inference in propensity score matching models. We use an analysis of real‐world technology invention data of public–private relationships to demonstrate the method and find that machine learning can provide an alternative approach to ad hoc matching. However, as with any method, it is also important to understand its limitations.Managerial SummaryThis article explores the use of machine learning to enhance decision‐making, particularly in addressing sample selection bias in large‐scale datasets. The rapid development of AI and machine learning offers new, powerful tools especially for digital ecosystems where complex data and causal relationships are complex to analyze. We offer managers and stakeholders insight into the effective integration of machine learning for selecting critical variables in propensity score matching models. Through a detailed examination of real‐world data on technology inventions within public–private relationships, we demonstrate the effectiveness of machine learning as a robust alternative to traditional matching methods.

Health selection into employment in a family health and time use context

AbstractUsing nationally representative data from Household Income and Labour Dynamics in Australia, this paper examines the impact of poor health, family impaired health and family time on the labour market participation of couples aged 25–64. We address sample selection bias and endogeneity bias by employing instrumental variable Tobit models. Our findings indicate that health selection into the labour market occurs not only due to individual's own health but also due to other family members' health, and family time—which is largely determined by various family factors such as a presence of preschool children (childcare) and partner's major life events. We find a significant trade‐off between family unpaid time and labour market participation, which is augmented after correcting for endogeneity of family unpaid time in market time consideration. This supports our argument that health selection should be considered within the family context, in which each member makes decisions conditional not only on their own but also other family members' health and time resources.

Addressing sample selection bias for machine learning methods

SummaryWe study approaches for adjusting machine learning methods when the training sample differs from the prediction sample on unobserved dimensions. The machine learning literature predominately assumes selection only on observed dimensions. Common approaches are to weight or include variables that influence selection as solutions to selection on observables. Simulation results show that selection on unobservables increases mean squared prediction error using popular machine‐learning algorithms. Common machine learning practices such as weighting or including variables that influence selection into the training or prediction sample often worsen sample selection bias. We propose two control function approaches that remove the effects of selection bias before training and find that they reduce mean‐squared prediction error in simulations. We apply these approaches to predicting the vote share of the incumbent in gubernatorial elections using previously observed re‐election bids. We find that ignoring selection on unobservables leads to substantially higher predicted vote shares for the incumbent than when the control function approach is used.

Does food safety perception matter to farmers’ happiness？ evidence from China

Although the determinants of individuals' happiness have been widely examined in the literature, little is known regarding whether and how food safety perception affects farmers' happiness. To fill this research gap, this paper examines the impact of food safety perception on happiness among Chinese farmers, utilizing open-access data collected through the Chinese Social Survey project in 2013, 2017 and 2021. This study focuses on Chinese farmers as the research subject, attempting to analyze the “happiness code” from the perspective of food safety, which supplements the literature on happiness and provides reference for protecting the rights of low-income groups and promoting food safety strategies in developing countries. To address sample selection bias, this paper employs the recursive bivariate ordered probit (RBOP) model and conditional mixed process (CMP) method. The results reveal that the perception of food safety exerts a positive and statistically significant impact on farmers' happiness in China. In addition, food safety perception is more important among middle-aged and elderly farmers and among those with higher education. Thus, the policy makers should continue to make up for the shortcomings of rural food safety work and extend regulatory measures to rural areas. They also need to take efforts to strengthen food safety promotion, enhance farmers' safety awareness, and safeguard farmers’ “safety on the tongue”.

Building Sustainability, Certification, and Price Premiums: Evidence from Europe

We examine how green certification impacts cashflows and transaction prices in the Western European office markets of France, Germany, Italy, the Netherlands, Spain, and the United Kingdom. We address sample selection bias by setting up a balanced panel matching sustainable-certified with noncertified buildings using various matching approaches. We find a statistically significant green building premium for properties in our sample. For certified properties, operating income and transaction prices are higher by approximately 12% and 10%, respectively. This premium is robust by country and higher for platinum- and gold-certified properties. Moreover, our examination at the market level indicates that the observed premium is not driven solely by markets with a limited supply of sustainable buildings. Additionally, the premium does not show a declining trend over time, but rather responds to market conditions. Notably, substantial heterogeneity in both cashflow and price premiums exists across countries and markets.

