arXiv subjects

Johan Ugander

Publications and source records attributed to Johan Ugander.

At least 37 records · Page 2Linked to original sources

arXiv2023

Counterfactual Evaluation of Peer-Review Assignment Policies

Peer review assignment algorithms aim to match research papers to suitable expert reviewers, working to maximize the quality of the resulting reviews. A key challenge in designing effective assignment policies is evaluating how changes to the assignment algorithm map to changes in review quality. In this work, we leverage recently proposed policies that introduce randomness in peer-review assignment--in order to mitigate fraud--as a valuable opportunity to evaluate counterfactual assignment policies. Specifically, we exploit how such randomized assignments provide a positive probability of observing the reviews of many assignment policies of interest. To address challenges in applying standard off-policy evaluation methods, such as violations of positivity, we introduce novel methods for partial identification based on monotonicity and Lipschitz smoothness assumptions for the mapping between reviewer-paper covariates and outcomes. We apply our methods to peer-review data from two computer science venues: the TPDP'21 workshop (95 papers and 35 reviewers) and the AAAI'22 conference (8,450 papers and 3,145 reviewers). We consider estimates of (i) the effect on review quality when changing weights in the assignment algorithm, e.g., weighting reviewers' bids vs. textual similarity (between the review's past papers and the submission), and (ii) the "cost of randomization", capturing the difference in expected quality between the perturbed and unperturbed optimal match. We find that placing higher weight on text similarity results in higher review quality and that introducing randomization in the reviewer-paper assignment only marginally reduces the review quality. Our methods for partial identification may be of independent interest, while our off-policy approach can likely find use evaluating a broad class of algorithmic matching systems.

cs.IR↗

arXiv2023

Model-Based Regression Adjustment with Model-Free Covariates for Network Interference

When estimating a Global Average Treatment Effect (GATE) under network interference, units can have widely different relationships to the treatment depending on a combination of the structure of their network neighborhood, the structure of the interference mechanism, and how the treatment was distributed in their neighborhood. In this work, we introduce a sequential procedure to generate and select graph- and treatment-based covariates for GATE estimation under regression adjustment. We show that it is possible to simultaneously achieve low bias and considerably reduce variance with such a procedure. To tackle inferential complications caused by our feature generation and selection process, we introduce a way to construct confidence intervals based on a block bootstrap. We illustrate that our selection procedure and subsequent estimator can achieve good performance in terms of root mean squared error in several semi-synthetic experiments with Bernoulli designs, comparing favorably to an oracle estimator that takes advantage of regression adjustments for the known underlying interference structure. We apply our method to a real world experimental dataset with strong evidence of interference and demonstrate that it can estimate the GATE reasonably well without knowing the interference process a priori.

stat.ME↗

arXiv2022

Estimating Geographic Spillover Effects of COVID-19 Policies From Large-Scale Mobility Networks

Many policies in the US are determined locally, e.g., at the county-level. Local policy regimes provide flexibility between regions, but may become less effective in the presence of geographic spillovers, where populations circumvent local restrictions by traveling to less restricted regions nearby. Due to the endogenous nature of policymaking, there have been few opportunities to reliably estimate causal spillover effects or evaluate their impact on local policies. In this work, we identify a novel setting and develop a suitable methodology that allow us to make unconfounded estimates of spillover effects of local policies. Focusing on California's Blueprint for a Safer Economy, we leverage how county-level mobility restrictions were deterministically set by public COVID-19 severity statistics, enabling a regression discontinuity design framework to estimate spillovers between counties. We estimate these effects using a mobility network with billions of timestamped edges and find significant spillover movement, with larger effects in retail, eating places, and gyms. Contrasting local and global policy regimes, our spillover estimates suggest that county-level restrictions are only 54% as effective as statewide restrictions at reducing mobility. However, an intermediate strategy of macro-county restrictions -- where we optimize county partitions by solving a minimum k-cut problem on a graph weighted by our spillover estimates -- can recover over 90% of statewide mobility reductions, while maintaining substantial flexibility between counties.

