3. Comparing reconviction rates across local authorities
Estimating reconviction rates for local authorities
In Reconviction rates in Scotland, we have historically only published reconviction rates for local authorities based on information for offenders convicted in courts that fall within that local authority area’s boundary. This is because it was the only information on local authority that we could obtain. However, the areas that courts serve do not exactly match local authorities; and offenders may be convicted in a court located in a different administrative area to where they live, yet they would be supervised in their area of residence (see Annex A12 and the footnote of Table 12). The characteristics of offenders are also likely to vary across local authority areas, therefore such comparisons between areas should be treated with caution, and it is suggested that a method which takes these factors into account should be employed (see section 3.1).
To improve estimates of reconviction rates for local authorities, we recently started to collect data on the first half of an offender’s home postcode from Police Scotland, for example EH1 or G1. This data can then be used to match an offender to their home local authority. This information will be particularly important for local authorities who use these statistics for planning purposes, such as schemes to reduce reoffending, or estimating the number of offenders that social workers need to supervise in their area. Local authority reconviction rates based on offender postcodes are published for the third time this year, but due to incomplete postcode coverage, we will still publish reconviction rates based on court area until it improves. We recommend that the figures based on court area are still used as the definitive local authority reconviction rates.
Local authority reconviction rates based on court area
Reconviction rates vary across local authority groups (based on the area covered by courts). Note that because some sheriff courts cover more than one local authority, we cannot distinguish between convictions in the different local authorities. Therefore some local authorities are grouped together, so that there are 24 groups rather than 32 separate local authorities. Index convictions in the High Courts are presented separately from local authority groups. High Court index convictions were included in the local authority where the High Court was located prior to the 2016-17 cohort bulletin so the figures here are not comparable with earlier bulletins (see revisions in Annex B32 for further information).
Table 12 shows that the highest reconviction rate in the 2017-18 cohort was for offenders whose index conviction was given in courts in Clackmannanshire (31.6%). Clackmannanshire also had had the highest number of reconvictions per offender on average along with Aberdeen and Aberdeenshire (0.59). Excluding convictions in the High Court, the lowest reconviction rate (12.3%), and lowest average number of reconvictions per offender (0.18), were for offenders whose index conviction was given at a court in the Shetland Islands. These are unadjusted figures which do not take account of underlying differences in population size and the characteristics of offenders in each area (see section 3.1 for comparisons of standardised reconviction rates which take these into account).
Reconvictions tend to fluctuate year to year for local authorities. Smaller local authorities tend to have larger fluctuations as they have small numbers of offenders. Small between-year fluctuations in the numbers of offenders reconvicted may lead to larger changes in the reconvictions in percentage terms compared to local authorities with larger numbers of offenders.
Local authority reconviction rates based on residence
Table 13 shows reconviction rates based on the local authority of offenders’ residence. This is achieved by matching the local authority to the first half of offenders’ postcodes. The local authority reconviction rates based on postcode are currently labelled as Experimental Statistics: Data being developed, as this analysis was only recently introduced and we did not have postcodes for a fifth of offenders (20%) with an index conviction in 2017-18. Postcodes may be missing because offenders have no fixed abode, but it may also be a recording issue.
DTTOs have a higher percentage of missing postcodes compared to other sentences, with 30% of offenders with an index disposal of a DTTO missing postcodes in 2017-18. This may relate to the personal circumstances of those given DTTOs. Custodial sentences also had a relatively high percentage of offenders missing postcodes, with 28% missing postcodes. This is not surprising as many custodial sentences counted here would have been recorded on the CHS before Police Scotland started sending us conviction data with postcode information. Sentences over 4 years have the highest percentage of missing postcodes, with 40% of offenders missing postcodes. Note that the data quality issues around the recording of postcodes only affects the local authority reconviction rates presented in Table 13 and does not affect any of the other reconviction rates presented in this publication.
