arXiv subjects

Naoki Egami

Publications and source records attributed to Naoki Egami.

10 recordsLinked to original sources

arXiv2026

Debiased Inference for AI-Generated Data without Gold-Standard Labels: Identification via Multiple Imperfect Measurements

An increasing number of scholars use AI to measure variables they subsequently include in downstream analyses. Although AI-measured variables are often analyzed as if observed without error, ignoring prediction errors in automated measurement leads to substantial bias and invalid confidence intervals in downstream analyses, even if AI measurement accuracy is high, e.g., above 90%. Existing solutions, such as design-based supervised learning and prediction-powered inference, combine error-prone AI-based measurements with gold-standard labels, which may be costly and difficult to obtain in some application areas. In this paper, we propose debiased inference with multiple imperfect measurements (DMM), a framework that combines multiple error-prone AI measurements to enable valid downstream inference without gold-standard labels. Building on the established results on CP decomposition, DMM assumes that these measurements are independent conditional on the latent true label and observed unit-level features, such as text features represented by embeddings. This framework allows for unknown misclassification rates to vary across annotation methods (e.g., large language models) and across units of annotation (e.g., texts). Under this assumption, we use semiparametric inference theory to prove that the DMM estimator is consistent and asymptotically normal, enabling valid inference for a wide range of downstream statistical analyses common in the social sciences. Our simulation results show that DMM yields valid inference and that adding accurate, though imperfect, measurements can improve efficiency. Focusing on common applications of large language model annotations, we also develop diagnostics to assess the conditional independence assumption.

stat.ME↗

arXiv2024

Beyond Reweighting: On the Predictive Role of Covariate Shift in Effect Generalization

Many existing approaches to generalizing statistical inference amidst distribution shift operate under the covariate shift assumption, which posits that the conditional distribution of unobserved variables given observable ones is invariant across populations. However, recent empirical investigations have demonstrated that adjusting for shift in observed variables (covariate shift) is often insufficient for generalization. In other words, covariate shift does not typically ``explain away'' the distribution shift between settings. As such, addressing the unknown yet non-negligible shift in the unobserved variables given observed ones (conditional shift) is crucial for generalizable inference. In this paper, we present a series of empirical evidence from two large-scale multi-site replication studies to support a new role of covariate shift in ``predicting'' the strength of the unknown conditional shift. Analyzing 680 studies across 65 sites, we find that even though the conditional shift is non-negligible, its strength can often be bounded by that of the observable covariate shift. However, this pattern only emerges when the two sources of shifts are quantified by our proposed standardized, ``pivotal'' measures. We then interpret this phenomenon by connecting it to similar patterns that can be theoretically derived from a random distribution shift model. Finally, we demonstrate that exploiting the predictive role of covariate shift leads to reliable and efficient uncertainty quantification for target estimates in generalization tasks with partially observed data. Overall, our empirical and theoretical analyses suggest a new way to approach the problem of distributional shift, generalizability, and external validity.

stat.AP↗

arXiv2023

Using Imperfect Surrogates for Downstream Inference: Design-based Supervised Learning for Social Science Applications of Large Language Models

In computational social science (CSS), researchers analyze documents to explain social and political phenomena. In most scenarios, CSS researchers first obtain labels for documents and then explain labels using interpretable regression analyses in the second step. One increasingly common way to annotate documents cheaply at scale is through large language models (LLMs). However, like other scalable ways of producing annotations, such surrogate labels are often imperfect and biased. We present a new algorithm for using imperfect annotation surrogates for downstream statistical analyses while guaranteeing statistical properties -- like asymptotic unbiasedness and proper uncertainty quantification -- which are fundamental to CSS research. We show that direct use of surrogate labels in downstream statistical analyses leads to substantial bias and invalid confidence intervals, even with high surrogate accuracy of 80-90%. To address this, we build on debiased machine learning to propose the design-based supervised learning (DSL) estimator. DSL employs a doubly-robust procedure to combine surrogate labels with a smaller number of high-quality, gold-standard labels. Our approach guarantees valid inference for downstream statistical analyses, even when surrogates are arbitrarily biased and without requiring stringent assumptions, by controlling the probability of sampling documents for gold-standard labeling. Both our theoretical analysis and experimental results show that DSL provides valid statistical inference while achieving root mean squared errors comparable to existing alternatives that focus only on prediction without inferential guarantees.

stat.ME↗

arXiv2022

Using Multiple Pre-treatment Periods to Improve Difference-in-Differences and Staggered Adoption Design

While a difference-in-differences (DID) design was originally developed with one pre- and one post-treatment period, data from additional pre-treatment periods are often available. How can researchers improve the DID design with such multiple pre-treatment periods under what conditions? We first use potential outcomes to clarify three benefits of multiple pre-treatment periods: (1) assessing the parallel trends assumption, (2) improving estimation accuracy, and (3) allowing for a more flexible parallel trends assumption. We then propose a new estimator, double DID, which combines all the benefits through the generalized method of moments and contains the two-way fixed effects regression as a special case. We show that the double DID requires a weaker assumption about outcome trends and is more efficient than existing DID estimators. We also generalize the double DID to the staggered adoption design where different units can receive the treatment in different time periods. We illustrate the proposed method with two empirical applications, covering both the basic DID and staggered adoption designs. We offer an open-source R package that implements the proposed methodologies.

stat.AP↗

arXiv2021

Leveraging Population Outcomes to Improve the Generalization of Experimental Results

