• extrapolating Maria's death toll

    From Rich Ulrich@21:1/5 to All on Mon Jun 4 12:14:11 2018
    The original, "official" death toll in Puerto Rico from Hurricane
    Maria was 64 - those were directly attributed.

    "Excess deaths" in 42 days after from September 20, 2017,
    came to 1052. There are a few other estimates, using different
    methods and different time frames.

    https://www.nytimes.com/2018/06/02/us/puerto-rico-death-tolls.html


    What has been in the news in the past week is a point
    estimate of 4645 deaths. That is based on a survey of 3000
    randomly selected households. Those households described
    38 deaths of members of the extended family. Using whatever
    demographic information information they had available, the
    surveyors extrapolated that to obtain a confidence interval
    of approximately (800, 8500) for the number of total deaths.
    The reported number, 4645, is the midpoint of the 95% CI.

    So - There is an estimate of 4645 deaths, extrapolated from
    an observed count of 38 deaths.

    What surprises me, a bit, is that the estimate is repeated so
    often as if it were a firm number; and no one points to the
    error range.

    I suppose that once you get into complications like the error
    range, you also might be pushed to discuss why the whole
    esimate is apt to be biased to the low end - which is also
    complicated.

    The same author, Sheri Fink, also had a May 30th article
    on the topic. I was going to post a comment based on
    that, until I Googled and found the newer report.

    --
    Rich Ulrich

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Stan Brown@21:1/5 to Rich Ulrich on Mon Jun 4 20:04:45 2018
    On Mon, 04 Jun 2018 12:14:11 -0400, Rich Ulrich wrote:
    the
    surveyors extrapolated that to obtain a confidence interval
    of approximately (800, 8500) for the number of total deaths.
    The reported number, 4645, is the midpoint of the 95% CI.

    So - There is an estimate of 4645 deaths, extrapolated from
    an observed count of 38 deaths.

    What surprises me, a bit, is that the estimate is repeated so
    often as if it were a firm number; and no one points to the
    error range.

    It doesn't surprise me. I always insisted that my students report
    intervals, not midpoint error, and for exactly this reason.

    (In this case, the midpoint is 4650 not 4645; was that a typo, or a
    mistake from another source?)

    What we don't know (at least I don't) is whether, in this case, the
    interval 800-8000 was given to the media and they computed or
    miscomputed the midpoint, or 46503850 was given to the newspapers,
    and they dropped the 3850.


    --
    Stan Brown, Oak Road Systems, Tompkins County, New York, USA
    http://BrownMath.com/
    http://OakRoadSystems.com/
    Shikata ga nai...

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Rich Ulrich@21:1/5 to the_stan_brown@fastmail.fm on Tue Jun 5 00:40:19 2018
    On Mon, 4 Jun 2018 20:04:45 -0400, Stan Brown
    <the_stan_brown@fastmail.fm> wrote:

    On Mon, 04 Jun 2018 12:14:11 -0400, Rich Ulrich wrote:
    the
    surveyors extrapolated that to obtain a confidence interval
    of approximately (800, 8500) for the number of total deaths.
    The reported number, 4645, is the midpoint of the 95% CI.

    So - There is an estimate of 4645 deaths, extrapolated from
    an observed count of 38 deaths.

    What surprises me, a bit, is that the estimate is repeated so
    often as if it were a firm number; and no one points to the
    error range.

    It doesn't surprise me. I always insisted that my students report
    intervals, not midpoint ± error, and for exactly this reason.

    (In this case, the midpoint is 4650 not 4645; was that a typo, or a
    mistake from another source?)

    What we don't know (at least I don't) is whether, in this case, the
    interval 800-8000 was given to the media and they computed or
    miscomputed the midpoint, or 4650±3850 was given to the newspapers,
    and they dropped the ±3850.


    See the article - I took the "approximate" description of the CI
    to mean that numbers were rounded (a) for convenience or
    (b) to avoid the implication of more precision than the numbers
    deserve. And 4645 would be rounded from the original point-estimate.

    I'm glad to see your comment. Now I think I see the source
    of something that was bothering me --

    I presume that the media worked from a press release, and
    the scientists got the coverage they intended. The "official"
    number of 64 was irresponsible, or, it is irresponsible to leave
    it hanging out there as if it were meaningful. There are several
    estimates of a bit over 1000 deaths, for a shorter period. And
    there were over 900 cremations with no report of causes of death.

    I think that the folks publishing in the New England Journal
    of Medicine were willing to over-emphasize the "4645" in a press
    release because, given all the other information, the whole range
    is set lower than it ought to be. So, they won't be shown to be
    grossly over the right figure, if better news comes out. And they
    thought that there was public value in being "impressive" rather
    than being more technically precise.

    - This is part of the awkward interface between science and
    the public assimilation of science.


    --
    Rich Ulrich

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)