Search site

School of Geography

Agent-based Modelling for Education Planning

Supervisors: Dr Alison Heppenstall and Prof John Stillwell

Providing excellent educational facilities is at the heart of local and national government policy. However, planning for education provision is complex school population changes each year, as does the number of school age children.number of school age children in Leeds need for rationalisation. However,demanding that the school network again expand. In the current financial climate planners are in need of sophisticated tools that can extract information from the School Census and create realistic projections for planning future ehow to model the flows of children (demand) to schools (supply). based model, a novel approach in this area, to produce secondary school pupil projections based on data from the Annual School Census. The resulting projections will be benchmarked against current projections produced by the Local Education Authority national population projections to allow recommendations on how the current system accommodate these demands.

The objectives of this research are:

  1. To explore the application of agent-based modelling techniques to an area where individual level information is readily available. This will be through the use of a modelling framework already developed in the Centre for Spatial Analysis and Policy (CSAP) in the Geography at the University of Leeds
  2. To create secondary school pupil projections for Leeds and compare these against the existing projections produced by the a Spatial Education Model developed in a previous CASE studentship with Education Leeds, sponsored by the  Economic and Social Research Council (PTA-033-2005-00045)
  3. To link the model to national population projections of school age children to assess how the school network will look in 10, 20 and 30 years time


The Annual School Census (form merly PLASC) is a highly detailed data set – app proximately 8 million individual records are added to o it each year (Jones and Elias, 2006). Processinng data from such a sizeable data set to be specificalcally used for educational planning purposes iis a computationally expensive operation. Furthermore, using data extracted from the School Census for creating future school projections is a complex task. Direct extrapolation techniques and cohort survival ratios are widely applied for this task by education planners, but carry the assumption that neither school systems nor pupil populations vary. Harland and Stillwell (2007a, b, c) have shown that this is an inaccurate assumption with variance evident in pupil numbers and in the social and ethnic complexion of the pupil population. The importance of the education sector is undeniable and in the current financial climate, it is critical that education planners have sophisticated and appropriate projection tools at their disposal to allow exploitation of resources (such as the School Census) to ensure education provision is commensurate with demand.

Work has been undertaken by Harland and Stillwell (2010) using spatial interaction modelling (SIM) in this sector. However, their approach was limited in the sense that advantage was not taken of the individual nature of the School Census data. This can be readily rectified through the use of an agent-based model (ABM). ABM is a technique that is increasingly used in the social sciences; through this approach, individuals and their characteristics can be readily modelled over space and through time. Use of ABM allows more detailed information on School selection behaviour by pupils and their families to be teased out of the School Census. This circumvents the need to use school preference data collated by each individual LEA. As preference data are not published as a national dataset, projections produced using a SIM for schools located close to LEA boundaries are hampered.

The research proposed will further the SIM approach developed by Harland and Stillwell through the addition of an ABM for extracting individual-level data. This data will be used to create new pupil projections for Leeds at secondary school level and predictions for the next 10, 20 and 30 years.
The following methodology will be taken:

  1. Data assembly: This will be achieved either through Education Leeds or from the DCSF
  2. Data cleaning: pre-processing education data is a lengthy process. Routines established by Harland and Stillwell (2007b) will be used when appropriate
  3. Application of an ABM (Heppenstall Grant Ref) to extract information from the cleaned data and produce projections of pupil numbers for Leeds secondary schools
  4. Application of a tested and established SI model to create projections of pupil numbers for secondary schools in Leeds
  5. Comparison of the outputs from both projection models with those produced by Education Leeds and with observed data to establish which methodology produces the most accurate projections
  6. Running of several scenarios, reflecting realistic changes to the school network in Leeds to assess the suitability of each modelling method for simulating ’what if‘ scenarios. These scenarios can be summarised as:
  • Linkage to national population projection models to assess how the school network could look in 10, 20 and 30 years time. What educational services will the population be demanding and how will the school system need to adapt to cope with these demands?
  • What-if? scenarios will be developed using the new projections to assess the environmental and social implications of policy changes such as the much publicised “admissions by lottery”.

Once the projections have been created, they will be compared against current methods and methods that practitioners, in this case Education Leeds, implement.

Ultimately, this work will be extended into an education planning support system that would be of use to LEAs nationwide.


For information on funding opportunities click here


For project related enquiries please contact the supervisors.
For application enquiries please contact Jacqui Manton