Sunday, December 15, 2019

SCAN

The article about Scientific Content Analysis, or SCAN,  certainly sounds alarming. While parts of the technique sound like they might be possible indicators of a person being deceptive, most sounded simply bogus overinterpretations, and some even sounded backwards.  I was gratified to learn farther into the article that an independent investigation had picked up something similar: two characteristics that SCAN counts as deceptive had actually been shown to be indicators of truth-telling in other research.

I noticed, for example, that substituting synonyms in a written account, using "kids" at one point in the story and "children" in another, was an indication of deception.  I was taught that it was merely more interesting writing.  When writing about a person, calling him "Jerry" in four straight sentences has a boring, singsong-y quality to it.  For smoother flow, one switches among "Mr. Arbus," "my neighbor," "the poor guy" and "Jerry Arbus." It's not deception, and anyone who grades it as such is a fool.

Still, the article was in ProPublica, a site that is not merely liberal, but quite radical. (I don't know about the South Bend Tribune.) They have an initial bias against the police and law enforcement in general, and so might be looking to trash a technique that is merely flawed, not invalid or useless.  I believe that people do eventually say what they mean whether they know it or not, and it is not ridiculous to think that intelligent observers have noticed things that are indeed meaningful. In particular, that a person might choose very careful true but deceptive statements, such as "I was not present for that discussion" when they were on speaker phone sounds plausible.  Looking for such things sounds worth pursuing.

I looked the technique up in other sources, then.  The journal Frontiers in Psychology has a research article showing that SCAN could not distinguish between known true and known false statements. The Skeptic's Dictionary calls it "too good to be true," and "gullibility and wishful thinking."

I'm thinking don't be impressed.

No comments: