Taming the beast of time by Freek Van de Velde

While historical studies in linguistics and elsewhere took part in the ‘quantitative turn’ of the past few decades, many studies treat ‘time’ as a regular independent variable, on a par with other explanatory factors. While this may be appropriate under specific circumstances, the approach is not without risks. Time is a weird beast, which cannot easily be tamed by enclosing it within a fence of a regression model covariate. In this seminar, I will look at a few techniques that are specifically tailored to dealing with time: time series analysis and survival analysis. For time series analysis, I will be concerned with a method to infer causation from correlation: Granger causality, in which two diachronic trends are compared to see which is causally prior. For survival analysis, I will show how the technique can be used to predict the longevity of words in a language.