For the reader: Learn what an unconditional probability tells you. When we hear about a mortality rate of 3.4%, what we are actually being told is that of all the people who have contracted COVID-19, 3.4% of them have died (past tense). It does not mean that in a room of 100 people with covid, 3 people will die. Unconditional probabilities can tell us about the past and can tell us simple statistics that leave out important and vital information. Unconditional probabilities should never be used to make claims about the future.

There may be biases in the sample which affects the reliability of the estimate, but its not generally true that unconditional probabilities can’t be used to make claims about the future.

If the sample has a similar composition to future populations, then the statistic can be used as an estimate for future probabilities.

Examples: The 3.4% of people who die is largely made up of elderly smokers, while total cases is spread somewhat randomly across a city.

• The probability of death of an individual who is young is probably lower than 3.4%
• The probability of death of an individual who is from a city with similar demographics (and with no further information given) might be somewhat close to 3.4%.
• The proportion of people who will die from an outbreak in a similar city which spreads randomly will be close to 3.4%.

I can see why the claim is being made. A lot of people will extrapolate from sample to future individuals, but when we talk about individuals in real-life, we usually have information on the individual which we can condition on. It is not common to talk about probabilities for individuals dying without knowing more information about that individual, and it is also possibly not useful (eg. 3.4% at the group level, 0% and 10% at sub-group levels. Neither subgroup is well-approximated by the group.).

But we shouldn’t fully dismiss using unconditional probabilities. We just have to be more aware of hidden variables.