Forgot to mention the really difficult part is correctly figuring out the range of those results. A good well controlled study will have a very narrow range. A study that has problems with reliability, sample size, etc, will have a very wide range. Another way to look at it is if the range of differences encompasses 0 by any substantial amount, most likely it means that the differences are not meaningful.
Speaking of such, I'm prepping a statistical criticism of the latest book byCharles Murray, author of the Bell Curve. Want to join in? On Wednesday, February 15, 2012, Larry C. Lyons <larrycly...@gmail.com> wrote: > You are not the only one. On my desk at home is a notebook with all my notes for the next version of my meta-analysis application. 150 pages and counting - most of which are botched formulae for calculating statistical power effect sizes and converting obtained probability values to effect sizes. Makes me wish at times I stayed with single case designs. > > 10 word or less that is really difficult. Can I go for 30? > > But you've essentially got the idea. I left out a lot, range estimation and correction for error andthat sort of thing, but yes. > > On Wednesday, February 15, 2012, Dana <dana.tier...@gmail.com> wrote: >> >> what not really -- the meaning of standard deviations? If so yeah you are >> right, I think but what Maureen and I said is an .... ok 10 words or less >> version. >> >> In this case p=0.011 so theoretically if they did everything else right, >> these results should replicate 99% of the time. And not, 1%. >> >> I realize that's it's not a given that the 1% is random or that it won't >> occur the next time you repeat the experiment, but I think that is a rather >> fine distinction for our purposes. Kinda like the difference between >> Springfield and Tyson's Corner, as seen from California, yanno? If I don't >> have that right then fine, tell me, but if you're going to crank up your >> statistical powers I'd rather hear an explanation of that leave one out >> thing they did a thousand times, because that part I do not understand at >> ALL. >> >> On Wed, Feb 15, 2012 at 6:21 PM, Larry C. Lyons <larrycly...@gmail.com >wrote: >> >>> >>> Not really. It depends on the stats that are used. When looking at >>> statistical results, the way to interpret statistical significance is as >>> follows. Let's say the researchers found the two groups showed a >>> significant difference of p < 0.05 . This means that if you replicated >>> the study an infinite number of times, 95% of these results would fall very >>> close to the difference found in the first study. How meaningful that >>> spread is depends on the standard error of the studies, and other factors. >>> It also mean that in order to show a significant difference with a smaller >>> sample you'd need a much larger difference to achieve statistical >>> significance. >>> >>> So you can make very accurate predictions based on fairly small samples. It >>> all depends on the statistical power of your experiment. I'm too burned out >>> to really discuss it now, but if interested Wikipedia has a pretty good >>> explanation of it - http://en.wikipedia.org/wiki/Statistical_power >>> >>> On Wednesday, February 15, 2012, LRS Scout <lrssc...@gmail.com> wrote: >>> > >>> > The sampling of 90 people is really really small. >>> > >>> > On Wed, Feb 15, 2012 at 7:29 PM, Dana <dana.tier...@gmail.com> wrote: >>> > >>> >> >>> >> feel free to run away, Sam, but you still haven't showed me any basis at >>> >> all for the crap you've been talking. >>> >> >>> >> On Wed, Feb 15, 2012 at 4:18 PM, Sam <sammyc...@gmail.com> wrote: >>> >> >>> >> > >>> >> > I give up and feel the fool for not heeding this advice sooner: >>> >> > >>> >> > Dont argue with idiots. They drag you down to their level and beat >>> >> > you with experience >>> >> > >>> >> > . >>> >> > >>> >> > On Wed, Feb 15, 2012 at 7:07 PM, Dana <dana.tier...@gmail.com> wrote: >>> >> > > >>> >> > >> >>> >> > >> Yes it is. It's the same study done three times. Two people, 90 >>> people >>> >> > >> and 28 people. >>> >> > >> >>> >> > > >>> >> > > Ah, here's the heart of the problem. No, Sam, it isn't. It's -- I'd >>> >> call >>> >> > it >>> >> > > two studies and an experiment I guess -- that tested the same >>> >> hypothesis. >>> >> > > According to your nomenclature here, all trials for the same drug >>> are a >>> >> > > single study. And mutually responsible for one another's >>> methodology. >>> >> > And, >>> >> > > according to you, everything anyone remotely affiliated with them >>> may >>> >> > have >>> >>> >> > > with ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~| Order the Adobe Coldfusion Anthology now! http://www.amazon.com/Adobe-Coldfusion-Anthology/dp/1430272155/?tag=houseoffusion Archive: http://www.houseoffusion.com/groups/cf-community/message.cfm/messageid:346965 Subscription: http://www.houseoffusion.com/groups/cf-community/subscribe.cfm Unsubscribe: http://www.houseoffusion.com/groups/cf-community/unsubscribe.cfm