Use of home and community-based services and loneliness in older people with functional limitations: a cross-sectional study

BackgroundLoneliness is one of the major health problems among older adults. Among this population, home and community-based services (HCBS) have become increasingly popular. Despite its health benefits, little is known about the relationship between HCBS use and loneliness in older people with functional limitations. We aim to explore the characteristics of loneliness among older people with functional limitations and examine the association between HCBS use and loneliness in China.MethodsWe used a cross-sectional data from the 2018 China Health and Retirement Longitudinal Study, which includes a nationally representative sample of Chinese residents aged 65 and older with functional limitations. Logistic regression models were used to examine the associations between HCBS use and loneliness, and we further used propensity score matching to address potential sample selection bias.ResultIn China, 46% of older people with functional limitations felt lonely and only 22% of older people with functional limitations reported using HCBS in 2018. Compared with participants who did not receive HCBS, those who received HCBS were less likely to report loneliness (OR = 0.81, 95% CI = 0.63, 0.99, p = 0.048), and the results remained significant after addressing sample selection bias using propensity score matching.ConclusionOur results showed that loneliness was common among Chinese older people with functional limitations, and the proportion of HCBS use was low. There was robust evidence to support that among older people with functional limitations, HCBS use was associated with decreased loneliness. Further policies should promote the development of broader HCBS use for older people with functional limitations to reduce their loneliness.

Deglobalization and the value of geographic diversification: evidence from Brexit

PurposeThis paper aims to explore the value of geographic diversification in the context of deglobalization, drawing evidence from a quasi-natural experiment – the Brexit referendum that took place on 23 June 2016 in the UK.Design/methodology/approachThis study applies an event study methodology to estimate the impact of the Brexit vote on a cross-section of firms with varying levels of geographic diversification – undiversified UK firms, UK firms with significant operations in the European Union (EU) and globally diversified UK firms. This study deploys a Heckman two-stage regression approach to address sample selection bias.FindingsThis study finds that undiversified UK firms experienced negative cumulative abnormal returns (CARs) around the Brexit referendum. The value of UK firms with majority sales within the UK declined by 0.9 percentage points, on average, in the three days centred on the Brexit referendum. In contrast, UK firms that are globally diversified, with the majority of sales within the EU are unaffected, while diversified firms in the rest of the world generated positive CARs of 1.8 percentage points over the same period. These results are robust to firm characteristics, selection bias and alternative measures of CARs and diversification.Research limitations/implicationsThis study is subject to some limitations that open avenues for future work. There are a few available proxies of diversification and further work on developing other proxies is much needed. Further work may also examine the long-term impact of diversification on UK firms. This study considered Brexit as a quasi-natural experiment, and this study could be applied to other deglobalization events like COVID-19 and can enhance the generalizability of diversification strategy in the deglobalized world. Findings may stimulate future work to explore how another form of diversification – product diversification has affected firm returns around Brexit. Finally, this study has focused on the UK as its base case. It may be interesting to corroborate the findings by exploring the impact of Brexit on European firms, who hitherto Brexit, had some operations in the UK.Practical implicationsThis work offers some insights for policymakers and regulators around the impact of deglobalization on local firms. Findings suggest that these trends significantly negatively impact the most vulnerable firms (smaller firms with less global reach), while their larger counterparts with significant global reach might be insulated. This finding is important for determining the nature of support needed by different firms in times of deglobalization. The work also offers insights to managers of firms operating in countries where there are real prospects of deglobalization. Specifically, the work highlights the importance of geographic diversification when free movement of goods, services and people is restricted.Originality/valueThis study shows that a certain group of globally diversified firms earned significantly higher returns from the prospect of the UK leaving the EU, thereby highlighting the value of geographic diversification in a time of deglobalization.

Business angels and firm performance: First evidence from population data.

