Exploring the Nature of Ethnic Disparities in Sentencing through Causal Inference

Few principles are more fundamental to a liberal society than equality under the law, and few public acts epitomise that principle more clearly than sentencing hearings. Understandably, the study of ethnic disparities in sentencing has attracted vast research efforts. Empirical studies have shown how offenders from ethnic minority groups tend to receive harsher punishments than white offenders after committing similar crimes. These disparities have been documented in great detail; being corroborated across jurisdictions, offence types, and sentence outcomes. However, one key question remains: can such disparities be taken as evidence of discrimination?

Research based on real cases cannot randomise offenders by their ethnicity, hence, how do we know that those disparities are not due to unobserved relevant case characteristics? For example, we would expect to see differences in sentence severity for similar crimes committed by black and white offenders if white offenders are shown to plead guilty more often than black offenders. The analytical response to this problem has been to 'control for' any relevant case characteristics. But what if those differences are based on case characteristics - such as offender dangerousness - that cannot be easily measured, nor controlled for? And, what if those case characteristics are not neutrally defined but subject to potential discriminatory practices? These are important methodological questions that remain unresolved. Here, we suggest using the new sentencing datasets made available by Administrative Data Research UK, and some of the latest sensitivity analysis techniques developed by epidemiologists to overcome this methodological impasse.

Sensitivity analysis is used to test the robustness of findings in contexts where key research assumptions are likely violated. The most famous example of this dates back to the 'smoking cause lung cancer' debate, which dragged on for decades because of the absence of experimental evidence. This was until Cornfield et al. (1959) demonstrated that the effect of any relevant unobserved factors (e.g. genes predisposing to nicotine craving while simultaneously carcinogenic) ought to be unrealistically high in order to explain the observed associations between lung cancer and smoking. We suggest following a similar approach here. Rather than uncritically dismissing ethnic disparities because of their inability to make perfect 'like with like' comparisons, we pose the following question: what should the strength of the unobserved relevant case characteristics be to explain away the ethnic disparities observed? Estimating that threshold will allow us to make a more informed judgement regarding whether the observed ethnic disparities represent evidence of discrimination.

Beyond their academic merit, the questions to be explored in this project are of fundamental interest to all criminal justice agencies. Various elements of the project, including its research questions, have been co-designed in collaboration with representatives of the England and Wales Bar Association, Sentencing Council, Crown Prosecution Service, and Sentencing Academy, which feature amongst the project's external partners. As a result, our findings will be directly relevant and available to the key policy-makers in charge of monitoring and redressing ethnic disparities in our jurisdiction. Ultimately, this project will help enhance the public debate around ethnic disparities in the criminal justice system throughout the nation. Regardless of the findings obtained, undertaking an independent and unprecedentedly robust examination of ethnic disparities in sentencing will increase transparency in the criminal justice system, signalling integrity and contributing to restore public trust.