Semiparametric bounds on treatment effects

We present a variety of semiparametric models that produce bounds on the average causal effect of a binary treatment on a binary outcome. The semiparametric assumptions exploit variation in observable covariates to narrow the bounds. In our main model, the outcome is determined by a generalized linear model, but the treatment may be arbitrarily endogenous. Our bounding strategy does not require the existence of an instrument, but incorporating an instrument narrows the bounds. The bounds are further improved by combining the semiparametric model with the joint threshold-crossing assumption of Shaikh and Vytlacil (2005).     

IDENTIFYING CAUSAL MECHANISMS (PRIMARILY) BASED ON INVERSE PROBABILITY WEIGHTING

This paper demonstrates the identification of causal mechanisms of a binary treatment under selection on observables, (primarily) based on inverse probability weighting; i.e. we consider the average indirect effect of the treatment, which operates through an intermediate variable (or mediator) that is situated on the causal path between the treatment and the outcome, as well as the (unmediated) direct effect. Even under random treatment assignment, subsequent selection into the mediator is generally non‐random such that causal mechanisms are only identified when controlling for confounders of the mediator and the outcome. To tackle this issue, units are weighted by the inverse of their conditional treatment propensity given the mediator and observed confounders. We show that the form and applicability of weighting depend on whether some confounders are themselves influenced by the treatment or not. A simulation study gives the intuition for these results and an empirical application to the direct and indirect health effects (through employment) of the US Job Corps program is also provided. Copyright © 2013 John Wiley & Sons, Ltd.     

A critique on the Corruption Perceptions Index: An interdisciplinary approach

The purpose of this study is to test the assumptions of the Corruption Perceptions Index (CPI) and examine the causal relationships among its data sources for improvement priorities. In doing so, we use a novel interdisciplinary methodology including cluster analysis, classification analysis, partial least squares structural equation modeling and importance-performance map analysis. Our methodology enables policymakers to identify the critical data sources that a given country should focus on in order to improve its position in the CPI ranking relative to other countries. Based on corruption perceptions of 176 countries in the 2016 CPI, our results provide evidence against the CPI's assumptions, as individual data sources have unequal effects on the CPI and exhibit the causal interrelations among one another. Corruption perceptions are not homogeneous across countries, with developed countries showing lower levels of perceived corruption than emerging countries. The presence of synergistic effects among the CPI's data sources suggests that national policymakers consider multiple data sources of the CPI for decision-making process rather than simply focus on any single one of these data sources or their equally-weighted aggregation. Moreover, policymakers should allocate the country's resources – which are often limited – with the first priority to improving the data source score of the Economist Intelligence Unit Country Risk Ratings, the critical driver of the CPI. Interestingly, the modified CPI which removes insignificant data sources outperforms the non-modified CPI in terms of the goodness-of-fit assessment, the unbiasedness and the association with the World Bank's Control of Corruption.     

Exogeneity in structural equation models

The practical relevance of several concepts of exogeneity of treatments for the estimation of causal parameters based on observational data are discussed. We show that the traditional concepts, such as strong ignorability and weak and super-exogeneity, are too restrictive if interest lies in average effects (i.e. not on distributional effects of the treatment). We suggest a new definition of exogeneity, KL-exogeneity. It does not rely on distributional assumptions and is not based on counterfactual random variables. As a consequence it can be empirically tested using a proposed test that is simple to implement and is distribution-free.     

On the existence of maximum likelihood estimates in random effects models for clustered multivariate binary data

We consider a class of random effects models for clustered multivariate binary data based on the threshold crossing technique of a latent random vector. Components of this latent vector are assumed to have a Laird–Ware structure. However, in place of their Gaussian assumptions, any specified class of multivariate distribution is allowed for the random effects, and the error vector is allowed to have any strictly positive pdf. A well known member of this class of models is the multivariate probit model with random effects. We investigate sufficient and necessary conditions for the existence of maximum likelihood estimates for the location and the association parameters. Implications of our results are illustrated through some hypothetical examples.     

