This touches on my biggest gripe with how analytics-based decision making is presented. Instead of comparing average outcomes of decisions, the exact same data could be used to tell us a decision needs to have an X% chance of success to be the optimal one.
Using this run expectancy situation as an example, you could use the exact same data used to come up with average run expectancy to determine that swinging away must be successful 28%* of the time to make the risk worthwhile. Once you have that, now you can look at the specific game situation to determine if you meet that threshold.
Now you have a framework that still relies on the same quantitative data as run expectancy while allowing you to adjust the dials to account for the realities of the current situation. That's how you marry quantitative and qualitative analysis to come up with the best decision. You don't throw the baby out with the bathwater and ignore all that valuable data and just revert to relying on the manager to gnaw on a blade of grass and do whatever his tummy tells him to do
As a secondary pet peeve, win probability should replace run expectancy as the bedrock of these types of decisions, especially later in the game. Doing so would do a better job of accounting for the fact that a run while down one in the eighth has more marginal value than a run while up two in the fifth, for example.
* Figure pulled out of thin air