Scottish Prison Population Projections: July 2026

This annex provides further technical detail on the microsimulation model used to produce the prison population projections. It explains how the model represents the prison population, how daily transitions are simulated, how probability-based decisions and repeated runs are used to generate projection intervals, and the main limitations of the approach.

Model structure and daily simulation process

The microsimulation model represents the prison population as a set of individual records. Each record has attributes used in the simulation, including legal status, sentence length, planned release date, arrival date, sentence type, whether the individual entered sentenced custody from remand, and other information needed to model transitions between remand, sentenced custody and the community. The model then updates these individual records one day at a time over the projection period.

At the start of each simulation run, the model is initialised using the latest available prison population data. For people already in custody, the model uses observed information where available. Where some attributes are not available directly, such as the start date of a current spell in custody, the model estimates these using available prison population history, sentence length and planned release information.

Each simulated day begins with the prison population carried forward from the previous day. The model then applies a series of daily processes that can add people to the prison population, move people between remand and sentenced custody, or remove people from custody. These include bail releases, court conclusions for people on remand, sentenced arrivals from the community, new remand arrivals, Home Detention Curfew decisions, sentenced releases, parole releases and returns to remand. The model records daily totals for remand, sentenced and total populations, as well as flows between community, remand and sentenced custody. Figure 18 provides a simplified overview of these daily transitions.

Figure 18. Overview of daily transitions represented in the microsimulation model.

Initial prison population

The initial population is generated from individual-level prison population data for the model start date. The model identifies each person’s legal status and classifies the sentenced population into sentence-type groups, including short-term sentences, long-term sentences, and cases where sentence length is missing or not applicable, such as life sentences and Orders for Lifelong Restriction.

For people on remand at the start of the projection, the model estimates the start of the current remand spell where possible by looking back through recent prison population data. For sentenced prisoners, the model uses sentence length, planned release date, sentence expiry information and previous legal status to estimate the current sentenced spell and whether the person entered sentenced custody from remand or directly from the community.

The model also estimates relevant future dates for some subgroups, such as parole review dates for long-term sentenced prisoners. These dates are then used during the daily simulation to determine when parole or release decisions may occur.

Daily simulation steps

For each simulated day, the model carries out a sequence of updates to the prison population. The order is important because decisions earlier in the simulated day can affect who is available for later decisions.

The main daily steps are:

• Court conclusion volumes are simulated. The model uses court disposals assumptions, adjusted for court capacity and scenario assumptions, to estimate the number of court conclusions by court type, conclusion type and custody status. These daily volumes are generated stochastically.

• Bail releases are simulated. The model estimates releases from remand to the community using recent bail liberation data, with separate estimates for summary and solemn procedure.

• Remand case conclusions are simulated. Where cases involving people on remand are concluded, the model selects individuals from the remand population and determines whether the conclusion results in a custodial sentence, whether all remand matters are resolved, and whether the person moves to sentenced custody, remains on remand, or leaves custody.

• Sentenced arrivals from the community are simulated. For people not already on remand, the model estimates whether court conclusions result in custodial sentences and then generates new sentenced arrivals from the community where appropriate.

• New remand arrivals are simulated. The model estimates daily remand arrivals using sampled historic remand arrival rates and assigns new individuals to the remand population.

• Release processes are applied. The model applies planned sentenced releases, Home Detention Curfew, parole and returns to remand. Individuals may leave the prison population or, in some cases, move from sentenced custody back to remand.

Decision points and probability models

Several modelled transitions require a probability-based decision. These include whether a court conclusion results in custody, the sentence length assigned to a custodial sentence, whether all remand matters are resolved, whether an eligible person is released on Home Detention Curfew, whether a person is released on parole, and whether a sentenced release results in a return to remand.

Some of these probabilities are estimated directly from recent historical data. For example, court disposals data is used to estimate the likelihood that a conclusion results in custody, broken down by procedure type, arrival status and conclusion type. Sentence length distributions are also derived from recent sentencing data, with different distributions used depending on whether the person was on remand or not, the procedure type, and whether the conclusion followed a trial or non-trial diet.

Other probabilities are estimated using decision-tree models. These are used where outcomes depend on combinations of characteristics. For example, decision-tree models are used to estimate whether a concluded remand case resolves all remand matters, and whether a person leaving sentenced custody returns to remand. These models allow the simulation to reflect observed relationships in the data without manually specifying every possible combination of factors.

