Potential confounding factors ATM phosphorylation at the Probation Services level include seasonality, probation staff may also be influenced by their perceptions/knowledge of individual factors above and this may in turn influence the allocation to care farm or comparator sites. As allocation decisions may be based on some of these factors, confounding by indication will need to be addressed in the planned follow on study. This will be carried out through either propensity (probability of being allocated to a care

farm) matching, or cases and control, or adjustment by propensity scores in the outcome models. The pilot data will assess feasibility of collecting information on these potential confounders and provide an initial examination of their relevance to the allocation decision by testing the propensity methods. Analyses Feasibility and acceptability outcomes will be reported descriptively. The correlation between CORE-OM and other secondary measure scores for the same person will be estimated from the pilot data. The estimate and

its variability of the primary outcome measure will be used in the sample size calculations for the follow-on study. Additionally, the differences in the outcomes between those offenders at care farms and other locations will be estimated from the pilot data. Two potential issues need to be addressed in the statistical analysis. First the outcomes are to be measured at multiple time points, therefore individuals may vary in their number of measurements due to attrition and there is likely to be correlation in an individual’s outcomes over time. Second, as the study includes three sites there is potential for clustering of outcomes and other factors for individuals within each site. To account for these issues multilevel models will be used with time points nested within individuals and individuals nested within sites. Using multilevel models therefore accounts for missing data at particular time points, correlation in outcomes for an individual and account for potential clustering between sites. Exploring the pilot data using these approaches provides

an estimate of the various relationships to inform the follow-on study analysis plan. If differences in outcomes are found between care farms, appropriate adjustment in the sample size of the main study will account for the clustering/site effect (ie, Batimastat the intracluster correlation coefficient (ICC)). The results from studies identified in the literature review will also be drawn on for sample size calculations (including ICC estimation) for the follow-on study, incorporating a sensitivity analysis framework to explore the impact of the variation of estimates from previous studies on the subsequent sample size calculation.54 Health economics component As this is a pilot study, the economic analysis will be exploratory.