(07-07-2017, 02:11 PM)SimmerDownBruhh Wrote:Very awesome read, bro! I love all of your analysis articles.

(07-07-2017, 04:17 PM)7hawk77 Wrote:I appreciate the kind words. I honestly don't know that much about statistics, so I was a little nervous throwing this out there.

(07-07-2017, 04:17 PM)7hawk77 Wrote:I appreciate the kind words. I honestly don't know that much about statistics, so I was a little nervous throwing this out there.

(07-07-2017, 03:27 PM)timeconsumer Wrote:You're off to a good start. If you want to go a little more in-depth I would examine your p-values to see if the relationship you are seeing with the correlation coefficient is considered significant. If you return a correlation with a p-value of something like 0.5 you're probably looking at a bunch of statistical noises. But if you see a p-value of 0.1 to 0.01 you could be on to something.

Also when it comes to correlation coefficients just because it's positive doesn't necessarily mean much. So when you found a coefficient of 0.16 for blocking it really didn't mean anything. 0 means absolutely no linear relationship at all and 1 means a completely linear relationship so if you look at the picture here:



You can see that 0.16 is basically a big blob. But finding something like a 0.5 or 0.6 (strength) is a good start if you have it associated with a significant p-value, and something you can definitely formulate a conclusion off of.

(disclaimer: not a statistician by any means, but I dabble)

(07-07-2017, 04:39 PM)7hawk77 Wrote:I gotcha. So some of my conclusions might still be right, but instead of me guessing at the statistical significance, I can use P-Values to determine that. I'll try and figure out how to calculate P-values on the above data when I find some time. Also, is there a simple way to do that in excel? I ended up using the data analysis to do an F-Test on Two sample for Variances but I think I just confused myself a bit.

(07-07-2017, 05:39 PM)7hawk77 Wrote:I gotcha. So some of my conclusions might still be right, but instead of me guessing at the statistical significance, I can use P-Values to determine that. I'll try and figure out how to calculate P-values on the above data when I find some time. Also, is there a simple way to do that in excel? I ended up using the data analysis to do an F-Test on Two sample for Variances but I think I just confused myself a bit.

(07-07-2017, 03:51 PM)timeconsumer Wrote:I'm a big fan of the regression analysis. The output will give you the r-squared value which tells you how much variance in the dependent variable can be explained by the independent variable(s).
Then you can check your ANOVA table underneath that to see the Significance F value to check the statistical significance of the whole test (and the associated r2).
Then finally your coefficients table gives you the p-value for each attribute while also showing you a coefficient attempting to explain what an increase of 1 in x does to y.

But here's the big issue with a pearson correlation and a regression (and most other tests we would use as amateurs) they rely on continuous data. And with my understanding of this sim attributes and their affect on the player are not continuous. This means that going from 70 to 80 strength does not give as much of a benefit as going from 90 to 100 strength. So instead we have to weight our attributes to account for this. I have a method of weighting that I use, and it's working okay for me, but unfortunately it's secret Otter property. Play around with weighting it, you might get a better system than me.

(07-07-2017, 03:47 PM)Molarpistols Wrote:Under the data analysis tool you could run a regression that'll give you a separate p-value for each variable (agi/str/etc/etc).  It'll also give you a p-value for the entire relationship, as well as F-values and r^2.  It likely won't tell you much unless you have quite a large data set for each variable however.

If you can't find the regression in the data analysis tool, you may have to download it by going into Excel->Options->Add-ins