Researchers investigating causal mechanisms in survey experiments often rely on nonrandomized quantities to isolate the indirect effect of treatment through these variables. Such an approach, however, requires a “selection-on-observables” assumption, which undermines the advantages of a randomized experiment. In this paper, we show what can be learned about casual mechanisms through experimental design alone. We propose a factorial design that provides or withholds information on mediating variables and allows for the identification of the overall average treatment effect and the controlled direct effect of treatment fixing a potential mediator. While this design cannot identify indirect effects on its own, it avoids making the selection-on-observable assumption of the standard mediation approach while providing evidence for a broader understanding of causal mechanisms that encompasses both indirect effects and interactions. We illustrate these approaches via two examples: one on evaluations of US Supreme Court nominees and the other on perceptions of the democratic peace.

This article examines how the presentation of information during a laboratory experiment can alter a study’s findings. We compare four possible ways to present information about hypothetical candidates in a laboratory experiment. First, we manipulate whether subjects experience a low-information or a high-information campaign. Second, we manipulate whether the information is presented statically or dynamically. We find that the design of a study can produce very different conclusions. Using candidate’s gender as our manipulation, we find significant effects on a variety of candidate evaluation measures in low-information conditions, but almost no significant effects in high-information conditions. We also find that subjects in high-information settings tend to seek out more information in dynamic environments than static, though their ultimate candidate evaluations do not differ. Implications and recommendations for future avenues of study are discussed.

Survey experiments often manipulate the description of attributes in a hypothetical scenario, with the goal of learning about those attributes’ real-world effects. Such inferences rely on an underappreciated assumption: experimental conditions must be information equivalent (IE) with respect to background features of the scenario. IE is often violated because subjects, when presented with information about one attribute, update their beliefs about others too. Labeling a country “a democracy,” for example, affects subjects’ beliefs about the country’s geographic location. When IE is violated, the effect of the manipulation need not correspond to the quantity of interest (the effect of beliefs about the focal attribute). We formally define the IE assumption, relating it to the exclusion restriction in instrumental-variable analysis. We show how to predict IE violations ex ante and diagnose them ex post with placebo tests. We evaluate three strategies for achieving IE. Abstract encouragement is ineffective. Specifying background details reduces imbalance on the specified details and highly correlated details, but not others. Embedding a natural experiment in the scenario can reduce imbalance on all background beliefs, but raises other issues. We illustrate with four survey experiments, focusing on an extension of a prominent study of the democratic peace.

Automated text analysis allows researchers to analyze large quantities of text. Yet, comparative researchers are presented with a big challenge: across countries people speak different languages. To address this issue, some analysts have suggested using Google Translate to convert all texts into English before starting the analysis (Lucas et al. 2015). But in doing so, do we get lost in translation? This paper evaluates the usefulness of machine translation for bag-of-words models—such as topic models. We use the europarl dataset and compare term-document matrices (TDMs) as well as topic model results from gold standard translated text and machine-translated text. We evaluate results at both the document and the corpus level. We first find TDMs for both text corpora to be highly similar, with minor differences across languages. What is more, we find considerable overlap in the set of features generated from human-translated and machine-translated texts. With regard to LDA topic models, we find topical prevalence and topical content to be highly similar with again only small differences across languages. We conclude that Google Translate is a useful tool for comparative researchers when using bag-of-words text models.

Data sets quantifying phenomena of social-scientific interest often use multiple experts to code latent concepts. While it remains standard practice to report the average score across experts, experts likely vary in both their expertise and their interpretation of question scales. As a result, the mean may be an inaccurate statistic. Item-response theory (IRT) models provide an intuitive method for taking these forms of expert disagreement into account when aggregating ordinal ratings produced by experts, but they have rarely been applied to cross-national expert-coded panel data. We investigate the utility of IRT models for aggregating expert-coded data by comparing the performance of various IRT models to the standard practice of reporting average expert codes, using both data from the V-Dem data set and ecologically motivated simulated data. We find that IRT approaches outperform simple averages when experts vary in reliability and exhibit differential item functioning (DIF). IRT models are also generally robust even in the absence of simulated DIF or varying expert reliability. Our findings suggest that producers of cross-national data sets should adopt IRT techniques to aggregate expert-coded data measuring latent concepts.

Words matter in politics. The rhetoric that political elites employ structures civic discourse. The emergence of social media platforms as a medium of politics has enabled ordinary citizens to express their ideological inclinations by adopting the lexicon of political elites. This avails to researchers a rich new source of data in the study of political ideology. However, existing ideological text-scaling methods fail to produce meaningful inferences when applied to the short, informal style of textual content that is characteristic of social media platforms such as Twitter. This paper introduces the first viable approach to the estimation of individual-level ideological positions derived from social media content. This method allows us to position social media users—be they political elites, parties, or citizens—along a shared ideological dimension. We validate the proposed method by demonstrating correlation with existing measures of ideology across various political contexts and multiple languages. We further demonstrate the ability of ideological estimates to capture derivative signal by predicting out-of-sample, individual-level voting intentions. We posit that social media data can, when properly modeled, better capture derivative signal than discrete scales used in more traditional survey instruments.

Instruments based on realizations of the endogenous variable in other units—for instance, regional or global weighted averages—are commonly used in political science. Such spatial instruments have proved attractive: they are convenient to obtain, typically have power, and are plausibly exogenous. We argue that the assumptions underlying spatial instruments remain poorly understood and challenge whether spatial instruments can satisfy the conditions required for valid instruments. First, when cross-unit dependence exists in the endogenous predictor, other cross-unit relationships—spillovers and interdependence—likely exist as well and risk violations of the exclusion restriction. Second, spatial instruments produce simultaneity in the first-stage equation, as left-hand side outcomes are included as right-hand side predictors. Because the instrument and the endogenous variable are simultaneously determined, the exclusion restriction is, necessarily and by construction, violated. Taken together, these concerns lead us to conclude that spatial instruments are rarely, if ever, valid.

This letter compares the performance of multiple imputation and listwise deletion using a simulation approach. The focus is on data that are “missing not at random” (MNAR), in which case both multiple imputation and listwise deletion are known to be biased. In these simulations, multiple imputation yields results that are frequently more biased, less efficient, and with worse coverage than listwise deletion when data are MNAR. This is the case even with very strong correlations between fully observed variables and variables with missing values, such that the data are very nearly “missing at random.” These results recommend caution when comparing the results from multiple imputation and listwise deletion, when the true data generating process is unknown.