Search

Policy evaluation is central to economic data analysis, but economists mostly work with observational data in view of limited opportunities to carry out controlled experiments. In the potential outcome framework, the panel data approach (Hsiao, Ching and Wan, 2012) constructs the counterfactual by exploiting the correlation between cross-sectional units in panel data. The choice of cross-sectional control units, a key step in its implementation, is nevertheless unresolved in data-rich environment when many possible controls are at the researcher's disposal. We propose the forward selection method to choose control units, and establish validity of the post-selection inference. Our asymptotic framework allows the number of possible controls to grow much faster than the time dimension. The easy-to-implement algorithms and their theoretical guarantee extend the panel data approach to big data settings., Comment: accepted for publication at the Journal of Econometrics

This paper investigates how migration and location choice decisions depend on a large set of location characteristics, with particular focus on measuring the importance and nature of non-monetary costs of moving. We employ a stated-preference approach to elicit respondents’ choice probabilities for a set of hypothetical choice scenarios, using two waves from the NY Fed’s Survey of Consumer Expectations. Our stated probabilistic choice approach allows us to recover the distribution of individual-level preferences for location and mobility attributes without concerns about omitted variables and selection biases that hamper analyses based on observed mobility choices alone. We estimate substantial heterogeneity in the willingness-to-pay (WTP) for location characteristics and in moving costs, both across and within demographic groups. We find moving costs to be strongly associated with attachment to the current neighborhood and dwelling and to social networks. Our results indicate evidence of sorting into current locations based on preferences for location attributes as well as a strong negative association between respondents’ non-monetary moving costs and their moving expectations and actual mobility decisions.

This paper examines the impact of Airbnb on the local rental housing market. Airbnb provides landlords an alternative opportunity to rent to short-term tourists, potentially leading some landlords to switch from long-term rentals and thereby, affecting rental housing supply and affordability. Despite recent government regulations to address this concern, it remains unclear how many and what types of properties are switching. Combining Airbnb and American Housing Survey data, we estimate a structural model of property owners’ decisions and conduct counterfactual analyses to evaluate various regulations. We find that Airbnb mildly cannibalizes the long-term rental supply. Cities where Airbnb is more popular experience a larger rental supply reduction, but they do not necessarily have a larger percentage of switchers. Affordable units are the major sources of both the negative and positive impacts of Airbnb. They cause a larger rental supply reduction, which harms local renters; they also create a larger market expansion effect, which benefits local hosts who own affordable units and may be less economically advantaged. Policy makers need to strike a balance between local renters’ affordable housing concerns and local hosts’ income source needs. We also find that imposing a linear tax is more desirable than limiting the number of days a property can be listed. We propose a new convex tax that imposes a higher tax on expensive units and show that it can outperform existing policies in terms of reducing cannibalization and alleviating social inequality. Finally, Airbnb and rent control can exacerbate each other’s negative impacts. This paper was accepted by Matt Shum, marketing. Supplemental Material: The data files and online appendix are available at https://doi.org/10.1287/mnsc.2021.4288 .

This paper studies the impact of foreign labour on the competitiveness of Italian dairy farms relying on the theory of foreign labour in profit functions. Application of an endogenous switching regression model identifies the drivers of adoption of the immigrant workforce. A counterfactual analysis performed on unit labour costs between farms employing and non-employing immigrants suggests the essential role of foreign farmworkers on dairy farms’ competitiveness, which provide a cheaper source of labour. The lower unit labour cost for immigrant workers resembles staunch support to the newly introduced ‘social conditionality’, for a CAP delivering also for farmworkers.

We assess the effects of land markets on misallocation and productivity both empirically and quantitatively. Exploiting variation from a land certification reform across time and space in Ethiopia, we find that certification facilitates rentals and improves agricultural productivity. We calibrate a quantitative macroeconomic model with heterogeneous household farms facing institutional costs to land markets using the micro panel data. The effect of a counterfactual reallocation from no rentals to efficient rentals increases zone-level agricultural productivity by 43 percent on average. While our estimated institutional costs are strongly associated with land certification across zones, there are nontrivial residual frictions to rental market activity, implying that land certification only partially captures the overall effects of rentals. A full certification reform accounts for about one-fourth of the overall productivity gains from land rentals. This result highlights the importance of comprehensive reforms alleviating frictions to land transactions beyond the granting of certificates.

Artist collaborations in music have been on the rise, and they tend to produce commercially and critically successful songs. We seek to uncover the effect of these collaborative projects on career trajectories and identify the factors that lift an artist’s profile in the short and long term. We develop a theory of collaboration based on the transfer of capital between the collaborating artists that facilitates spillovers across time. To validate the theory, we use weekly radio plays of individual songs across 25 European countries between the years 2011 to 2018, together with a multiattribute Spotify data set of songs and Hofstede’s cultural dimensions in relation to artist origins. We create pairs of similar artists who released a collaboration and a solo song in the same week and measure the impact of collaborations based on the difference-in-differences methodology. We find that releasing a collaboration song, in comparison with a solo song, increases the number of plays of an artist in the future by +4.6%. This lift can be broken down into +9.6% for the current song and +7.7% for subsequently released songs, whereas past songs are unaffected. The effect is moderated by the difference in economic, social, and cultural capitals and is significantly larger when one’s partner has higher economic and social capital or is highly dissimilar along the cultural dimension. Our theoretical and empirical exploration of such strategic alliances uncovers several underlying mechanisms at play in the success of these pairings and can serve as the basis for future work targeted at prescriptive contributions. This paper was accepted by David Simchi-Levi, operations management. Funding: The research was supported by the Spanish Ministry of Science and Innovation (Ministerio de Ciencia e Innovación) [Grant Ref. PID2020-116135GB-I00/AEI/10.13039/501100011033]. Supplemental Material: The online appendix and data are available at https://doi.org/10.1287/mnsc.2023.4743 .

