Developing and evaluating statistical methodology for quantitative bias analysis
Poor reproducibility of research results (i.e., the inability to obtain the same results on repeating the same study) has caused alarm, raising concerns about the reliability of scientific findings and the wasted resources, both human and financial. Key contributing factors include the explosion in the amount of data available to researchers, inadequate statistical training, and a lack of statistical methods and software to produce reliable results. Our aim is to develop statistical methods, software and accompanying guidelines to enable non-technical data analysts to assess the sensitivity of their results to key sources of bias in health research.
We will use the UK Biobank data to evaluate the performance of the methods and software tools that we will develop, and to communicate our findings to non-technical analysts. Whilst some data analysts provide a qualitative discussion of the effect of such biases on their results, few provide a quantitative assessment; thereby, leading to over-confidence in their scientific findings. Our proposed research will make a major contribution to addressing this reproducibility problem, leading to improved public confidence in health-science.