Understanding how to draw causal conclusions in markets where participants' actions directly affect each other—what economists call “interference”—is one of the field’s most challenging frontiers. Traditionally, causal inference assumes that what happens to one individual or unit doesn’t spill over to others. But real-world markets, from equity crowdfunding to online advertising auctions, are full of interference. The question is: how can we rigorously uncover cause-and-effect relationships in these tangled environments, especially when we have the power to design the rules of the market itself? This is not just an academic puzzle; it shapes how we judge everything from fundraising strategies to policy interventions in digital platforms.
Short answer: Causal inference in markets with interference can be performed by designing specific market mechanisms—such as randomizing key features or rules—that deliberately structure how interference occurs, so that the resulting data can be analyzed with modern statistical methods (like mediation analysis or surrogate indices) to tease apart direct and indirect effects. By carefully crafting the mechanism, researchers can create “experiment-like” conditions that allow for valid causal insights even when interference is unavoidable.
Why Interference Complicates Causal Inference
Let’s first clarify the problem. In standard causal inference, the “Stable Unit Treatment Value Assumption” (SUTVA) requires that each unit’s outcome depends only on its own treatment, not on the treatments assigned to others. In a typical clinical trial, this means your health outcome depends only on whether you got the drug, not on whether your neighbor did. But in markets, this assumption fails: if one entrepreneur asks for a large fundraising target on a crowdfunding platform, it can affect how much investors are willing to give to others. Similarly, in auctions, the price one bidder pays depends on the actions of all other bidders. This “interference” or “spillover” is ubiquitous.
According to the NBER working paper “Be Careful What You Ask For: Fundraising Strategies in Equity Crowdfunding,” such interference is not just a theoretical nuisance, but a central feature of how markets operate. The paper shows, for instance, that what one entrepreneur asks for in a crowdfunding campaign can shape investor behavior across the platform, influencing not only their own success but also the fortunes of others. As they note, “entrepreneurs not only set investment goals, but also choose when to close their campaigns,” and these decisions ripple through the market (nber.org).
So, how can we study causality when interference is everywhere? Here’s where “designed mechanisms” come in. In market design, we can set the rules—who interacts with whom, how information flows, which actions are allowed, and under what conditions. By engineering these rules, we gain leverage over the interference structure itself.
One approach is to randomize certain elements of the market mechanism. For example, in a crowdfunding platform, researchers might randomly assign different fundraising goals or randomly vary the information shown to investors about other campaigns. By doing so, they can observe how these randomized changes affect not only the targeted entrepreneur’s outcome but also the outcomes for others, thus mapping out the causal web of interference.
The NBER lectures by Raj Chetty and Kosuke Imai, referenced in the same source, highlight advanced statistical tools that become crucial here. Mediation analysis, for example, allows us to decompose the total effect of a treatment (like a change in auction rules) into direct effects on the treated unit and indirect effects mediated through others. Surrogate indices can serve as stand-ins for complex outcomes, capturing the combined impact of interference in a single measure.
Concrete Examples: Crowdfunding and Beyond
Take the equity crowdfunding case from the NBER paper: more experienced and educated founder teams tend to “ask for more” and, as a result, “raise more money.” But these actions don’t happen in a vacuum. If a prominent team sets a high fundraising target, it may anchor investor perceptions, affecting how much others can raise. Female founder teams, interestingly, “ask for less, are equally successful, yet raise significantly less,” and they “wait longer before closing campaigns”—possibly as a response to observed interference from the market (nber.org). By randomizing the information or timing of campaigns, researchers can disentangle whether these patterns are due to intrinsic characteristics or to market spillovers.
Similar logic applies to online ad auctions, labor markets, or any setting where one participant’s actions shape the payoffs of others. For example, in digital advertising, changing the bidding rules or the visibility of competitors’ bids can be randomized to study how one firm’s spending influences the outcomes for others.
Statistical Methods and Their Role
Once the mechanism is designed to introduce random variation in features that generate interference, modern statistical methods can be applied. Mediation analysis, as discussed by Chetty and Imai, helps identify not just if a treatment works, but how it works—crucially, whether its effect is channeled through direct action or through changes in the behavior of others. For example, did a new fundraising rule increase total money raised because it made each campaign more attractive, or because it shifted investor attention between campaigns?
Surrogate indices, another technique mentioned in the NBER lectures, allow researchers to summarize the effects of complicated, interconnected outcomes—such as the overall health of a market—into a single, causally interpretable score. These methods are particularly valuable in the presence of interference, where traditional outcome measures may not capture the full picture.
Limitations and Challenges
Despite these advances, several challenges remain. Randomizing market features is not always feasible, especially in large, mature markets where participants may resist changes. Moreover, the complexity of interference—how many participants affect each other, and in what ways—can make statistical analysis tricky. The NBER sources acknowledge that “all errors are ours,” signaling the inherent difficulty in getting clean causal estimates even with careful design.
Furthermore, as the sciencedirect.com and cambridge.org excerpts suggest, access to comprehensive, high-quality data is crucial but not always straightforward. Researchers must often rely on collaboration with platform operators or businesses willing to experiment with their rules, as was the case with the SEEDRS data mentioned in the NBER paper.
Bringing It All Together: The State of the Art
In sum, causal inference in markets with interference is possible, but it requires a blend of creative market design and sophisticated statistical tools. By deliberately structuring the rules—randomizing key features, controlling information flow, and tracking outcomes—researchers can create conditions where interference can be measured and controlled for, rather than being an obstacle. As the NBER sources demonstrate, this approach is already yielding insights in areas like crowdfunding, where the interplay between entrepreneur requests, investor responses, and platform rules can be mapped out with precision.
The key, as highlighted by the NBER lectures and research, is to “exploit unique features” of the market and to “carefully craft experimental designs” (nber.org). This allows researchers to move beyond the limitations of standard causal inference and to tackle the messy, interconnected reality of real-world markets.
Looking forward, as more digital platforms open up their mechanisms to experimentation, the toolkit for causal inference with interference will only grow richer. The lessons learned in equity crowdfunding are likely to shape how economists, data scientists, and market designers study everything from gig work to online retail, making causal answers more robust and actionable even in the presence of pervasive interference.
