Changing Inequality In The Profile Of Victims In Scotland From 2008-09 To 2017-18

A Crime Drop For Whom? Changing Inequality In The Profile Of Victims In Scotland From 2008-09 To 2017-18 Presented by: Mr. Matthews Date: 26 April 2022 Implications - Data hard to find - case for simplification - No evidence of cronyism. But: - • We only included current appointments • We only included director appointments • We did not (could not?) include personal relationships A Director Daisy Chain: • Independent from management, but not from system. Motivating question - The last thirty years has seen a huge drop in crime in Scotland, the rest of the UK, the USA, Australia and across Europe (van Dijk et al. 2012) - Victimization has fallen by almost 40% in Scotland between 2008/09 and 2017/18 (Scottish Government, 2019) - One of the big questions: has crime fallen for all groups in society equally (Hunter and Tseloni, 2016; McVie et al., 2020)? - Or put another way, has victimization inequality changed over the crime drop? Measuring inequality is hard - By victimization inequality, we basically mean that some groups in society have higher victimization rates than other groups - But any attempt to measure inequality between two (or more) groups relies on a lot of design decisions: - • Symmetrical measure? • Absolute or relative risk? • If an asymmetrical measure, which group should be the reference category? • Do we want to adjust for other confounding factors? This all gets harder when your outcome is the product of a statistical model - you need to care about the uncertainty in your results Why Care About Uncertainty? - We only have a sample of people who respond to victimization surveys, so we're extrapolating to the population as a whole - In general, any 'best guess' result - so, any statistical results - should be accompanied by an estimate of the precision of the guess - If we don't do this, our analysis of changing inequality can only really tell us if change is up, down or exactly equal - Often the most faithful representation of the evidence we have is "we don't know" or "we need more information" - This focus on uncertainty sounds like a bad thing, but can open up avenues for new research In review - We know that victimization has fallen overall in Scotland - Have patterns of victimization inequality in Scotland changed between 2008/09 and 2017/18, factoring in estimation uncertainty? Measuring victimization inequality in Scotland - We take survey responses to the Scottish Crime and Justice Survey for each sweep between 2008/09-2017/18 - The survey is managed by Scottish Government and recruits a nationally representative sample of people in Scotland, asking them questions about their experiences of victimization and the criminal justice system - Prior to 2008 the survey used a different design and so the data are not comparable Measuring victimization inequality: Dimensions of inequality - Inequality is multi-dimensional, not just financial - we need to look across multiple dimensions (Hunter and Tseloni, 2016) - Literature is pretty fragmented, but important dimensions include age, gender, ethnicity, disability, area characteristics (deprivation and region), financial hardship and tenure (see variously Nilsson and Estrada, 2004; Hunter and Tseloni, 2016; Herranz de Rafael and Fernández-Prados, 2019; Ganpat et al., 2020; Grove et al., 2012; Rossetti et al., 2016) Data - Data: combined SCJS sweeps from 2008/09-2017/18 (n = 79,608) - Outcome: number of victimizations of all crime types (0 = 82%, 1 = 11%, 2+ = 7%) - Predictors: • Age (16-24, 25-44, 45-64, 65+; 33, 32, 21, 14% of sample respectively) • Gender (male/female; 48, 52%) • Ethnicity (white/minority ethnic; 97, 3%) • Disability (no/yes; 74, 26%) • SIMD (15% most deprived/other; 85, 15%) • Urban/rural (82, 18%) • Financial hardship (a big problem or impossible to access £100 in another; 89, 9%, 2% missing) • Tenure (other/social renting; 78, 21, 1% missing) Estimating victimization rates - We use statistical models to 'predict' the average victimization rate for each sociodemographic group in each year (Gelman and Pardoe, 2007) - Social groups may be different in lots of ways: for example, people who live in areas of high deprivation may also be more likely to experience financial hardship, and both might be related to victimization - We can use a statistical model to estimate an 'independent' effect of the relationship between e.g. living in an area of high deprivation and victimization by statistically 'adjusting' for differences in other characteristics - We calculate 'adjusted' rates, as well as unadjusted comparisons - For the adjusted victimization rates we make a statistical counterfactual for each person in the dataset, for each variable, and then calculate the average victimization rate for each group we're interested in (I'm happy to say more about this in the questions!) - We measure victimization inequality as the ratio of these estimated victimization rates between the levels of each predictor; i.e. each of our socio-demographic groups What have we learned? - Victimization has pretty clearly fallen for every group - except for minority ethnic people where there's a lot of uncertainty in our estimates - This uncertainty is a serious limitation of our current understanding of victimization in Scotland I think - Young people have seen the largest crime drop. That's the same thing we see in convictions data (Matthews and Minton, 2018) - Large effect of the statistical adjustment on estimates by disability and ethnicity confounding by age? Reflections, limitations and future directions - All this is conditional on our model specifications! High uncertainty in victimization estimates for minority ethnic people: an analytical problem as much as a data problem? - Area deprivation, tenure and financial hardship all have a lot smaller 'independent' effects: because in real life they're not independent? - Ironically, all else being equal fewer victims of crime mean that it's harder to understand victimization through the SCJS, particularly for minority groups