The death penalty and racial bias in the US

Eliana La Ferrara, Alberto Alesina 27 May 2011

a

A

African Americans are disproportionately represented among people condemned to death in the US. While they make up 12% of the national population, they account for more than 40% of the country's current death row inmates, and one in three of those executed since 1977. -- Amnesty International, USA Death by Discrimination - The Continuing Role of Race in Capital Cases, April 2003, page 1.

While factually correct, statements like the above can hardly be interpreted as evidence of racial bias because violent crime rates are higher among minorities than whites. Ideally we would want to find two identical crimes, one committed by a minority defendant and one committed by a white defendant and check whether the courts treat the two events differently. The problem is that it is virtually impossible to define two crimes as identical. Homicides can differ along so many dimensions that this empirical strategy is a non-starter.

The alternative is to define a test for bias which does not require controlling for all the different features of different crimes. This is what we do in a recent working paper (Alesina and La Ferrara 2011).

In the US, death sentences are imposed first by lower courts but each trial is automatically reviewed by superior courts. The first review is called direct appeal and if the death sentence is affirmed, the appeal can proceed to subsequent stages until it reaches the final review by federal courts (the habeas corpus stage).

Our test checks for racial bias at the trial stage, i.e. when the death sentence is first imposed. The idea is that superior courts are unbiased (or less biased) and they reverse judicial errors by trial courts. It is rare for higher courts to decide that an innocent person was sentenced to death. Most mistakes concern the severity of the punishment, typically failure to take into account mitigating factors that would lead to a sentence of life without parole, for instance, instead of death. Thus the reversal implies a lesser penalty than death.

The basis for our test is that, in the absence of bias, one should not observe systematic patterns in the distribution of mistakes and reversals depending on the race of the defendant and that of the victim. To be sure, the first court can make mistakes due to lack of accurate information, but these mistakes should not be systematically more frequent for certain combinations of defendant and victim race.

Instead we find that the number of reversals of is higher for cases involving a minority defendant and a white victim. Thus minority defendants are not discriminated against in general but only in cases involving white victims. This result holds strong both for the cases that reach the final stage of revisions and appeal, the habeas corpus, and for the full sample of cases in the first appeal stage, the direct appeal. For habeas corpus cases involving a minority defendant, the error rate was 37.5% if the victim was white and 28.4% if it was not white. The difference, equal to 9 percentage points, is statistically significant. The effect is driven by southern states, where the corresponding difference in error rates is 16 percentage points.

The validity of our test relies upon several assumptions. The first is about the behaviour of the higher courts. If these courts are unbiased and make mistakes uncorrelated with the race of defendant and victim, then our test is exactly specified. If higher courts are also racially biased in the same direction of the lower courts but less so, our test underestimates the amount of racial bias of first degree courts. Our test would overestimate the level of bias if higher courts actively discriminated in favour of minority defendants who killed white victims. We assess the plausibility of this interpretation empirically, exploiting differences in ideology across appeal courts. We build upon the premise that the judges who would be most likely to reverse discriminate in favour of minority defendants would be those more "left-leaning". Using various measures of political orientation of higher courts' judges, we do not find any evidence of this effect. Both left- and right-leaning judges exhibit the same pattern of reversal of first degree sentences: higher reversal for minority defendants who killed white victims compared to those who killed non-white victims, but not so for white defendants who killed white victims compared to white defendants who killed non-white victims.

Another assumption upon which our test rests is that possible unobservable characteristics, such as characteristics of the crime or quality of the evidence, are not systematically different across victims' races, given the defendant's race. To assess the plausibility of this assumption, we investigate whether our results are driven by easily observable characteristics of the crime or of the trial which differ between defendant-victim pairs, such as the type of crime (e.g., robberies versus other crimes), the characteristics of the victim (gender, status in the community), and the quality of legal counsel. We do not find any evidence of this.

In summary, testing for racial bias when crimes rates and other unobservable characteristics vary across races is extremely difficult. We have had to make several assumptions to develop a test, and such test is only valid under these hypotheses. Conditional on this cautionary remark, we do find evidence of racial bias against minorities killing white victims in southern states of the US.

References

Alberto Alesina and Eliana La Ferrara (2011) “A test of racial bias in capital sentencing” NBER Working Paper 16981.

Amnesty International, USA Death by Discrimination - The Continuing Role of Race in Capital Cases, April 2003, page 1.

a

A

Topics:  Frontiers of economic research Poverty and income inequality

Tags:  US, Racism, death penalty

Nathaniel Ropes Professor of Political Economy, Harvard University; and Research Fellow, CEPR

Professor of Economics, Bocconi University and CEPR Research Fellow