They studied 100 sites, but said only 12 showed declining populations. Only for
one site did they give a percentage change. From the article:
[in]"one of their earliest trapping sites from 1989, the total mass of their catch had fallen by nearly 80%"
What about the other 88 sites? What about the percentage decline in the other 11 declining sites? How were the sites chosen? How have the sites changed over the years?
This is alarmist ecological drama. Some readers on this website are extrapolating the results from a single experiment at a single site to the entire world. And I think Science magazine did a poor job of reporting this correctly.
Here's a link to another alarmist post about this study. Note the url:
http://www.bouldercountybeekeepers.org/insect-biomass-declin...
Here's a link to AFAICT a copy of the paper that started this brouhaha. Read it and be questionably impressed:
http://www.bouldercountybeekeepers.org/wp-content/uploads/20...
The authors of the paper are doing good work, but the alarmism here on HN far exceeds the scope of any validity of the original paper. People should be wary of their confirmation bias.
And if a HN reader cannot make a reasonable judgment about the scope of this study, I doubt that (s)he would be successful in an enterprise such as Machine Learning, given that science and statistics are the very basis of ML.
OTOH Woody Allen once said, and I begrudgingly admit he has a strong point, "Showing up is 80 percent of life".