Distributional Effects of a Continuous Treatment with an Application on Intergenerational Mobility

This paper considers the effect of a continuous treatment on the entire distribution of outcomes after adjusting for differences in the distribution of covariates across different levels of the treatment. Our methodology encompasses dose-response functions, counterfactual distributions, and ‘distributional policy effects’ depending on the assumptions invoked by the researcher. We propose a three-step estimator that consists of (i) estimating the distribution of the outcome conditional on the treatment and other covariates using quantile regression; (ii) for each value of the treatment, averaging over a counterfactual distribution of the covariates holding the treatment fixed; (iii) converting the resulting counterfactual distribution into parameters of interest that are easy to interpret. We show that our estimators converge uniformly to Gaussian processes and that the empirical bootstrap can be used to conduct uniformly valid inference across a range of values of the treatment. We use our method to study intergenerational income mobility where we consider effects of parents’ income on features of their child's income distribution such as (i) the fraction of children with income below the poverty line; (ii) the variance of child's income; and (iii) the inter-quantile range of child's income–all as a function of parents’ income.     

Focus: The Effects on the Forecast of Oil Prices and Economic Policy

Focus: The Effects on the Forecast of Oil Prices and Economic Policy
The central forecast depends on the underlying assumptions which, because of developments in the oil market, are particularly uncertain. In order to show how the forecast might change if our assumptions on oil prices and monetary and fiscal policy were to prove incorrect, we have analysed the effect on our October 1985 forecast of changes to the key assumptions. In addition to the effects of these changes, the simulations show that our forecast for GDP has been revised up.     

A PLS path model to investigate the relations between institutions and human development

The paper studies the relations between institutions and human development, in particular the causal effects of the different types of institutions on different components of human development. We assume development to be created by aggregate demand; in particular that aggregate demand determines the material components of human development. We thus divide institutions into those that create demand and those that are determined by the whole process of development. Similarly we divide human development in its three traditional components (economic development, health, knowledge). Both human development and institutions are assumed to be multidimensional constructs; all the main components of these constructs are defined as latent variables, and the relations between them as structural relations. A partial least squares (PLS) path model is developed: it is the aggregation (and simultaneous estimation) of an outer model relating observed or manifest variables to their own latent variable and of a structural model (inner model) relating some endogenous latent variable to other latent variables. From the goodness of fit point of view, our results seem to validate our theoretical assumptions.     

Selection correction and sensitivity analysis for ordered treatment effect on count response

In estimating the effect of an ordered treatment τ on a count response y with an observational data where τ is self‐selected (not randomized), observed variables x and unobserved variables ε can be unbalanced across the control group (τ = 0) and the treatment groups (τ = 1, …, J). While the imbalance in x causes ‘overt bias’ which can be removed by controlling for x, the imbalance in ε causes ‘covert (hidden or selection) bias’ which cannot be easily removed. This paper makes three contributions. First, a proper counter‐factual causal framework for ordered treatment effect on count response is set up. Second, with no plausible instrument available for τ, a selection correction approach is proposed for the hidden bias. Third, a nonparametric sensitivity analysis is proposed where the treatment effect is nonparametrically estimated under no hidden bias first, and then a sensitivity analysis is conducted to see how sensitive the nonparametric estimate is to the assumption of no hidden bias. The analytic framework is applied to data from the Health and Retirement Study: the treatment is ordered exercise levels in five categories and the response is doctor office visits per year. The selection correction approach yields very large effects, which are however ruled out by the nonparametric sensitivity analysis. This finding suggests a good deal of caution in using selection correction approaches. Copyright © 2004 John Wiley & Sons, Ltd.     

Panel data analysis of U.S. coal productivity

We analyze labor productivity in coal mining in the United States using indices of productivity change associated with the concepts of panel data modeling. This approach is valuable when there is extensive heterogeneity in production units, as with coal mines. We find substantial returns to scale for coal mining in all geographical regions, and find that smooth technical progress is exhibited by estimates of the fixed effects for coal mining. We carry out a variety of diagnostic analyses of our basic model and primary modeling assumptions, using recently proposed methods for addressing ‘errors-in-variables’ and ‘weak instrument bias’ problems in linear and nonlinear models.     