cs.CY↗

arXiv2022

Ballot Length in Instant Runoff Voting

Instant runoff voting (IRV) is an increasingly-popular alternative to traditional plurality voting in which voters submit rankings over the candidates rather than single votes. In practice, elections using IRV often restrict the ballot length, the number of candidates a voter is allowed to rank on their ballot. We theoretically and empirically analyze how ballot length can influence the outcome of an election, given fixed voter preferences. We show that there exist preference profiles over $k$ candidates such that up to $k-1$ different candidates win at different ballot lengths. We derive exact lower bounds on the number of voters required for such profiles and provide a construction matching the lower bound for unrestricted voter preferences. Additionally, we characterize which sequences of winners are possible over ballot lengths and provide explicit profile constructions achieving any feasible winner sequence. We also examine how classic preference restrictions influence our results--for instance, single-peakedness makes $k-1$ different winners impossible but still allows at least $Ω(\sqrt k)$. Finally, we analyze a collection of 168 real-world elections, where we truncate rankings to simulate shorter ballots. We find that shorter ballots could have changed the outcome in one quarter of these elections. Our results highlight ballot length as a consequential degree of freedom in the design of IRV elections.

cs.MA↗

arXiv2021

How to Quantify Polarization in Models of Opinion Dynamics

It is widely believed that society is becoming increasingly polarized around important issues, a dynamic that does not align with common mathematical models of opinion formation in social networks. In particular, measures of polarization based on opinion variance are known to decrease over time in frameworks such as the popular DeGroot model. Complementing recent work that seeks to resolve this apparent inconsistency by modifying opinion models, we instead resolve the inconsistency by proposing changes to how polarization is quantified. We present a natural class of group-based polarization measures that capture the extent to which opinions are clustered into distinct groups. Using theoretical arguments and empirical evidence, we show that these group-based measures display interesting, non-monotonic dynamics, even in the simple DeGroot model. In particular, for many natural social networks, group-based metrics can increase over time, and thereby correctly capture perceptions of increasing polarization. Our results build on work by DeMarzo et al., who introduced a group-based polarization metric based on ideological alignment. We show that a central tool from that work, a limit analysis of individual opinions under the DeGroot model, can be extended to the dynamics of other group-based polarization measures, including established statistical measures like bimodality. We also consider local measures of polarization that operationalize how polarization is perceived in a network setting. In conjunction with evidence from prior work that group-based measures better align with real-world perceptions of polarization, our work provides formal support for the use of these measures in place of variance-based polarization in future studies of opinion dynamics.

cs.SI↗

arXiv2021

To Recommend or Not? A Model-Based Comparison of Item-Matching Processes

Recommender systems are central to modern online platforms, but a popular concern is that they may be pulling society in dangerous directions (e.g., towards filter bubbles). However, a challenge with measuring the effects of recommender systems is how to compare user outcomes under these systems to outcomes under a credible counterfactual world without such systems. We take a model-based approach to this challenge, introducing a dichotomy of process models that we can compare: (1) a "recommender" model describing a generic item-matching process under a personalized recommender system and (2) an "organic" model describing a baseline counterfactual where users search for items without the mediation of any system. Our key finding is that the recommender and organic models result in dramatically different outcomes at both the individual and societal level, as supported by theorems and simulation experiments with real data. The two process models also induce different trade-offs during inference, where standard performance-improving techniques such as regularization/shrinkage have divergent effects. Shrinkage improves the mean squared error of matches in both settings, as expected, but at the cost of less diverse (less radical) items chosen in the recommender model but more diverse (more radical) items chosen in the organic model. These findings provide a formal language for how recommender systems may be fundamentally altering how we search for and interact with content, in a world increasingly mediated by such systems.

cs.CY↗

arXiv2021

Choice Set Confounding in Discrete Choice

Standard methods in preference learning involve estimating the parameters of discrete choice models from data of selections (choices) made by individuals from a discrete set of alternatives (the choice set). While there are many models for individual preferences, existing learning methods overlook how choice set assignment affects the data. Often, the choice set itself is influenced by an individual's preferences; for instance, a consumer choosing a product from an online retailer is often presented with options from a recommender system that depend on information about the consumer's preferences. Ignoring these assignment mechanisms can mislead choice models into making biased estimates of preferences, a phenomenon that we call choice set confounding; we demonstrate the presence of such confounding in widely-used choice datasets. To address this issue, we adapt methods from causal inference to the discrete choice setting. We use covariates of the chooser for inverse probability weighting and/or regression controls, accurately recovering individual preferences in the presence of choice set confounding under certain assumptions. When such covariates are unavailable or inadequate, we develop methods that take advantage of structured choice set assignment to improve prediction. We demonstrate the effectiveness of our methods on real-world choice data, showing, for example, that accounting for choice set confounding makes choices observed in hotel booking and commute transportation more consistent with rational utility-maximization.

