During my 1-year PhD placement, as a Model Developer within the Working Age Modelling and Forecasting Division (WAMForD), part of the DWP Analytical Community, I had the privilege to contribute to some of the Government’s most high-profile analysis outputs. The model and data-development work conducted in this division underpins key government analyses, including Budget forecasts that account for over £200 billion in annual expenditure on welfare benefits and reform programs.

​

I was part of the INFORM2 team within WAMForD, which is responsible for maintaining and developing the division’s main Java-based dynamic microsimulation model. This model uses administrative Universal Credit (UC) data at the household level to predict the volume of claimants that lead to the departments Average Managed Expenditure (AME) forecast. The microdata generated by this model provides a monthly profile of UC households for over 5 years, with projections based on transition probabilities for various elements such as health status, children, and housing tenure. Each element is modelled individually to obtain transition probabilities, which are inputted into the overall microsimulation that ages household members by one month and simulates data accordingly. The INFORM2 team’s forecast summary is then passed to subsequent teams for official fiscal forecasting, with key stakeholders including the Office for Budget Responsibility (OBR). Given the sensitive nature of the individual-level data used in INFORM2, the model inputs remain unseen by the OBR.

Key Projects

Throughout my placement, I led several projects, including:

• Off-flows Element Modelling: Improved the accuracy of the existing logistic model concerned with individuals leaving the benefit system and ran the UC microsimulation model incorporating these enhancements.

• Decision Trees and Random Forest Modelling: Replicated the off-flows modelling as a decision tree and as a random forest model, drawing on my experience with machine learning techniques. Assessed the feasibility of integrating this into the microsimulation model.

• Housing Tenure Modelling: Transformed the existing housing Discrete Probability Matrix into a multinomial logistic regression model.

• Housing Postcode Transitions: Analysed transitions between housing postcodes to evaluate the feasibility of adding this feature to the model.

• Regional and Urban/Rural Variables Integration: Sourced and linked regional and urban/rural variables to the microsimulation sample dataset. Adapted the microsimulation model to accommodate these additional variables and analysed the predicted results, which were subsequently sent to the AME team to understand whether this improved expenditure forecasting. Introduced the use of R to the team.

• Induction Material Creation: Developed induction materials tailored to new joiners of the INFORM2 team, providing an overview of DWP as a department, the WAMFORD division, and the INFORM2 model.

Skills Gained

During this role, I acquired a variety of skills and insights, including:

• Microsimulation Methodology: Gained a deep understanding of microsimulation methodology and the workings of the INFORM2 model, building on my knowledge of agent-based modelling. This experience complemented my PhD research, where I use longitudinal panel data to study the school-to-work transition.

• SAS Programming: Learned coding in SAS using big data, building on my existing knowledge of R and Python through DWP-taught courses and practical experience through projects completed.

• Adapting to Organisational Practices: Quickly adapted to DWP’s way of working, gaining knowledge of the benefit system, fiscal events, and the process flow leading to the Autumn Budgets and Spring Statements. This included understanding how our modelling work and forecasted UC caseload contribute to the broader UK government budget.

• Model Version Control and Quality Assurance: Learned practical implementation of model version control and quality assurance in a government setting, including attending quality assurance challenge meetings with the OBR.

• Teamwork and Leadership: Enjoyed working in an agile team environment, with opportunities to lead projects and support new team members. I particularly valued the autonomy in making modelling decisions and the increased responsibility this provided.

I was also a part of the DWP Modelling & Forecasting Centre of Excellence, a group composed of PhD placement students, DWP-sponsored PhD students and academic fellows. This community provided valuable support and allowed discussions on PhD-related topics, such as transitioning back to academia after the placement.

​

Overall, I thoroughly enjoyed my year at DWP, where I applied my data analytics and data science skills to meaningful projects. The experience provided me with numerous new skills that I will take back to my PhD and future roles. Moreover, this placement has enhanced my confidence and better equipped me to pursue my goal of securing an industrial role after completing my PhD.