Multicell experiments for marginal treatment effect estimation of digital ads

2023-02-27

AI Summary

In digital advertising randomized experiments, one-sided noncompliance prevents conventional two-arm RCTs from identifying the marginal treatment effect (MTE), thereby impeding critical operational decisions—such as “how many users to reach” and “how much budget to allocate.” To address this, we propose a novel multi-unit experimental paradigm tailored for dense-margin decision-making, integrating multi-unit randomization design, MTE modeling, and structural causal inference. Our approach identifies the optimal ad exposure intensity without requiring additional budget. Calibrated on real Facebook advertising data, simulation experiments demonstrate that our method significantly outperforms direct optimization baselines across diverse partial-compliance scenarios. It robustly estimates both the optimal consumer coverage and budget allocation policy, overcoming fundamental limitations of traditional RCTs in real-world advertising environments where noncompliance is pervasive and treatment intensity is continuous.

Abstract

Randomized experiments with treatment and control groups are an important tool to measure the impacts of interventions. However, in experimental settings with one-sided noncompliance extant empirical approaches may not produce the estimands a decision maker needs to solve the problem of interest. For example, these experimental designs are common in digital advertising settings but typical methods do not yield effects that inform the intensive margin: how many consumers should be reached or how much should be spent on a campaign. We propose a solution that combines a novel multicell experimental design with modern estimation techniques that enables decision makers to solve problems with an intensive margin. Our design is straightforward to implement and does not require additional budget. We illustrate our method through simulations calibrated using an advertising experiment at Facebook, demonstrating its superior performance in various scenarios and its advantage over direct optimization approaches.

Problem

• Estimates marginal treatment effects in digital ads
• Addresses one-sided noncompliance in experiments
• Optimizes ad reach and spending decisions

Innovation

• Novel multicell experimental design for ads
• Modern estimation techniques for intensive margin
• No additional budget required for implementation