Scottish Prison Population Projections: February 2024 Edition

4 Introduction to Prison Population Projections

This section provides an overview of the assumptions and scenarios developed to project the prison population and explains how the model’s previous projections are validated for robustness against actual prison population figures in recent months. The first issue of the Scottish prison population projections^[21] provides a more detailed overview on the modelling technique used.

Model Assumptions and Scenarios

The scenarios on which the projections are based on a variety of assumptions about how the rate of transitions into and out of the prison population might change. These can be influenced by trends and planned changes to the court system, including: court capacity changes, increased court conclusions rate, increased remand arrivals, remand/bail mix and prioritisation of remand case progression through courts. Departures from the sentenced population are estimated using a combination of court disposal data and snapshots of the prison population.

Due to the level of the inflows and outflows which contributed to growth in the sentenced and remand populations in early 2023, and the abrupt seasonal population changes which usually occur between December and the beginning of January ^[22], additional scenarios were developed and added to the November 2023 projections. In the first issue of the Scottish prison population projections^[23] publication only low, central and high court throughput scenarios were considered. In addition to these three court throughput scenarios, six new scenarios featuring varying remand arrivals (low, central and high) combined with court throughput variation were introduced in the last publication^[24]. A full breakdown of all nine scenarios is given in table 1. This set of scenarios has been developed to account for uncertainty in future remand arrivals, especially in early 2024. Additionally, case registrations in Sheriff Summary and Sheriff Solemn were higher in Q2 2023/24 (July to September 2023) than during Q2 2022/23⁷ (July to September 2022) and increased case registration can contribute to elevated remand arrivals.

Three variations of remand arrivals assumptions have been used for the scenarios: central, higher and lower. The central remand arrivals scenario assumes remand arrivals will be similar over the next few months to what they were between January 2023 and December 2023. The higher remand arrivals scenario assumes the rate will be similar to the levels from March 2023 to August 2023, and the lower scenario assumes remand arrivals around the lower levels between October 2022 and March 2023.

Furthermore, to help with understanding how sensitive the prison population may be to variations in court case conclusion rate, three variants have again been included in the modelling - central, higher, and lower. The “central” court throughput scenario variant assumes that the conclusion rate per court will be similar over the next few months as it has been between January 2023 to December 2023. The “higher” scenario assumes the average case throughput per court will be slightly greater than it has been over the same period, and the “lower” scenario assumes that the average case throughput per court will be slightly smaller. The nine scenario variants shown in table 1 include all the possible combinations of the above variations of court throughput and remand arrivals.

Since the progression of the justice system’s recovery (e.g., rate of reduction of scheduled trials) impacts the model’s assumptions and longer-term predictive power, the projections cover a limited period, from January 2024 to June 2024.

A review of previous trends shows that a general seasonal pattern tends to occur at the end of the year. This results in atypical court throughput and remand arrivals during the months of December and January. To account for this, in the latest projections the assumptions for the month of January 2024 are sampled from January 2023.

Model Quality Assurance

The modelling scenarios from the previous publication can be compared against the actual population data to check the accuracy of the projections and confirm the model’s suitability for providing reliable projections. Figure 12 shows the October 2023 projections from the microsimulation, which were published in November 2023^[25]. Even after accounting for a wide variety of uncertain dynamics in the system, it may be seen in Figure 12 that the projected levels of remand, sentenced and total populations have reasonably accurately aligned with the actual numbers. It is important to note the projections cover the end of the year where a seasonal pattern in the populations is observed. In this case the sentenced and total populations both increased throughout December and then decreased toward the end of December. The remand population generally decreased throughout December but then increased as the month concludes.

Solemn procedure throughput was high in recent months^[26] and contributed to the growth in the sentenced population in particular trending towards the upper projection bound during mid-December, before dipping towards the centre of the range at the very end of December.

The remand population ended December at a point which was slightly lower than the centre of the projection range.

The actual total population at the end of 2023 was close to the middle of the range (since the projected range and the actual population both fell in December). The data and assumptions, based on December 2022, were representative of actual demand and throughput for December 2023.

Several key modelling assumptions about inflows and outflows are defined as the average over a month and hence can’t reflect the abrupt changes in the population which tend to occur from one day to the next around the end of December. Nevertheless, the model is suitable for projections over several months. The assumptions (e.g., remand arrivals) for December 2023 were modelled directly from the average of the previous December in 2022. The full projection period is still to be completed but so far it is aligning satisfactorily, inclusive of the seasonal effects.

To check the model further, back-casting is used to retrospectively compare the actual prison population for the past few months against a projection generated by the model based on actual monthly court throughput data. The back-cast eliminates uncertainty about the majority of the assumptions, so if there was a difference between the back-cast and the actuals it may indicate technical deficiencies in the model. The recent back-cast projections presented in Figure 13, show that the back-cast from July 2023 to December 2023 is largely accurate.

Microsimulation Model Limitations

The model relies on the availability of a large amount of frequently refreshed high-quality data about court activity and prison populations, some of which can be resource intensive to obtain and process.

The model does not currently simulate flows for different crime-types, so crime-based trends are not explicitly modelled. However, there are plans to develop the model further and include case-mix in future modelling.