> Editor's note: Facebook is among NPR's financial supporters and since publishing her book, The Art of Screen Time, Kamenetz's husband took a job with Facebook. He works in an unrelated division.
Ok, here's one. Check out this paragraph, emphasis mine.
> According to Facebook's own annotations of the leaked slides, the finding broadly reported as "30% of teen girls felt Instagram made them feel worse about their bodies" was based on 150 respondents out of a few thousand Instagram users surveyed. They only answered the question about Instagram's role if they had already reported having body image issues. So the finding does not describe a random sampling of teenage girls, or even all the girls in the survey. It's a subset of a subset of a subset.
Can you see how insane this paragraph is? It's like Big Tobacco getting caught with a study that shows that smoking causes cancer. But then they annotate their leaked internal slides saying "well it only cause cancer in 150 out of like thousands of people.. trust us, we annotated the slides." My emphasis is to show how biased the author is -- are we seriously giving credence to the very company that the Wall Street Journal reporting was criticizing? And now we're trusting an employee that has a material stake (stock options, her salary, reputation) in protecting the company? I'm not sold that social media is the "worst thing ever," but NPR really needs to have higher journalistic standards.
> This type of innuendo based commentary is a staple of Fox and conservative media.
It really isn't, and I think this is a really unfair criticism of GP. It's pretty much just being skeptical 101.
I think a better analogy to smoking would be "30% of smokers felt that cigarettes caused them to cough chronically" in a survey of people who already had a chronic cough.
That's not the same thing as "smoking causes chronic cough in 30% of people" - rather "smoking makes coughing worse in 30% of people who already had a chronic cough." Maybe they got that from smoking, maybe they didn't, but the data doesn't show that, and it cannot be extrapolated to the general population. More data is needed.
It is in fact a subset of a subset.
[edit] btw 150 people can be enough to draw statistically significant conclusions but you need to properly design your survey and you need a suitably random sampling of your target population. Not sure this has either?
From page 4: The research presented instagram users with a survey about a select negative experience they may have had. Only if the user reported having had such an experience did they get a random "deep dive" ...
So the "reported" here just mean answered a previous question and does not precede using Instagram or smoking.
So the annotation of the slides indicates that the slides have misrepresented the study. For one thing, before pushing any agenda, it would be good to recognize this fact.
(From my perspective, I'm a statistician by training, the misrepresentation is egreggious IMHO. )
Innuendo? It's public knowledge that the author has competing interests. A Bayesian approach to the situation demands consideration of the possibility that the author is corrupt.
Bayesianism can't decide relevance (relevance isnt a statistical concept).
Indeed, "fallacies" are just discrete constraints on a problem like this.
Ie., that the author's relationship to facebook makes no difference to the truth of their claims shows that we need evidence relevant to their truth first.
In otherwords, you're effectively conditioning on the claims being false when you include this relationship.
Your point would be stronger if you attacked her arguments directly.