Another reason why models can't be trusted

For ANYTHING, let alone the effects of pollution on populations.

The money quote, and the reason why a lot of modeling doesn't predict future or past events:

In the case of the CMA’s model, the selected pollutants are ozone and particulate matter of 2.5 microns. Boadway explains these two factors were chosen among all other possibilities on the recommendation of an international panel of experts. Choosing your explanatory variables ahead of time and then looking for significant links is called model selection.

McKitrick uses a different approach, something called Bayesian model averaging. Such a method is necessary, he claims, due to the sheer number of possible variables involved. “There are literally tens of billions of potential combinations,” McKitrick observes. Besides numerous different forms of air pollution, researchers may also include time-delayed measurements, as well as independent factors such as weather, lifestyle and income. Allowing authors to arbitrarily pick the variables through model selection creates uncertainty and opens the potential for cherry-picking the most desirable results, he says.

It's the same with "climate change" as with this study, in the end, no two people have the same circumstances, and so we can only study correlations, not direct links.