There is a lot of information out there about COVID-19.
Instead of repeating “facts” to you that are true from any given perspective, here is some information you can use to determine if something you hear is accurate, or needs more evidence to back it up, or is an attempt to mislead you.
The public is being bombarded with lots of statistics, but most people have not been trained on how statistics work or to understand what they really mean.
First off, there is a massive difference between “correlation” and “causation,” meaning that just because two things occur simultaneously doesn’t automatically mean one of them caused the other.
If you have heard the old saying, “There are three types of lies: Lies, damn lies, and statistics!” you may be familiar with this.
Most of the statistics you hear about in the media are one of two types:
1. A survey, usually showing what some people think about something at a moment in time.
2. “Regression analysis,” usually a “simple linear regression analysis.”
In the first example, you often hear that “X% of people believe Y is a problem.”
For example, 75% of Canadians believe they will be homeless because of the economic impact of coronavirus in the next three months.
Hearing that, you might conclude, “Things must be really hard in Canada, this is a crisis!”
But maybe the survey sample included a lot of homeless people.
Maybe the people interviewed are scared but likely won’t be homeless in three months.
Maybe, for some, “homeless” means they will be renters instead of owners.
All you can really conclude from a survey like this is that these people “think” they will be homeless, not that they will actually “be” homeless.
The second example, regression analysis, is a statistic that looks at how “related” things are. It can only determine correlation, or the likelihood that these things relate to each other.
It cannot determine causation or “how” these things are related.
The problem is people, including many “experts,” regularly use these statistics to prove their “how” or “causation” arguments.
Here is a well-known example:
Did you know that ice cream consumption and drowning have a really close correlation? Some studies have shown a greater correlation between ice cream and drowning than smoking and lung cancer.
So here is the question: Does eating ice cream cause you to drown, or does drowning cause you to eat ice cream?
Statistics do not know the difference and have no way of proving that one causes the other.
It could be that drowning and ice cream consumption are both related to how hot and sunny it is outside, when more people go swimming, but a statistic won’t give you that answer.
The same is true for almost all statistics. They provide a very specific piece of probable information in an extremely narrow context.
Statistics cannot prove smoking causes lung cancer, decapitation causes death, the cause of poverty, or really, anything.
They can justify funding to do other types of studies that try to determine the “causal relationship” between things the statistics found are “correlated.”
There are other more complicated components of statistics that help you identify how reliable the claim is.
Looking into the “P-value,” meaning the probability value, the “confidence interval” and other information that is often misrepresented or left undisclosed to the public in statistical surveys may help you detect when you are being “taken for a ride.”
Some day, when people hear a statistical claim, let’s hope they will ask, “Does this statistic actually prove what it is claiming?”
If people are eating a lot of ice cream and drowning, perhaps what we actually want to study is why so many people are drowning. We should probably do a scientific study on that.
Otherwise, we might as well accept that it is a “proven fact” that if you drown, you will become an ice-cream craving zombie.
— Alex Vezina is the CEO of Prepared Canada Corp. and has a graduate degree in Disaster and Emergency Management
