Experimental statistical evidence about eyewitness identification, intersecting reference classes, and trial by jury

On a list today I said the following on the topic of "experimental statistical evidence about eyewitness identification, intersecting reference classes, and trial by jury":
The problem [XXX] points to is [an example of] the familiar problem of intersecting reference classes -- which occurs in this context because every eyewitness and every eyewitness situation are (at least) always "slightly" different and there will always be some features about an eyewitness or an eyewitness situation whose significance or non-significance [has] not been empirically tested.

Does this mean that generalizations about [] eyewitnesses and eyewitness situations are irrelevant, or useless? No, not necessarily. What it means is that until better evidence comes along, "subjective human judgment" must determine whether this or that empirically-untested feature matters. This is the way it always is - there will always be such features - even though we fans of the rule of law and rational inference feel uncomfortable about this inevitable situation. Perhaps this problem (of the inevitable presence of empirically untested features) is one very good reason for trial by jury [of 12 or more persons].

&&&
The dynamic evidence page
Evidence marshaling software MarshalPlan