Junk Charts takes on “March mildness”

No, not the weather! We’re talking about bracketology. Apparently this was a year running low on upsets, and the NYT(New York Times) wanted to make a point. By improving on the NYT(New York Times) chart, Junk Charts makes some interesting discoveries: 10 beat 7 upsets occurred almost as frequently in recent history as 9-8 upsets. In fact, in 1999, all 9 and 10 seeds won!

11-6 and 12-5 upsets are not that uncommon (bye bye Duke!). 13-4 is relatively uncommon, and 14-3 upsets are rare (unfortunately including a nasty splotch on UNC’s record). 15-2 upsets have occurred once, and 16-1 upsets have never occurred as you hear over and over every march.

Technorati Tags: ,

Hey, Swivel’s up!

Share your data analysis all you want. Swivel‘s up!

Technorati Tags: ,

Law of the iterated logarithm, and a reminder on the limitations of statistics

We trust confidence intervals, and they are reported with just about every study in existence. However, they are often wrong, not just because they are wrong 5% of the time but also because they are often incorrectly computed.

In doing applied statistics day after day, it’s difficult to remember all of the theoretical statistics I learned in graduate school. (I got my degree at UNC-Chapel Hill, which is known for its theoretical statistics program.) So I brought home my copy of Van der Vaart’s “??Asymptotic Statistics??”:http://www.amazon.com/Asymptotic-Statistics-Statistical-Probabilistic-Mathematics/dp/0521784506/sr=1-1/qid=1164382703/ref=pd_bbs_sr_1/002-3849726-6409614?ie=UTF8&s=books. It’s a very nice combination of applied and theoretical statistics, with lots of examples and reminders about why the statistical methods we use in the applied world work the way they do. With this study came a stark reminder about some of the dangers in overinterpreting the results of statistical methods.

There’s this very odd looking result called the law of the iterated logarithm: lim supn→∞ (Y1 + … + Yn)/√(n log log n) = √2 for a sequence of random variables Y1, … with mean 0 and variance 1.

So what does this mean. lim sup means maximum “in the limit,” i.e. what is the maximum value of that strange-looking expression if you ignore the first 10, 100, or even 1,000,000 sums. Now, what about that that strange looking expression? It’s the average of Y1, …, Yn, multiplied by √(n/log log n). If you’ve happened to have gone through a few statistics classes and an advanced calculus class or two, you might recognize that the average goes to 0 (because that’s the mean of all the Yn) so “in the limit” that strange-looking expression is an indeterminate 0×∞. It converges to 0, but infinitely often it’s close to √2.

This is a result that applied statistics don’t use very often, but it still has important implications. I won’t go into the full argument, but an example illustrates how this means that confidence intervals are wrong.

Let’s say that we are measuring the average height in a population (and let’s say there are an infinite number of people). We measure 100 people and construct a 95% confidence interval AVE-2/√n – AVE+2/√n (I use 2 to make this simple, though statisticians might rather see 1.965.) And then say we do a sequential testing scheme: we add another person into the study and redo the average and confidence interval. Then add another person and redo the average and confidence interval again.

The confidence interval covers the true average height only if (AVE-true mean)*√(n/log log n) < 2/√log log n. But because of the law of the iterated logarithm, this fails infinitely often in our sequential testing scheme above.

Two take home lessons:
# 95% confidence intervals are wrong 5% of the time, and refining them to make them smaller doesn’t change that fact
# approaching sequential testing in a naive way leads, infinitely often, to confidence intervals that are wrong

Technorati Tags:

The evil twin brother of Number Needed to Treat

Some weeks ago I posted an entry on the NNT(Number Needed to Treat), which is essentially the expected/average number which you would have to give a treatment (surgery, pharmaceutical, or device) at the labeled dose/frequency to receive the labeled benefit.

When you are talking about adverse event risk, the number is NNH(Number Needed to Harm), which is the expected/average number who would have to take a treatment at the labeled dose/frequency to receive the noted adverse effect. You want these numbers to be large.