cs.LG↗

arXiv2021

Adaptive normalization for IPW estimation

Inverse probability weighting (IPW) is a general tool in survey sampling and causal inference, used both in Horvitz-Thompson estimators, which normalize by the sample size, and Hájek/self-normalized estimators, which normalize by the sum of the inverse probability weights. In this work we study a family of IPW estimators, first proposed by Trotter and Tukey in the context of Monte Carlo problems, that are normalized by an affine combination of these two terms. We show how selecting an estimator from this family in a data-dependent way to minimize asymptotic variance leads to an iterative procedure that converges to an estimator with connections to regression control methods. We refer to this estimator as an adaptively normalized estimator. For mean estimation in survey sampling, this estimator has asymptotic variance that is never worse than the Horvitz--Thompson or Hájek estimators, and is smaller except in edge cases. Going further, we show that adaptive normalization can be used to propose improvements of the augmented IPW (AIPW) estimator, average treatment effect (ATE) estimators, and policy learning objectives. Appealingly, these proposals preserve both the asymptotic efficiency of AIPW and the regret bounds for policy learning with IPW objectives, and deliver consistent finite sample improvements in simulations for all three of mean estimation, ATE estimation, and policy learning.

stat.ME↗

arXiv2021

Pairwise Choice Markov Chains

As datasets capturing human choices grow in richness and scale -- particularly in online domains -- there is an increasing need for choice models that escape traditional choice-theoretic axioms such as regularity, stochastic transitivity, and Luce's choice axiom. In this work we introduce the Pairwise Choice Markov Chain (PCMC) model of discrete choice, an inferentially tractable model that does not assume any of the above axioms while still satisfying the foundational axiom of uniform expansion, a considerably weaker assumption than Luce's choice axiom. We show that the PCMC model significantly outperforms the Multinomial Logit (MNL) model in prediction tasks on both synthetic and empirical datasets known to exhibit violations of Luce's axiom. Our analysis also synthesizes several recent observations connecting the Multinomial Logit model and Markov chains; the PCMC model retains the Multinomial Logit model as a special case.

stat.ML↗

arXiv2020

Randomized Graph Cluster Randomization

The global average treatment effect (GATE) is a primary quantity of interest in the study of causal inference under network interference. With a correctly specified exposure model of the interference, the Horvitz-Thompson (HT) and Hájek estimators of the GATE are unbiased and consistent, respectively, yet known to exhibit extreme variance under many designs and in many settings of interest. With a fixed clustering of the interference graph, graph cluster randomization (GCR) designs have been shown to greatly reduce variance compared to node-level random assignment, but even so the variance is still often prohibitively large. In this work we propose a randomized version of the GCR design, descriptively named randomized graph cluster randomization (RGCR), which uses a random clustering rather than a single fixed clustering. By considering an ensemble of many different cluster assignments, this design avoids a key problem with GCR where a given node is sometimes "lucky" or "unlucky" in a given clustering. We propose two randomized graph decomposition algorithms for use with RGCR, randomized 3-net and 1-hop-max, adapted from prior work on multiway graph cut problems. When integrating over their own randomness, these algorithms furnish network exposure probabilities that can be estimated efficiently. We develop upper bounds on the variance of the HT estimator of the GATE under assumptions on the metric structure of the interference graph. Where the best known variance upper bound for the HT estimator under a GCR design is exponential in the parameters of the metric structure, we give a comparable variance upper bound under RGCR that is instead polynomial in the same parameters. We provide extensive simulations comparing RGCR and GCR designs, observing substantial reductions in the mean squared error for both HT and Hájek estimators of the GATE in a variety of settings.

stat.ME↗

arXiv2020

Prioritized Restreaming Algorithms for Balanced Graph Partitioning

Balanced graph partitioning is a critical step for many large-scale distributed computations with relational data. As graph datasets have grown in size and density, a range of highly-scalable balanced partitioning algorithms have appeared to meet varied demands across different domains. As the starting point for the present work, we observe that two recently introduced families of iterative partitioners---those based on restreaming and those based on balanced label propagation (including Facebook's Social Hash Partitioner)---can be viewed through a common modular framework of design decisions. With the help of this modular perspective, we find that a key combination of design decisions leads to a novel family of algorithms with notably better empirical performance than any existing highly-scalable algorithm on a broad range of real-world graphs. The resulting prioritized restreaming algorithms employ a constraint management strategy based on multiplicative weights, borrowed from the restreaming literature, while adopting notions of priority from balanced label propagation to optimize the ordering of the streaming process. Our experimental results consider a range of stream orders, where a dynamic ordering based on what we call ambivalence is broadly the most performative in terms of the cut quality of the resulting balanced partitions, with a static ordering based on degree being nearly as good.

