Change Blindness and not seeing the rarely present

[L.J.Steele]

An NPR story:

A study published in the journal Current Biology looks into the problems involved in visually searching for exceedingly rare targets and comes to an unsettling conclusion: "If you don't find it often, you often don't find it," says study author Jeremy Wolfe. In other words, we are not very good at finding things that are rarely there.

* * *

Among people who study visual search, there's apparently widespread acceptance of what's now known as the "prevalence effect" — the idea that people often miss a rare thing, at least when tested in the lab.

The real question, says Kyle Cave, a cognitive researcher at Amherst University, is whether or not the effect is as large in the real world — whether a well-trained, full-time expert like radiologist Julianne Greenberg will make the same mistakes when she sits in her little closet day after day.

http://www.wbur.org/npr/122561355

[clpexco]

Lisa,

This is interesting, and I have seen this in work I did with remote examiners a couple of years ago. One of the complaints from many of the examiners was looking through literally thousands and thousands of comparisons without finding an ident. There were quite a few missed identifications by individual examiners in that project (all of them were caught by a second examiner as far as we know). But one of the lessons learned was that examiners need to see identifications often to remain on alert, so to speak. Even the examiners ask for this. So we have integrated this into the current version of our consultants work queue. We should have enough data soon to compare the missed identification rate of the first project with a similar volume of other work in the new system.

-Kasey

[Gerald Clough]

I wonder if this is really an example of an examiner essentially "forgetting" the intention to find a match. That sounds silly at first, but there's substantial experimental evidence that subjects can have their initial intention restored to more reliable function when they encounter partial matches. One set of experiments used the task of locating the names of a animals beginning with the letter "L" as the written names sequenced, "L" animal names being infrequently displayed. Performance was improved by including a number of other words beginning with "L." (Subjects rarely misidentified a cue as a true match, even though the response was just to push a button.) The experiment was repeated, but instead of partial match words, a similar number of textual reminders appeared, such as, "Remember to push the button when you see the name of an animal beginning with L." Both kinds of cues were equally effective. And subjects had considerable recall of the L-word cues and little of the non-cue words. Interestingly, increasing the number of textual cues did not further improve performance.

As I recall, there were other variations run, but that's the idea. The required task was one at which all the subjects could be considered competent. They could, at any time, instantly tell you what task they had been assigned. It wasn't gone from memory. But there seems to have been a break between intention and attention that functionally acts like forgetting. I think most of us would guess that a "partial match," one that has enough cues to trigger renewed attention to the task is probably as effective as a true match through a certain number of observations. But the examination task may diverge from the experiment design when it continues for a great many observations. To be of interest here, the next evolution of the experiment would be to see how long partial matches remained effective as cues when a true match had not been found in a long time. Kasey's reported experience suggests that when the task goes on for a long time without a true match, partial matches lose their ability to cue the task, since I would expect the examiners to regularly encounter partial matches in any large set.