See “here”:http://www.jr2.ox.ac.uk/bandolier/booth/glossary/NNH.html for more info.

(h/t “Pharmagossip”:http://pharmagossip.blogspot.com)

Technorati Tags:

The bar is higher for Democrats

Andrew Gelman (a fellow statistician) and colleagues analyze the probability that the Democrats will retake the House. He has a few interesting insights about the past few elections as well. Remember that Gore won the popular vote but lost the electoral college vote in 2000? There’s been a similar phenomenon in congressional candidates in most elections since 1994. At any rate, realize that the linked article is a bit heavy on the statistics, but you should be able to get the gist.

Merck’s MK-0557 — statistical vs. clinical relevance strikes again

Merck got their “p-value”:http://www.healthday.com/view.cfm?id=535281, but no one cares about losing three more pounds in _one year_ over placebo. And certainly no one wants to take a drug once a day for 365 days to do it. It looks like MK-0557 won’t be making it in the anti-obesity field.

Other fun issues with the drug and its trial:

* 832 out of 1661 enrolled subjects completed the trial (just over 50%)
* A prominent doctor comments

“In my view, this trial suggests not that a cocktail of drugs will be needed, but that for the most part, drugs are not the right answer at all,” said Dr. David L. Katz, an associate professor of public health and director of the Prevention Research Center at Yale University School of Medicine.

“Though rare case of obesity may warrant medication as part of a comprehensive treatment plan, the hope that drugs will save most of us the trouble of addressing weight control through lifestyle practices is misplaced,” he said.

* The target receptor of the drug was completely blocked, but does not appear to have the same biological activity as researchers once thought. Or shutting it down puts a higher load onto other systems. Obesity is a complex issue.
* Except in a relatively few instances, diet and exercise are the best ways to prevent and alleviate obesity.

Technorati Tags: , ,

Slate has a good article on some good consumer critical thinking about statistics

Slate has a good “article”:http://www.slate.com/id/2150354/?nav=ais on how to think about whether you should take a drug. In the confusing world of “relative risk vs. absolute risk”:http://www.randomjohn.info/wordpress/2006/05/03/lying-with-statistics-relative-risk-vs-absolute-risk/, it’s really hard to know the effect of a drug.

Enter the NNT(Number Needed to Treat). The idea behind this number is the _expected_ number of people that you would have to treat so that _one_ person would realize the benefit of the treatment. For example, if the NNT is 3, then you would expect one out of every three people to benefit from the treatment.

Let’s take the Pravachol (a statin, like Lipitor) example from the article. In a 1995 study in ??NEJM(New England Journal of Medicine)??, researchers reported a 31% reduction in the risks of heart attack in men who took one Pravachol every day for five years. 7.5% in the placebo experienced a heart attack vs. 5.3% in the Pravachol group — a 31% relative reduction in risk or a 2.2% absolute reduction. The NNT (see more “here”:http://www.cebm.utoronto.ca/glossary/nntsPrint.htm) is 1/2.2% = 45.5. So you would expect to have to give over 45 men Pravachol once a day for five years to prevent one heart attack. Turned around, we expect that over 44 of them would not avoid a heart attack (either would not experience one any way, or would not be prevented).

I’ll leave all commentary aside about whether drug companies want you to think that way. The data coming from premarketing approval has to be made public (as a certain company just found out), and anyone with a calculator and absolute risk in hand can calculate an NNT.

Slate has a few interesting NNTs:

|_Drug_|_Indication_|_NNT_|
|cortisone|painful shoulder|3|
|amoxicillin|shorten fever for ear infection|20|
|Proscar – 4 yr|Avoid surgery for enlarged prostate|18|
|Aspirin|Avoid heart attack|208|

Think about it. Think about how much you spend each year on some of these drugs, and think about what the chance is they help.

(h/t “insider”:http://pharmagossip.blogspot.com)

Technorati Tags: ,