Saturday, June 16, 2012

Spoke not a word, though it meant my life

On the question of justice, the contemporary Western world [or Anglosphere, anyway] is collectively more repulsed by a type I error than it is by an error of the second kind, and our judicial system is organized in a way reflecting that. The bar is set high when it comes to demonstrating a person's guilt, the consequences being that people are more likely to go unpunished for committing a crime than they are to be punished for a crime they didn't commit.

But dissent from that societal more is not equally distributed across all groups. On four occasions, the GSS has asked respondents which miscarriage of justice is worse--convicting the innocent or allowing the guilty to go free. Blacks and liberals tilt more heavily in their preference for allowing some guilty people to walk as a cost of ensuring that the innocent aren't convicted (76.3% and 74.9%, respectively, saying that convicting the innocent is worse than letting the guilty off is) than whites and conservatives (71.6% and 70.7%) do.

As Steve Sailer regularly reminds us and the Trayvon Martin case illustrates, though, the "Who? Whom?" question is of paramount importance. In a Florida poll conducted last month, 79% of blacks but only 21% of whites were confident in George Zimmerman's guilt. On the political front, 49% of Democrats and 13% of Republicans asserted that he is guilty of murder. Better lock Zimmerman up to ensure the guilty won't walk!

Getting back to the philosophical, ideological realm, which people hastily evacuate when "Who? Whom?" becomes relevant, it is by intelligence rather than political orientation, sex, or race where divergence in opinion on the question is most pronounced. The percentages of people, by IQ grouping*, who would prefer to see a guilty man let go than an innocent man locked up rather than the other way around:

IQFree
Real dumbs66.0%
Pretty dumbs68.1%
Normals67.0%
Pretty smarts75.3%
Really smarts88.6%

No, dummy, Barabbas' release wasn't the real tragedy!

GSS variables used: VERDICT, WORDSUM, POLVIEWS(1-3)(4)(5-7), RACE

* Respondents are broken up into five categories in a way that approximates a normal distribution; Really Smarts (wordsum score of 9-10, comprising 13% of the population), Pretty Smarts (7-8, 26%), Normals (6, 22%), Pretty Dumbs (4-5, 27%), and Real Dumbs (0-3, 12%).

10 comments:

Jokah Macpherson said...

I guess they figure, given the tendencies of people to behave the same way over time, that most Type II errors will sort themselves out eventually.

pat said...

I used to do a couple lectures every semester on all this when I taught statistics. It's called the Neymann-Pearson Lemma. In descriptive statistics there is a recognition that Type I and Type II errors are conjoined such that eliminating all "false positives" "true negatives" and vice versa. There is no simple solution to this problem. You have to make a choice.

In research academics want to publish so there must be some standards else the journals fill up with garbage. In the social sciences this standard is generally a risk of incorrectly accepting the null hypothesis by chance at no more than 5%. In the medical sciences the risk is usually held to under 1%.

It is considered important enough that we don't fill the journals with false positives that we risk ignoring real findings. In point of fact a recent study of medical findings reported that fully 50% of all new findings in medical journals are never replicated. This is why the news continuously reports "breakthroughs"and "cures" and nothing is ever heard of them again. Half of all new findings published in medical journals are wrong. In the social sciences where standards are lower, I imagine it's even worse.

In the court room our stance is also against false positives. That would be convicting someone who is innocent. With appeals it is almost impossible in America these days to convict an innocent man and have him stay convicted. Yet we know it still happens. Convicted rapists keep being freed by DNA evidence.

There is nothing that can be done about this short of freeing everyone. Try to convict the guilty and you will eventually convict an innocent man.

In the courtroom it's not just crime but the appearance of crime that counts. If there is a man on trial for murder and you let him go there is another risk. He may kill again. But the court doesn't "see" these crimes in the same way. Everyone is aware of a public execution but few notice that there are a large number of deaths which are never solved. At least some of these unsolved or even unknown murders are the result of the actions of men who had been tried but not convicted. It's easier to kill the sacond time around.

To get the math right we should probably consider not just past murders but future murders. That statistic is tricky. Probably if a wife kills her husband of thirty years she poses no future risk to the community. But if a sixteen year old kills someone while robbing a convenience store, he should probably be executed on much weaker evidence.

Unfortunately the law gets this backwards. They release the kids and convict the old coots.

Albertosaurus

Audacious Epigone said...

Albertosaurus,

There's a recent econtalk podcast with Ed Yong where just that issue is discussed, although the perspective that allowing false negatives as a cost of avoiding true negatives wasn't shared by either Yong or Roberts. Instead, Yong basically insinuates that lots of dubious stuff gets published and isn't replicated because it took 20 trial runs just to reach 99% confidence. Interesting.

Audacious Epigone said...

That is, 20 trial runs, with the first 19 showing no relationship between variables, and then the 20th finally showing the desired result. Consequently, replication shows the published result to be BS.

Bill Drissel said...

Singular of mores is mos.

SFG said...

I wonder if brighter people tend to be from the middle and upper classes and tend to figure they can't handle themselves in jail? Certainly I know I wouldn't last 10 minutes.

Also, the preference for the Type I error comes out of a British cultural matrix and reaction against government overreach. In France, from what I understand, you are guilty until proven innocent...

Audacious Epigone said...

SFG,

Good point. It would have been more accurate for me to have written "the contemporary Anglosphere is collectively more repulsed by..."

Audacious Epigone said...

In civil courts in France though, there is the presumption of innocence, I think. Or is that incorrect?

Anonymous said...

Aren't we the only country with juries?

Audacious Epigone said...

Anon,

Jury systems are used throughout much of Europe, though notably not in Germany. I didn't think they were used in Japan, either, but apparently in serious criminal cases, they are as of just a few years ago.