Business angels dominate early-stage investment in firms, but research on their effects on firms is scarce and limited by sample selection. To address sample selection, we propose using population data and we develop an algorithm for identifying business angel investments in such data. We illustrate this novel approach by applying it to detailed and longitudinal total population data for individuals and firms in Sweden. In our application, we focus on a subset of business angels-active business angels who are themselves successful entrepreneurs with a profitable exit. We then study active business angels' effects on firm performance, using population data. Employing a quasi-experimental estimator, we find that the business angels invest in firms that already perform above par. There is also a positive effect on subsequent growth compared with control firms. However, contrary to previous research on business angels, we cannot find any impact on firm survival. Overall, the paper underlines the need to address sample selection when studying business angels and suggests using population data for identification.

Unleash liquidity constraints or competitiveness potential: The impact of R&D grant on external financing on innovation

This paper uses Korea's internal data on R&D programs to analyze how the receiving of R&D impacts the amount of subsequent external financing for technological innovation received by R&D firms. We address sample selection and endogeneity issues that arise when estimating the impact of R&D policy by utilizing a matching method using unique features of Korea's R&D grant programs. Our empirical results show that firms that receive R&D grants further receive 22%-32% less external financing compared to those who do not receive R&D grants, though there are some differences in the statistical significance of the results based on the specification of the regression model. This is consistent with the view that R&D grants reduce firms’ liquidity constraints, or that government support effectively crowds out funding from the financial market.

Family ownership and cooperative R&D: the moderating effect of political ties

PurposeThis study aims to investigate the impact of family ownership on cooperative research and development (R&D). Drawing on the ability and willingness paradox framework in family business research, the authors suggest that family ownership influences cooperative R&D via two opposing mechanisms: power concentration and wealth concentration. It also deepens the current understanding of the boundary conditions of informal institutions for the impact of family ownership on cooperative R&D by investigating the moderating role of political ties.Design/methodology/approachThe authors analyze a panel of 610 Chinese manufacturing family firms and 2,127 firm-year observations from 2009 to 2017. Fixed effects regression analysis is used to test the hypotheses, with the two-stage Heckman model to address sample selection bias.FindingsThe research findings indicate that family ownership has an inverted U-shaped relationship with cooperative R&D and political ties moderate the relationship in such a way that the inverted U-shaped relationship will be steeper in firms with more political ties than in firms with fewer political ties.Practical implicationsFamily ownership influences firms’ cooperative R&D through the positive effect of power concentration and the negative effect of wealth concentration. Family owners should, therefore, take advantage of concentrated power, for instance, by adapting quickly and committing sufficient resources to cooperative R&D opportunities, while controlling path-dependent relationship development caused by concentrated family wealth. The effect of political ties on the relationship between family ownership and cooperative R&D is found to be a double-edged sword.Originality/valueThis study extends the ability and willingness paradox framework and provides novel insights into cooperative R&D in family businesses by integrating power concentration and wealth concentration associated with family ownership. Moreover, this study provides a contingency perspective and introduces the moderating role of political ties in shaping cooperative R&D in family firms.

Gender Segregation in Education: Evidence From Higher Secondary Stream Choice in India.

This paper investigates gender-based segregation across different fields of study at the senior secondary level of schooling in a large developing country. We use a nationally representative longitudinal data set from India to analyze the extent and determinants of gender gap in higher secondary stream choice. Using fixed-effects regressions that control for unobserved heterogeneity at the regional and household levels, we find that girls are about 20 percentage points less likely than boys to study in science (STEM) and commerce streams as compared with humanities. This gender disparity is unlikely to be driven by gender-specific differences in cognitive ability, given that the gap remains large and significant even after we control for individuals' past test scores. We establish the robustness of these estimates through various sensitivity analyses: including sibling fixed effects, considering intrahousehold relationships among individuals, and addressing sample selection issues. Disaggregating the effect on separate streams, we find that girls are most underrepresented in the study of science. Our findings indicate that gender inequality in economic outcomes, such as occupational segregation and gender pay gaps, is determined by gendered trajectories set much earlier in the life course, especially at the school level.

