Guidance

Risk assessment methodology for good state-funded schools

Updated 14 July 2023

Applies to England

This is the risk assessment process that Ofsted uses when scheduling inspections of good state-funded primary and secondary schools. This normally takes place in time for the start of the third school year after the most recent inspection.

Introduction

We use risk assessment to make sure our approach to inspection is proportionate and to focus our efforts where they can have the greatest impact. The risk assessment model helps us to identify schools graded as good that may potentially decline at their next inspection, and to help proportionally allocate graded and ungraded inspections. Schools at greater risk of decline are more likely to receive a graded inspection. It is important to note that the risk assessment process is not used to pre-judge inspection outcomes.

The risk assessment process

Risk assessment has 2 stages:

• stage 1 involves an assessment of each school based on analysis of school-level performance and contextual data
• stage 2 involves a review of a wider range of available information

At stage 2 Senior His Majesty’s Inspectors (SHMI) review the potential types of inspection to make sure the most appropriate type (if any) is carried out.

Stage 1: analysis of school-level data

Machine learning

We use ‘supervised machine learning’ to predict the probability of a good school declining to ‘less than good’ at its next inspection. Machine learning is a way of getting computers to make decisions that have not been explicitly programmed, by learning from past experiences. A common application is classifying items into two or more groups.

In a typical application of machine learning, there will be a large dataset called ‘training data’, for which we already know the groups that the items belong to. We use this to train the machine learning model to distinguish between unknown items. For example, a ‘spam’ filter can be trained by giving it emails those users have marked as spam and non-spam emails, and the algorithm works out the differences.

Machine learning applied to inspection outcomes

To develop the machine learning model, we use a training dataset that includes inspection outcomes from the previous academic year. The model then retrospectively predicts the previous year’s known inspection outcomes from data relating to the schools’ characteristics.

Data sources in the training dataset include the latest available:

• schools contextual data
• performance (progress and attainment) and subject entry data
• school workforce census data
• qualifying complaints about schools
• data collected from Ofsted’s Parent View questionnaire

The machine learning model ‘learns’ from the training dataset, creating a series of decision trees to classify schools into those that are more likely to decline to ‘less than good’ and those more likely to remain good or improve to outstanding. We use the most accurate of these decision tree models to predict the outcomes of inspections. We then test this model on ‘test data’ consisting of inspection outcomes from a different time period to the ‘training data’, to ensure its validity in predicting inspection outcomes.

Once the model is trained, we apply the latest school performance and contextual data to create risk scores. These estimate the likelihood that a school will decline at its next inspection. This is our ‘raw risk score’, which takes a value between 0 and 1.

It is important to note that:

• the risk model is only used at stage 1 of the risk assessment process and SHMI reviews follow on from this before we finalise the selections
• risk model scores do not influence inspection judgements

Inclusion criteria for stage 1 of the risk assessment

Stage 1 applies to good primary and secondary schools only. It does not apply to nursery schools, pupil referral units, alternative provision or special schools. Schools with fewer than 11 pupils at the relevant key stage are also not included in stage 1 of the risk assessment.

Stage 2: further review

SHMIs in each region review the information provided by stage 1 of the risk assessment process. Before finalising their selection, they also consider:

• the outcomes of any inspections that we have carried out since the last routine inspection
• qualifying complaints about the school referred to us by parents
• statutory warning notices
• data on potential gaming by schools, including details of schools with exceptionally high levels of pupil movement
• any other significant concerns that are brought to our attention

Timing of inspections

For further information on how schools are selected for inspection and the timing of inspections, please refer to the school inspection handbook.