On Racial Bias in Criminal Sentencing

Sometimes, people say things like “Black Americans get harsher criminal sentences than white Americans even when race is the only thing differentiating the black and white defendants”. This fact is then pointed to as evidence that racial bias is at work in the American criminal justice system, and American society at large. In this post, I will argue that the available empirical research does not justify this claim and that, in fact, the empirical evidence more favors the view that there is no racial bias in criminal sentencing.

Pre-trial Outcomes

The standard sort of research that is cited in this area consists of studies which compare legal outcomes for criminals after controlling for observable differences between them such as the type of crime they committed and their criminal history. These studies can be divided into those that examine pre-trial and post-trial outcomes.

With respect to pre-trial outcomes, Wu (2016) meta-analyzed 36 studies on the effect that race has on the probability that a defendant would be fully charged. Wu argues that these pre-trial decisions are very importance since 80% of state cases and 90% of federal cases never actually go to trial. Wu finds that black defendants are 9% more likely than white defendants to be charged.

In the moderator analysis, Wu produces several interesting findings. First, this effect is only found in the South. This finding is consistent with out intuitions about how racism is distributed across America. However, several other findings from the moderator analysis cast doubt on the idea that this meta-analysis is detecting real bias.

First, the strength of this effect has not changed with time contrary to what we might expect if racial animus was its cause.

Second, it was found that controlling for crime type and criminal history actually increased the size of this effect, a finding that is hard to make any sense of.

Thirdly, and most importantly, no bias was found in studies that reported their standard error. The standard error is a statistic needed to put an result into this meta-analysis. Some studies reported them and others did not. When the standard error was not reported, Wu estimated what the standard error probably was. As Wu writes “Estimated standard errors likely overestimated or underestimated the effect of race and ethnicity on prosecutorial outcomes.”

The results of this meta-analysis suggest that Wu’s estimation process inflated the racial bias effect. The bias reported in the meta-analysis is driven by the 11 studies that failed to report their standard error. The best explanation for the fact that no bias was found in that 25 studies that reported their standard error is that the finding of bias when all the studies are combined is a statistical error. Thus, I don’t think research on pre-trial outcomes is good evidence for the existence of racial bias.

Post-trial Outcomes

Turning to post-trial decisions, Mitchell (2005) conducted a meta-analysis in which various sentencing outcomes were considered (e.g. prison vs no prison, sentence length, etc.) and converted to a common statistical measure. The results were then divided into those concerning state sentencing and federal sentencing.

With respect to state crimes, black Americans were 28% more likely than white Americans to receive a harsh sentence. However, this gap was reduced to 14% in studies that controlled for things like criminal history and crime type. There was also evidence that studies finding larger effects were more likely to be published since the elevated risk in unpublished studies was only 14% even before applying any controls.

Turning to federal sentencing, the total elevated risk for a harsh outcome was 15% and this effect was not statistically significant. In this case, the effect was actually larger in unpublished studies, but, unlike in the case of state sentencing, these studies were highly unreliable exhibiting a confidence interval that stretched from 7% to 136%. Given this, this result is not serious evidence of publication bias in the research on federal sentencing.

For these findings to be compelling, we would need to see a meta-analysis which both controls for publication bias and only includes studies that control for obvious differences in crime type and criminal history. Since both of these moves seem to reduce the significance of the effect, it is remains unknown whether the result would remain statistically, or practically, significant if both moves were done at once. Given this, until a better meta-analysis is conducted, I don’t think we can form confident views about what this research is actually finding.

Moreover, research on both pre and post trial outcomes suffer from a more fundamental flaw which is that they omit many variables which might differ between black and white criminals and which might explain their differing legal outcomes.

For instance, this research fails to control for differences in the way in which crimes were carried out within the same crime type. As an example, assaults can differ both in their degree of seriousness and in the motives that caused them. These variables might in-turn differ, on average, between black and white criminals in a way that leads to white criminals receiving preferable legal outcomes.

Another omitted variable is courtroom behavior. How criminals act in a courtroom, and especially how remorseful they seem, might influence their sentencing. Yet, this is uncontrolled for in studies on racial bias.

Mock Juries

Mock jury experiments get around these problems, though not without bringing in new ones. These experiments have people act as juries and decide the sentencing outcomes of hypothetical defendants. Typically, people are assigned to two groups and in one group the defendant is black while in the other the defendant is white. No other difference between these defendants exists and so there can be no omitted variables.

Mitchell et al. (2005) analyzed data from 34 studies in which people acted as jurors and voted on whether a given defendant was guilty.  It was found that whites have nearly no bias in such decisions while the black people exhibit an in-group bias that is 15 times larger than the minuscule bias seen among whites.

With respect to deciding sentence length, bias among white people was still practically insignificant, while the bias among blacks was 7.6 times greater.

Devine and Caughlin (2014) conducted a similar meta-analysis finding that white jurors had no bias against black defendants but did have a modest bias against Hispanic defendants. Black jurors, once again, exhibited a pro-black or anti-white bias.

So, these studies might be taken as suggesting that even if bias is found in observational data, this is because observational data cannot control for all the differences between black and white criminals while experimental research can.

On the other hand, someone might question whether this experimental research is applicable to the real world. It could be that the sort of people included in these experiments differ from the sorts of people who make up real world juries in a way that is relevant to their degree of racial bias. Or is might be that people simply don’t act the same way in experiments as they do in the real world, again, in a way that is relevant to racial bias. These are logical possibilities, but I am not aware of any evidence suggesting that they are true.

Black Judges and Black Lawyers

Another line of evidence worth mentioning concerns research on racial bias among black judges and black lawyers. These studies offer more evidence that racial bias is not at play in the criminal justice system.

Turning first to lawyers, Libgober (2019) conducted experiments showing that black sounding names received fewer call backs from lawyers than did white sounding names, a problem which could impact a criminal’s legal outcomes, but this tendency was found to be the same among white and black lawyers.

Moving on to judges. Steffenmeier et al. (2001) analyzed data on 40,000 sentences that were given in Pennsylvania between the years of 1991 and 1994. The impact that being black had on a person’s sentence was found to not significantly differ between black and white judges.

Similarly, Ulhman (1978) analyzed data from 35,000 trails which took place between 1968 and 1974. It was found that black and white judges exhibited equal degrees of racial bias. This was true both with respect to whether a defendant was found guilty and with respect to their sentence length.

The fact that the supposed racial bias seen among white judges is also found among black judges strengthens the case that this so called “bias” is actually due to variables which differ between black and white criminals but which are omitted from the standard research literature. Of course, it is possible that black judges are racist against black people to the exact same degree that white judges are, but there is no empirical evidence suggesting that this is so, and in the absence of evidence we should assume this is false since black people in general do not exhibit an anti-black bias.

Conclusion

In summary, research on pre-trial outcomes does not seem to favor the view that racial bias is at play. Meta-analyses may claim otherwise, but a close look at their moderator analyses suggests that the bias they claim to detect is actually a statistical error. Research on post-trial outcomes might suggest a small anti-black bias but we can’t tell both because there are no meta-analyses that utilize strong controls and correct for publication bias at once, and because this literature in general ignores variables which might contribute to racial differences in legal outcomes. By contrast, these issues do not plague mock jury research, or research comparing the behaviors of black and white judges, and both of these lines of evidence suggest that no racial bias is at play. Thus, I think the available empirical evidence makes the existence of significant anti-black bias in sentencing less probable than its non-existence.