cs.SI↗

arXiv2020

Evaluating stochastic seeding strategies in networks

When trying to maximize the adoption of a behavior in a population connected by a social network, it is common to strategize about where in the network to seed the behavior, often with an element of randomness. Selecting seeds uniformly at random is a basic but compelling strategy in that it distributes seeds broadly throughout the network. A more sophisticated stochastic strategy, one-hop targeting, is to select random network neighbors of random individuals; this exploits a version of the friendship paradox, whereby the friend of a random individual is expected to have more friends than a random individual, with the hope that seeding a behavior at more connected individuals leads to more adoption. Many seeding strategies have been proposed, but empirical evaluations have demanded large field experiments designed specifically for this purpose and have yielded relatively imprecise comparisons of strategies. Here we show how stochastic seeding strategies can be evaluated more efficiently in such experiments, how they can be evaluated "off-policy" using existing data arising from experiments designed for other purposes, and how to design more efficient experiments. In particular, we consider contrasts between stochastic seeding strategies and analyze nonparametric estimators adapted from policy evaluation and importance sampling. We use simulations on real networks to show that the proposed estimators and designs can increase precision while yielding valid inference. We then apply our proposed estimators to two field experiments, one that assigned households to an intensive marketing intervention and one that assigned students to an anti-bullying intervention.

stat.ME↗

arXiv2020

Scaling Choice Models of Relational Social Data

Many prediction problems on social networks, from recommendations to anomaly detection, can be approached by modeling network data as a sequence of relational events and then leveraging the resulting model for prediction. Conditional logit models of discrete choice are a natural approach to modeling relational events as "choices" in a framework that envelops and extends many long-studied models of network formation. The conditional logit model is simplistic, but it is particularly attractive because it allows for efficient consistent likelihood maximization via negative sampling, something that isn't true for mixed logit and many other richer models. The value of negative sampling is particularly pronounced because choice sets in relational data are often enormous. Given the importance of negative sampling, in this work we introduce a model simplification technique for mixed logit models that we call "de-mixing", whereby standard mixture models of network formation---particularly models that mix local and global link formation---are reformulated to operate their modes over disjoint choice sets. This reformulation reduces mixed logit models to conditional logit models, opening the door to negative sampling while also circumventing other standard challenges with maximizing mixture model likelihoods. To further improve scalability, we also study importance sampling for more efficiently selecting negative samples, finding that it can greatly speed up inference in both standard and de-mixed models. Together, these steps make it possible to much more realistically model network formation in very large graphs. We illustrate the relative gains of our improvements on synthetic datasets with known ground truth as well as a large-scale dataset of public transactions on the Venmo platform.

cs.SI↗

arXiv2020

Choosing to Grow a Graph: Modeling Network Formation as Discrete Choice

We provide a framework for modeling social network formation through conditional multinomial logit models from discrete choice and random utility theory, in which each new edge is viewed as a "choice" made by a node to connect to another node, based on (generic) features of the other nodes available to make a connection. This perspective on network formation unifies existing models such as preferential attachment, triadic closure, and node fitness, which are all special cases, and thereby provides a flexible means for conceptualizing, estimating, and comparing models. The lens of discrete choice theory also provides several new tools for analyzing social network formation; for example, the significance of node features can be evaluated in a statistically rigorous manner, and mixtures of existing models can be estimated by adapting known expectation-maximization algorithms. We demonstrate the flexibility of our framework through examples that analyze a number of synthetic and real-world datasets. For example, we provide rigorous methods for estimating preferential attachment models and show how to separate the effects of preferential attachment and triadic closure. Non-parametric estimates of the importance of degree show a highly linear trend, and we expose the importance of looking carefully at nodes with degree zero. Examining the formation of a large citation graph, we find evidence for an increased role of degree when accounting for age.

