Wednesday, May 26, 2010

Origin Of Genital Crabs



back in vogue again the controversy on alternative medicine. Paul's last post dedicated to Clara Palomba, among other supporters of homeopathy, occasionally peeping guy saying "with me and it worked people x". Now, no doctor will not tell you anything more placebo effect than it does already good WeWee . But I would like to relate a little incident when I was studying "experimenters to Physics First", known to friends as "a fisichetta. It was an exercise.

Suppose you have invented a new wax ski, and want to test to see if it works better than the old. Take two equal slopes, with the two waxes waxing them and make them get up a snowy slope. The result of ten tests, eight times with the ski wax comes before the new skis with the old wax. This is enough to say that the new wax is better than the old one?

Instinctively, you are taken to answer yes. 80% is well over 50%, right? "It can not be a coincidence." Well, if you properly apply the scientific method, however, it is concluded that this result is that you can safely discard the new wax. Here's why.

first point: if we want to be sure to give ok to the new wax only if it is better than the old, we must "think of evil": that is, we assume that the new wax is NOT better than the old. Only the net did absolutely force us to change your mind, but when we do, the risk of doing so will be minimal to no avail.

So this is the so-called "null hypothesis . In this case, it means that the probability of a cross before the other arrivals were 50% (note, we are not saying, we're assuming). So, the first attempt, we will have a 50% chance for the new wax and ditto for the old.

With two attempts, whenever possible, will be: 25% always wins the old wax, 50% a win apiece, the new 25% wins.

With three attempts, the odds are (V = old, N = new)
N 0 wins (12.5% \u200b\u200bchance): VVV

a victory for N (37.5%):
NVV
VNV
VVN
N 2 wins (37.5%):
NNV

NVN NNV N
3 wins (12.5%): NNN


This is called the binomial distribution , in particular, the binomial with probability p = 0.5 (50%) and number of tests n 2 (or 3) in short-B (0.5, 2) and B (0.5, 3). The also found between the functions of OpenOffice.

Returning to our example with three tests, we see that the probability of three victories for V is 12.5% \u200b\u200bwhile the probability of having 2 or more is 50%. This, remember, the assumption that the two waxes are "identical."

progresses the number of attempts, with a little 'accounts of trust that we leave to our computers, once in ten attempts we have the following:


Wins Chance N cumulative
0 0.1% 0.1%
1 1.0% 1,1%
2 4,4% 5,5%
3 11,7% 17,2%
4 20,5% 37,7%
5 24,6% 62,3%
6 20,5% 82,8%
7 11,7% 94,5%
8 4,4% 98,9%
9 1.0% 99.9%
10 0.1% 100.0%

Looking at the column of cumulative probability, we see that seven wins in 94 or less are possible , 5% of cases. That is, repeating groups of ten trials in 94.5% of the cases I get seven or fewer victories with the new wax, though this is absolutely identical to the old wax.

At this point we ask: how likely is it to eight or more hits with the same wax? Add up the last three values \u200b\u200band is the remaining 5.5% probability. This means that when we do an experiment with ten tosses with two identical wax, there is a 5.5% chance of getting eight wins or more new wax ALTHOUGH NOT THAT MINIMALLY better than the old.

It is 5.5% or so little? Here it becomes arbitrary, but clearly, if I have to entrust their lives, 5.5% is damn too. Would you get on a plane taking off every twenty can not make it? Would you trust a mechanic in 5% of cases fitted the brakes the wrong way?

Clearly, the ideal is to have zero probability. Unfortunately, with a similar value, I never change wax, especially if the experiments instead of skiing, we make them on the sick and, therefore, with the strong limits on sample size, its homogeneity and the time available for testing. The scientific community has decided to establish two levels of significance : 1% and 5%. If there is significance at 1%, the new wax is better than the old one, if it is between 1 and 5% should investigate further, otherwise you can throw it all away.

Why 1%? And if we reason to the contrary, just 99% of defeats? And to say "the new wax is better than the old" is tantamount to saying "the new wax is different from the old"? Now things become philosophical, very specialized and I stop. What I wish to emphasize is how the phrase "eight out of ten" means just as a demonstration of the effectiveness of something. Eight out of ten cases, moreover, are not the same as cases out of 80 percent, or 800 thousand. The statistic is a more exact science than you would have us believe, but the statistics are to be submitted complete: not enough to give a percentage, are the sample size, protocol, etc. distributions. That's why I have many doubts when I read this pdf for example effectiveness of acupuncture.

0 comments:

Post a Comment