But as you have said, even ignoring the outlier you would still expect a 1-1.5% decrease in mortality. Taking the 2018 deaths and extrapolating forward completely ignoring 2019 still gives you double the excess deaths compared to the mean/median
-
-
Ideally, what you'd do is a Bayesian inference that takes into account the uncertainty of both the initial downtrend and the outlier of 2019, which I suspect is what the SCB has done, but even your crude example shows how useless it is to simply use the mean
1 reply 0 retweets 0 likes -
I agree that ignoring a possible downward trend in mortality was an oversight on my part. Something like 93-96k sounds about right for excess of 2-5k.
2 replies 0 retweets 0 likes -
Replying to @TedPetrou @zorinaq
96k would be ignoring the downtrend tho. By your simple regression model, it should be around 94k, and my guess is a more complex inference would revise that downward somewhat after taking into account the uncertainty of 2019 and previous years
1 reply 0 retweets 0 likes -
Im not ignoring it at all. There is large variance here and we don't know if the downtrend would continue. The 5 years of 2014-2018 were relatively flat. Thats why I gave a range. 96k would be at the high end and unlikely.
1 reply 0 retweets 0 likes -
Replying to @TedPetrou @zorinaq
I mean, it's a strong trend that has been present for the last 50 years not just in Sweden but across most/all developed nations. 96k requires you to assume that such a trend would not only discontinue but, given the 2019 figures, reverse!
1 reply 0 retweets 0 likes -
The trend has been slowing down and its very difficult to extrapolate. There is a probability of increase. Just like life expectancy in the US decreased for men in 2019 (I believe). Strange things happen. This is how most people define ranges -formally it is a confidence interval
1 reply 0 retweets 0 likes -
Replying to @TedPetrou @zorinaq
Actually, that is not the correct definition of a confidence interval. It sounds more like you're trying to create a predictive interval taking into account uncertainty around the predicted decline, except it's very one-sided
1 reply 0 retweets 2 likes -
If you were to take a realistic predictive interval taking into account the fact that 2019 is *likely* but not *definitely* an outlier, your true result would probably include 96k but also dip down into ~88k or lower
3 replies 0 retweets 2 likes -
I'd suggert that
@TedPetrou validates his method by using it to calculate the expected deaths *of the previous years* (like 2016, 2017, 2018, 2019) using his method of considering rates from a suitable but fixed number of years prior, similar to what he did for 2020.2 replies 0 retweets 2 likes
Yes, validating the method by looking at previous years would be one method of seeing how large the errors are
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.