Objective: The objective of this study was to evaluate England's Best Practice Tariff (BPT) and consider potential implications for Medicare patients should the US adopt a similar plan.Summary Background Data: Since the beginning of the Affordable Care Act, Medicare has renewed efforts to improve the outcomes of older adults through introduction of an expanding set of alternative-payment models. Among trauma patients, recommended arrangements met with mixed success given concerns about the heterogeneous nature of trauma patients and resulting outcome variation. A novel approach taken for hip fractures in England could offer a viable alternative.Methods: Linear regression, interrupted time-series, difference-in-difference, and counterfactual models of 2000–2016 Medicare (US), HES-APC (England) death certificate-linked claims (≥65y) were used to: track US hip fracture trends, look at changes in English hip fracture trends before-and-after BPT implementation, compare changes in US-versus-English mortality, and estimate total/theoretical lives saved.Results: 806,036 English and 3,221,109 US hospitalizations were included. Following BPT implementation, England's 30-day mortality decreased by 2.6 percentage-points (95%CI: 1.7–3.5) from a baseline of 9.9% (relative reduction 26.3%). 90- and 365-day mortality decreased by 5.6 and 5.4 percentage-points. 30/90/365-day readmissions also declined with a concurrent shortening of hospital length-of-stay. From 2000–2016, US outcomes were stagnant (p > 0.05), resulting in an inversion of the countries’ mortality and > 38,000 potential annual US lives saved.Conclusions: Process measure pay-for-performance led to significant improvements in English hip fracture outcomes. As efforts to improve US older adult health continue to increase, there are important lessons to be learned from a successful initiative like the BPT.

Deep learning models have achieved high performance across different domains, such as medical decision-making, autonomous vehicles, decision support systems, among many others. However, despite this success, the inner mechanisms of these models are opaque because their internal representations are too complex for a human to understand. This opacity makes it hard to understand the how or the why of the predictions of deep learning models. There has been a growing interest in model-agnostic methods that make deep learning models more transparent and explainable to humans. Some researchers recently argued that for a machine to achieve human-level explainability, this machine needs to provide human causally understandable explanations, also known as causability. A specific class of algorithms that have the potential to provide causability are counterfactuals. This paper presents an in-depth systematic review of the diverse existing literature on counterfactuals and causability for explainable artificial intelligence (AI). We performed a Latent Dirichlet topic modeling analysis (LDA) under a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework to find the most relevant literature articles. This analysis yielded a novel taxonomy that considers the grounding theories of the surveyed algorithms, together with their underlying properties and applications to real-world data. Our research suggests that current model-agnostic counterfactual algorithms for explainable AI are not grounded on a causal theoretical formalism and, consequently, cannot promote causability to a human decision-maker. Furthermore, our findings suggest that the explanations derived from popular algorithms in the literature provide spurious correlations rather than cause/effects relationships, leading to sub-optimal, erroneous, or even biased explanations. Thus, this paper also advances the literature with new directions and challenges on promoting causability in model-agnostic approaches for explainable AI.

Embeddings of textual data containing location names (e.g., social media posts) have essential applications in various contexts such as marketing and disaster management. In these downstream implementations, social biases behind location names are highly prone to introduce unfair results through their embeddings; for example, emergent text messages with swapped location names might result in varied rescue responses. Hence, it is critical to address social biases encoded in location names and to seek its mitigation. Prevalent works addressing biases in embeddings mainly focus on individual attributes like gender or ethnicity. Yet, a large number of social attributes behind location names (e.g., income level and population density) makes it challenging to originate the source of biases. Existing mitigation methods based on finding attribute subspaces cannot be simply applied to address social biases. Moreover, bias mitigation tends to simultaneously remove necessary semantics from embeddings, making it difficult to achieve a balance between mitigation performance and semantics retention. In this article, we first employ the concept of counterfactual fairness to investigate the social biases encoded in training data. Then, we quantify the biases in the contextual embeddings (BERT and ELMo). We report a high correlation between biases in the training data and embeddings. Next, we introduce a novel bias mitigation algorithm that customizes bias representations for any location names. The method yields debiased location name vectors for various social attributes simultaneously. The proposed algorithm achieves a better mitigation performance on overall attributes compared with a prevalent postprocessing method, while maintaining correctness by retaining semantic information.

The increasing use of machine learning in practice and legal regulations like EU’s GDPR cause the necessity to be able to explain the prediction and behavior of machine learning models. A prominent example of particularly intuitive explanations of AI models in the context of decision making are counterfactual explanations. Yet, it is still an open research problem how to efficiently compute counterfactual explanations for many models. In this contribution, we investigate how to efficiently compute counterfactual explanations for an important class of models, prototype-based classifiers such as learning vector quantization models. In particular, we derive specific convex and non-convex programs depending on the used metric. Typically counterfactual explanations deliver a feedback in terms of changes of the input features which lead to a different output – one application scenario is the link of these required changes to actionable items to change the desired outcome. Yet, rather than minimum changes of the input, it is interesting to address minimum changes of the model itself, which are required to lead to a different result. rather than a change of its inputs. We phrase this question as a counterfactual of the model prescription rather than the data points. We focus on distance-based classifiers (in particular learning vector quantization models), where model changes correspond to changes of metric parameters, and we develop efficient optimization techniques to generate such counterfactual metric changes depending on the chosen model.

This paper investigates the impacts of road investments in secondary markets, which the authors label wider local impacts. The impacts are studied using four indicators: commuting, population, new firms, and employment. We use the synthetic control method to study the counterfactual problem, namely what would have happened if a given project had not been realised. The method is used to compare municipalities that had been given a new road with municipalities that had not had a new road. The study sample consists of ten road projects that opened for traffic between 2000 and 2010 and the impacts of the projects are examined at municipal level. The results do not provide a clear answer as to whether road projects are a suitable tool for fulfilling political objectives of improving the local economy. Apart from possibly one exception, none of the projects scored positively on all indicators. We identify several examples of significant negative impacts as a result of road investments, and conclude that although the impacts have been positive in many areas, there is no evidence that road investments are generally a potent tool for achieving positive wider local impacts.

