free range statistics

An observational study claiming to be an RCT might have something to say but there are far too many discretionary researcher choices taken to believe its findings. But I use this as a chance to play with statistical inference after estimating a regression via lasso.

Exploration of change in occupations in the Australian health industry, and economy more broadly, from 1986 to the present.

There is a fast growing body of knowledge and tools to help estimate effective reproduction number of an epidemic in real time; I have a go at applying the latest EpiNow2 R package to data for Covid-19 cases in Victoria, Australia.

Science isn't broken, but journals are. A joint solution is emerging for disparate problems of access, quality control and replicability in scientific publishing.

My forecasts for the 2020 New Zealand general election are out, and predict a comfortable win for Jacinda Ardern's Labour Party either alone or in coalition.

Surgisphere, a tiny startup that claims to be providing large real world data for scientific health studies, is probably fabricating data at scale.

It turns out to be quite easy in R to reorder your bars within each clump, to produce a bad bar chart like the unfortunate example from Georgia doing the rounds.

Even when you adjust for test-positivity rates, the number of new COVID-19 cases per day in Texas is going up, although not as rapidly as the unadjusted numbers imply.

I look at several different ways of accounting for the information given us by high positive testing rates for COVID-19 and look at the impact on estimates of effective reproduction number at a point in time.

A pragmatic way of generating prediction intervals from a generalized linear model with a quasi-likelihood response, if you're prepared to make an additional assumption about the distribution of the response.