The keynote speaker for the seminar, Dr Jonathan Minton’s main substantive areas of interest are in demography and public health, and his methodological interests are in promoting good data science practices in quantitative research, and in complex exploratory data visualization. Thus, his examples of mapping population data were drawn largely from despair-related deaths in deprived areas of Scotland: datasets of incidences of mortality from alcoholism, drug taking and suicide.
Starting from the principle that “learning statistical modelling involves learning to say something in a rarefied language”, he argued that population data should be thought about as forming complex three dimensional structures, and visualized as such. Whereas spatial maps show how elevation varies with latitude and longitude, demographic ‘maps of time’, known as ‘Lexis surfaces’, show how quantity or risk (like risk of dying) varies with both age and year.
Lexis surfaces are useful at various stages throughout examining population data and have value at each stage. Minton proposes that we think of the Lexis surface as a “way of uniquely revealing something.” Standard ways of visualizing population data either tend to involve collapsing many numbers into a single number (e.g. standardised mortality rates and life expectancies), or cutting the data into thin slices along the age or year ‘plane’. In either case, most of the available data is missing to the researcher, and the often complex ways that age and time interact over the data structure are hidden from view. Standard (‘variable-based’) ways of modelling data, in which the effects of predictor variables are assumed to be independent in how they influence a response variable, further distort and hide the complex structures and relationships that population data tends to form and contain.
By contrast, using the cartographer’s toolkit to produce maps of risk over age and time allows the full structure of the data to be revealed, and for tens of thousands of values relating to a population to be thought of as constituting a single ‘case’ (rather than many different variables to be ‘controlled for’). This allows more effective understanding of the population processes to be developed, and more appropriate statistical models to be specified when required. As a result, Minton advocates that, from initial exploration of the data, to the development, testing, and presentation of statistical models and outputs, seeing population data in structures provides an intuitive approach to graphical inference and, what is more compelling, allows an individual to reason about the dataset because the Lexis surface provides an intuitive clue to statistical inference. Thus structures are potentially the most useful way of visualizing population data for policy makers because actionable insights are revealed.
Using examples of all-cause and cause-specific mortality in Scotland, Minton went on to demonstrate that intuitive inferences could be made about the relationship between suicide and drug-related deaths. In particular, he illustrated why, in Scotland, the similarity in incidences of suicide and drug-related deaths (see the bottom right population structures in the figure here) make it more likely that they share a common set of determinants, than when compared with alcohol-related deaths. For a policy-maker looking at these population structures, he concluded, this might for example prompt their decision to provide drugs rehabilitation concurrently with mental health support.
This talk was delivered as part of the LIDA Seminar: Maps of Time. Dr Jonathan Minton was the keynote speaker, presenting alongside Dr Roger Beecham who discussed Maps as Statistics? A call to Adventure for Perception Research in (geo)visualization.