Moving Averages
Moving averages are taking away what little visibility we still have about what SARS-CoV-2 is doing to humanity.
Pandemic surveillance, vaccination and treatment for COVID are disappearing in both of my countries. Even much of the wastewater monitoring is being discontinued. The World Health Organization’s and USA federal government’s decisions to end pandemic emergency status doesn’t mean the pandemic has ended. It only means that emergency measures against it are ending. Indeed, WHO warns that an estimated 1 out of 10 infections (not people, not symptomatic cases, but infections) will result in a need for long term health care. It’s a mass disabling event.
Among other things, ending the emergency declarations will end cost-free access to test kits in the USA. (The UK ended that ages ago.) Although the false negative rate on rapid tests has become so high that a negative rapid test result doesn’t mean much, a positive result still has meaning. More accurate PCR tests are hard to get and expensive now.
Not having a positive test result can lock people out of help in the future. They can be shut out of a Long COVID clinic if they can’t prove they had COVID. They can’t claim their post-COVID health problems as a workplace injury if they caught it at work but can’t show a positive test result. Filing for disability support requires convincing the insurer or government department that they have something that causes disability. Without a positive test result, many disability evaluators and even many doctors claim the patient’s problems are psychological. Numerous studies prove Long COVID is physiological, but tests to determine whether someone has long term post-COVID health problems are not yet part of standard medical practice.
Without well-run testing and pandemic surveillance programs, I cannot believe the case and COVID death numbers being reported in the media for either of my countries. That data is mushy at best.
Without testing, contact tracing, reportability and so on, we are left with excess deaths from all causes as our main insight into whether we’re in trouble due to the pandemic, and whether it’s worse or improving or about the same. It’s a lagging indicator, telling us where we were instead of where we are now, but at least people’s deaths are tallied.
That’s where moving averages step in to hide reality in plain sight.
Examples of Good Use of the Data
The chart at the beginning of this post is much better than what I’m seeing more and more often in the media. It shows death rates in the pandemic years so far as a percentage of the average death rates in the five years before the pandemic swept across the world. Plotting the percentage instead of raw numbers allows us to compare countries that have very different population sizes.
Notice that sometimes a country had, for a while, fewer deaths than the pre-pandemic average. Although lockdowns made it harder to get medical help for problems other than COVID, lockdowns also kept other diseases from spreading as easily as usual and kept people from doing as many activities that could result in an injury.
Moving averages are not in play there.
Another way to look at it comes from The Economist through Our World In Data, showing estimated cumulative excess deaths from all causes from the start of 2020 to now:
This is a good way to see how countries that started out careful but eventually opened up still tend to be doing better than countries that started out with incomplete or poor public health measures. Using a per-capita approach is another way to compare countries that have disparate population sizes. Moving averages aren’t in play here, either.
Articles in mainstream media use data like this less and less. I’m seeing more of the news toss out numbers about excess mortality rates calculated in comparison with the previous five years.
Those news stories are using moving averages. Why do I say we shouldn’t do that?
How We Analyze Data Matters
Solid, valid data can look very different, depending upon how you look at it.
Most people listen to the radio or podcasts, watch news on television, casually browse their favorite news media’s online articles or read print media (newspapers and news magazines).
Most people haven’t dealt with moving averages as part of how they make a living. If they notice the media mentions the excess deaths data is based on the past five years, it sounds nerdy so it must be right.
The data is correct, but moving averages gradually obscure what is most important for us to see. Is the virus still causing many people to die prematurely? That’s what we need to know. Moving averages wipe out our ability to see the answer to that question.
What Is a Moving Average?
If you want more detail than the simple introduction I’m about to offer, a moving average is also called a running average or rolling average. It is heavily used in factory automation and by stock market traders. You can use a moving average to get a sense of a trend. There is a reasonably clear explanation at Statistics How To.
You may be a visual learner, so I’m going to lay out how you can do this more visually than the usual tutorial.
Jot down the years 2015 through 2023. Beside each year, put a number. For purposes of understanding how a moving average can make a big jump in death rates look normal, the numbers beside 2015 through 2019 should be smaller than the numbers beside the other years. The real data has that pattern too, but the real numbers are big. For this exercise, you can use small numbers.
Use a couple of cards or pieces of paper to cover all but the first five years. Beside the most recent year in the set of five, write down the average of the numbers for those years. Now move your window later by one year and write down the average for those five years. Keep moving your window by one more year and calculating the average for those five years until you have run out of years to include. What you end up with for five year moving averages looks similar to my example here, with whatever numbers you chose. I’ve shown how mine graphs, too.
Elevated Numbers of Deaths Look Normal
Don’t expect this to track exactly with real mortality numbers, which will vary from one country to another, although I did set a pattern loosely based on where I live. I say loosely because UK excess mortality in 2022 was among the worst in 50 years and I didn’t go that far, but the general pattern of the example is much like the real pattern in most countries.
See what’s happening? The raw number at 2019 isn’t far from the moving average of the five years 2015-2019. The 2020 number is far above the moving average for 2016-2020. But as we keep shifting the window for the moving average, the years with high numbers nudge the moving average up. That makes the years with high numbers start to be not far above the moving average.
The moving average becomes based more and more on years that have elevated numbers. In my example, it looks like 2022 isn’t high at all. It’s right on top of the moving average. But 2022 is quite a bit higher than what was typical before everything kicked off in 2020.
As I said, actual mortality figures produce a similar pattern. I used small numbers to make it easier to grasp how the mathematics work.
To see how we’re doing with the pandemic, we need to compare years from 2020 onward with the last few years before 2020, the way The Economist and Our World In Data did.
When news stories talk about mortality figures in terms of five year moving averages, they make elevated levels of death sound normal. It’s misleading. It can trick us into lowering our guard when it isn’t safe to do so. We shouldn’t fall for it.