Donald B. Rubin

Across the globe, many youths experience a problematic education-to-work transition, particularly episodes of being not in employment, education or training (NEET). Even though youth unemployment is low in East Asia in worldwide comparison, youth joblessness still represents an urgent topic of social inequality as youths are more affected than prime-aged workers (ILO, 2017). This study focuses on Taiwan, which experienced an increase in youth unemployment over the last decades (Craig et al., 2019; Huang, 2013). After long periods of economic growth and a successful transition to a service economy, Taiwan suffered from economic turmoil in the 1990s. Despite a quick recovery after the 1997 financial crisis there was an ...

Causal Inference for Statistics, Social, and Biomedical Sciences - by Guido W. Imbens April 2015

Causal Inference for Statistics, Social, and Biomedical Sciences - by Guido W. Imbens April 2015

Investigators in observational studies have no control over treatment assignment. As a result, large differences can exist between the treatment and control groups on observed covariates, which can lead to badly biased estimates of treatment effects. Propensity score methods are an increasingly popular method for balancing the distribution of the covariates in the two groups to reduce this bias; for example, using matching or subclassification, sometimes in combination with model-based adjustment. To estimate propensity scores, which are the conditional probabilities of being treated given a vector of observed covariates, we must model the distribution of the treatment indicator given these observed covariates. Much work has been done in the case where covariates are fully observed. We address the problem of calculating propensity scores when covariates can have missing values. In such cases, which commonly arise in practice, the pattern of missing covariates can be prognostically important, and then propensity scores should condition both on observed values of covariates and on the observed missing-data indicators. Using the resulting generalized propensity scores to adjust for the observed background differences between treatment and control groups leads, in expectation, to balanced distributions of observed covariates in the treatment and control groups, as well as balanced distributions of patterns of missing data. The methods are illustrated using the generalized propensity scores to create matched samples in a study of the effects of postterm pregnancy.

The problem of equating a new standardized test to an old reference test is considered when the samples for equating are not randomly selected from the target population of test takers. Two problems with equating from biased samples are distinguished: (a) bias in the equating function arising from nonrandom selection of the equating sample, and (b) excessive variance in the equating function at scores that are relatively underrepresented in the equating sample relative to the target population. A theorem is presented that suggests that bias may not be a major problem for equating, even when the marginal distributions of scores are distorted by selection. Empirical analysis of data for equating the Armed Services Vocational Aptitude Battery (ASVAB) based on samples of recruits and applicants supports this contention. Analysis of ASVAB data also indicates that excessive variance in the equating function is a more serious issue. Variance-reducing methods, which smooth the test score distributions using extended beta binomial and loglinear polynomial models before equating by the equipercentile method, are presented. Empirical evidence suggests that these smoothing models are successful and yield equating functions that improve on both equipercentile and linear equating of the raw scores.

We present a procedure for computing significance levels from data sets whose missing values have been multiply imputed data. This procedure uses moment-based statistics, m ≤ 3 repeated imputations, and an F reference distribution. When m = ∞, we show first that our procedure is essentially the same as the ideal procedure in cases of practical importance and, second, that its deviations from the ideal are basically a function of the coefficient of variation of the canonical ratios of complete to observed information. For small m our procedure's performance is largely governed by this coefficient of variation and the mean of these ratios. Using simulation techniques with small m, we compare our procedure's actual and nominal large-sample significance levels and conclude that it is essentially calibrated and thus represents a definite improvement over previously available procedures. Furthermore, we compare the large-sample power of the procedure as a function of m and other factors, such as the dimensionality of the estimand and fraction of missing information, to provide guidance on the choice of the number of imputations; generally, we find the loss of power due to small m to be quite modest in cases likely to occur in practice.