Treatment effect bounds: An application to Swan–Ganz catheterization

We reanalyze data from the observational study by Connors et al. (1996) on the impact of Swan–Ganz catheterization on mortality outcomes. The study by Connors et al. (1996) assumes that there are no unobserved differences between patients who are catheterized and patients who are not catheterized and finds that catheterization increases patient mortality. We instead allow for such differences between patients by implementing both the instrumental variable bounds of Manski (1990), which only exploits an instrumental variable, and the bounds of Shaikh and Vytlacil (2011), which exploit mild nonparametric, structural assumptions in addition to an instrumental variable. We propose and justify the use of indicators of weekday admission as an instrument for catheterization in this context. We find that in our application, the Manski (1990) bounds do not indicate whether catheterization increases or decreases mortality, where as the Shaikh and Vytlacil (2011) bounds reveal that at least for some diagnoses, Swan–Ganz catheterization reduces mortality at 7 days after catheterization. We show that the bounds of Shaikh and Vytlacil (2011) remain valid under even weaker assumptions than those described in Shaikh and Vytlacil (2011). We also extend the analysis to exploit a further nonparametric, structural assumption–that doctors catheterize individuals with systematically worse latent health–and find that this assumption further narrows these bounds and strengthens our conclusions. In our analysis, we construct confidence regions using the methodology developed in Romano and Shaikh (2008). We show in particular that the confidence regions are uniformly consistent in level over a large class of possible distributions for the observed data that include distributions where the instrument is arbitrarily “weak”.     

Nonparametric bounds on the returns to language skills

This paper applies the theoretical literature on nonparametric bounds on treatment effects to the estimation of how limited English proficiency (LEP) affects wages and employment opportunities for Hispanic workers in the United States. I analyse the identifying power of several weak assumptions on treatment response and selection, and stress the interactions between LEP and education, occupation and immigration status. I show that the combination of two weak but credible assumptions provides informative upper bounds on the returns to language skills for certain subgroups of the population. Adding age at arrival as a monotone instrumental variable also provides informative lower bounds. Copyright © 2005 John Wiley & Sons, Ltd.     

Testing identifying assumptions in bivariate probit models

This paper considers the bivariate probit model's identifying assumptions: linear index specification, joint normality of errors, instrument exogeneity, and relevance. First, we develop sharp testable equalities that detect all possible observable violations of the assumptions. Second, we propose an easy-to-implement testing procedure for the model's validity using existing inference methods for intersection bounds. The test achieves correct empirical size and performs well in detecting violations of the conditions in simulations. Finally, we provide a road map on what to do when the bivariate probit model is rejected, including novel bounds for the average treatment effect that relax the normality assumption.     

Economics,Darwinism, and the Case of Disciplinary Imports

The problem of causality in economics is still contended by various epistemological alternatives. The article builds on the received view of Darwinism in economics and examines the way in which economics and biology find common ground in concepts and assumptions that reflect causal commonalities of the natural and the social world. We claim that the role the contingent pattern plays in understanding socioeconomic change provides reasons to concede corrections to a rule‐based causal mechanism. The article concludes on the merits of advancing the ontological equivalent of interdisciplinary studies as one possible standard in reference to which to judge the epistemic adequacy of any import.     

Enactment,Sensemaking and Decision Making: Redesign Processes in the 1976 Reorganization of US Intelligence

Weick’s theories of organizing and sensemaking help enrich the assumptions in the organization design school. This study builds on Weick’s theories of sensemaking to illustrate how three fundamental organization design assumptions — dominant variables, causal laws and executive dictates — were found to be restrictive in the explanation of redesign processes in the 1976 reorganization of the US intelligence community. The assumption of dominant variables was challenged by the appearance of a multitude of events, or enactments, which were selected by organization members for further attention. The assumption of causal laws was challenged by the appearance of individual‐level cause maps which were filtered, through sensemaking processes, into organization‐level workable realities. The assumption of executive dictates was challenged by the appearance of attempts to punctuate redesign processes as organizational decisions. The study suggests value in moving from simple organization design assumptions to more reliable findings drawn from detailed observations of redesign processes.     

