Evaluating managers is one of the greatest challenges of sabermetrics. We expect a good manager to help his team win more games, but when a team does win more games than expected, it is unclear whether that overachievement is the result of excellent management, flawed projections, randomness, or a variety of other factors. Definitively, what statistics can do is evaluate manager decision-making. By aggregating the outcomes of strategic decisions in similar situations historically, we can evaluate whether a manager is making a good decision based on trends, rather than on the results in that one instance.
Let's take a look at manager decision-making in multiple categories. I've split the AL and NL because of the strategic differences between the two leagues due to their different rules.