Artificial intelligence methods such as deep neural networks promise unprecedented capabilities in healthcare, from diagnosing diseases to prescribing treatments. While this can eventually produce a valuable suite of tools for automating clinical workflows, a critical step forward is to ensure that the predictive models are reliable and to enable a rigorous introspection of their behavior. This has led to the design of explainable AI techniques that are aimed at uncovering the relationships between discernible data signatures and decisions from machine-learned models, and characterizing strengths/weaknesses of models. In this context, the so-called counterfactual explanations that synthesize small, interpretable changes to a given query sample while producing desired changes in model predictions to support user-specified hypotheses (e.g., progressive change in disease severity) have become popular. When a model’s predictions are not well-calibrated (i.e., the prediction confidences are not indicative of the likelihood of the predictions being correct), the inverse problem of synthesizing counterfactuals can produce explanations with irrelevant feature manipulations. Hence, in this paper, we propose to leverage prediction uncertainties from the learned models to better guide this optimization. To this end, we present TraCE (Training Calibration-based Explainers), a counterfactual generation approach for deep models in medical image analysis, which utilizes pre-trained generative models and a novel uncertainty-based interval calibration strategy for synthesizing hypothesis-driven explanations. By leveraging uncertainty estimates in the optimization process, TraCE can consistently produce meaningful counterfactual evidences and elucidate complex decision boundaries learned by deep classifiers. Furthermore, we demonstrate the effectiveness of TraCE in revealing intricate relationships between different patient attributes and in detecting shortcuts, arising from unintended biases, in learned models. Given the widespread adoption of machine-learned solutions in radiology, our study focuses on deep models used for identifying anomalies in chest X-ray images. Using rigorous empirical studies, we demonstrate the superiority of TraCE explanations over several state-of-the-art baseline approaches, in terms of several widely adopted evaluation metrics in counterfactual reasoning. Our findings show that TraCE can be used to obtain a holistic understanding of deep models by enabling progressive exploration of decision boundaries, detecting shortcuts, and inferring relationships between patient attributes and disease severity.

This paper studies the impact of a minimum wage policy in a labor market with a private and a public sector. We develop a two-sector search and matching model with minimum wage and heterogeneous workers in their human capital. We structurally estimate the model using data for Chile, a country with a large fraction of employment in the public sector and a binding minimum wage. Counterfactual analysis shows that institutional features of public sector employment reduce labor market frictions and mitigate the negative effect of the minimum wage on unemployment and welfare.

We study the optimal design of demand response and peak-time rebate programs in the electricity sector. Load-serving entities (LSEs) frequently use demand response or peak-time rebate programs to reduce consumption during peak hours. However, the LSE has imperfect information, so it typically estimates a ‘baseline’, a customer’s counterfactual energy consumption, and it offers (weakly positive) incentives for reductions below this threshold. This approach leads to generous incentives, even to customers who make no effort to reduce their demand, due to inherent variability or unpredictability of demand. We show that the estimated baseline is not necessarily the optimal threshold for calculating rebates in demand response programs. We consider a customer demand model and investigate a natural objective function for the LSE. Our methodology helps the LSE learn about customer demand patterns and also customers’ propensity for reductions over time. The LSE balances multiple objectives to determine the appropriate rebate threshold for customers. By incorporating data from a ToU pricing experiment in London, we show that compared to a conventional demand response program, our approach significantly improves the targeting of demand response payments.

Do healthcare providers pick their patients? This paper uses patient-level administrative data on skilled nursing facilities in California to estimate a structural model of selective admission practices in the nursing home industry. I exploit within-facility covariation between occupancy and admitted patient characteristics to distinguish admission patterns attributable to selective admission practices from those attributable to heterogeneous patient preferences. In spite of anti-discrimination laws, I find strong evidence of selective admission practices that disproportionately harm Medicaid-eligible patients with lengthy anticipated stays. Counterfactual simulations show that enforcing a prohibition on selective admissions would increase access for these residents at the cost of crowding out short-stay non-Medicaid patients from their preferred facilities. I simulate two additional policies intended to mitigate selective admissions: raising the Medicaid reimbursement rate and expanding capacity. I find the latter to be less costly and more effective than the former.

The demand for long-term care (LTC) services is growing strongly, mostly due to population aging. Historically, the German LTC market was dominated by non-profit nursing homes, but the recent entry wave was tilted towards for-profit competitors. Using a rich administrative dataset on all LTC facilities in Germany, we examine strategic interaction between these two ownership types in a static entry model. The estimates of competitive effects imply that non-profit and for-profit homes are substitutes, but competition is much stronger within-type, suggesting that they provide differentiated products. For-profit homes in particular act as if they operate in a different market segment, but over time their entry behavior has converged to that of the more established non-profits. Counterfactual simulations of proposed changes in government policy suggest that even small changes favoring either type could have a large impact on the fraction of markets that remain unserved or only served by a single type.

Population and public health are in the midst of an artificial intelligence revolution capable of radically altering existing models of care delivery and practice. Just as AI seeks to mirror human cognition through its data-driven analytics, it can also reflect the biases present in our collective conscience. In this Viewpoint, we use past and counterfactual examples to illustrate the sequelae of unmitigated bias in healthcare artificial intelligence. Past examples indicate that if the benefits of emerging AI technologies are to be realized, consensus around the regulation of algorithmic bias at the policy level is needed to ensure their ethical integration into the health system. This paper puts forth regulatory strategies for uprooting bias in healthcare AI that can inform ongoing efforts to establish a framework for federal oversight. We highlight three overarching oversight principles in bias mitigation that maps to each phase of the algorithm life cycle.

The Kyoto Protocol has received much criticism for its effectiveness as well as the spillover effect (i.e. carbon leakage and competitiveness loss). This paper provides the first evidence for the effect of the Kyoto Protocol on the bilateral trade in environmental goods, which can mitigate and prevent environmental damage. Using the generalized synthetic control method, I construct the counterfactual of trading pairs with Kyoto commitment and show that the export of environmental goods by Kyoto countries increases by 29–30% after the protocol enters into force. The paper further examines the top exporters (Germany and Japan) individually and also reports positive effects. Technology transfer induced by project-based mechanisms under the Kyoto Protocol, in particular Clean Development Mechanism, is found to be the potential driving force.

