A new research poster runs a boundary-detection method over thousands of simulated disease maps and finds it will usually mark a line across a gradual change in risk, though rarely in the same place twice.
A simulation study from the School of Underlying Conditions finds Bayesian areal wombling draws a line on most gently graded maps, in a different place each time
A method that statisticians use to trace boundaries across disease maps will usually find one even where the risk beneath it only slopes, according to A Line Somewhere on the Ramp, a new research poster in statistical epidemiology from the School of Underlying Conditions.
The work is led by Dr Brynjar Obiora, Senior Research Fellow, with Dr Vigdis Emenike, Senior Lecturer, and Dr Bram Ntuli, Lecturer and Convenor of Improvement Grand Rounds. Across 3,000 simulated maps of an invented region, the team found the method recovered a sharp step almost every time, and still marked a boundary on most gentle ramps, its position wandering across the slope from one map to the next.
Every map we make at the School starts with someone deciding that one thing has ended. It has been good to spend a season on the moment just before that decision, and I expect we will spend a good many more there.
— Dr Brynjar Obiora, Senior Research Fellow, School of Underlying Conditions
Working entirely from synthetic data, the poster sets established boundary-detection practice against a structured programme of simulation, strengthening the evidence base that underpins spatial reasoning wherever a map is asked to carry a decision.
It is the kind of crossing the University hoped its earth sciences school would make early: a field surveyor’s working question, put with some care to a statistician’s map. Dr Ntuli said the findings would go with him to the next session of Improvement Grand Rounds, “where I suspect they will find a room full of people who have drawn a line or two of their own”.
The full poster is available from the University’s research repository under an open licence, doi:10.5555/slop.r3zao7.