Mathematical models are used to interpret and forecast human behaviour in many disciplines, ranging from patient safety to energy, transport and marketing. In the past few decades, researchers in Statistics and Choice Modelling have developed increasingly complex modelling techniques to achieve a more accurate representation of the complexity of real-life behaviour. Most of such efforts are ultimately aimed at helping policy-makers predict the acceptance/adoption of new technologies or policies, as well as improving the design of such policies thanks to a better understanding of the underlying decision processes.
One of the behavioural processes that has received the least attention is that of decisions made jointly by multiple agents. Theoretical models developed in Game Theory have rarely if ever been applied to real-world data, a shortfall which is partially due to the inherent complexity of the issue and the difficulty in observing and recording the process of two or more people bargaining and reaching a decision together. Nevertheless, it is evident how its omission can lead to reduced insights and inferior ability to make predictions of future behaviour. Joint decisions as well as social influences are relevant to most aspects of life, from household management to therapy choices in the case of illness, to management decisions in business contexts. Modelling the role of different people involved in a decision process allows an analyst to predict how such processes can be affected, for example how the role of doctors in patient decisions can be enhanced, or how power balance in work relationships can be modified for an improved work environment.
In recent years, Machine Learning has emerged as a key analytical tool for representing decision making, and has arguably made greater strides than traditional statistical approaches when it comes to capturing interactions between agents and joint decision making. However, unlike Choice Modelling, Machine Learning lacks an econometric and psychological foundation – the outputs cannot be used for welfare analysis and little is learned about the behavioural processes beyond being able to predict outcomes.
Choice modellers are now making initial steps in combining (rather than contrasting) Machine Learning and behavioural modelling, however with a focus completely on individual decisions. The proposed PhD project seeks to leverage the power and flexibility of Machine Learning together with the behavioural validity of choice models in the context of modelling joint decision making and other interactions between separate decision makers.
The aim of this project is four-fold. First of all, a review of the studies of mathematical models of joint and collective decision making will be produced. Such a review is needed and will be a contribution in itself, as the existing work is sparse in terms of methods and application fields. Secondly, informed by the first step, a new statistical framework, which can accommodate joint decision making while ensuring behavioural interpretability, will be developed. Such a framework will be consistent with economic principles of rational decision making while incorporating the complex dynamics of social interaction. The developed model(s) will then be applied in different areas including transport and health. The specific application will be defined also on the basis of the candidate’s background and interests, but possible options are residential location, allocation of joint resources to travel, health-related decisions in case of illness or others’ health (e.g. children’s vaccination or treatment or relatives which are cared for). Existing data sources will be exploited, and recommendations for future data collection will be made. Finally, the forecasting ability and welfare implications of these methods will be compared to other techniques such as machine learning and AI.
The project is in competition for a 3.5 years EPSRC DTP 2020 Environment scholarship which will include: tuition fees (£4,500 for 2019/20), tax-free stipend (£15,009 for 2019/20), and a research training and support grant.