Academics have proposed hybrid products with equity features for the financing of housing. In spite of their risk-sharing benefits these products have not become mainstream. This paper studies an important exception, a UK government scheme which in the five years since its inception has provided almost £10 billion of equity financing. The analysis of the origination and prepayment behavior of households who have used the scheme highlights housing affordability constraints. A difference-in-difference analysis of an increase in the maximum government equity limit shows that households took advantage of the increase to buy more expensive properties, and not to reduce their mortgage debt and house price risk exposure. A counterfactual study of homebuyers who, instead of using the equity available, relied on high loan-to-value mortgages shows that their financing choices can be rationalized by an expected rate of house price appreciation of 7.7% per year. We draw general implications for how households approach their house purchase and financing decisions, taking advantage of the fact that the shared equity mortgages that we study allow the separation of the consumption and investment dimensions of housing.

Environmental regulation can interact with agricultural markets to produce underappreciated competitiveness and leakage effects. This paper measures effective carbon tax stringency by structurally recovering the domestic supply schedule for a trade-exposed beef cattle industry such that elasticities and carbon tax rates change with product prices (i.e., due to the curvature of the supply function). Two basic propositions from the economics of taxation -- that excess burdens increase in elasticities and tax rates -- are shown to cause the stringency of uniform carbon policy to vary nonlinearly with output prices. Based on the domestic supply function, the relationship between marginal excess burden, a measure of policy stringency from the industry's perspective, and product prices is estimated. Several policy-relevant counterfactual scenarios are explored. Results show that with moderately high output prices, supply elasticities are small and the efficiency cost of a $40/tCO2e carbon tax (gross of environmental benefits) is less than $0.01 per dollar tax revenue. As prices decline, supply curves become increasingly elastic and marginal excess burdens grow rapidly.

The theoretical literature on price-matching guarantees (PMGs) finds that this policy has both a competition-softening and a competition-enhancing effect. Which effect dominates depends on market structure. This paper is the first to propose a structural framework to measure the impact of PMGs on market competition through a counterfactual analysis. The structural model proposed here can be estimated using price data alone. I estimate the model using data from the automotive tire market, and I find that the competition-softening effect is stronger than the competition-enhancing effect. PMGs keep transaction prices between 1% and 8% higher than they would be in the absence of such policy. PMGs exert the strongest effect on price-sensitive consumers, who tend to be the poorest. This consumer segment pays up to 10% higher prices in the presence of PMGs. The tire market has some unique features that facilitate the competition-enhancing effect of PMGs. Hence, that the competition-softening effect dominates even in the tire market suggests that PMGs may increase prices in many other markets, too.

This paper investigates the how government outsourcing affects efficiency and expenditures by considering how outsourcing decisions are determined along two dimensions: (i) cost differences between private firms and government suppliers of public goods and (ii) dynamics arising from cost complementarities and capacity constraints. I formulate and estimate a dynamic model of government outsourcing using project-level data from the dredging industry. Model estimates indicate substantial cost savings due to outsourcing but also that government presence in the market yields cost reduction. A counterfactual policy featuring direct competition between government and private sector firms finds a total expenditure reduction of 15.7 percent. (JEL D44, H41, H57, L84)

This paper uses new, large-scale vehicle registry data from Norway and a two-sided market framework to show nonneutrality of different subsidies and estimate their impact on electric vehicle adoption when network externalities are present. Estimates suggest a strong positive connection between electric vehicle purchases and both consumer price and charging station subsidies. Counterfactual analyses suggest that between 2010 and 2015, every dollar spent on station subsidies resulted in more than twice as many additional electric vehicle purchases than the same amount spent on price subsidies. However, this relation inverts with increased spending, as station subsidies’ impact tapers off faster. (JEL D12, D62, D85, H25, L62, Q54)

The success of Artificial Intelligence at solving real-world problems poses the need for interpretable models, especially in human-centered applications. The multi-class scenario is often present in these environments; however, the majority of research on interpretability has focused on binary classification. In this work, a novel method based on hierarchical decompositions to obtain interpretable multi-class models is introduced. The proposal, named Threshold Control for Nested Dichotomies (TC-ND) method, creates a binary-based hierarchical class structure. Then, it discards meta-classes at each dichotomy of the structure according to a certain level of confidence, pursuing a modular and more comprehensible decomposition of the multi-class problem. The approach presents internal parameters that are optimized using the Differential Evolution algorithm. The goodness of our proposal is assessed using a twofold approach: performance is evaluated by comparing against other state-of-the-art multi-class methods; interpretability is supported by practical examples with counterfactual explanations and a discussion of the advantages that the TC-ND method presents regarding its transparency and auditability.

This paper analyzes the effects of credit frictions in a trade model where heterogeneous firms select both into exporting and into two types of external finance. In our framework, small producers face stronger credit frictions, pay a higher borrowing rate and rely on bank finance, whereas large firms have access to cheaper bond finance. We show that an increase in credit frictions induces firms to select into bank finance, which attenuates the negative implications on product variety and welfare. In the open economy, the presence of effective financial intermediation increases the welfare gains from trade. In a counterfactual analysis, we exploit that our framework nests a model with credit frictions and one type of finance as a special case, and we show that endogenous selection into external finance is an important channel of adjustment.

Summary This paper uses a panel data approach to assess the evolution of economic consequences of the drastic lockdown policy in the epicenter of COVID‐19 ‐‐‐the Hubei Province of China during worldwide curbs on economic activity. We find that the drastic 76‐day COVID‐19 lockdown policy brought huge negative impacts on Hubei’s economy. In 2020:q1, the lockdown quarter, the treatment effect on GDP was about 37% of the counterfactual. However, the drastic lockdown also brought the spread of COVID‐19 under control in little more than two months. After the government lifted the lockdown in early April, the economy quickly recovered with the exception of passenger transportation sector which rebounded not as quickly as the rest of the general economy.

