Over the past 15 months or so I have spent time looking at Covid data. I hesitate to call any of it information.
The one thing I have learned is that statistical inference is much more subtle than I knew, and I thought I was careful. Be reluctant to treat your self-created judgements as reasonable.
One of the things I was smacked with is the assumption that all other things are equal.
Every principle is more impressive if it has a Latin expression. We all overuse the idea of all else equal, but we promptly forget the principle as soon as we have some outcome that matches our biases or preferred answer. Paying attention to Covid data will soon teach you to be more careful.
For example, there is Simpson’s Paradox. It is the case where a clear relationship exists in summarized data, but disappears, or even reverses, if you break it down into categories — age range for example, or gender. That result is so counterintuitive that we can lead ourselves far astray.
I did an article on that years ago. You can see it here. Intuitively Obvious Is Usually Wrong
It’s easier than you think to go astray. A February 2021 paper authored by Julius van Kulgen and others dealt with a comparison. Italy versus China. In every age demographic, case fatality rates in Italy were lower, but overall they were higher. Say what? We tend to believe the sum of the parts equals the whole and it does in many cases in life, but not necessarily in superficial statistics.
Why? Because all else equal never happens. What if Italy or China’s age mix is not identical? That mixing is a factor in creating the whole. Time is a factor too. Did the survey cover the same period exactly? If not other conditions may have been different.
It is easy to arrive at a weak conclusion when you deal with superficial observations.
Observing a fallacy is not very serious. Acting on it can be, so we should be more thorough.
Suppose we know that many people who sleep in their clothing wake up with a headache. Can we assume that sleeping in your clothing causes a headache? We could, but are you sure?
We also know that some people who do not sleep in their clothing wake up with a headache. When you see inconsistent things, it behooves you to think a little deeper.
By digging a bit deeper we discover that morning headaches are correlated with wearing your clothing to sleep, but there are other possible correlations. One is having had too much too drink the previous evening. In this case wearing clothing to sleep and waking up with a headache are both correlated to the same thing. Drinking to excess.
Dig a little deeper, even when the data seems clear.
This graph shows the death rate for people between age 10 and 59 based on their vaccinated status over the period mid-March 2021 to mid-September 2021.Gold = vaccinated, Blue = not vaccinated
Your conclusion? Obviously vaccines are more harmful than none. Your skeptic should say, wait a minute, what else is there to know? Good skepticism!
Are these all deaths or just Covid deaths? Let’s assume Covid related only.
What does Covid related mean?
The point not presented and likely important and true, is older parts of the age cohorts were vaccinated first. The have higher mortality rates by a wide factor. If we investigated and found that few people under 20 were vaccinated before the last third of the graph’s time, would you not expect a graph like this one. There is a better explanation than I can give you here.
All else is never equal
Each of us, including many trained people, are incapable of extracting the exact meaning from a dataset. The reason is meaning is a deeper idea than presentation. Presentation can easily fall afoul of the idea of all else is equal, when in fact it is not.
The main difference between the trained and the rest of us, trained people know enough to look deeper while we amateurs tend to be follow the surface details.
All else is never equal
It is easy enough to find data, yet near impossible to use it to create a viable plan of attack.
You cannot do your own research because all the obvious data is incomplete or shaded a certain way to support some narrative or other.
The first step in defence is recognizing the data as incomplete and addressing how you can change that condition. You usually cannot.
On the internet if a graph presents an obvious inference, assume it is incomplete and mistaken, or propaganda.
It’s is far easier to derive false inferences than you think.
Assess the presenter’s objectives and assume they are correlated with some preferred cause story.
Avoid neuroticism while being highly skeptical.
Inferences are not facts. Acting on them is dangerous.
I prefer rational but I don’t know how to use Covid data to get there either.
Help me please. If you have found this useful, please subscribe and forward it to others.
I build strategy and fact-based estate and income plans. The plans identify alternate ways and alternate timing to achieve both spending and estate distribution goals. In the past I have been a planner with a large insurance, employee benefits, and investment agency, a partner in a large international public accounting firm, CEO of a software start-up, a partner in an energy management system importer, and briefly in the restaurant business. I have appeared on more than 100 television shows on financial planning, have presented to organizations as varied as the Canadian Bar Association, The Ontario Institute of Chartered Accountants, The Ontario Ministry of Agriculture and Food, Banks – from CIBC to the Business Development Bank.
Be in touch at 705-927-4770 or by email to email@example.com