I studied the crew reports, which sometimes described large deviations from the tracks that the bombers were supposed to fly. For the majority of crews, who reported no large deviations, there was no way to tell how close to their assigned tracks they actually flew. My best estimate of the density of bombers was uncertain by a factor of 10. This made the collision formula practically worthless as a predictive tool. But it still had value as a way to set an upper bound on the collision rate. If I assumed maximum values for all three factors in the formula, it gave a loss rate due to collisions of 1 percent per operation. One percent was much too high to be acceptable, but still less than the overall loss rate of 5 percent. Even if we squeezed the bomber stream to the highest possible density, collisions would not be the main cause of losses.
How common, really, were collisions? Observational evidence of lethal crashes over Germany was plentiful but unreliable. The crews frequently reported seeing events that looked like collisions: first an explosion in the air, and then two flaming objects falling to the ground. These events were visible from great distances and were often multiply reported. The crews tended to believe that they were seeing collisions, but there was no way to be sure. Most of the events probably involved single bombers, hit by antiaircraft shells or by fighter cannon fire, that broke in half as they disintegrated.
In the end I found only two sources of evidence that I could trust: bombers that collided over England and bombers that returned damaged by nonlethal collisions over Germany. The numbers of incidents of both kinds were reliable, and small enough that I could investigate each case individually. The case that I remember best was a collision between two Mosquito bombers over Munich. The Mosquito was a light, two-seat bomber that Bomber Command used extensively for small-scale attacks, to confuse the German defenses and distract attention from the heavy attacks. Two Mosquitoes flew alone from England to Munich and then collided over the target, with only minor damage. It was obvious that the collision could not have been the result of normal operations. The two pilots must have seen each other when they got to Munich and started playing games. The Mosquito was fast and maneuverable and hardly ever got shot down, so the pilots felt themselves to be invulnerable. I interviewed Pilot-Officer Izatt, who was one of the two pilots. When I gently questioned him about the Munich operation, he confessed that he and his friend had been enjoying a dogfight over the target when they bumped into each other. So I crossed the Munich collision off my list. It was not relevant to the statistics on collisions between heavy bombers in the bomber stream. There remained seven authentic nonlethal collisions between heavy bombers over Germany.
For bombers flying at night over England in training exercises, I knew the numbers of lethal and nonlethal collisions. After more than 60 years, I can’t recall them precisely, but I remember that the ratio of lethal to nonlethal collisions was three to one. If I assumed that the chance of surviving a collision was the same over Germany as over England, then it was simple to calculate the number of lethal collisions over Germany. But there were two reasons that assumption might be false. On the one hand, a badly damaged aircraft over Germany might fail to get home, while an aircraft with the same damage over England could make a safe landing. On the other hand, the crew of a damaged aircraft over England might decide to bail out and let the plane crash, while the same crew over Germany would be strongly motivated to bring the plane home. There was no way to incorporate these distinctions into my calculations. But since they pulled in opposite directions, I decided to ignore them both. I estimated the number of lethal collisions over Germany in the time since the massive attacks began to be three times the number of nonlethal collisions, or 21. These numbers referred to major operations over Germany with high-density bomber streams, in which about 60,000 sorties had been flown at the time I did the calculation. So collisions destroyed 42 aircraft in 60,000 sorties, a loss rate of .07 percent. This was the best estimate I could make. I could not calculate any reliable limits of error, but I felt confident that the estimate was correct within a factor of two. It was consistent with the less accurate estimate obtained from the theoretical formula, and it strongly confirmed Smeed’s belief that collisions were a smaller risk than fighters.
For a week after I arrived at the ORS, the attacks on Hamburg continued. The second, on July 27, raised a firestorm that devastated the central part of the city and killed about 40,000 people. We succeeded in raising firestorms only twice, once in Hamburg and once more in Dresden in 1945, where between 25,000 and 60,000 people perished (the numbers are still debated). The Germans had good air raid shelters and warning systems and did what they were told. As a result, only a few thousand people were killed in a typical major attack. But when there was a firestorm, people were asphyxiated or roasted inside their shelters, and the number killed was more than 10 times greater. Every time Bomber Command attacked a city, we were trying to raise a firestorm, but we never learnt why we so seldom succeeded. Probably a firestorm could happen only when three things occurred together: first, a high concentration of old buildings at the target site; second, an attack with a high density of incendiary bombs in the target’s central area; and, third, an atmospheric instability. When the combination of these three things was just right, the flames and the winds produced a blazing hurricane. The same thing happened one night in Tokyo in March 1945 and once more at Hiroshima the following August. The Tokyo firestorm was the biggest, killing perhaps 100,000 people.
The third Hamburg raid was on the night of July 29, and the fourth on August 2. After the firestorm, the law of diminishing returns was operating. The fourth attack was a fiasco, with high and heavy clouds over the city and bombs scattered over the countryside. Our bomber losses were rising, close to 4 percent for the third attack and a little over 4 percent for the fourth. The Germans had learnt quickly how to deal with WINDOW. Since they could no longer track individual bombers with radar, they guided their fighters into the bomber stream and let them find their own targets. Within a month, loss rates were back at the 5 percent level, and WINDOW was no longer saving lives.