Does the study on mortality rates in different countries take into account the different methods of tracking and defining "Covid death" in each country, and in each state in the US?
Great question! For those not as familiar with this, it is not always clear what someone has died of when they die. Imagine, for example, a COVID-recovered patient who dies two weeks after leaving the hospital from a cardiac event. Are they a COVID death? This all depends on the way that data are classified in the country and locality where they died, and also the evidence that COVID-19 was responsible for their death. It’s actually rather complicated to decide what someone died of.
The researchers in this study used statistical methods that attempt to get around this kind of kind of variation. Specifically, the estimates of mortality were conduced using a regression analysis based on something called a fixed-effects model. I am not an expert in statistics, but my understanding of fixed-effects models is that they are used under conditions when it is expected that certain data might be in error for reasons that are uniform across the dataset; ie, where it is unlikely that variations in the data occur at random, but instead occur due to some fixed effect.
The application of national and locality medical criteria to adjudicating deaths potentially due to COVID-19 could be seen as a fixed effect. There would be some sort of random variation between different individual medical practitioners ruling on cause of death, but at the national level one might expect that discrepancies between countries would be nonrandom and would occur due to differences in these rules.
I am not sure that this fully accounts for differences between countries in terms of adjudicating deaths; it is just my interpretation of what is written in the methods section of the paper. If any readers see a serious error in what I’ve written here, please let me know.
That said, I would summarize all questions about modeling and estimation of this type by noting that no model is ever perfect. The question is whether the imperfections made the model lose usefulness. I don’t believe that in this case they did. I think that OECD countries have relatively similar methods of adjudicating deaths, and even with minor variations I wouldn’t expect such disparities as ended up appearing between the US and other countries in the analysis we talked about.
Does the study on mortality rates in different countries take into account the different methods of tracking and defining "Covid death" in each country, and in each state in the US?
Great question! For those not as familiar with this, it is not always clear what someone has died of when they die. Imagine, for example, a COVID-recovered patient who dies two weeks after leaving the hospital from a cardiac event. Are they a COVID death? This all depends on the way that data are classified in the country and locality where they died, and also the evidence that COVID-19 was responsible for their death. It’s actually rather complicated to decide what someone died of.
The researchers in this study used statistical methods that attempt to get around this kind of kind of variation. Specifically, the estimates of mortality were conduced using a regression analysis based on something called a fixed-effects model. I am not an expert in statistics, but my understanding of fixed-effects models is that they are used under conditions when it is expected that certain data might be in error for reasons that are uniform across the dataset; ie, where it is unlikely that variations in the data occur at random, but instead occur due to some fixed effect.
The application of national and locality medical criteria to adjudicating deaths potentially due to COVID-19 could be seen as a fixed effect. There would be some sort of random variation between different individual medical practitioners ruling on cause of death, but at the national level one might expect that discrepancies between countries would be nonrandom and would occur due to differences in these rules.
I am not sure that this fully accounts for differences between countries in terms of adjudicating deaths; it is just my interpretation of what is written in the methods section of the paper. If any readers see a serious error in what I’ve written here, please let me know.
That said, I would summarize all questions about modeling and estimation of this type by noting that no model is ever perfect. The question is whether the imperfections made the model lose usefulness. I don’t believe that in this case they did. I think that OECD countries have relatively similar methods of adjudicating deaths, and even with minor variations I wouldn’t expect such disparities as ended up appearing between the US and other countries in the analysis we talked about.