The effect of household appliances on female labor force participation: Evidence from microdata

We estimate the effect of household appliance ownership on the labor force participation rate of married women using micro-level data from the 1960 and 1970 U.S. Censuses. In order to identify the causal effect of home appliance ownership on married women's labor force participation rates, our empirical strategy exploits both time-series and cross-sectional variation in these two variables. To control for endogeneity, we instrument a married woman's ownership of an appliance by the average ownership rate for that appliance among single women living in the same U.S. state. Single women's labor force participation rates did not increase between 1960 and 1970. We find evidence in support of the hypothesis that the diffusion of household appliances contributed to the increase in married women's labor force participation rates during the 1960's.     

Family income and children's education: Using the Norwegian oil boom as a natural experiment

Parental income is positively correlated with children's educational attainment. This paper addresses the causality of this observed link. We have a unique data set for Norwegians born in the period 1967–1969, with a measure of permanent family income in the children's adolescence. This enables us to examine the long-term effect of family income on children's educational attainment. The Norwegian oil shock in the 1970s is used as an instrument, because this – in some regions but not in others – implied a general increase in income unrelated to parents' abilities. This variation in income is used to estimate the causal effect of family income on children's educational attainment. We find no such causal relationship. This result is robust with respect to different specification tests.     

Interval Estimation of Treatment Effects In Randomized Trials: When do Confidence Intervals Have Nominal Coverage?

In the classical theory of randomized trials (RTs), the treatment effect is defined as a mean difference of potential outcomes. To achieve nominal coverage probability for confidence intervals (CIs) on treatment effects in RTs, certain assumptions are necessary. Specifically, one must either make assumptions about the joint distribution of potential outcomes or enroll subjects in the trial by random sampling of the target population on which the treatment effect is defined. In practice, no such sampling usually takes place and assumptions about the joint distribution of potential outcomes cannot be verified based on observed data. Furthermore, the most common of these assumptions, such as treatment‐unit additivity (TUA) or independence are biologically implausible in most RTs involving human subjects. Hence, it is not usually possible to construct CIs on treatment effects with nominal coverage probability. However, for any joint distribution of potential outcomes, the standard estimator of the variance of the difference of two independent sample means produces CIs with asymptotic coverage at least at the nominal level. This interpretation of CIs as conservative bounds may not always hold in conventional regression models applied to RT data.     

The impact of the National School Lunch Program on child health: A nonparametric bounds analysis

Children in households reporting the receipt of free or reduced-price school meals through the National School Lunch Program (NSLP) are more likely to have negative health outcomes than observationally similar nonparticipants. Assessing causal effects of the program is made difficult, however, by missing counterfactuals and systematic underreporting of program participation. Combining survey data with auxiliary administrative information on the size of the NSLP caseload, we extend nonparametric partial identification methods that account for endogenous selection and nonrandom classification error in a single framework. Similar to a regression discontinuity design, we introduce a new way to conceptualize the monotone instrumental variable (MIV) assumption using eligibility criteria as monotone instruments. Under relatively weak assumptions, we find evidence that the receipt of free and reduced-price lunches improves the health outcomes of children.     

Partial observability in bivariate probit models

This study investigates random utility models in which the observed binary outcome does not reflect the binary choice of a single decision-maker, but rather the joint unobserved binary choices of two decision-makers. Under the usual normality assumptions, the model that arises for the observed binary outcome is not a univariate probit model, but rather a bivariate probit model in which only one of the four possible outcomes is observed. Estimation and identification issues are discussed, and the implications for sample selectivity problems are noted.