Afternoon Seminar on Statistics in Law on January 29th 2018

The Statistical Society is organizing an afternoon seminar on “Statistics in Law” in collaboration with the Law School at the University of Eastern Finland. The seminar is held at the Language Centre (Kielikeskus) at the University of Helsinki (Fabianinkatu 26, 3rd floor) in the Juhlasali lecture hall and is open to all interested parties.


To ensure adequate catering, please send a short e-mail to confirming your participation by the 22nd of January.


12.00 opening of the seminar
12.10 presentation I + discussion
12.50 presentation II + discussion
13.30 presentation III + discussion

14.10 coffee break

15.00 presentation IV + discussion
15.40 presentation V + discussion
16.20 closing of the seminar.


Mikko Aaltonen: Social disadvantage and crime in Finland: contrasting results from between-individual and within-individual models

Finnish criminal policy has traditionally relied on the notion that social disadvantage causes crime, and thus prevention of social exclusion should be one of the focal points of effective crime prevention. Analyses of socioeconomic backgrounds of convicted offenders appear to confirm this hypothesis: offenders often lack educational qualifications and have weak ties to the labor market, leading to low incomes and debt problems. However, for a variety of reasons, the causal nature of these associations has proven difficult to establish. In this presentation, I present key results from a series of published and ongoing articles that have used individual-level register-based panel data and attempted to overcome some selection issues inherent to the study of these topics. Overall, these studies suggest that within-individual change in employment and social disadvantage tend to predict changes in property crime, but not violent crime. I conclude by discussing some recent studies that have dealt with selection bias in a convincing manner, and whether such studies could be feasible in Finland.

Pekka Pere (joint with Tuomas Lahti and Mika Sutela): Sentences and prosecutors’ demands for aggravated drunk driving in Finland.

Sentences and prosecutors’ demands for aggravated drunk driving are categorised into three classes: The sentence is more lenient than, is compatible with, or is harsher than the prosecutor’s demand. The probability of a sentence falling into one of the three ordered categories is explained by a cumulative logit model. The following circumstances affect the probability of a more lenient or harsher sentence, in decreasing order of importance: driving a truck, facing at least four counts, having a legal assistant, and being present in the trial. The hypothesis that factors known by the prosecutor, at the time of writing the demand, should not systematically affect sentences is refuted. The judges assess circumstances differently than the prosecutors. The prosecutors’ role is nevertheless prominent in the sense that the sentences follow, to a great extent, their demands. Notable gender effects of the actors in the courtroom are found.

Mika Sutela: Multilevel modelling of sentencing: Variation between individual judges and prosecutors in the severerity of punishments.

According to section 6 of the Constitution of Finland, everyone is equal before the law. In sentencing equality means that, in principle, the same criminal act should be sentenced by the same punishment regardless of where the act has been committed and who gives the judgment. In this multilevel study, criminal sentencing decisions is analyzed with linear mixed models. In particular, the variation in the severity of punishments between individual judges and prosecutors is examined. The research data consists of the aggravated drunk driving (1.20- ‰) cases (N = 477) in Finnish district courts during years 2006–2010. With the mixed models it is possible to take into account the hierarchical and nested structure of sentencing data. For instance, individual judges or prosecutors are nested with courts, and criminal cases are nested with individual judges and prosecutors. In the analysis, judges and prosecutors are handled as random effects, as well as district courts and years. Individual characteristics of courtroom actors and procedural/organizational/community-level factors are handled as fixed effects. Preliminary results show that legal factors are significant but also the variation between prosecutors is rather large.

Brian Johnson. ‘Natural selection’: Conceptual issues and empirical strategies for treating sample selection bias in sentencing research

Analytical issues involving sample selection are pervasive in sentencing research.  This lecture will provide a basic conceptual overview of sample selection bias in criminological research, along with an introductory treatment of common statistical modeling approaches that are designed to deal with different types of sample selection.  It will focus primarily on the use of Heckman’s selection model and its commonly used alternatives in the context of sentencing research, and it will provide a brief discussion of ongoing and emergent issues in this area.

Jose Pina-Sanchez: Exploring disparities in sentencing using multilevel modelling: opportunities and pitfalls.

Unwarranted judicial disparities undermine trust in the criminal justice system. To tackle this problem a growing number of jurisdictions have followed the example of the US and implemented guidelines schemes seeking to constrain judicial discretion and promote consistency in sentencing. In this talk I will present the many opportunities that multilevel modelling techniques afford us to explore the nature and extent of sentencing disparities that cannot be explained by legitimate case characteristics; how findings from such models can be used to rethink sentencing guidelines; but also how the underlying assumptions made by multilevel models need to be carefully considered.

More technically, I will present a variety of multilevel models that I have implemented over the last four years to explore the topic of unwarranted disparities in sentencing. Ranging from the standard random intercepts model (used to explore the share of unobserved variability in sentence length due to systematic disparities between courts) to the more complex location-scale model (used to estimate different levels of unexplained within court variability), and cross-classified models (capable of distinguishing unobserved variability stemming from the judge level, the court level, and the interaction between them).