Considerations

Although Abraham Wald did bring up an important and influential point, I must say I am not entirely satisfied the computation.

A few issues here. To begin with, the degree of protection is probably not directly proportional to the thickness of the armor - at some point there should be a plateau of the relation curve.

Second, it was assumed that the bullets were fired randomly and evenly, so that the density of bullet holes on each part of a surviving plain was inversely proportional to the vulnerability of that part. For obviously reasons we know that's not correct. To go one step further, with the data and knowledge of the survey, pilots of the adversary air force would try to focus more often on the vulnerable parts, resulting in a negative feedback.

Moreover, there are parts that are intrinsically protected and therefore have a lower chance of being hit by bullets. For example, the medial surface of the engine could not be easily hit. If you want an extreme example, the wheels and tire could hardly be hit because they are hidden inside the plane while flying. Alas, you won't put armor on the tire, will you?