When Amazon reviews predict waves of covid

Like the waves of Covid-19, the phenomenon is cyclical: on Amazon, Yankee Candle-brand American and scented candles regularly drown in negative reviews, posted by customers disappointed by their too-weak fragrance.

All in all, apparently logical when we know that one of the main symptoms of Covid is anosmia, that is, a more or less lasting loss of smell, the correlation has been noted many times, on the web by some researchers such as Kate Petrova, or in the media.

This was also the case for Nick Beauchamp, a political science professor and researcher at Boston’s Northeastern University, who had fun matching the curve of negative Amazon reviews of Yankee Candles to the appearance and distribution, in the United States, of Yankee Candles. the Omicron variant in December 2021.

From the vague joke to the precise study, there is only one (admittedly big) step that the American did not hesitate to take, as our colleague from Arrêt sur images Loris Guémart pointed out on Twitter. “It wouldn’t be hard to do this more cleanly”also notes Nick Beauchamp.

The American thus ended up studying the phenomenon in a more precise and statistical way, and publishing an article entitled “This candle doesn’t smell like anything”: detecting the effects of anosmia using the autoregressive model of the Bayesian vector”, presented there a few days ago in the International Conference on Web Documents and Social Networks.

“A recent hypothesis has hypothesized that complaints in Amazon reviews about the lack of scent of scented products may be an unintended indicator of anosmia.”explains the researcher in the presentation of his work.

“This paper presents a new Bayesian model developed to test this hypothesis and shows that ‘no smell’ reviews on Amazon reflect the number of COVID cases in the United States.”

Joke, correlation, causality.

To create his model and do his statistical checks, Nick Beauchamp used a Chrome extension that downloaded 9,837 Amazon reviews of Yankee Candles over 172 weeks between 2018 and 2021.

His conclusion? Although great care must be taken, especially since the results and media coverage of his study could “contaminate” the sincerity of the comments, a greater occurrence of negative comments could help predict the arrival of a next wave of Covid.

“Nick Beauchamp discovered that instinctual correlation was actually causation”It is written in the newspaper of the university of the Northeast. “Covid-19 cases seemed to predict the arrival of ‘no smell’ comments. For every 100,000 covid cases in a week, it found that these ‘no smell’ comments increased by a quarter the following week.

In their first analyses, the cases of Covid-19 therefore preceded the negative comments published on Amazon. But the researcher then added data relating to the Omicron wave to his data set and, a miracle of science so to speak, things reversed: the comments began to precede the wave.

“I added that six more months, including the Omicron wave, and now it’s predictive in the sense that in theory the feedback gives us a slight edge.”he explains.

However, it concludes on an important note: what has “started with a tweet and turned into a punchline” is supported by an article which, if it gives interesting clues, is still considered by its own author to be “the extension of a joke on Twitter”.

Leave a Reply

Your email address will not be published. Required fields are marked *