Speaker: Lauren Stockley (University of Plymouth)

Abstract:

The Office for National Statistics use four subjective measures to understand well-being in the UK. The aim of the project was to gain a further understanding as to whether other, objective, factors influence a person's well-being and if so, which of these are the most important.

Principal Components Analysis was used to combine the four well-being variables for subsequent analysis. A variety of different regression techniques for modelling the data were then considered, namely classical linear regression, quantile regression and Poisson regression. The combined well-being variable was used as the response variable and the remaining variables in the dataset were used to explain it. Various techniques were considered in the linear regression analysis to select the best subset of predictors to explain a person’s well-being. The chosen model was then interpreted to determine how each variable affects well-being using the three different regression models. All models considered found very similar results when comparing how each variable explains well-being. Cross-validation was used to measure the performance of the linear and Poisson models through prediction. The linear model was found to produce well-being predictions closer to the true values than the Poisson model when evaluating the models using mean squared error.

A non-parametric model, random forest regression was fit to the data and used as a way of testing the importance of predictors in the model. This analysis showed that the most important variables for explaining well-being are disability, marital status and age of respondent.