Could 'no smell' candle complaints help sniff out future COVID-19 waves?
Study compares negative reviews of scented candles on Amazon with U.S. COVID case counts
Nick Beauchamp wasn't sure if the viral tweet he read — linking negative candle reviews on Amazon to COVID-19 case numbers — passed the smell test, so he wanted to know more.
The Northeastern University political scientist and statistician, in Boston, published a paper in May that analyzed roughly 10,000 online reviews of scented products.
It probed whether or not the "no smell" reviews could assist in predicting future waves of the virus. And although that exact correlation wasn't made clear, the findings do indicate the reviews "reflect changes in U.S. COVID cases."
Loss of smell, or anosmia, is a symptom people worldwide have experienced from COVID-19. The belief is some Amazon reviewers couldn't smell the candles because they lost their sense of smell and unknowingly had the virus.
fresh wave of bad reviews for yankee candles <a href="https://t.co/1mlandB78I">pic.twitter.com/1mlandB78I</a>—@drewtoothpaste
Initially, he considered the investigation as a sort of an extended joke. However, since the paper's release, the online candle reviews went from mirroring COVID numbers, to predicting them, to now moving in the opposite direction from reported cases — with "no smell" reviews going up as COVID numbers tracked down.
"I think the big question was whether it was just cherry-picking humorous reviews or whether it was something real going on," said Beauchamp.
Here's part of his conversation with host Saroja Coelho.
This is a really quirky way to find out health information. Did you find it funny when you read it?
I think that the reason it caught my eye is the same reason it caught everybody else's eye. It was a fairly funny phenomenon, although it's a little hard to put your finger on exactly what makes it funny — people yelling at their candles when perhaps the fault is in themselves.
But then you decided to take this on and take the project forward. Why did you want to do that?
Partly it was just an accident. When I scraped a bunch of these reviews and looked at them, they were mentioning the smell — and I plotted that curve. It looked quite a lot like the COVID curves over the last couple of years.
I made a tweet that had already been somewhat viral on Twitter. Then my tweet becomes viral.
And then like the next couple of days, you know, you're fielding all these questions saying, "what about seasonality?" and "what about other products and so forth?"
Next thing you know, it sort of turned into an actual little project and an actual little paper, but it was all very much originally accidental.
How did you decide which candles to track?
Basically, I just looked for the ones that were most popular, but I don't think that I had the most systematic way of doing that.
I sort of looked through all of the scented candles, found the ones that had tens of thousands of reviews, and figured those were good ones to track.
What can you learn about COVID from these reviews that you can't learn anywhere else?
I think the first thing I'd say is to not take this data too seriously. I did the best I could with the two data series: one is about complaints in the reviews, and the other is COVID cases.
I do my best to try to weed out other things that might be accidentally causing the connection between these two things.
Maybe it's seasonality? I try to eliminate that. When I look at perfumes, I find the same pattern. I looked at the flu, and I found that it doesn't work for the flu.
But at the end of the day it's still a fairly limited set of data. I think the main utility of this is just one more piece of data in our arsenal.
I think the interest these days is because official measures have become less reliable. There's been sort of increased attention to these more indirect measures, which may include wastewater measurement.
"No smell" reviews are now going up, while reported COVID cases, numbers reported, and COVID case numbers are going down. What do you think is going on there?
The relationship between these two things has shifted a little bit [compared] to when I first did it. I was wondering whether the reviews would be predictive of COVID cases as a useful tool.
And I found that, no, they weren't really predictive. They lag behind COVID cases, but they were tightly related to COVID cases, which suggests it's at least possible, although still speculative, the complaints were actually caused by people having lost their sense of smell.
But what was interesting is that when I revisited this in June, I found that when you added a wave and a half of Omicron to the data, the reviews tended to go up before the COVID cases did. So, it actually became more predictive of the COVID cases than originally.
Now, when I revisited again in October, things have shifted even more. At this point, the two curves for the first time since the start of COVID have started to diverge significantly with the COVID cases being steady or going down in the U.S., while the [no smell] reviews and complaints are going up over the last couple of months.
👀 Yankee Candle reviews indicate that COVID is about surge again <a href="https://t.co/b6j7qhaFob">pic.twitter.com/b6j7qhaFob</a>—@DataDrivenMD
How might the results of what you're finding here be useful as we head into the fall and winter?
Again, I would caution taking these sorts of results too seriously. I think their main utility is in conjunction with other direct and indirect indicators.
If a bunch of these things [such as wastewater counts] are going up then maybe all of them together can suggest that official cases will start going up sometime soon.
There's a lot of sense among those who are still worried about COVID that attention is wandering, and maybe this is to draw people's attention back to the ongoing pandemic.
People who leave these reviews seem to be genuinely unaware that they could have COVID, and maybe they don't. Maybe some of them have a poor sense of smell or something wrong was actually happening with that candle.... What sorts of insight can this give you that you can't get anywhere else?
Occasionally the virality of this has spoiled the signal, at least briefly, where you'll get a bunch of people mentioning in the reviews that they don't have COVID; that they're not blaming the candle because they have COVID; that it's not their fault, it's the candle's fault.
Arguably, they're not going to be as biased by what people think they should do or think they should say. We've seen data saying that people tend to misrepresent whether they have COVID, and I imagine they even misrepresented themselves a little bit.
So, perhaps the advantage of this most indirect measure is that it sort of captures a symptom without it being necessarily contaminated by people's expectations. And in that sense it's closer to the truth.
Produced by Mickie Edwards. Q&A has been edited for length and clarity.