Agricultural mechanization and cropland abandonment in rural China

ABSTRACT This study examines the impacts of the adoption of three mechanized farming strategies (no-mechanized, semi-mechanized, and full-mechanized farming) on cropland abandonment, using the 2016 China Labour-force Dynamics Survey data. An innovative multivalued treatment effects model is employed to address sample selection bias. The results show that relative to no-mechanized farming, the adoption of semi-mechanized farming increases the probability of cropland abandonment. However, relative to either no or semi-mechanized farming, the adoption of full-mechanized farming significantly decreases the likelihood of cropland abandonment. Thus, farmers’ decisions of cropland abandonment depend on whether mechanization can partially or fully substitute farm labour. Our findings suggest that promoting agricultural mechanization via scale and topography-appropriate farm machines and enhancing land transfer for consolidation can help alleviate cropland abandonment.

Copula Models for Addressing Sample Selection in the Evaluation of Public Health Programmes: An Application to the Leeds Let's Get Active Study.

Sample selectivity is a recurrent problem in public health programmes and poses serious challenges to their evaluation. Traditional approaches to handle sample selection tend to rely on restrictive assumptions. The aim of this paper is to illustrate a copula-based selection model to handle sample selection in the evaluation of public health programmes. Motivated by a public health programme to promote physical activity in Leeds (England), we describe the assumptions underlying the copula selection, and its relative advantages compared with commonly used approaches to handle sample selection, such as inverse probability weighting and Heckman's selection model. We illustrate the methods in the Leeds Let's Get Active programme and show the implications of method choice for estimating the effect on individual's physical activity. The programme was associated with increased physical activity overall, but the magnitude of its effect differed according to adjustment method. The copula selection model led to a similar effect to the Heckman's approach but with relatively narrower 95% confidence intervals. These results remained relatively similar when different model specifications and alternative distributional assumptions were considered. The copula selection model can address important limitations of traditional approaches to address sample selection, such as the Heckman model, and should be considered in the evaluation of public health programmes, where sample selection is likely to be present.

Gender-based judicial ingroup bias in sex crime sentencing: Evidence from Belgium

Exploring data on adjudication of sex crimes at a major Belgian court, we study the presence of gender-based judicial ingroup bias in criminal sentencing. Using a difference-in-difference approach and addressing sample selection concerns, we find that defendants who are randomly assigned to a panel featuring a presiding judge of matching gender receive comparatively more severe sentences. The resulting effect is discernible especially when the crime involves a female victim. This suggests that the underlying driver of the negative gender-based ingroup bias is judicial enforcement of a salient ingroup norm, in itself an expression of affinity with one's gender-based ingroup.

FIRMS’ EXPORT PERFORMANCE: A FRACTIONAL ECONOMETRIC APPROACH

Export activities have become crucial to firms’ competitiveness, with determinants of export performance being a challenging field of research, since there is no consensus regarding the explained and explanatory variables or on the econometric methods to be used. Using a panel data of Portuguese wine firms, this paper aims to contribute to this debate, combining both resource- and institutional-based views of the firm. This paper tries to overcome the methodological hurdle, addressing sample selection issues and considering the fractional response nature of export performance. Given the pros and cons of each econometric approach, the Heckman selection model, the fractional probit model and the two-part fractional response model are estimated, and the results compared. From a public policy perspective, the results show that policies that promote wine firm size, labor productivity and wine promotion in third countries have a positive impact on export performance at firm-level. Age does not appear as a key factor on the internationalization of Portuguese wine firms

The Formation of Hidden Negative Capital in Banking: A Product Mismatch Hypothesis

This paper investigates hidden negative capital (HNC) in banks. We develop a simple theoretical model that links HNC to assets charge-offs and shows how product mismatch raises the HNC. In the empirical part, we further test the product mismatch hypothesis using unique data on HNC for U.S. and Russian banks. To address sample selection and endogeneity issues, we apply the Heckman selection approach with instruments. Our results indicate that in both countries product mismatch causes formation of HNC and, conditionally on selection, increases its size. Our results may facilitate improvements in the prudential regulation of banking activities in different countries.

