Opinion – Pedro Hallal: Correlation and causality

One of the great challenges of science has always been to distinguish simple correlations between two phenomena from cause and effect relationships. It is elementary in science to know that not all correlations are causal. That’s why, decades ago, one of the greatest epidemiologists in history, Sir Austin Bradford Hill, published a text with nine criteria that help us distinguish simple correlations from causal relationships. Sir Hill is the same one who, along with Sir Richard Doll, discovered that smoking increased the risk of lung cancer. That’s the level of scientist we’re talking about.

Well, for more than a year I have been showing that there is a very strong and linear correlation between the percentage of votes obtained by the current president in the last elections and mortality by Covid-19, using official data from more than 5,500 Brazilian cities. The uncomfortable graphic uses data from the Superior Electoral Court and divides each of the Brazilian cities into ten groups, according to the percentage of votes obtained by the current president in the second round of the 2018 elections. On the vertical axis, the graphic uses updated data from the Ministry of Health on the accumulated mortality by Covid-19 for each group of 100 thousand inhabitants.​

If there were no relationship between the votes obtained by the current president and the mortality from Covid-19, the blue line would be basically straight, obviously with small fluctuations caused by the variability of the data. However, since the beginning of the pandemic, each time we run this analysis, we get exactly the same result: the higher the vote of the current president, the higher the mortality from Covid-19.

Whenever I present this data, the reactions are the most varied:

• The most radical group, whose guru ended up dying of the disease that he preached did not exist, and for which there is already a vaccine, immediately starts cursing.
• A second group adopts a more “professional” tone, trying to teach me epidemiology and the reasons why this analysis is “misguided”.
• A third group adopts a more cautious tone, being surprised by the astonishing result, and trying to understand the relationship.
• The fourth group already adopts a more indignant tone, simply not being able to leave unpunished the denialism that has already killed family and friends of all the readers of this column.

Obviously, there is a lot of room for the arguments presented by groups 2 and 3. Science lives on questions. Group 2 usually starts looking for alternative explanations for the observed correlation, other than the fact that denialism really is the cause of higher mortality from Covid-19.

The main ones are:

(a) The current president had more votes among older people, who are also more vulnerable to the risk of death from Covid-19.

(b) The current president had more votes among the wealthiest people and cities, and this can skew the results.

(c) The current president had more votes in cities with greater population density, and this can skew the results.

The truth is that all the analyzes we have done testing these alternative explanations do not change the result presented at all.

Perhaps we should then test Sir Hill’s nine causality criteria to see if there is reason to suspect a causal relationship between denialism and Covid-19 mortality. Yes, we’ve already done that, and the result is frightening, though not surprising. As if it were the answer of a perfect test, Sir Hill’s checklist is filled out, item by item, in the case of the relationship between denialism and mortality from Covid-19. There is evidence supporting each of the nine criteria listed by Sir Hill many decades ago.

A basic tenet of epidemiology is that deaths do not happen by chance. Brazil should be much more outraged to observe that it has a mortality from Covid-19 between four and five times higher than the world average. Denialism kills, and those who pay with their lives are our friends and family.