Experimental researchers routinely make a difficult choice before data collection even begins: which outcome, stimulus, task, scenario, product, or behavior should stand in for the broader construct they care about?
The hidden assumption in a single-proxy experiment
When an experiment relies on a single proxy, that proxy is effectively treated as fixed. If the experimental effect depends on specific characteristics of that operationalization, the study cannot reveal this because there is no variation across proxies to observe.
Why this matters especially for behavior
Broad behavioral constructs can be expressed in many different ways. Pro-environmental behavior, for example, can mean choosing tap water instead of bottled water, using public transport, repairing a device, eating plant-based food, or sorting waste. These behaviors may all belong to the same broad construct while differing in cost, effort, normativity, identity signaling, and many other characteristics.
A finding obtained with one of these behaviors is informative about that behavior. The problem arises when we want to make a broader claim while the study design provides no evidence about how the effect varies across other valid proxies.
What a multi-proxy experiment changes
Instead of choosing one proxy and giving it to everyone, researchers build a pool of proxies and randomly assign participants to one of them. The experimental manipulation is still randomized in the familiar way. The additional randomization determines which proxy is used to operationalize the outcome.
What you see: one experimental effect for one operationalization.
What you can see: whether an effect generalizes across proxies and how much it varies.
Variation is not only a nuisance
The most interesting result may not be that an effect varies, but why. If proxies are characterized on meaningful features, those characteristics can be tested as moderators of the experimental effect. This turns variation across proxies into a source of theoretical insight.