We'll Be Having Fun All Sumner Long
"For instance, Romer argued that the 1929 stock market crash sharply reduced consumer confidence, and that this was a major factor depressing aggregate demand. But the quite similar stock market crash in 1987 seemed to have no impact at all on economic growth, suggesting that the direct impact of stock prices on real output is certainly very small." -- Scott Sumner, The Midas Paradox, p. 39
What we have here is an oscillation between an historical mode of explanation and a scientific one. Science deals in generalizations, and has nothing to say about specific events other than citing them as instances of its generalizations. ("The arrow will fall back to earth, as it was shot in the air with less than escape velocity.") History, on the other hand, deals with particulars, and a resort to scientific generalizations is at most a stopgap in an historical explanation, filling in where we lack actual historical knowledge. ("Although we have no evidence for what happened in that battle, the Egyptians probably won, as armies with such a preponderance of force typically do.")
But a stopgap generalization like that above cannot be used to defease an historical narrative based on actual evidence: it is silly to write, "It is unlikely that some obscure Macedonian named Alexander conquered the Persian Empire, as armies facing such an imbalance in force are almost always defeated." Or: "Since there were many assassinations of notable political figures between 1880 and 1920, it is unlikely that a world war broke out in 1914 after one more such assassination." We are dealing with extensive evidence that these things did happen, and we can only reasonably dispute that evidence with historical evidence showing, for instance, a widespread conspiracy to fake the existence of Alexander.
So even if, somehow (based on only two cases!) Sumner has established a scientific generalization that "generally, stock market crashes have very little impact on real output," that says nothing about the actual historical evidence for whether, in the particular case of The Great Depression, one had a large impact. What starts a panic in one crowd in one place on one day may be shrugged off nonchalantly by another crowd in another place on another day; events are what they are interpreted to be by human beings, and human interpretation is not constant.