Question regarding statics and correlation..

**Justin7** · 07-18-10, 08:40 PM

In addition to raw correlation, you need to know the confidence interval. Generally, you want to have a sample size big enough to get at least a 95% confidence interval.

How low to go? It depends on what you are using the analysis for.

**bztips** · 07-18-10, 09:14 PM

Depending on the purpose, simple pairwise correlations may not tell you much at all. If all 10 things are potentially relevant, then you'd need to do something more sophisticated (like multiple regression). There is both an art and a science to it, determining which variables are or are not significant; you may well have "multicollinearity" issues where the 10 variables you're looking at are correlated WITH EACH OTHER in varying degrees -- this makes the analysis trickier, and there are potentially lots of other statistical issues to think about in such a situation. In all honesty, a real understanding of it would require some formal training in probability, stats and econometrics.

**jakeandba** · 07-18-10, 11:06 PM

justin and , thanks for the replies...since i am new to stats ...looks like i have a ton of reading to do to try and get a grip on this subject

**uva3021** · 07-18-10, 11:27 PM

sample size is crucial, anything can be correlated purely be being randomly configured to fit a certain period of time, like the color of my shoes and the phases of venus

but if after years of observation and accumulating data the correlation persists, then it becomes statistically significant, so like justin said its sample size

if observed correlations are low, that doesn't mean they aren't statistically significant, but for the sample as a whole on average a relationship can't be found

this is true for statistics in football like penalties or special teams, or rushing and passing statistics (think Navy and Texas Tech), so you would have to isolate the correlations, and possibly serry the teams into a fashion determined by style of play or tendency, but doing so reduces the sample size, its a very complex and enigmatic process

**jakeandba** · 07-19-10, 07:44 PM

thanks for the additional input..have been some serious reading on correlation and it is in sync with uva....was reading something interesting..dont know how accurate this is but if u square the correlation results, it would give a percentage.

has anyone heard of that...so if the results are .50 then it would equal 25%?