For the past few weeks, Brado has monitored Google search trends around COVID-19. The goal was to answer this question:
Do spikes in symptom-related Google searches strongly correlate to the growth in Coronavirus case counts across the United States?
In short, the answer we found was a resounding, “Yes.” Below, you will see all the data we uncovered with takeaways.
To confirm the data in Google Trends, we reached out to SEMrush for additional click-stream data.
Using SEMrush’s March click-stream data, we took a granular look at that correlation. For expediency, we focused on search behavior in Oklahoma, Michigan, and New York.
|New York has a total population of about 20 million||138,863||5,489|
|Michigan has a total population of about 10 million||18,970||845|
|Oklahoma has a total population of about 4 million||1,472||67|
While New York is potentially nearing the apex of the virus, and Michigan is likely a few weeks out, we chose to examine Oklahoma because they have not implemented shelter-in-place orders yet.
SEMrush Click-Stream Data Findings
The rise in searches for terms related to lack of smell or anosmia (uh·naaz·mee·uh) and terms related to loss of taste shows a distinct correlation to the rise of COVID-19 cases. The same symptom-based searches are spiking year over year in all three states.
Each image below is linked to more data.
Analysis — SEMrush and Google Trends
To make it easier to analyze, we imported the raw data into Google Data Studio. Image one shows SEMrush data, and image two shows Google Trends data. You can select and compare multiple symptom-based searches by state and rise in case count.
Click image above for interactive report
Loss Of Smell and Taste-Related Symptom Searches Show a Strong Correlation To COVID-19
Link to Google Data Studio Dataset
The above graph shows a strong correlation between searches for loss of taste and smell and surges of COVID-19 cases. But we do not see any correlation between eye-related symptoms and COVID-19 cases from 2017 to March 2020.
Eye-related symptom searches show no clear correlation to COVID-19.
Are Google Searches A Window or A Mirror?
Obviously, correlation is not causation. And we need to ask ourselves how much the media influences search behavior. Searches for anosmia may help us answer this question.
Anosmia searches were spiking prior to media coverage.
The media started to widely report on the loss of smell and taste as potential symptoms of COVID-19 on March 20th.
However, data from Oklahoma, Michigan, and New York shows searches for anosmia surged before March 20th. Seth Stephens-Davidowitz, writer and quantitative analyst, published strikingly similar results in the New York Times. These findings suggest Google search, is in fact, much more of a window than a mirror reflecting back media reporting on symptoms.
March 2020 Searches
Search data from Oklahoma, Michigan, and New York shows significant spikes for anosmia, dry cough, and loss of taste and smell in mid-March compared to the start of the month. This same spike can be seen when comparing data from 2017, 2018, and 2019 for the same regions as well.
Each image below is linked to more data.
Potential Uses For Anosmia/Symptom-Related Search Spikes
- To help prioritize or ration scarce medical supplies, such as ventilators.
- To anticipate second- or third-wave infection spikes.
- To share with public health model builders, who could potentially use this data to target appropriate control measures (such as when to tighten social-distancing and shelter-in-place restrictions).
- Brado will update Google Data Studio with information on all 50 states. Bookmark it!
- We will continue to work with SEMrush to mine for potential symptom spikes by state through the month of April.
- We would love to hear ideas and trends we can add to the analysis or any findings you or your team have uncovered that could be helpful.
While we can’t prove causation, we believe this strong correlation merits further inspection from the search and data mining community as well as the healthcare community.
From a trend perspective, these searches could be vital in prioritizing medical supplies, like ventilators, to the “hottest” areas. As noted above, predictive search data could also help anticipate second- or third-wave virus spikes.
Maybe the most effective use isn’t just medical; perhaps, equally, it is cultural and political – in creating standards and infrastructure for public health modeling – when to initiate social distancing and shelter-in-place restrictions.
Our point? Our job…meaning, as Brado data scientists, is forever changed. This is the pivot and new paradigm for our company and our clients. Besides compassion and care, we possess tools to predict and possibly help play a role in preventing the spread of future viruses.