Marcela’s Reviews > The Data Detective: Ten Easy Rules to Make Sense of Statistics > Status Update

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

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antidepressants 48-3 positive turned out to be 38-37 because s22 had negative result and unpublished, 11 had negative results to FDA Erick Turner NEJM 2008,Jan17.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2008 study of anti-depressants 48 trials pos & 3 none,researchers found 23 unpublished trials w/22 neg where drug didn't help,11 of trials pos in articles had produced neg results in FDA summary, cherrypick good news and hand-wave bad news about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.

survivor bias distorts studies of investment performance, if ignore all funds that quietly disappear, apparent performance twice as good as actual perf. Result is to persuade ppl to invest in actively managed funds which often charge high fees when better served by low-cost passive whole stock market fund, multibillion $ mistake. Contest cow same as stockbrokers

Even cases of COVID-19 are subject to selective attention:people who feel terrible go to the hospital and are tested for the disease; people who feel fine do now.As a result, the disease looks even more dangerous than it really is.Even though statisticians understand this problem well there's no easy way to solve it without systematic testing,elusive in early stage of pandemic when diff policy decisions being made

How to evaluate a statistical claim: 1rst chap:try to notice your feelings about the claim, 2nd chap: constructively sense-checking the claim against your personal experience, 3rd chap:asking yourself if you really understand what the claim means.4th:Step back and look for information that can put the claim into context. Try to get a sense of the trend.