cs.SI↗

arXiv2020

Measuring Directed Triadic Closure with Closure Coefficients

Recent work studying triadic closure in undirected graphs has drawn attention to the distinction between measures that focus on the "center" node of a wedge (i.e., length-2 path) vs. measures that focus on the "initiator," a distinction with considerable consequences. Existing measures in directed graphs, meanwhile, have all been center-focused. In this work, we propose a family of eight directed closure coefficients that measure the frequency of triadic closure in directed graphs from the perspective of the node initiating closure. The eight coefficients correspond to different labelled wedges, where the initiator and center nodes are labelled, and we observe dramatic empirical variation in these coefficients on real-world networks, even in cases when the induced directed triangles are isomorphic. To understand this phenomenon, we examine the theoretical behavior of our closure coefficients under a directed configuration model. Our analysis illustrates an underlying connection between the closure coefficients and moments of the joint in- and out-degree distributions of the network, offering an explanation of the observed asymmetries. We also use our directed closure coefficients as predictors in two machine learning tasks. We find interpretable models with AUC scores above 0.92 in class-balanced binary prediction, substantially outperforming models that use traditional center-focused measures.

cs.SI↗

arXiv2020

Discovering Context Effects from Raw Choice Data

Many applications in preference learning assume that decisions come from the maximization of a stable utility function. Yet a large experimental literature shows that individual choices and judgements can be affected by "irrelevant" aspects of the context in which they are made. An important class of such contexts is the composition of the choice set. In this work, our goal is to discover such choice set effects from raw choice data. We introduce an extension of the Multinomial Logit (MNL) model, called the context dependent random utility model (CDM), which allows for a particular class of choice set effects. We show that the CDM can be thought of as a second-order approximation to a general choice system, can be inferred optimally using maximum likelihood and, importantly, is easily interpretable. We apply the CDM to both real and simulated choice data to perform principled exploratory analyses for the presence of choice set effects.

cs.LG↗

arXiv2020

Fundamental Limits of Testing the Independence of Irrelevant Alternatives in Discrete Choice

The Multinomial Logit (MNL) model and the axiom it satisfies, the Independence of Irrelevant Alternatives (IIA), are together the most widely used tools of discrete choice. The MNL model serves as the workhorse model for a variety of fields, but is also widely criticized, with a large body of experimental literature claiming to document real-world settings where IIA fails to hold. Statistical tests of IIA as a modelling assumption have been the subject of many practical tests focusing on specific deviations from IIA over the past several decades, but the formal size properties of hypothesis testing IIA are still not well understood. In this work we replace some of the ambiguity in this literature with rigorous pessimism, demonstrating that any general test for IIA with low worst-case error would require a number of samples exponential in the number of alternatives of the choice problem. A major benefit of our analysis over previous work is that it lies entirely in the finite-sample domain, a feature crucial to understanding the behavior of tests in the common data-poor settings of discrete choice. Our lower bounds are structure-dependent, and as a potential cause for optimism, we find that if one restricts the test of IIA to violations that can occur in a specific collection of choice sets (e.g., pairs), one obtains structure-dependent lower bounds that are much less pessimistic. Our analysis of this testing problem is unorthodox in being highly combinatorial, counting Eulerian orientations of cycle decompositions of a particular bipartite graph constructed from a data set of choices. By identifying fundamental relationships between the comparison structure of a given testing problem and its sample efficiency, we hope these relationships will help lay the groundwork for a rigorous rethinking of the IIA testing problem as well as other testing problems in discrete choice.

math.ST↗

arXiv2019

An Experimental Study of Structural Diversity in Social Networks

Several recent studies of online social networking platforms have found that adoption rates and engagement levels are positively correlated with structural diversity, the degree of heterogeneity among an individual's contacts as measured by network ties. One common theory for this observation is that structural diversity increases utility, in part because there is value to interacting with people from different network components on the same platform. While compelling, evidence for this causal theory comes from observational studies, making it difficult to rule out non-causal explanations. We investigate the role of structural diversity on retention by conducting a large-scale randomized controlled study on the Twitter platform. We first show that structural diversity correlates with user retention on Twitter, corroborating results from past observational studies. We then exogenously vary structural diversity by altering the set of network recommendations new users see when joining the platform; we confirm that this design induces the desired changes to network topology. We find, however, that low, medium, and high structural diversity treatment groups in our experiment have comparable retention rates. Thus, at least in this case, the observed correlation between structural diversity and retention does not appear to result from a causal relationship, challenging theories based on past observational studies.

cs.SI↗