Person:

Miratrix, Luke

In this paper we propose a general framework for topic-specific summarization of large text corpora and illustrate how it can be used for the analysis of news databases. Our framework, concise comparative summarization (CCS), is built on sparse classification methods. CCS is a lightweight and flexible tool that offers a compromise between simple word frequency based methods currently in wide use and more heavyweight, model-intensive methods such as latent Dirichlet allocation (LDA). We argue that sparse methods have much to offer for text analysis and hope CCS opens the door for a new branch of research in this important field.

For a particular topic of interest (e.g., China or energy), CSS automatically labels documents as being either on- or off-topic (usually via keyword search), and then uses sparse classification methods to predict these labels with the high-dimensional counts of all the other words and phrases in the documents. The resulting small set of phrases found as predictive are then harvested as the summary.

To validate our tool, we, using news articles from the New York Times international section, designed and conducted a human survey to compare the different summarizers with human understanding. We demonstrate our approach with two case studies, a media analysis of the framing of “Egypt” in the New York Times throughout the Arab Spring and an informal comparison of the New York Times’ and Wall Street Journal’s coverage of “energy.” Overall, we find that the Lasso with L2 normalization can be effectively and usefully used to summarize large corpora, regardless of document size.

Users make lasting judgments about a website's appeal within a split second of seeing it for the first time. This first impression is influential enough to later affect their opinions of a site's usability and trustworthiness. In this paper, we demonstrate a means to predict the initial impression of aesthetics based on perceptual models of a website's colorfulness and visual complexity. In an online study, we collected ratings of colorfulness, visual complexity, and visual appeal of a set of 450 websites from 548 volunteers. Based on these data, we developed computational models that accurately measure the perceived visual complexity and colorfulness of website screenshots. In combination with demographic variables such as a user's education level and age, these models explain approximately half of the variance in the ratings of aesthetic appeal given after viewing a website for 500ms only.

We consider the conditional randomization test as a way to account for covariate imbalance in randomized experiments. The test accounts for covariate imbalance by comparing the observed test statistic to the null distribution of the test statistic conditional on the observed covariate imbalance. We prove that the conditional randomization test has the correct significance level and introduce original notation to describe covariate balance more formally. Through simulation, we verify that conditional randomization tests behave like more traditional forms of covariate adjustmet but have the added benefit of having the correct conditional significance level. Finally, we apply the approach to a randomized product marketing experiment where covariate information was collected after randomization.

“M-Bias”, as it is called in the epidemiological literature, is the bias introduced by conditioning on a pretreatment covariate due to a particular “M-Structure” between two latent factors, an observed treatment, an outcome, and a “collider”. This potential source of bias, which can occur even when the treatment and the outcome are not confounded, has been a source of considerable controversy. We here present formulae for identifying under which circumstances biases are inflated or reduced. In particular, we show that the magnitude of M-Bias in Gaussian linear structural equation models tends to be relatively small compared to confounding bias, suggesting that it is generally not a serious concern in many applied settings. These theoretical results are consistent with recent empirical findings from simulation studies. We also generalize the M-Bias setting to allow for the correlation between the latent factors to be nonzero, and to allow for the collider to also be a confounder between the treatment and the outcome. These results demonstrate that mild deviations from the M-Structure tend to increase confounding bias more rapidly than M-bias, suggesting that choosing to condition on any given covariate is generally the superior choice. As an application, we re-examine a controversial example between Professors Donald Rubin and Judea Pearl.

In randomized experiments, randomization forms the “reasoned basis for inference.” While randomization inference is well developed for continuous and binary outcomes, there has been comparatively little work for outcomes with nonnegative support and clumping at zero. Typically outcomes of this type have been modeled using parametric models that impose strong distributional assumptions. This article proposes new randomization inference procedures for nonnegative outcomes with clumping at zero. Instead of making distributional assumptions, we propose various assumptions about the nature of response to treatment. Our methods form a set of nonparametric methods for outcomes that are often described as zero-inflated. These methods are illustrated using two randomized trials where job training interventions were designed to increase earnings of participants.

