[FRIAM] Statistical poser (aka fact checking is hard)

[Robert Holmes]

Early this week I came across a recent press release from NM Dept of
Health: "Governor Martinez Announces Continued Improvement in Drug Overdose
Death Rankings" <https://nmhealth.org/news/information/2018/12/?view=728>.
I've been tinkering round with opioid statistics, so thought it might be
worth fact checking the release. The results were… interesting. If nothing
else they've shown me how difficult it must be to communicate public health
statistics.

So here are some of the key figures from the release:

   - New Mexico’s national ranking has improved from the second highest
   death drug overdose death rate in the United States in 2014 to 17th highest
   in 2017
   - New Mexico previously reported a 4 percent decline in death rates in
   2017 due to overdose of commonly prescribed opioids such as oxycodone
   compared to 2016. In addition, deaths due to heroin decreased by 9 percent
   and deaths due to synthetic opioids such as fentanyl decreased by 6 percent
   over the same time period.

The first of these claims passed my sniff test: I know NM's ranking has
been improving, even though individual counties rank the wort in the
nation. And sure enough, if you pull the underlying CDC data
<https://wonder.cdc.gov/mcd.html> you can confirm these exact numbers
(ignoring DC).

The second claim is the one that gave me pause. Those reductions in
individual opioids look kinda high. Yes, NM's ranking is improving but it's
because our rate is essentially stable while other states rocket past. And
when I check the above CDC data, yes the reduction in death rate appears to
be about 2%

So there's the poser: if NM's reduction in opioid deaths (2016-2017) is 2%,
how can this be consistent with individual opioid reductions of 9%
(heroin), 4% (natural & semi synthetic, inc. oxycodone), and 6% (synthetic,
inc. fentanyl)?

—Robert

P.S. I'll post my best guess later. Oh, and it's not that they omitted
methadone: deaths due to that are down 19% in the same period.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20190102/3b8a395d/attachment.html>