It's a tricky one. You can't infer any sort of causality by just looking at real stores, because the policy varies by store or chain. So there are a zillionty (to be precise) other variables changing along with whether they use a common queue or not. Any difference you see could be due to any of them. And just taking an existing store and trying to randomize the policy is still tricky, since the form of queue you use influences the physical layout of the store and you might get weird effects from a mismatch.

But doing it in a lab means you have no ecological validity. People know they're being watched closely, much more closely than they would be in a real life scenario, and that will affect their behavior. So you really can't be sure your results will generalize to the real world.

Possibly the best you can do is take a few stabs at both approaches and hope that all the studies point in roughly the same direction.

> there are a zillionty (to be precise) other variables.

I hit all of them every time I go. It's awful. The old person returning a head of broccoli, the kid pulling down a display stand into my shopping on conveyer, the milk bottle leaks onto cashier. Every time I go it's a new kind of pain.

At least the lab data should be confirmed by observing comparable real-life checkout scenarios. Managing variables would be challenging in a genuine retail setting, but it probably can be accomplished well enough.

The fact that the outcome will vary in different store environments isn't a trivial factor in evaluating the claims of the experiment. IOW the findings may apply in some cases but not others, which is an important kind of detail to consider.

Exactly my thoughts. Why rig up some online simulation of a store when they could just get a reak store to try both ways while timing it.