We propose a general framework for topic-specific summarization of large text corpora, and illustrate how it can be used for analysis in two quite different contexts: an OSHA database of fatality and catastrophe reports (to facilitate surveillance for patterns in circumstances leading to injury or death) and legal decisions on workers’ compensation claims (to explore relevant case law). Our summarization framework, built on sparse classification methods, is a compromise between simple word frequency based methods currently in wide use, and more heavyweight, model-intensive methods such as Latent Dirichlet Allocation (LDA). For a particular topic of interest (e.g., mental health disability, or carbon monoxide exposure), we regress a labeling of documents onto the high-dimensional counts of all the other words and phrases in the documents. The resulting small set of phrases found as predictive are then harvested as the summary. Using a branch-and-bound approach, this method can be extended to allow for phrases of arbitrary length, which allows for potentially rich summarization. We discuss how focus on the purpose of the summaries can inform choices of tuning parameters and model constraints. We evaluate this tool by comparing computational time and summary statistics of the resulting word lists to three other methods in the literature. We also present a new R package, textreg. Overall, we argue that sparse methods have much to offer text analysis, and is a branch of research that should be considered further in this context.

Two common concerns raised in analyses of randomized experiments are (i) appropriately handling issues of non-compliance, and (ii) appropriately adjusting for multiple tests (e.g., on multiple outcomes or subgroups). Although simple intention-to-treat (ITT) and Bonferroni methods are valid in terms of type I error, they can each lead to a substantial loss of power; when employing both simultaneously, the total loss may be severe. Alternatives exist to address each concern. Here we propose an analysis method for experiments involving both features that merges posterior predictive p-values for complier causal effects with randomization-based multiple comparisons adjustments; the results are valid familywise tests that are doubly advantageous: more powerful than both those based on standard ITT statistics and those using traditional multiple comparison adjustments. The operating characteristics and advantages of our method are demonstrated through a series of simulated experiments and an analysis of the United States Job Training Partnership Act (JTPA) Study, where our methods lead to different conclusions regarding the significance of estimated JTPA effects.

Understanding and characterizing treatment effect variation in randomized experiments has become essential for going beyond the "black box" of the average treatment effect. Nonetheless, traditional statistical approaches often ignore or assume away such variation. In the context of a randomized experiment, this paper proposes a framework for decomposing overall treatment effect variation into a systematic component that is explained by observed covariates, and a remaining idiosyncratic component. Our framework is fully randomization-based, with estimates of treatment effect variation that are fully justified by the randomization itself. Our framework can also account for noncompliance, which is an important practical complication. We make several key contributions. First, we show that randomization-based estimates of systematic variation are very similar in form to estimates from fully-interacted linear regression and two stage least squares. Second, we use these estimators to develop an omnibus test for systematic treatment effect variation, both with and without noncompliance. Third, we propose an $R^2$-like measure of treatment effect variation explained by covariates and, when applicable, noncompliance. Finally, we assess these methods via simulation studies and apply them to the Head Start Impact Study, a large-scale randomized experiment.

Experimenters often use post-stratification to adjust estimates. Post-stratification is akin to blocking, except that the number of treated units in each stratum is a random variable because stratification occurs after treatment assignment. We analyse both post-stratification and blocking under the Neyman–Rubin model and compare the efficiency of these designs. We derive the variances for a post-stratified estimator and a simple difference-in-means estimator under different randomization schemes. Post-stratification is nearly as efficient as blocking: the difference in their variances is of the order of 1/n2, with a constant depending on treatment proportion. Post-stratification is therefore a reasonable alternative to blocking when blocking is not feasible. However, in finite samples, post-stratification can increase variance if the number of strata is large and the strata are poorly chosen. To examine why the estimators’ variances are different, we extend our results by conditioning on the observed number of treated units in each stratum. Conditioning also provides more accurate variance estimates because it takes into account how close (or far) a realized random sample is from a comparable blocked experiment. We then show that the practical substance of our results remains under an infinite population sampling model. Finally, we provide an analysis of an actual experiment to illustrate our analytical results.

Risk-limiting post-election audits limit the chance of certifying an electoral outcome if the outcome is not what a full hand count would show. Building on previous work, we report on pilot risk-limiting audits in four elections during 2008 in three California counties: one during the February 2008 Primary Election in Marin County and three during the November 2008 General Elections in Marin, Santa Cruz and Yolo Counties. We explain what makes an audit risk-limiting and how existing and proposed laws fall short. We discuss the differences among our four pilot audits. We identify challenges to practical, efficient risk-limiting audits and conclude that current approaches are too complex to be used routinely on a large scale. One important logistical bottleneck is the difficulty of exporting data from commercial election management systems in a format amenable to audit calculations. Finally, we propose a bare-bones risk-limiting audit that is less efficient than these pilot audits, but avoids many practical problems.