Re: Don't Believe the Hype -- world

[ Show ]

Support VoyForums

[ Shrink ]

VoyForums Announcement: Programming and providing support for this service has been a labor of love since 1997. We are one of the few services online who values our users' privacy, and have never sold your information. We have even fought hard to defend your privacy in legal cases; however, we've done it with almost no financial support -- paying out of pocket to continue providing the service. Due to the issues imposed on us by advertisers, we also stopped hosting most ads on the forums many years ago. We hope you appreciate our efforts.

Show your support by donating any amount. (Note: We are still technically a for-profit company, so your contribution is not tax-deductible.) PayPal Acct: Feedback:

Donate to VoyForums (PayPal):

04/26/25 2:13:56am

[ Login ] [ Main index ] [ Post a new message ] [ Search | Check update time | Archives: 1, 2, 3, 4, 5 ]

Subject: Re: Don't Believe the Hype

Author:
krz

[ Next Thread | Previous Thread | Next Message | Previous Message ]

Date Posted: 02/ 4/06 3:02:33pm
In reply to: tjm 's message, "Don't Believe the Hype" on 02/ 3/06 9:46:20pm

I'm not familiar with the book he cites, but am familiar with the statistical techniques he references (I use many of the same in my work). G and I have also been "reading" (actually listening to an audio version) of the book Freakonomics - which attempts to statistically model things like "motive" (to cheat for example, or to work hard to sell a home).

What was interesting to me coming from the statistical perspective is that when researching if we find a model that explains 25% of the variance in a complicated set of data - we are thrilled. Think of it this way - with all we know about the probability of having a heart attack (genetics, lifestyle, age, other health risks, emotional balance, marital status, social status, education -- you get the point), if I could explain without question 25% of the risk - then I can reduce the population risk by a similar amount (assuming the factor is controllable).

What I always struggle with personally is that statistically I manipulate the data from groups of people and it takes groups of people to have data powerful enough to find an effect of something. Groups data rarely explains individual performance, so the statistics help me explain a phenomena but don't always assist in good decision making with a patient in front of me. It's an interesting conundrum.

Annoys me to tears that reporters attempt to fortell the future with data from the past. We are so careful when writing that we are incapable of predicting future events - any statistical analysis is only meant to represent a present state of the data and is not meant to suggest a prediction. Even when done - what reporters rarely do is present something called the R2 ("R squared") otherwise known as the coefficient of determination which tells the reader how much of the change in a set of data the resultant analysis explains. You can have a very statistically strong predictor equation that only explains about 5% of the variance - so not meaningful even if significant.

Reporters should be obligated to take a basic intro course to stats and a basic intro course in scientific rhetoric (not to mention mandatory ethics courses).

Have you gotten to podcasting yet? Science Friday and the show Nature are excellent examples of reporting the state of the work and not extrapolating to predictions.

interesting article t

[ Next Thread | Previous Thread | Next Message | Previous Message ]