Saturday, July 14, 2007

Does social class explain the race-crime link? I'm getting tired of these so-called experts who say that the relationship between race and street crime is explained by social class as if that has ever been demonstrated by research. It hasn't. I'm quite familiar with the data and the impact of race, especially on violence, is much stronger than class. The amazing dilemma is that very few studies have examined this question directly. Everyone repeats the mantra, but no one bothers to actually check it out. Well, the General Social Survey has a way of calling bullshit on all the squishy talk.

Using logistic regression, I examined data on 7,477 black and white Americans. The dependent variable is whether you have been arrested or not. When race alone is included in the model, its coefficient is .432, and it is significant at the .001 level. If conventional wisdom is correct, this coefficient should be reduced to zero when parents' class is included. I used father's completed years of school as a measure of this. When it is included in the model along with race, its coefficient is -.026 (significant at the .01 level). (Do not compare the coefficients since the measures are in different metrics, but trust me, race is more powerful).

So does race drop to zero with class thrown in? Absolutely not--in fact, it increases to .461. Class explains nothing.


  1. dearieme4:27 AM

    "I used father's completed years of school as a measure of this." 1) Is the range of that variable wide enough (for black criminals' fathers) to make the exercise meaningful?
    2) Given what I read about the USA black way of life, wouldn't mother's completed years of school be more relevant?

  2. dearime: 1) Of course the range is wide enough: in this sample, black father's educ. ranges from 0 to 20+ years of education. By the nature of the variable, father's education is not restricted in range, race is--only two categories--yet it is much more powerful than class. 2) When I use mother's educ., the coefficient is -.034, so it is stronger than dad's educ., but once again, including it in the model only increases the race coefficient from .432 to .498.

  3. Ron,

    A jaw dropper of a post. Really interesting, and a little worrying.

  4. Would it be more meaningful to look at conviction rates rather than arrest rates?

  5. peter: Criminologists conventionally like measures closest to the actual illegal behavior (like self-reports), but they often use arrest data for convenience and feel comfortable doing it since validation research shows it's a pretty good measure.

    The GSS data, however, is self-reported arrest, and there is research (that compares self-reports to police records)suggesting that blacks underreport getting arrested. Evidently, they are likely to view an interviewer as The Man, and are reluctant to admit things.

  6. Anonymous5:43 PM

    And if you put IQ (WORDSUM, I suppose) in there?

  7. Is it logical that age has a negative coefficient if the question is "have you ever been arrested?"? If anything the coefficient should be positive since "ever" is your entire previous life.

    Or is the question, "Have you been arrested in the last year ... or ten years?"

  8. robert: The question is ever arrested. You're generation turns out to be much more important than risk due to time. This is a sample of adults, and most people who will ever be arrested will have it happen before 25 or 30, so exposure beyond that just doesn't matter.

  9. That makes sense. It's a generation effect, not an age effect.

    There certainly is an age effect on committing crime, but this question does not get at it.

    Perhaps there is another question that would get at the age effect and you could get an even better estimate of the race effect.

  10. This is interesting. B. S. Centerwall did a study comparing murder rates to socioeconomic status. He used household crowding as an indicator of socioeconomic status. He found that, once you control for household crowding, there is no statistically significant difference between blacks and whites in murder rates. How do you explain these results?

  11. Pres: Since when is population density a measure of SES? San Francisco is the densest area in CA for example, but it is not the poorest by any stretch. SES data are readily available--why didn't he ues that? Probably because in study after macrolevel study, race is more powerful than poverty.

    Centerwall's study is anomalous: the body of macrolevel research shows that race explains are least half of the variation in homicide. Of course, researchers do everything they can think of to reduce the race-crime link to nothing. Crowding doesn't pass the smell test as a crucial cause. Hong Kong and Japan are crowded for example, but their rates of homicide are low.

    A finding that is much more common than what Centerwall reports is that controlling for concentration of single-parent families reduces or eliminates the race-violence link. Of course, researchers assume this proves social causation. They don't even address that possibility that family structure could be linked to biology.


Meta-analysis of clinical trials: Eat walnuts

I am always looking for easy eating choices that are good for you. This new meta-analysis of 26 clinical trials looked to see if walnuts ma...