It would be nice if we had more studies like that. Tech's insistence that productivity is unquantifiable has allowed a large amount of superstitions - sorry, "best practices" - to flourish.

Code review centered around maintainability isn’t comparable to the effect marketing has on customers. It’s more comparable to how marketing manages its process of ad creation, connections in the industry, etc. Indeed human expertise is useful for having a starting point and determining what directions one should go in to fix things. But without data, how are you supposed to tell that your changes have actually had an intended effect? Gathering data around marketing would be more comparable to having monitoring on your software systems. Much like not having any way to know if the code changes you just made have actually fixed a performance issue you were seeing in the system, not having any data to judge whether a change you made to your marketing had the intended effect makes making such changes pointless.

Experts are wrong all the time regardless. The best of them tend to like being able to check their work against reality.

The data driven decision is 'passing the smell test'. If you have it one way or another do you find it easier or harder to work with as a first order approximation. Do your coworkers find it easier or harder as a second order approximation to be sure that in the relevant set that you aren't an outlier who happens to know pointer math really well but everyone else finds far harder.

For the code review task they don't need precise statistical data they just need to know which direction the arrow points with a decision in the main case.

^