Generalizing causal estimates in randomized experiments to a broader target population is essential for guiding decisions by policymakers and practitioners in the social and biomedical sciences. While recent papers developed various weighting estimators for the population average treatment effect (PATE), many of these methods result in large variance because the experimental sample often differs substantially from the target population, and estimated sampling weights are extreme. To improve efficiency in practice, we propose post-residualized weighting in which we use the outcome measured in the observational population data to build a flexible predictive model (e.g., machine learning methods) and residualize the outcome in the experimental data before using conventional weighting methods. We show that the proposed PATE estimator is consistent under the same assumptions required for existing weighting methods, importantly without assuming the correct specification of the predictive model. We demonstrate the efficiency gains from this approach through simulations and our application based on a set of job training experiments.

stat.ME↗

arXiv2021

Identification and Estimation of Causal Peer Effects Using Double Negative Controls for Unmeasured Network Confounding

Scientists have been interested in estimating causal peer effects to understand how people's behaviors are affected by their network peers. However, it is well known that identification and estimation of causal peer effects are challenging in observational studies for two reasons. The first is the identification challenge due to unmeasured network confounding, for example, homophily bias and contextual confounding. The second issue is network dependence of observations, which one must take into account for valid statistical inference. Negative control variables, also known as placebo variables, have been widely used in observational studies including peer effect analysis over networks, although they have been used primarily for bias detection. In this article, we establish a formal framework which leverages a pair of negative control outcome and exposure variables (double negative controls) to nonparametrically identify causal peer effects in the presence of unmeasured network confounding. We then propose a generalized method of moments estimator for causal peer effects, and establish its consistency and asymptotic normality under an assumption about $ψ$-network dependence. Finally, we provide a network heteroskedasticity and autocorrelation consistent variance estimator. Our methods are illustrated with an application to peer effects in education.

stat.ME↗

arXiv2021

Covariate Selection for Generalizing Experimental Results: Application to Large-Scale Development Program in Uganda

Generalizing estimates of causal effects from an experiment to a target population is of interest to scientists. However, researchers are usually constrained by available covariate information. Analysts can often collect much fewer variables from population samples than from experimental samples, which has limited applicability of existing approaches that assume rich covariate data from both experimental and population samples. In this article, we examine how to select covariates necessary for generalizing experimental results under such data constraints. In our concrete context of a large-scale development program in Uganda, although more than 40 pre-treatment covariates are available in the experiment, only 8 of them were also measured in a target population. We propose a method to estimate a separating set -- a set of variables affecting both the sampling mechanism and treatment effect heterogeneity -- and show that the population average treatment effect (PATE) can be identified by adjusting for estimated separating sets. Our algorithm only requires a rich set of covariates in the experimental data, not in the target population, by incorporating researcher-specific constraints on what variables are measured in the population data. Analyzing the development experiment in Uganda, we show that the proposed algorithm can allow for the PATE estimation in situations where conventional methods fail due to data requirements.

stat.ME↗

arXiv2020

Spillover Effects in the Presence of Unobserved Networks

When experimental subjects can interact with each other, the outcome of one individual may be affected by the treatment status of others. In many social science experiments, such spillover effects may occur through multiple networks, for example, through both online and offline face-to-face networks in a Twitter experiment. Thus, to understand how people use different networks, it is essential to estimate the spillover effect in each specific network separately. However, the unbiased estimation of these network-specific spillover effects requires an often-violated assumption that researchers observe all relevant networks. We show that, unlike conventional omitted variable bias, bias due to unobserved networks remains even when treatment assignment is randomized and when unobserved networks and a network of interest are independently generated. We then develop parametric and nonparametric sensitivity analysis methods, with which researchers can assess the potential influence of unobserved networks on causal findings. We illustrate the proposed methods with a simulation study based on a real-world Twitter network and an empirical application based on a network field experiment in China.

stat.ME↗

arXiv2020

Identification of Causal Diffusion Effects Under Structural Stationarity

Although social and biomedical scientists have long been interested in the process through which ideas and behaviors diffuse, the identification of causal diffusion effects, also known as peer and contagion effects, remains challenging. Many scholars consider the commonly used assumption of no omitted confounders to be untenable due to contextual confounding and homophily bias. To address this long-standing problem, we examine the causal identification under a new assumption of structural stationarity, which formalizes the underlying diffusion process with a class of dynamic causal directed acyclic graphs. First, we develop a statistical test that can detect a wide range of biases, including the two types mentioned above. We then propose a difference-in-differences style estimator that can directly correct biases under an additional parametric assumption. Leveraging the proposed methods, we study the spatial diffusion of hate crimes against refugees in Germany. After correcting large upward bias in existing studies, we find hate crimes diffuse only to areas that have a high proportion of school dropouts.

stat.ME↗

arXiv2018

How to Make Causal Inferences Using Texts

New text as data techniques offer a great promise: the ability to inductively discover measures that are useful for testing social science theories of interest from large collections of text. We introduce a conceptual framework for making causal inferences with discovered measures as a treatment or outcome. Our framework enables researchers to discover high-dimensional textual interventions and estimate the ways that observed treatments affect text-based outcomes. We argue that nearly all text-based causal inferences depend upon a latent representation of the text and we provide a framework to learn the latent representation. But estimating this latent representation, we show, creates new risks: we may introduce an identification problem or overfit. To address these risks we describe a split-sample framework and apply it to estimate causal effects from an experiment on immigration attitudes and a study on bureaucratic response. Our work provides a rigorous foundation for text-based causal inferences.

stat.ML↗