Given data, does applying probability and statistics necessarily yield reliable scientific results?

[Boojatta]

Alternatively, are there different methodologies, principles, rules, or tools of probability and statistics for different kinds of data?

For example, if some data is associated with not merely arbitrary correlations, but with actual causation, then is there any scientifically acceptable motivation to give special treatment to the data?

[orpupilofnature57]

1. I put my trust in a pot smoking star gazer a long time ago ,and it never fails.

http://users.tpg.com.au/users/tps-seti/baloney.html

[rucky]

2. Bookmarked!

thanks

[phasma ex machina]

5. Spin more than one hypothesis - don't simply run with the first idea that caught your fancy.

A former coworker of mine always gave more weight to the second, following, counterintuitive idea.

[orpupilofnature57]

6. Exactly !! people are to anxious for that Eureka moment ,unfortunately

in science it's Dangerous.

[hlthe2b]

3. Well, there is an old adage... "garbage in, garbage out..."

Which means, the analysis is only as good as the data and the methods used to collect the data.

That said, causation is not proven by correlation alone, no matter how statistically strong the findings. At least in epidemiology studies, it is inferred from the consistency of multiple studies, biologic plausibility of the association (often from animal studies), dose response when present (likelihood of outcome increases with increasing dose or exposure) and the strength of the statistical findings....

And yes, there are different methodologies, principles, rules and statistical tools for the analysis of different kinds of data-- enough to keep countless statisticians, methodological epidemiologists, computer analysts, and mathematicians busy for countless careers.

[Jim__]

4. Speaking to probability and statistics in general and not to specific scientific uses, ...

I'd say probability is very reliable, statistics, depending on what you're looking for, is more difficult.

Probability is about predicting sample populations from a known population. A simple example of a probability problem, if you have a box containing 30 red balls, 20 green balls and 50 yellow balls, and you randomly select 10 balls from this population, what is the probability that your sample will contain 3 red, 2 green and 5 white balls. A computation of the probability will usually come very close to an actual count from a large number of tests.

Statistics is about going the other way. Same example, given that you have picked 10 balls from a box containing a mixture of 100 red, green, and yellow balls, and you picked 3 red, 2 green, and 5 yellow, what is the probabilty that the percentage of red, green and blue balls is 30%, 20%, and 50%.

When they sample voters to give estimates of what election results would be if they were held today, that's statisitics.

As to the question about correlations, when I was doing statistics, and I found "curious" correlations, I investigated them to find out what the actual relationship was. For instance if 2 variables had a .35 correlation, how were they actually related. My experience (testing specific types of data) was that there was usually a subset of the data for which these 2 variables were strongly correlated (say above .9). Then, investigatign these subsets could tell you a lot about how your system performed under certain conditiions (the conditions of this particular subset). By checking various weakly correlated pairs, you could learn to predict and react to adverse events, either early, or act to prevent them.

[Boojatta]

7. Kick

[Boojatta]

8. Kick because the concept of causation is worth discussing

[dimbear]

9. Obviously not, and here is the why. Statisticians have only a few

tools in their bag, all of which depend on the assumption that reality conforms to a specific distribution, usually the Gaussian distribution. That one gets picked because its easy to calculate. (If not that specifically, there are a very few others.) Statisticians pick one and run with it and never look back.

[DetlefK]

10. How do you know whether your data deserves special treatment?

Every data is always treated the same:
1. You bin it into a probability density function.
2. Theory: From the measurement mechanism you have to derive the pdf your data SHOULD have. (binomial, Breit-Wigner, Lorenz, Chi², Poisson, t-, Gamma, Gaussian or in some very rare cases some very exotic and difficult ones)
3. If your data fits to the pdf, your measurement is of good quality. If not, it's bad quality.

All data is created equal. But then discrimination kicks in... And your only award is, you gonna get used. ;-)