The model also includes assumptions about specific release processes, including Home Detention Curfew, parole, bail backdating and changes to short-term prisoner release arrangements. These assumptions are applied probabilistically where the individual-level eligibility or outcome is not fully observed in the model data.

Stochastic simulation and prediction intervals

The model is stochastic, meaning that some inputs and decisions vary between simulation runs. For example, daily remand arrivals and court conclusion volumes are sampled from probability distributions based on recent observed data. Individual outcomes, such as whether a case results in custody or whether a person is released on parole or Home Detention Curfew, are also sampled probabilistically.

Each scenario is run repeatedly. In each run, the same scenario assumptions are used, but the stochastic elements can produce different population paths. The distribution of results across runs is then used to calculate central projections and prediction intervals for the remand, sentenced and total prison populations.

The 50%, 75% and 95% prediction intervals show increasingly wide ranges of modelled outcomes across repeated runs. The 50% interval shows the central range of outcomes, while the 95% interval shows a wider range that contains a larger proportion of simulated outcomes under the assumptions used. The wider intervals therefore reflect greater modelled uncertainty, but they should not be interpreted as guarantees that the actual population will fall within those ranges. The published projection range combines uncertainty within each scenario with the differences between the scenario assumptions.

Monthly projection figures are presented as average daily populations. This means that the monthly estimate is calculated from the modelled daily population values within that month, rather than from the population on a single date.

Scenario generation and court throughput assumptions

Scenario assumptions are generated before the microsimulation is run. For this publication, scenarios vary the assumed rate of court case conclusions, while remand arrival assumptions are not varied across the scenarios. The scenario generator creates combinations of projection length, conclusion-rate scenario and remand-arrival scenario, and writes these assumptions into scenario-specific files used by the microsimulation.

The central scenario is based on court conclusion rates observed over the recent twelve-month period. The higher and lower scenarios apply higher or lower conclusion-rate assumptions to assess how sensitive the prison population is to changes in court throughput. The scenario generation process also incorporates court capacity assumptions, including whether the number of courts is assumed to change over the projection period.

Court conclusion assumptions are converted into model inputs by estimating expected conclusion volumes by stage, procedure type, conclusion type and custody status. These are then used during the daily simulation to generate court conclusion events.

Policy and release-date adjustments

The model includes specific adjustments for recent short-term prisoner release policy changes. For STP40 and STP30, the model applies changes to planned release dates for eligible short-term prisoners, including adjustments for validation runs that cross the implementation dates of those policy changes.

For sentenced arrivals generated within the simulation, release dates are calculated using sentence length, arrival date, sentence type and the applicable short-term prisoner release arrangements. Where relevant, the model applies assumptions about whether an individual is affected by short-term prisoner release changes and whether bail backdating applies.

Home Detention Curfew is represented by identifying eligible short-term sentenced prisoners on each simulated day and applying an assumed likelihood of release. Individuals considered for HDC are recorded so that they are not reconsidered repeatedly in the same run.

Model limitations

The model necessarily simplifies some aspects of the justice system. It represents key flows between community, remand and sentenced custody, but it does not currently model every subgroup or case characteristic separately. For example, it does not currently simulate separate flows by crime type, gender, age group, sheriffdom, individual prison establishment or detailed case mix.

The model also combines datasets collected by different organisations for different operational purposes. Some court data is event- or case-based, while the microsimulation estimates person-level prison population flows. As a result, modelling assumptions and adjustments are required to translate these inputs into consistent flows within the simulation. The data preparation scripts include comparisons between court disposals and observed prison population flows, with adjustments applied where necessary to improve consistency between data sources.

Some probabilities used in the model are estimated from recent historical data and are assumed to remain broadly applicable over the projection period. If behaviour changes, for example through operational changes, policy changes, changes in case mix, or changes in court throughput, the model may not fully capture these effects unless they are explicitly included in the assumptions.

The model estimates demand for prison places under the stated assumptions. It does not explicitly model operational constraints within the prison estate, such as establishment-level capacity, population management decisions or practical limits on the number of people who can be held safely. Projected population levels should therefore be interpreted as estimates of potential demand rather than predictions that those levels will necessarily be reached in practice.

Uncertainty increases over longer projection horizons. This is because daily uncertainty accumulates over time, and because assumptions about future court throughput, remand arrivals, sentencing outcomes and release behaviour become less certain further into the future. The extension to a nine-month projection horizon reflects recent methodological development and validation work, but the later months should still be interpreted with greater caution than the earlier months.