Categories
current events death mathematics numbers politics probability science statistics trust

Why Stories Circulate about Covid-19 Deaths

Share

I’ve seen several posts on Facebook claiming that deaths of relatives or friends have been falsely attributed to covid-19 when in fact they were due to some other cause. These anecdotes represent a misunderstanding of the way statistics work and how data for statistics is collected. Of course, researchers want as accurate a count as possible for the number of deaths caused by Covid-19. But that kind of accuracy is harder than it sounds.

At first researchers were counting only deaths where the person who died had tested positive for Covid-19. They soon realized however, that they were under-counting the number of Covid-19 deaths. How did they realize that? They knew what the death rate in a particular place was prior to the pandemic. For example, if a city typically had 1,000 deaths in 30 days, and suddenly the number jumps to 3,000 but only 1,500 of those were due to patients who tested positive for Covid-19, then that left 500 deaths unaccounted for. So researchers decided to broaden the criteria for recording deaths as attributable to Covid-19. They decided to included deaths where symptoms were similar to those caused by Covid-19. They also included deaths even when the patient tested negative.

Why would someone who tested negative for covid-19 still be listed as a victim of it? Testing is not 100% accurate. Data on accuracy of the most widely used Covid-19 test is not publicly available, but some estimates range as high as 30% for false negatives, meaning that 3 out of 10 people who test negative for the disease actually have it. Even with a test that is 100% accurate under ideal conditions, real-world conditions can skew results. Many conditions can affect the amount of virus in a specimen collected by a swab. The most widely used test has close to a 100% accuracy for positive results, the the accuracy for negative results is uncertain and can vary depending on many factors. This is why some people who have died after testing negative for covid-19 are nevertheless listed as victims of covid-19. As long as they had symptoms consistent with the infection, they might very well have covid-19 listed on their death certificate. Of course, casting a broader net for data also means that there will be instances of people being listed as having died from covid-19 who actually died of other causes. Researchers make every effort to ensure this does not happen, but no procedure is foolproof. However, if the number of deaths identified as having been caused by Covid-19 matches the uptick in deaths overall, then it’s a pretty safe assumption that the data is pretty clean.

Because many people are suspicious of our government or the media or liberal elites—none of which are actually sufficiently monolithic to carry off a genuine conspiracy—and of expert authority in general, these types of stories gain currency on social media. Some may be true, but they usually do not contain sufficient detail to validate them. Even if they are true, they are generally offered by people who are not experts in determining cause of death.

So before you share one of these anecdotes about a suspicious Covid-19 death, consider not just whether it is true, but also whether it undermines the very institutions we have put in place to help us deal with infectious disease epidemics. While there are plenty of politicians ready to make hay out of crisis events, the experts and researchers who do the actual work genuinely care about producing good quality studies that advance our understanding of the virus and how it spreads. They are not out to get you.

Share
Categories
drunk driving economics sexism statistics

Meaningless Statistics

Share

Once again the Minnesota Department of Transportation is putting the word out that drunk drivers cause 1 in 4 traffic deaths. I begin with this example of a meaningless statistic, not because it is especially egregious, but because it exemplifies what makes statistics meaningless.

Of course, it is not entirely meaningless. We all have a gut feeling that drunk drivers do not drive 1 out of 4 miles driven in America. We strongly suspect that the vast majority of our fellow travelers are not drunk even at 1:00 AM. So it doesn’t take much thought to realize that 1 out of 4 traffic deaths is out of all proportion to the number of miles drunk drivers actually drive. In fact, the National Highway Traffic Safety Administration estimates that drunk drivers drive 1 out of every 140 miles driven on America’s highways. So drivers doing only 1/140th of the driving are responsible for 1/4th of the fatalities. That’s 35 times the expected number.

But most people seeing the signs have no idea what the context is. They do not know what fraction of miles driven are driven drunk. Statistically speaking there is no difference between “Drunk drivers cause 1 in 4 traffic deaths” and “Sober drivers cause 3 in 4 traffic deaths.” Yet the latter statement seems to make a case for drinking before driving! The lack of context is what empties the statistic of its meaning.

In the same way, there is an oft-quoted statistic that women earn $0.77 for every $1.00 men earn that also suffers from lack of context. (Apparently in 2015 the pay gap went down $0.02. Women now earn $0.79 on average for every $1.00 men earn.) The pay gap is an aggregate of all the income women earn compared to all the income men earn. It is commonly used as evidence of continuing sexism in corporate America. But as evidence it fails because there are so many other factors involved. Missing from the statistic are a lot of facts. For example, men work more hours than women. Women also tend to be over-represented in care-giving and hospitality occupations, which do not pay as well as more male-dominated occupations. This may be due to cultural sexism, but it’s hard to see what actions businesses or governments could take to close whatever portion of the gap is due to this kind of income difference. The truth is most companies in America already have policies prohibiting gender discrimination.

Statistics always present an aggregate view of data. That is what makes statistics valuable. However, aggregating data always also loses some information. The reports on which popularized statistics are based are usually careful to include methodology and context and indicate other possible interpretations of the data. But when the statistic shows up in Facebook meme or a highway sign, all that context is lost. The power of statistics is in simplifying complex data into a few numbers. We understand by simplifying. We should not, however, mistake our understanding for a grasp of the truth.

Share