Status Updates From How to Make the World Add U...

Google Flu turned out to be winter detector, high school basketball predicted flu, "You Look Like A Thing and I Love you" algorithm shown healthy skin and skin cancer which had ruler in photograph, "Big Data" by Cukier and Mayer-Schonberger used Google Flu Trends before collapse "Weapons of Math Destruction"

Google Flu turned out to be winter detector, high school basketball predicted flu, "You Look Like A Thing and I Love you" algorithm shown healthy skin and skin cancer which had ruler in photograph, "Big Data" by Cukier and Mayer-Schonberger used Google Flu Trends before collapse "Weapons of Math Destruction"

"Dark data"missing responses when try to conduct polls or census--must as k who's missing, can lead to sampling bias. more answers doesn't mean more accurate-Gallup poll with 3000 ppl beat poll with 2 mill, also algorithmic bias if shown skewed sample

"Invsible women"-women ignored,police vests ignore vests,Apple health app ignores menstruation,Asch conformity experiments at Swarthmore only men, Milgram's only men,thaliomide resulted in women excluded from clinical trials, COVID data didn't aggregate cases by gender so couldn't say why men dying more

Role of science journalists to translate scientific knowledge, can be superficial retread of press release or can explain facts, put them into context and speak truth to power.ask how large effect, was research done with humans,mice or petri dish, make space to explain. Cochrane Collaboration does systematic research review

antidepressants 48-3 positive turned out to be 38-37 because s22 had negative result and unpublished, 11 had negative results to FDA Erick Turner NEJM 2008,Jan17. "Selective Publication of Antidepressant Trials and Influence on Apparent Efficacy"

publication bias where only interesting results get publish, negative or studies that do not replicate prior study don't get published, Kickstarter has few standouts that get funded but many don't get any money

2008 study of anti-depressants 48 trials pos & 3 none,researchers found 23 unpublished trials w/22 neg where drug didn't help,11 of trials pos in articles had produced neg results in FDA summary, cherrypick good news and hand-wave bad news about drug not effective. Corrected score not 48-3, but 38-37.