Research

Home | Research | Teaching | CV | Blog

My Research

Two-Sided Prioritized Ranking: A Coherency-Preserving Design for Marketplace Experiments [Paper]

Job Market Paper
Based on joint work with Mahyar Habibi and Negar Ziaeian.

Presented at CODE@MIT 2024 (October 2024)
Poster presentation at Frontiers of Online Advertising at ACM EC 2025 (July 2025)

Abstract: In online two-sided marketplaces, users must see a coherent platform environment in which all items remain accessible and item attributes, such as prices or features, are displayed consistently across users. Many interventions, especially those involving price changes, require such coherency to avoid confusion or perceptions of unfair treatment. These constraints complicate causal inference because users interact with shared items, and treatment applied to one item can spill over onto others, creating interference bias in standard A/B tests. We propose a new experimental design, Two-Sided Prioritized Ranking (TSPR), which uses the marketplace’s ranking algorithm as an experimental instrument. TSPR assigns treatment through prioritized exposure while preserving coherency and limiting spillovers. Our goal is to estimate the lift, defined as the proportional change in total outcomes between counterfactual worlds in which all items are treated or untreated. Using Monte Carlo simulations based on large-scale search impression data from an online travel agency, we show that TSPR produces lift estimates that are substantially less biased than those from coherent item-side A/B tests. We also compare TSPR to cluster-randomized designs, which can reduce interference when clusters isolate exposure but require detailed knowledge of the underlying network. TSPR provides a practical and scalable alternative for platforms that must preserve coherency while obtaining reliable estimates of treatment effects.

Distinguishing Biases from Personal Preferences: An Honest Machine Learning Approach [Paper]

Mahyar Habibi, Zahra Khanalizadeh, Negar Ziaeian

Abstract: This study proposes a new method for estimating biases at the micro-level in scenarios with multiple bilateral interactions, where the presence of individual preferences and correlated characteristics complicates the analysis. The proposed method comprises two stages. In the first stage, the method introduces a novel approach to extract preferences and characteristics, employing Collaborative Filtering with an ‘honest’ design. This technique is designed to separate preferences and self-induced outcomes from the constructed embeddings of interacting units. In the second stage, the method utilizes a Double Machine Learning estimator to identify biases at the unit level, based on the embeddings generated in the first stage. Monte Carlo simulations demonstrate that HCF+DML substantially outperforms naive OLS estimation under confounding, reducing RMSE by up to 50% and maintaining high correlation with true parameters even under embedding misspecification and non-random selection. We apply the method to nearly 150,000 film reviews from professional critics to estimate gender-based bias in evaluations of female-directed films. Naive comparisons suggest that 29% of critics exhibit statistically significant favoritism toward female directors. After controlling for the match between critic preferences and film characteristics, this figure drops to under 1%. The apparent pro-female pattern largely reflects critics’ preferences for genres where female directors are disproportionately represented, rather than gender-based favoritism per se.

Extending the Reproductive Horizon: Exploring Beliefs, Preferences, and Decision-Making

Joint work with Negar Ziaeian
[Extended Abstract]
Scheduled poster presentation at PAA 2026 Annual Meeting (May 2026)

Abstract: Limited fertility windows create a prominent tradeoff between career development and motherhood timing. Emerging assisted reproductive technologies, such as IVF, extend this window to some degree, and the share of IVF-conceived births continues to rise worldwide. Yet little is known about how women incorporate access to these technologies into decisions about when to have children. This project studies women’s beliefs about natural fertility, age-related decline, and the extent to which IVF can compensate for biological constraints, and asks whether misperceptions about these factors influence willingness to postpone childbearing. We field a survey and discrete choice experiment with women aged 20–30 in the United Kingdom to measure how beliefs about fertility and IVF shape preferences over motherhood timing and career investment. We estimate the income premium that makes respondents indifferent between early childbearing and postponement and examine how information about IVF success and costs shifts these tradeoffs. A real-stakes outcome complements stated preferences.