Corrupt government hiring is common in developing countries. This paper uses original data to document the operation and consequences of corrupt hiring in a health bureaucracy. Hires pay bribes averaging 17 months of salary, but contrary to conventional wisdom, their observable quality is comparable to counterfactual merit-based hires. Exploiting variation across jobs, I show that the consequences of corrupt allocations depend on the correlation between wealth and quality among applicants: service delivery outcomes are good for jobs where this was positive and poor when negative. In this setting, the correlation was typically positive, leading to relatively good performance of hires. (JEL D73, I11, J16, J24, J45, M51, O17)

Artificial intelligence systems are becoming ubiquitous in everyday life as well as in high-risk environments, such as autonomous driving, medical treatment, and medicine. The opaque nature of the deep neural network raises concerns about its adoption in high-risk environments. It is important for researchers to explain how these models reach their decisions. Most of the existing methods rely on softmax to explain model decisions. However, softmax is shown to be often misleading, particularly giving unjustified high confidence even for samples far from the training data. To overcome this shortcoming, we propose Bayesian model uncertainty for producing counterfactual explanations. In this paper, we compare the counterfactual explanation of models based on Bayesian uncertainty and softmax score. This work predictively produces minimal important features, which maximally change classifier output to explain the decision-making process of the Bayesian model. We used MNIST and Caltech Bird 2011 datasets for experiments. The results show that the Bayesian model outperforms the softmax model and produces more concise and human-understandable counterfactuals.

Graduate Practice is in Slovakia one of the measures of active labour market policy aimed at decreasing the unemployment rate of young graduates. It is one of the most frequently used interventions by young jobseekers under the age of 26; similar measures are used in other European Union (EU) member states. The specific objective of this intervention is to provide young jobseekers to gain the first contact with the labour market, first work experience and work habits that may be attractive to potential employers. The intervention is financed from the European Social Fund (within the Operational Program – Human Resources and the priority axis – Youth Employment Initiative) and the state budget. This fact requires ad 1) to evaluate the eligibility of drawing these resources and ad 2) to evaluate the effectiveness of the use of resources in terms of employability and sustainability of participants of the intervention in the labour market, the second of which is the primary objective of this study. The European Commission has increasingly required the member states of the EU to carry out impact evaluations of funded interventions and to develop evidence-based policies that will implement the results of such evaluations. The aim of this study is, therefore, to evaluate and review the effects of the Graduate Practice on the employability of young jobseekers in conditions of Slovakia. This evaluation is carried out using a counterfactual approach by comparing the results of treated and non-treated individuals by applying the three most used methods: regression adjustment, instrumental variable, and propensity score matching method. The results indicate significant effects of the Graduate Practice on the employability of young jobseekers and their sustainability in employment. The results of the evaluation are valuable mainly for policymakers, i.e. in setting the rules of the intervention and its eligibility to make the measure more effective or potentially in preparing other measures aimed at youth unemployment. Implications for Central European audience: The findings of the evaluation may inspire policymakers in other countries than Slovakia to introduce similar or new programs for the youth unemployed. The results of this study may also be used for an international comparison of the effectiveness of interventions if a similar program for youth unemployed operates in another country, which may be appropriate mainly from the point of view of being a program(s) co-financed by the European Social Fund.

Background The coronavirus disease 2019 (COVID-19) pandemic disrupted access to and uptake of hepatitis C virus (HCV) care services in the United States. It is unknown how substantially the pandemic will impact long-term HCV-related outcomes. Methods We used a microsimulation to estimate the 10-year impact of COVID-19 disruptions in healthcare delivery on HCV outcomes including identified infections, linkage to care, treatment initiation and completion, cirrhosis, and liver-related death. We modeled hypothetical scenarios consisting of an 18-month pandemic-related disruption in HCV care starting in March 2020 followed by varying returns to pre-pandemic rates of screening, linkage, and treatment through March 2030 and compared them to a counterfactual scenario in which there was no COVID-19 pandemic or disruptions in care. We also performed alternate scenario analyses in which the pandemic disruption lasted for 12 and 24 months. Results Compared to the “no pandemic” scenario, in the scenario in which there is no return to pre-pandemic levels of HCV care delivery, we estimate 1060 fewer identified cases, 21 additional cases of cirrhosis, and 16 additional liver-related deaths per 100 000 people. Only 3% of identified cases initiate treatment and Conclusions A recommitment to the HCV epidemic in the United States that involves additional resources coupled with aggressive efforts to screen, link, and treat people with HCV is needed to overcome the COVID-19-related disruptions.

Dynamic treatment regimes (DTRs) include a sequence of treatment decision rules, in which treatment is adapted over time in response to the changes in an individual’s disease progression and health care history. In medical practice, nested test-and-treat strategies are common to improve cost-effectiveness. For example, for patients at risk of prostate cancer, only patients who have high prostate-specific antigen (PSA) need a biopsy, which is costly and invasive, to confirm the diagnosis and help determine the treatment if needed. A decision about treatment happens after the biopsy, and is thus nested within the decision of whether to do the test. However, current existing statistical methods are not able to accommodate such a naturally embedded property of the treatment decision within the test decision. Therefore, we developed a new statistical learning method, Step-adjusted Tree-based Reinforcement Learning, to evaluate DTRs within such a nested multi-stage dynamic decision framework using observational data. At each step within each stage, we combined the robust semi-parametric estimation via Augmented Inverse Probability Weighting with a tree-based reinforcement learning method to deal with the counterfactual optimization. The simulation studies demonstrated robust performance of the proposed methods under different scenarios. We further applied our method to evaluate the necessity of prostate biopsy and identify the optimal test-and-treat regimes for prostate cancer patients using data from the Johns Hopkins University prostate cancer active surveillance dataset.

