|
Posted by Wildbilly on January 26, 2010, 11:56 am
> x-no-archive: yes
>
> Michael Eades discusses saturated fat studies:
>
> http://tinyurl.com/yklgays
>
>
> As I¹ve discussed before on these pages, meta-analyses are not my
> favorite types of studies. I¹ve attacked them when they¹ve been used to
> Œprove¹ the low-fat diets are better, so I can¹t very well embrace
> meta-analyses when they present a conclusion I agree with. And I really
> can¹t embrace meta-analyses when they are compilations of observational
> studies, which are themselves next to worthless.
>
> For those who don¹t know, meta-analyses are compilation studies in which
> researchers comb the medical literature for papers on a particular
> subject and then combine all the data from the individual studies
> together into one large study. This combining is often done to bring
> together a collection of studies, none of which contain data that has
> reached statistical significance, to see if the aggregate of all the
> data in the studies reaches statistical significance. I think these
> types of meta-analyses are highly suspect, because they can lead to
> conclusions not warranted by the actual data.
>
>
> To give you an example of what I mean, let¹s assume that we have a study
> looking at a flipped coin. If a researcher flips a coin 10 times and
> comes up with 6 heads and 4 tails, runs this through a program checking
> for statistical significance, he/she will discover that the 6-4 ratio
> isn¹t a statistically-significant difference because of the low number
> of overall flips (10). Now, let¹s say that 50 researchers did the same
> kinds of study and some found that their coins came up heads 6 times out
> of 10 or 4 times out of 10, etc. If a researcher then wants to Œprove¹
> that heads comes up more times than tails on a coin flip, he/she can
> gather all the studies showing heads come up more times than tails, add
> them together in a meta-analysis and come up with 25 studies, each with
> 10 flips, showing that heads came up 63 percent of the time. Now we¹re
> talking 250 flips and we would probably reach statistical significance.
> We know that over the long run a flipped coin is going to come up
> heads about 50 percent of the time and that the more the times it is
> flipped the more likely the number of heads will close in on the 50
> percent figure. But, the meta-analysis that selected the studies
> showing the 63 percent heads is statistically significant because the
> studies were cherry picked.
>
> Researchers using meta-analyses set up selection criteria to pick which
> studies will be included in their final product, which leaves the door
> open for all kinds of mischief. For example, let¹s say a researcher
> wants to make the case that low-fat diets reduce cancer. He/she would
> create a set of criteria, do a literature search for all the studies
> that meet those criteria, then do a statistical analysis of all the
> data. If the data demonstrate that low-fat diets are linked to lower
> rates of cancer to a statistically significant degree, the researchers
> submit their paper for publication. But let¹s say that when the data is
> crunched, it doesn¹t show any such relationship? It¹s easy to go
> through all the studies and find which ones strongly show the opposite
> of what the researchers want to show and then figure out how to change
> the study-selection criteria in such a way as to keep those studies from
> being selected, run the whole process again, and repeat until enough
> studies are found to make the meta-analysis show the link between
> low-fat diets and lower rates of cancer.
>
> Sad to say, this is often how it is done. Which is why I don¹t give a
> lot of credence to meta-analyses.
>
> But having said all this, I¹m still happy to see a researcher with the
> academic credentials of Ron Krauss coming out with a meta-analysis
> showing no correlation between saturated fat intake and cardiovascular
> disease risk. And getting it published in the AJCN, probably the
> world¹s most prestigious nutritional journal, no less. It¹s called
> putting your money where your mouth is. Many academics whom I¹ve spoken
> with admit that there is no correlation, but wouldn¹t risk their
> academic reputations doing a meta-analysis to Œprove¹ it.
>
> I¹ve had many people tell me that it¹s really nice to finally see some
> studies coming out vindicating saturated fats. Or at least not
> attacking them.
>
> I have to tell them that pro-saturated fat studies have been around for
> years. Not just observational studies or meta-analyses, but real
> controlled studies looking at death rates from heart disease as a
> function of fat intake.
>
> Let¹s look at a couple.
And then there is
http://www.biblelife.org/stefansson2.htm
Don't let the name of the web site weird you out.
--
"Fascism should more properly be called corporatism because it is the
merger of state and corporate power." - Benito Mussolini.
http://news.yahoo.com/s/ap/20100119/ap_on_re_mi_ea/ml_israel_arresting_activists
http://www.democracynow.org/2010/1/19/headlines
| 
Yahoo! 
Google 
Windows Live 
del.icio.us 
digg 
Netscape 
XML Sitemap