Amazon Reviews, Google Trends, and COVID-19

A friend was telling me about an interesting book he was reading titled “Everybody Lies”, which cites evidence of how people lie to each other, to pollsters, and to themselves. But one place they don’t lie is where they can be online and anonymous and no one is conducting a survey.

Two examples of such places are Amazon reviews and Google search data.

Here are two interesting looks at how some data analysis may have revealed alternative ways to track the trend in the number of COVID cases.

Yankee Candles Reviews

Terri Nelson, a Portland-based science illustrator, first suggested a potential correlation between the pandemic and negative candle reviews in a Nov. 24 tweet: “There are angry ladies all over Yankee Candle’s site reporting that none of the candles they just got had any smell at all,” she wrote.

The observation spurred Kate Petrova, a research assistant with the Harvard Study of Adult Development at Bryn Mawr (Pa.) College, to analyze about 20,000 Amazon reviews for scented and unscented candles as a personal research project.

She found reviews for the top-rated scented candles fell by about one full star rating in 2020 compared to prior years. Ms. Petrova did not find a similar trend for unscented candle reviews. The amount of scented candle reviews containing phrases like “can’t smell” or “no scent” increased from 2 percent in January to 6 percent in November.

The finding offers “strongly suggestive” evidence that Americans experiencing a loss of smell due to COVID-19 may be writing more negative candle reviews this year, according to the Post. However, Ms. Petrova said her analysis should be taken as “a fun exercise at data visualization — not a peer-reviewed study.”

Google Trends in Search Terms

Also in the second half of 2020,  author Dan Sinker came across a correlation with some serious consequences. He compared Google searches for “loss of taste” in the United States with the number of coronavirus cases confirmed each day. Loss of taste and smell are, by now, recognized as common symptoms of infection with the novel coronavirus. Should someone lose one of those senses, it’s understandable why they’d head to Google to figure out what was happening. The correlation between the two seems as though it probably has a causal link.

So Washington Post reporter Philip Bump decided to explore whether there was a consistent relationship between searches for those terms (loss of taste, loss of smell) and COVID case totals nationally or in states. Using Google’s online Trends tool and The Washington Post’s coronavirus data set, he compared the two.

He writes: “Sometimes data analysis yields a truly stunning result. This was such a time.”

Bump refers to this as just a remarkable overlap. Except, of course, for that first spike. But he says that’s easily explained: The United States was barely doing any testing in March and April, as documented by the COVID Tracking Project. There were fewer confirmed cases back then because there were fewer conducted tests.

This seems to point to a fertile field of research to examine more closely online customer reviews and search trends.

But be warned – it may reveal more about us than we care to know.

42 thoughts on “Amazon Reviews, Google Trends, and COVID-19

  1. I question the finding that people don’t lie when they’re online and anonymous. I think it’s farfetched to think they suddenly all start telling the truth.

    Liked by 1 person

    1. in some cases, that may be true. but I think people asking questions on google are looking for something that they may not normally share with someone. and while there are many fake reviews on Amazon, I would think that if there is a sudden change in the nature of reviews, that could be revealing something new.

      Liked by 1 person

  2. very interesting correlation. i have to disagree about amazon reviews always being honest. i have had a number of companies, whose products i’ve ordered from amazon, offer me a discount or free reorder in exchange for a favorable review. when reading reviews, some will disclose this in their review and i’m guessing most do not, though it’s hard to know which are ‘real’ and which just wanted the good deal. one thing that is generally telling are a series of negative reviews about one item, unless people are disgruntled with a company or, as above, there may be some other reason, such as loss of smell or taste, and cannot fairly judge the product.

    Liked by 2 people

  3. Online data, whether honest or not, is pulled from such a vast group of samples that trends become much easier to recognize. I, too, love when the information is used for such educational purposes. We may lie in a review, but we aren’t lying about what we type in our search boxes. I find these correlations intriguing. Good post, Jim!

    Liked by 2 people

    1. I agree with you about google searches, and as you point, even if many people lie in reviews, if a trend becomes noticeable, that may be telling us something…

      Liked by 1 person

  4. I always thought about people lying on HR questionnaires. I sometimes would get called out for being too honest. I knew others would put down what they thought HR wanted to hear in the interest of staying under the radar. Then there are the people who exaggerate their accomplishments on their annual self evaluations. They leave me speechless. Boy I am glad I don’t have to fill out HR surveys and forms any more. I preferred the days when my work did the talking.

    Liked by 1 person

  5. I guess there is more than one way to figure out what’s happening in a country. I think intelligence agencies use similar methods to estimate stuff indirectly.

    The correlations are fascinating. I may disagree with Bump’s explanation for the first spike. If there was a huge initial spike in undiagnosed cases as he says the Google search data suggests, there would also be a huge initial spike in hospitalizations and deaths. That data would support or refute his explanation.

    Liked by 1 person

      1. My recollection is that hospitalizations and daily deaths in April were still relatively few and not several times higher than they were during subsequent spikes which is what the search data would imply. I don’t have an explanation for the initial spike, but it might be a simple as people just being interested in understanding Covid symptoms when the pandemic was declared in March. There is a definite correlation between searches and Covid infections from May forward.

        Liked by 1 person

  6. I wonder if politicians change their ways when doing online surveys, or if their normal behaviour is too ingrained to do things differently?

    Liked by 1 person

      1. The problem with a lot of them is that they share their true beliefs, and they are repugnant. Until they are required to answer a straight question…

        Liked by 1 person

  7. That’s pretty interesting. Given that political polling has become notoriously inaccurate, I wonder is someone could use Google searches as some form of replacement. Or who knows, maybe they already have.

    Liked by 1 person

    1. that’s a good suggestion. or perhaps they could track the number of people who make a donation to a political candidate. I would think such data is already available…

      Liked by 1 person

  8. This is another interesting bit of data, it makes me think of Google’s ability to identify diseases symptoms quickly by what people are typing in a Google search (I’m not up for another pandemic to test this though!)

    Liked by 1 person

Comments are closed.