Complex statistical modelling of socio-economic variables in public health

Preview

PDF WRAP_Theses_Eyre_2018.pdf - Submitted Version - Requires a PDF viewer. Download (9MB) | Preview

Request Changes to record.

Abstract

The statistical inference of socio-economic variables in public health is key to the design of interventions to address the many health inequalities that exist across the world. However, such inferences are achieved commonly using a small standardised library of statistical methods. Meanwhile other fields such as computer science and systems biology have seen the development of many new methods allowing for more varied and useful analyses.

Here we present analyses in three important contextual areas of socio-economic variables in public health, bringing in modern and sophisticated methods in order to develop highly useful and flexible results and further expand the library of statistical methods in public health.

In the first, we further develop and apply a non-linear temporal model to analyse the spread of health aspects such as mood and weight over US adolescent friendship networks by a process known as social contagion. The use of this model improves our ability to more realistically reflect patterns we expect to see result in the data from contagion. This was achieved using analysis of the Add Health dataset.

In the second, we use the flexibility and complex features of Gaussian processes to analyse two different aspects of pregnancy in rural South Africa using the Agincourt HDSS dataset. First, the modelling of fertility-patterns over combinations of variables where some have established models and others do not, allowing us to incorporate such variables into our model without risking the enforcement of unjustified assumptions. Second, analysing social contagion of pregnancy risk behaviour where no social network data exists, demonstrating how the use of sophisticated methods can enable us to attempt complicated research questions.

Finally, in the third we build three possible Bayesian belief network models of household food security in the Agincourt study area. The structural features of these models make them potentially highly useful causal tools that enable us to model a wide range of interventions on our system.

Through these analyses we demonstrate the importance of expanding the library of statistical methods in public health to include the many modern and sophisticated methods being developed in other fields, whilst also producing findings and tools of great robustness, flexibility, and utility.

Request changes or add full text files to a record

Repository staff actions (login required)

View Item