The assessment of causality is fundamental to epidemiology and biomedical sciences. One well-known approach to distinguishing causal from noncausal explanations is the nine Bradford Hill viewpoints. A recent article in this journal revisited the viewpoints to incorporate developments in causal thinking, suggesting that the sufficient cause model is useful in elucidating the theoretical underpinning of the first of the nine viewpoints—strength of association. In this article, we discuss how to discern the causal mechanisms of interest in the sufficient cause model, which pays closer attention to the relationship between the sufficient cause model and the Bradford Hill viewpoints. To this end, we explicate the link between the sufficient cause model and the potential-outcome model, both of which have become the cornerstone of causal thinking in epidemiology and biomedicine. A clearer understanding of the link between the two models provides significant implications for interpretation of the observed risks in the subpopulations defined by exposure and confounder. We also show that the concept of potential completion times of sufficient causes is useful to fully discerning completed sufficient causes, which leads us to pay closer attention to the fourth of the nine Bradford Hill viewpoints—temporality. Decades after its introduction, the sufficient cause model may be vaguely understood and thus implicitly used under unreasonably strict assumptions. To strengthen our assessment in the face of multifactorial causality, it is significant to carefully scrutinize the observed associations in a complementary manner, using the sufficient cause model as well as its relevant causal models.

The randomized controlled trial is the quintessential scientific tool to evaluate the effectiveness and safety of medications. While early trials of drugs used for the treatment of chronic obstructive pulmonary disease (COPD) and other respiratory diseases were generally unambiguous, more recent studies have been controversial. It has become evident that the conduct, design and analysis of these trials were highly variable and may have been responsible for incoherencies in results and interpretation. With the advent of new studies, the need for guiding principles for the conduct of future randomized trials has become manifest. We describe the concept of the counterfactual principle as it applies to the treatment of patients and to the randomized trial. We then present ten methodological tenets for the design and statistical aspects of randomized controlled trials evaluating the effectiveness of drugs used in the treatment of several respiratory diseases. They include eight study design and two statistical analysis principles: 1) Study question; 2) Intervention; 3) Study population; 4) Blinding; 5) Run-in period; 6) Follow-up; 7) Outcome; 8) Safety; 9) Intent-to-treat; 10) Covariate adjustment. These tenets are described using mainly examples from trials of pharmacological treatments for COPD, as well as some from asthma and idiopathic pulmonary fibrosis, conducted over the last 30 years. The careful application of these principles in the conduct of randomized trials will provide rigorous studies and improve the validity of results. The ensuing clearer interpretation of findings will permit their well-founded contribution to treatment guidelines and optimal clinical management.

Research on eye-movement control during natural scene viewing has investigated the degree to which the duration of individual fixations can be immediately adjusted to ongoing visual-cognitive processing demands. Results from several studies using the fixation-contingent scene quality paradigm suggest that the timing of fixations adapts to stimulus changes that occur on a fixation-to-fixation basis. Analysis of fixation-duration distributions has revealed that saccade-contingent degradations and enhancements of the scene stimulus have two qualitatively distinct types of influence. The surprise effect begins early in a fixation and is tied to surprising visual events such as unexpected stimulus changes. The encoding effect is tied to difficulties in visual-cognitive processing and occurs relatively late within a fixation. Here, we formalize an existing descriptive account of these two effects (referred to as the dual-process account) by using stochastic simulations. In the computational model, surprise and encoding related influences are implemented as time-dependent changes in the rate at which saccade timing and programming are completed during critical fixations. The model was tested on data from two experiments in which the luminance of the scene image was either decreased or increased during selected critical fixations (Walshe & Nuthmann, Vision Research, 100, 38–46, 2014). A counterfactual method was used to remove model components and to identify their specific influence on the fixation-duration distributions. The results suggest that the computational dual-process model provides a good account for the data from the luminance-change studies. We describe how the simulations can be generalized to explain a diverse set of experimental results.

PurposeThe purpose of this paper is to examine the impact of agricultural cooperative membership on potato production and technical efficiency.Design/methodology/approachA combination of propensity score matching technique and sample selection stochastic frontier framework that addresses potential selection bias due to observable and unobservable attributes is used to estimate the effect of participation between cooperative members and non-members. Using a stochastic meta-frontier approach, the technical efficiency of farmers was estimated and compared.FindingsThe empirical results show that the effect of participation in agricultural cooperatives is associated with increased yield and technical efficiency. A comparison of group-specific frontiers indicates that cooperative members perform better than non-members. Cooperative membership decisions is significantly associated with household and farm characteristics (e.g. education, participation in off-farm work, total farmland, distance to market and geographic location).Practical implicationsThe findings of this study demonstrate that cooperative organisations can be an important tool to enhance the productivity and efficiency of smallholder farmers. Successful cooperative models together with training programs designed to enlighten farmers on the importance and tangible benefits of collective action should be used to enlarge participation in cooperative organisations. In addition, governments and development agencies should implement targeted investment and capacity building programs related to irrigation management, gender-sensitive awareness and development of the internal institutional mechanisms in cooperatives for the transfer of knowledge and mutual learning so that all members benefit from cooperatives.Originality/valueDespite the pervasive evidence of the impact of cooperatives on productivity and technical efficiency in the Asian region, this study is probably the first attempt in the crop sector in Mongolia. It provides a rigorous empirical analysis of the impact of agricultural cooperative membership on potato production and technical efficiency through a counterfactual design.

Local government funding of sports stadiums is frequently justified as stimulating economic activity despite consensus contrary findings in the academic literature. Though some studies have identified positive neighborhood effects from stadiums on nearby residential property, scant research exists regarding impacts on highly-localized commercial activity that is hypothesized to spur further local economic development. This analysis exploits the recent relocation of a professional baseball team from downtown Atlanta to a pre-existing Georgia Business Improvement District (Community Improvement District or CID) in suburban Cobb County to estimate the impact of the stadium development on commercial property values. The existence of multiple CIDs in the metro-Atlanta area provides the opportunity to estimate a counterfactual comparison absent the stadium to draw causal inference regarding the stadium’s impact on economic activity using the synthetic control method. Estimates do not identify increased commercial property values in the district following the stadium opening, which is consistent with past studies that find little to no positive impact on economic activity. The results indicate that previous findings of positive impacts on residential property may not be applicable to commercial property.

