by David Lazer & Christoph Riedl
Most of the time, doing research and trying to produce solid science is a grueling march full of setbacks and disappointments. On rare occasions, however, you come across something truly unexpected. We are particularly excited about our (just out) paper on diffusion processes. It is probably the one time in our careers that matched Asimov’s great quote regarding scientific discovery:
“The most exciting phrase to hear in science, the one that heralds new discoveries, is not ‘Eureka!’ (I’ve found it!), but ‘That’s funny…’”
This was, in fact, a broken experiment: an experiment meticulously crafted over weeks that took an unexpected turn. But the broken experiment turned out to be way more interesting than what we had originally planned.
The experiment had a pretty simple set up. We collaborated with Telenor, a multinational telecommunications company, to study spreading processes among its customers. They distributed 70,000 unique voucher codes for a free data plan to their customers, and we watched them spread, meticulously tracking the diffusion of each of the 70,000 codes.
And they did spread, to almost one million people over two weeks, at which point Telenor stopped the promotion. So, what do you think the rate of spread was during this period? Most people familiar with theories of the diffusion of innovation or spread of infectious diseases in social networks would predict the traditional S-shaped diffusion curve. This has been long known to marketing scholars who study diffusion of innovations.
The answer? Drum roll please …
Linear — with some diurnal bumpiness. That’s not something you see every day in the vast diffusion literature. In fact, we could not find a single example in our review of the literature (but, please reply if you have one). This was an entirely unexpected result that had us think hard about what possible mechanism could lead such a diffusion process.
We found one more major and unexpected quirk in the data. Of the 70,000 voucher codes, the spreading process was wildly uneven across the codes. It was even more uneven than you’re thinking after you read that sentence. The top single code accounted for ~80% of adoptions, and the next few accounted for the vast majority of the rest.
Why the linear pattern and concentrated success? The answer is obvious (but only in retrospect), simple, but profound (we think):
Some people posted their codes online.
The most successful voucher code — the one that attracted the most adopters — must have been indexed within minutes (because its rapid spreading started 15 minutes after we started the experiment).
The spreading process was linear because the background search rate is roughly constant day to day for discounts. We found almost identical daily patterns for every single day of the experiment (the figure below is not scaled in any way). Furthermore, any viral element in the diffusion process turned out to be quite small. Those codes that were not posted, that just spread person to person, had an infection rate below the critical take off threshold. In a Google-less world, the codes would have spread for a little while to a few people, and then stopped. In our research we estimate that the viral element of the most successful product accounted for only 11% of its popularity, while the other 89% are the result of a background rate of interest in discounts.
While many codes were posted online only, a few dominated because of the power of ranking on search engines — it’s good to be #1 on a ranked list of results.
The implication is profound, because it means anything that is posted online and is findable via search will spread — although it will spread in a way quite different from how we thought.
That’s the good news. The bad news is that current search technologies will single out particular answers way above all other answers. In this case, there were 70k other possible answers, hundreds of which were posted online and equally discoverable using search engines.
But one answer won, not because of merit, but because it was in the right place at the right time for whatever the dominant search technology was. And it’s not healthy for the global information ecology for one answer to so dominate.
The broader scientific message is: sometimes you discover something interesting on the way to other places.
— — —
This piece is based on recent research published in the Journal of the Royal Society Interface available without restrictions at http://dx.doi.org/10.1098/rsif.2017.0751.
David Lazer is a Distinguished Professor of Political Science and Computer and Information Science at Northeastern University.
Christoph Riedl is an Assistant Professor of Information Systems at the D’Amore-McKim School of Business and the College of Computer and Information Science at Northeastern University.