April 11, 2007

Quantifying certainty and uncertainty

Recently we've seen reports about the United Nations' blue-ribbon international panel of climatologists noting that this august body described the existence and effects of global warming with words rather than statistics. The U.N. panel says that global warming is an unequivocal fact "very likely" caused by human activity. The word, "very," is an upgrade from its 2001 report, which said it was "likely" that humans were the cause of this.

Climate-change scientists often rely on words to describe the likelihood of danger to the world but six years ago they began to use numerical expressions in order to convince skeptics that they weren't being subjective or just making stuff up. This reminded me of the problems expert witnesses often face when they report the results of their analyses at trial.

In one case I worked on  the major evidence against a Louisiana youth named Michael Carter was a  confession statement that he had signed. In it he admitted shooting and killing a policeman in a botched burglary attempt. Fortunately for Michael, the police made a tape-recording of the interrogation leading up to the confession statement. Several times on the tape Michael denied killing the cop, all the while crying and  finally becoming so ill that the police had to shut the tape off. But they typed up a confession statement and got him to sign it. We can never know what happened during the time the tape wasn't running.

Things looked bad for Michael. But since one detective testified that the confession statement simply reported exactly what Michael had said, using his own  words and sentences, this opened the door for me to compare the language of the detective with the language in the statement that Michael had signed. I found huge differences. I also tape recorded Michael's speech while he was incarcerated and waiting for trial in order to get another speech sample to compare with the written statement. I won't describe here all the differences but if you're interested you can read about this case in my book, The Language of Confession, Interrogation, and Deception (Sage, 1998).

The point I'm slowly getting to is what happened when I took the witness stand. My direct examination concerned my methodology, the data I used, and my comparison of Michaels language with that of the detective. When I finished with this, the defense attorney asked me:

Q: Does the sentence structure that appears on the written confession statement resemble the sentence structure used by Michael Carter when Michael Carter speaks?

A: Not at all.

Q: Whose sentence structure does it resemble in this scenario involving these people?

A: Well, I don't mean to pick on the detective, but since he said he wrote down what Michael Carter said, it's closer to his sentence structure than it is to Michael Carter's.

Later, at the end of my direct testimony, the defense attorney asked me:

Q: Just generally based on everything you've said, Dr. Shuy, and based on your examination of all these items and documents, how likely is it that the words that appear on that written statement are the words of Michael Carter?

A: It's very unlikely that these were Michael Carter's words.

Q: How likely is it that the words that appear on the written statement are instead the words of the detective?

A: It is much more likely that these were his words, especially in light of the fact that he said that he was the one who wrote them down.

Instead of using terms like "a lot," "very unlikely," and "much more likely," it would have been very nice to be able to say something more quantitatively representative. In my analysis I had used percentages to compare sentence types of Michael and the detective but I could not quantify the likelihood of authorship in the way that the defense attorney wanted. Like the climate-change scientists, for that I had to resort to a non-statistical assessment that used words.

Today the quantification of certainty or doubt is beginning to be used by climate-change scientists. They believe that using words for  the language of uncertainty is easy to misrepresent and that politicians seem to want something more specific --  numbers. This change began when researchers Richard Moss and Stephen Schneider urged the U.N. panel of climate scientists to strengthen their report language with hard numbers (here). Now if a report says something is "virtually certain," it means 99 percent certain. If it's "very likely," it's between 90 percent and 99 percent. "More likely than not" refers to 50 percent and "unlikely" means somewhere between 10 and 33 percent. To review other verbal expressions that  have been assigned numeric values, see this chart.

I wish I had heard of this back in 1989 when I testified in Michael Carter's case but, as it turns out, I didn't really need it. After the prosecutor heard my testimony, he dropped all the charges.

Posted by Roger Shuy at April 11, 2007 11:04 AM