This paper explores the adoption choice of electronic medical records by U.S. hospitals, which could exhibit strategic complements or substitutes. I find complementarities in adoption through a reducedform analysis with instruments for unobserved market characteristics. I further develop a dynamic oligopoly model to allow for strategic timing incentives that are missing in the static model. Adopting a dominant local vendor could increase perperiod profits from adoption by 9.2% over choosing a marginal vendor. A counterfactual analysis suggests an incentive program rewarding coordination, not just adoption, is more effective in achieving interoperability, especially before the widespread adoption of the technology.

Banks use trading as a vehicle to take risk. Using unique high-frequency regulatory data, we estimate the sensitivity of weekly bank trading profits to aggregate equity, fixed-income, credit, currency and commodity risk factors. Our estimates imply that U.S. banks had large trading exposures to equity market risk before the Volcker Rule, which they curtailed afterwards. They also have exposures to credit and currency risk. The results hold up in a quasi-natural experimental design that exploits the phased-in introduction of reporting requirements to address identification. Heterogeneity and placebo tests further corroborate the results. Counterfactual stress-test analyses quantify the financial stability implications.

In cancer randomized controlled trials, surrogate endpoints are frequently time-to-event endpoints, subject to the competing risk from the time-to-event clinical outcome. In this context, we introduce a counterfactual-based mediation analysis for a causal assessment of surrogacy. We use a multistate model for risk prediction to account for both direct transitions towards the clinical outcome and indirect transitions through the surrogate outcome. Within the counterfactual framework, we define natural direct and indirect effects with a causal interpretation. Based on these measures, we define the proportion of the treatment effect on the clinical outcome mediated by the surrogate outcome. We estimate the proportion for both the cumulative risk and restricted mean time lost. We illustrate our approach by using 18-year follow-up data from the SPCG-4 randomized trial of radical prostatectomy for prostate cancer. We assess time to metastasis as a surrogate outcome for prostate cancer-specific mortality.

I study a college-admissions model with two need-blind colleges and heterogeneous students. In a game in which colleges can choose a financial aid policy and either binding, nonbinding, or no early admissions, a unique equilibrium outcome exists. In equilibrium, the more prestigious and wealthier college is more selective, has a more generous financial aid policy, and offers nonbinding early admissions, whereas the other college offers a binding program. Compared to the counterfactual in which only regular admissions are offered, early admissions make the more prestigious college worse off, the other college better off, and students, in the aggregate, better off.

Background The SEARCH study provided community-based HIV and multi-disease testing and antiretroviral therapy (ART) to 32 communities in East Africa and reported no statistically significant difference in three-year HIV incidence. We used mathematical modelling to estimate the effect of control arm viral suppression and community mixing on SEARCH trial outcomes. Setting Uganda and Kenya. Methods Using the individual-based HIV modeling software EMOD-HIV, we configured a new model of SEARCH communities. The model was parameterized using demographic, HIV prevalence, male circumcision, and viral suppression data, and calibrated to HIV prevalence, ART coverage, and population size. Using assumptions about ART scale-up in the control arm, degree of community mixing, and effect of baseline testing, we estimated comparative HIV incidence under multiple scenarios. Results Prior to the trial results, we predicted that SEARCH would report a 4-40% reduction between arms, depending on control arm ART linkage rates and community mixing. With universal baseline testing followed by rapidly expanded ART eligibility and uptake, modelled effect sizes were smaller than the study was powered to detect. Using interim viral suppression data, we estimated three-year cumulative incidence would have been reduced by up to 27% in the control arm and 43% in the intervention arm compared to a counterfactual without universal baseline testing. Conclusions Our model suggests that the active control arm substantially reduced expected effect size and power of the SEARCH study. However, compared to a counterfactual "true control" without increased ART linkage due to baseline testing, SEARCH reduced HIV incidence by up to 43%.

We use a simple structural matching model with unobserved heterogeneity to produce counterfactual marriage patterns of two kinds: counterfactuals that hold the match surplus constant across markets and counterfactuals that hold the joint utility constant. These counterfactuals allow us to decompose matching patterns into an intensive margin (who marries whom) and an extensive margin (how many and who remain single). We apply this approach to US data from 1962 to 2017 and show that marital patterns can explain about 1/3 of the rise in income inequality, the intensive margin contributing for 5%, the extensive margin for the remaining 95%. The extensive margin is itself driven for 2/3 by a change in the total number of singles and 1/3 by a change in the distribution of types among singles.

OBJECTIVE: To re‐evaluate the effect of Medicaid on poverty using a poverty measure that accounts for health insurance needs and benefits and an evaluation approach that reflects disparities in access to alternative coverage. DATA SOURCES: The Current Population Survey (CPS) for calendar year 2015. STUDY DESIGN: We estimate the effect of losing Medicaid on poverty, combining two previous approaches: (1) A propensity impact, which simulates a no‐Medicaid counterfactual incorporating changes to health insurance and medical out‐of‐pocket spending, using the Supplemental Poverty Measure (SPM). This measure does not reflect a need for health care access nor how health benefits meet that need. (2) An accounting impact, which assumes that those losing Medicaid remain uninsured and does not incorporate any behavioral changes, using the health‐inclusive poverty measure (HIPM). This measure includes a need for health insurance in the threshold and health insurance benefits in resources. DATA COLLECTION/EXTRACTION METHODS: Not applicable. PRINCIPAL FINDINGS: Using the propensity‐matched approach, we attributed a 2.5 percentage point reduction in health‐inclusive poverty among those younger than age 65 to the Medicaid program, between the 1.0‐point SPM propensity‐match impact and the 3.9‐point HIPM accounting impact. Medicaid's antipoverty impact and HIPM‐SPM differences are greater among those who would become uninsured. HIPM propensity‐matched estimates reveal much larger impacts of Medicaid on poverty disparities linked to race/ethnicity and single parenthood than SPM‐based propensity estimates. CONCLUSIONS: Both the poverty measure and the method used to estimate the counterfactual make substantial, policy‐relevant differences to estimates of Medicaid's impact on poverty. A poverty measure that fails to incorporate health insurance needs and benefits substantially underestimates Medicaid's effect. Failing to consider adjustments in insurance coverage and out‐of‐pocket spending substantially overestimates Medicaid's effect and underestimates its reduction of disparities.

Recommendations are prevalent in Web applications (e.g., search ranking, item recommendation, advertisement placement). Learning from bandit feedback is challenging due to the sparsity of feedback limited to system-provided actions. In this work, we focus on batch learning from logs of recommender systems involving both bandit and organic feedbacks. We develop a probabilistic framework with a likelihood function for estimating not only explicit positive observations but also implicit negative observations inferred from the data. Moreover, we introduce a latent variable model for organic-bandit feedbacks to robustly capture user preference distributions. Next, we analyze the behavior of the new likelihood under two scenarios, i.e., with and without counterfactual re-weighting. For speedier item ranking, we further investigate the possibility of using Maximum-a-Posteriori (MAP) estimate instead of Monte Carlo (MC)-based approximation for prediction. Experiments on both real datasets as well as data from a simulation environment show substantial performance improvements over comparable baselines.

Until its highly publicised downfall, Theranos was a so-called unicorn; with a compelling proposition to popularise routine blood testing by making it more convenient, using a smaller sample of blood, and testing at a lower cost than conventional, often fear-inducing needles. Holmes and Theranos were reliant on their development of new technology to bring their idea to life, however they instead became perhaps one of the most infamous recent example of failed product innovation.In this paper we contend that, although character failings and alleged criminal activity could not have been realistically mitigated, valuable strategic and operational lessons can be learned for future technology start-ups. By conducting a counterfactual thought experiment and examining Theranos through the lens of Design Innovation (DI) we provide evaluations and predictions across strategic and operational constructs from both an internal and external perspective. A DI approach also applies methods and tools used by designers across a company as a whole – this mirrors an inventor-founder’s path from narrow product influence to broader environmental influences. We show that criticisms through the DI lens have specific and insightful bearing on Theranos because of its markedly poor strategic planning; manifest as a reliance on ideas and spin in place of tangible products and services. We use Theranos to demonstrate the value of DI to future technology-driven entrepreneurs, proposing alternative approaches to medical device start-ups.

This paper presents a legal-economic analysis of the World Trade Organization's Article 22.6 arbitration report on the dispute over certain United States’ antidumping methodologies. The Arbitrator sought to quantify the damages suffered by China due to US non-compliance with an earlier ruling. The case covered 25 antidumping duty determinations for which at least one of three methodologies (weighted average-to-transaction; single rate presumption; and zeroing) was incorrectly applied. Damage calculations rely heavily on how the counterfactual is defined – what would have been the duty had it not been for the inconsistent measures? The Arbitrator deemed a zero-duty counterfactual to be appropriate, but the justifications were in our view weak and illustrate the danger of an Arbitrator essentially performing re-litigation of violations that may or may not have occurred in the administrative investigations. We conclude that the Arbitrator may have gone above and beyond its mandate in this determination.

The literature has typically found that the distribution of socioeconomic factors like education, labor status and income does not account for the remarkable wealth inequality disparities between countries. As a result, their different institutions and other latent factors receive all the credit. Here, we propose to focus on one type of wealth inequality, the inequality of opportunities (IOp) in wealth: the share of overall wealth inequality explained by circumstances like inheritances and parental education. By means of a counterfactual decomposition method, we find that imposing the distribution of socioeconomic factors of the USA into Spain has little effect on total, financial and real estate wealth inequality. On the contrary, these factors play an important role when wealth IOp is considered. A Shapley value decomposition shows that the distribution of education and labor status in the USA consistently increase wealth IOp when imposed into Spain, whereas the opposite effect is found for the distribution of income.

University decisions can have lasting effects on students in the labor market; however, little is known about how these decisions are made. This paper develops a new framework for empirically analyzing course offerings at a sample university. The framework is based on the idea that course offerings directly affect student utilities and the probabilities that students choose courses in a given field. As such, administrators deciding which courses to offer are always implicitly trading off the number of students choosing courses in each field and total student utility. By measuring the marginal effects of offering additional courses in each field on field enrollments and student utility, one can quantify these implicit tradeoffs between student utility and field enrollments. In my empirical application, I find that a marginal dollar of spending on social science courses produces 2.5 times as much student utility as a marginal dollar of spending on business or occupational courses at a sample university. From this, I conclude the university is implicitly sacrificing student utility to draw students out of social science courses and into business or occupational courses. If this is intentional, then the university has a preference for business and occupational enrollment which may affect how their course offerings respond to changes in policies or student composition. Counterfactual analyses show that ignoring these responses can lead to understating the effects of changes in student composition on field enrollments by a factor of three.

In this paper, I show that a common way that platforms display firms' quality ratings incentivize firms to strategically take costly short-run actions that improve their ratings. Most review platforms display star ratings of goods and services rounded to a half star, rather than display the exact average rating. Since the true average rating is not shown, firms have an incentive to remain just above the rounding threshold in order to have a higher displayed rating. However, once a firm's rating passes the rounding threshold, the incentive to improve the ratings drops as their rating moves farther from the threshold. I study this phenomena in the context of auto repair. I find that there is an excessive amount of bunching around ratings thresholds. The firms' actions toward improving their ratings are typically unobserved, but due to my novel data and the discontinuity of displayed ratings, I can model and infer firm behavior. Specifically, I provide evidence that firms change the services they provide and exert extra effort when they are close to rounding thresholds. Finally, I provide a theoretical framework in order to quantify the actions and provide optimal policies for firm actions depending on their rating and number of reviews in a variety of counterfactual settings.