Re: Study data for dummies

[Guest guest]

You're looking for an introductory college level book on statistics.

http://www.amazon.com/s/ref=nb_ss_gw/002-8518248-4389613?url=search-alias%3Daps & field-keywords=introductory+statistics

or start easier

http://www.amazon.com/s/ref=nb_ss_gw/002-8518248-4389613?url=search-alias%3Daps & field-keywords=statistics

We live in a world without absolutes, but:

a study of 15 people which reports a small improvement may be the result of random chance, even at a 95% confidence level I'm sorry to say, which I know is confusing;

barring systematic bias, we can be more confident of results if a study includes more people (more than 65 eliminates a lot of potential problems), more observations, and reports a large difference.

P and T confidence intervals express how much confidence you have in the results. The usually accepted level of confidence is 95% or (P= .05), although 99% or (P= .01) is a more confident standard. http://www.tufts.edu/~gdallal/p05.htm

Confidence levels are often used in conjunction with an error range.

The last time I checked, the accepted error range for a Viral Load Test is 0.5 log (P= .05). If your viral load is 5,000 that is 3.7 logs. Your statistical estimate is 5,000 but you can be 95% confident that your actual viral load is between 3.2 and 4.2 logs, or somewhere between 1,581 and 15,811. That's what 5,000 really means.

So when you drink Lipton Tea and your viral load is now reported as 2,500 - we can have virtually no confidence that Lipton Tea is an antiviral because this difference in numbers could have resulted from two different tests of the same blood sample due to the inherent error in the testing method.

Likewise, statistics is the cause of a lot of "viral blips". Lets say we magically knew your viral load was 49, one less than "undetectcable". This 49 viral load would be reported as some number between 16 and 155 - or between undetectable and 155. Its really 49, but you have no way of knowing this. All you see are periodic viral blips.

Likewise a change from a T4 percentage of 23% to less than 25% is not statistically valid at a 95% confidence level, due to the error range of a flow cytometer.

You will also see many other statistical tests of relevence used such as Chi Square.

http://www.medscape.com/viewarticle/416762_5

http://en.wikipedia.org/wiki/Statistical_significance

http://en.wikipedia.org/wiki/Confidence_interval

http://en.wikipedia.org/wiki/Poisson_distribution

http://en.wikipedia.org/wiki/Chi-square_distribution

>> I've got a confession to make. I'm scientifically illiterate. I've been a member of this list since it's inception, and was a member of the predecessor list before that, but I still don't know how to interpret study data. Of course, I get the gist of the study by reading the abstract, summary and conclusion, but when I want to delve deeper to understand the data used to support the conclusion, I get lost. For example what does it mean when data is followed by "(P=.012)"? Is that the margin of error? Is there a website or article that could be posted for the benefit of all of us scientific illeterates (I'm sure there are many of us on this list) that explains how to read study data? If not, perhaps one of you more scientifically literate members could deconstruct a study report.> > Thanks,> > >