The statistics behind the spread of ideas
Everyone loves a good idea. It’s even better when the idea becomes a tangible innovation, a better mousetrap that anyone can use and every mouse should fear. The awkward truth, however, is that even in a polished form, good ideas can be slow to spread.
Anaesthetic and antiseptic offer an instructive contrast. Both were developed in the mid-1800s. Anaesthetic spread faster than a hula-hooping craze. Atul Gawande explained in the New Yorker, “within seven years, virtually every hospital in America and Britain had adopted the new discovery”. Antiseptic, in contrast, took a generation to catch on.
“The puzzle is why,” noted Dr Gawande, before conceding that it is not a puzzle at all. Anaesthetic solves an immediate problem: a patient screaming and writhing in agony. Antiseptic is a defence against an invisible killer, infection, that acts only with a delay.
Unfortunately, many innovations are more like antiseptic than anaesthetic: they solve problems that can only be seen through a statistical lens. People are slow to embrace what they cannot see. A few years ago, researchers at the OECD looking at the diffusion of global productivity gains concluded that there was a growing gap between productive companies and the laggards. The gulf was huge — typically a fivefold productivity gap per worker, even after adjusting for differences in the equipment available.
Whether the innovation is a hardier variety of seed, a safer pharmaceutical compound or a more reliable manufacturing process, the benefits will rarely be as obvious as slumbering through surgery. Such ideas often spread all too slowly.
There are other barriers to the diffusion of innovation. If people feel they can’t adapt a new idea to their own purposes, or try it out on a small scale, they will resist. One major obstacle is social: evangelists for innovation are often rather different kinds of people from their audience. Agronomists are not farmers; pharmaceutical sales representatives are not general practice doctors; inventors are different from the rest of us. We will gladly imitate our peers, although that still raises the question of who will go first. One influential early study of hybrid corn in Iowa between 1926 and 1941 found that a few farmers would experiment with the new seed in small quantities to see how things worked out. Even the early adopters took things cautiously, while others watched. Farmers would then eventually copy their neighbours.
It is tempting to shrug and conclude that this is simply a tough problem. But there is no need to despair. Late last year, the British Medical Journal published a study that caught my attention, in part because of the cross-disciplinary team of authors: Alex Walker and Ben Goldacre (epidemiologists), Felix Pretis (an economist) and Anna Powell-Smith (a data scientist) — but also because those authors were looking at the diffusion of innovation in an innovative way.
The study examined how quickly National Health Service general practice clinics in England caught up with best practice in prescribing two types of drug. In one case, the birth-control pill Cerazette came off patent in 2012, at which point patients should generally have been prescribed cheaper generic versions of the drug, desogestrel. In the other, national guidelines were changed to recommend a different antibiotic for urinary tract infections.
NHS England publishes anonymised data, every month, describing the drugs being prescribed by GPs across 8,000 clinics. If you have time, you can noodle around on OpenPrescribing.net — a platform developed by Ms Powell-Smith and Dr Goldacre — looking for patterns.
And since that sounds like hard work, the BMJ study uses a statistical tool to spot whenever a clinic seems to have changed its clinical practice, and whether they did so promptly or gradually, or suddenly but after a delay, or not at all. The patterns are clear to the naked eye once pulled out of the mass of data: here’s a clinic that swiftly and sharply switched to the cheaper generic drug; here’s a clinic that never read the email. A follow-up study performs a similar analysis for statins.
What’s remarkable about all this is how unremarkable it really is. The diffusion of innovation could once only be studied in small settings and by taking considerable pains. But this is the 21st century: the NHS has made the data available to allow us to watch a good idea spreading across the nation, or not, almost in real time.
This is, of course, an atypical situation. It is unusual to be able to collect such a large set of high-quality data, showing who has or has not embraced a new idea. And it is unusual to have such sharply defined innovations: either the doctor prescribes the new drug to patient X or she does not. Still, being able to observe leaders and laggards in the NHS is no small thing. It should be straightforward to prod the laggards — and to ask the leaders how they do it. And Dr Goldacre’s group have published their statistical tools. Hopefully, it won’t take too long for the idea of using them to spread.
Written for and first published in the Financial Times on 7 February 2020.