Are Prosumer Households that Much Different? Evidence from Stated Residential Energy Consumption in Germany

This paper discusses the effect of residential energy prosuming on households’ energy consumption behavior with the objective to find evidence for a “prosumer rebound effect” on energy consumption. Prosuming is the partial self-consumption of distributed energy production. We hypothesize that prosumer households are systematically different from consumer households regarding their housing situation and socio-economic characteristics. We address sample selection bias by using a quasi-experimental technique called propensity score matching as our identification strategy. We employ data from a nationwide online survey among homeowners in Germany. While the data shows a correlation of lower energy consumption and prosumer households, we find no significant difference of energy consumption behavior between prosumers and non-prosumers when controlling for sample selection bias. Instead, the lower energy consumption of prosumer households is attributed to more energy-efficient technical equipment and thus to purchasing behavior. Our results show neither evidence for negative nor positive externalities of prosuming on residential energy consumption behavior and therefore we conclude that there is no need for additional governmental measures in the form of taxation or subsidies to address behavioral changes of energy prosuming.

Do Firms Adjust Their Acquisition Strategies and Payment Options in Response to Changes in Financial Reporting Incentives?

This paper focuses on U.S firms’ acquisition strategy and payment options in response to changes in financial reporting requirements. We study the impact of Statement of Financial Accounting Standards (SFAS) 141, the accounting treatment of business combination, and its subsequent revision, SFAS 141(R), on firms’ participation in M&A activities and choice of payment methods. SFAS 141 eliminated the pooling interest method of accounting. The Financial Accounting Standards Board (FASB) revised the standards in 2007, with the SFAS 141(R), in an attempt to enhance the ability of financial reports to reflect firms’ financial situation. The revision strengthens the rules of accounting for contingent liabilities by requiring the fair value of earnouts to be recorded at acquisition date. In comparison, under the previous standards, earnouts were only recognised if, and, when, the payment contingency is resolved. Our results provide evidence that the frequency of stock deals (earnout deals) sharply decreases (increases) since SFAS 141, and this decrease (increase) is strongest among financially constrained bidders, who also decrease their participation in M&A markets. In addition, our results show that SFAS 141(R) diminishes the growth of earnout usage. A Heckman probit model is implemented to address sample selection bias. The implications of these findings for deal design and M&A motivations are discussed.

Medical Malpractice Litigation and the Market for Plaintiff‐Side Representation: Evidence from Illinois

How concentrated is the market for medical malpractice (med mal) legal representation? Do successful plaintiffs’ lawyers start off with better cases to begin with, do they add more value to the cases they handle, or both? How do top plaintiffs’ lawyers market their services, and where did they go to school? How large are the “wages of risk”—the compensation to plaintiffs’ lawyers for working on contingency? How often do plaintiffs proceed pro se, and with what results? We address these questions using a data set of every insured med mal case closed in Illinois during 2000–2010. We show that most plaintiffs have a lawyer. We quantify the market share, case mix, and amounts recovered by the 1,317 law firms that handled med mal cases in our sample, stratify the firms into four tiers, and assess differences across tiers. We find that the market for plaintiff‐side med mal representation is both unconcentrated and highly stratified. At all firms, a small number of cases account for a heavily disproportionate share of total recoveries. We use the extensive covariates in our data to (imperfectly) address sample selection, and to estimate the effect of having a lawyer and law firm tier on outcomes. Controlling for observable claim characteristics, having a lawyer predicts a large increase in the probability of prevailing and the estimated recovery. Higher‐tier firms have only modestly higher success rates, but substantially higher estimated recoveries. However, the differences shrink and are statistically insignificant when we compare first‐tier to second‐tier firms. This suggests that there are substantial benefits to having a lawyer—and a higher‐tier lawyer—but diminishing marginal returns at the top of the market. Assuming that there is some unobserved case selection, our findings provide a plausible upper bound on the “value added” by different tiers of plaintiffs’ lawyers.