Annex Table B1 shows the number and percentage of offenders with missing postcodes in each local authority group, based on the location of the court they were convicted in. It also shows the percentage of offenders living in the local authorities that are covered by the court areas, and the percentages that live in different local authorities to those covered by the court areas. This shows that there are significant percentages of offenders who are convicted in a court that covers a different local authority to where they live. Stirling had the highest percentage (31%) of offenders living in a different local authority to the court area where they were convicted. Note that percentages may be higher in other local authority groups but this cannot be determined due to the missing postcode data.
The group with missing postcodes had a higher reconviction rate and average number of reconvictions (29.9% and 0.56) compared to the national rates for Scotland as a whole (26.3% and 0.46). This suggests that there is possibly some bias in the recording of postcodes.
Reconviction rates based on postcode data varied between 32.8% for Aberdeen to 11.8% for the Shetland Islands (Table 13). However, as Annex Table B1 shows, there is variation in the percentages of missing postcodes between local authorities, so direct comparisons between local authorities should be treated with caution. The missing data may mean that the reconviction rates are over or underinflated, but we do not have enough information to know fully know the effects of the missing data on the rates. Also, different local authorities may have different mixes of offender characteristics, and small local authorities may experience greater fluctuations, which should be considered when comparing local authorities. The next section discusses these considerations in more detail (although those comparisons of local authorities are based on court area, the same factors would apply here).
3.1 Accounting for the variability between local authorities
Reconviction rates could be used to rank performance across different local authorities. However, there is an inherent problem in using this approach since it implicitly assumes that a difference in reconviction rates reflects a ‘real’ difference between local authorities. In reality, all systems within which these local authorities operate, no matter how stable, will produce variable outcomes in the normal run of events. In particular, outcomes in local authorities with smaller sized populations tend to vary more than those in local authorities with larger populations. The question we need to answer is therefore: Is the observed variation more or less than we would normally expect?
In this respect, it is better to use a method of comparison that takes account of inherent variability between local authorities. The funnel plot is a simple statistical method that takes into account the variability of different sized populations and so highlights whether there are differences that may be attributed to some other special cause.
Table 12 shows the average number of reconvictions per offender and reconviction rates for each local authority group (based on court area of conviction) and Chart 11 shows these reconviction rates against the number of offenders. The plot takes into account the increased variability of the local authority groups with smaller populations, where a small increase in the number of reconvictions may lead to a large percentage change in the reconviction rate. Rates for local authority groups which lie inside the funnel are not significantly different from the national rate, and we can then usefully focus on possible explanations for rates which deviate significantly from the national figure. In this case, the cut-off level for statistical significance is 95% (or two standard deviations from the mean): if there were no difference between local authorities groups apart from that which could reasonably be attributed to random variation, we would expect that 5% of the authorities (i.e. only 1 of them) would lie outside the funnel.
Chart 11 shows that Clackmannanshire, Falkirk, and Aberdeen and Aberdeenshire lie above the funnel, and so have a higher reconviction rate than expected. Argyll & Bute, Na h-Eileanan Siar, Shetland Islands, Stirling, and West Lothian lie below the funnel and so have lower rates than expected. Whilst this is useful for highlighting that there are practical differences in reconviction rates between each local authority group, even after taking into account differences in population sizes, it does not allow us to identify if this disparity is due to variation in the characteristics of offenders in each area or a variation in practices between different local authority groups. Different offender characteristics between local authority groups could include: age, gender, crime, disposal, deprivation, etc.
Chart 12 is standardised to take into account some of the differences between local authority groups attributable to the characteristics of offenders, such as the number of previous offences, sentence, gender, and age. It provides the standardised reconviction rates against the observed number of offenders minus expected number of offenders. Since all local authorities groups are within the funnel it suggests that the apparent differences in reconviction rates in Chart 11 are primarily attributable to either the variation in the characteristics of the offenders, the type of crime they committed, or the sentence they received, rather than differences in ‘performance’ between the local authority groups. This overall conclusion for all local authorities on the 2017-18 cohort is consistent with findings in the previous Reconviction Rates in Scotland publications.