# Thread: Can I trust this regression analysis?

1. ## Can I trust this regression analysis?

Hello all!

I did some mini research and decided to run a regression analysis
to establish a relationship between raid5 read performance (MB/s) and
raw cpu power. (pass mark points).

This is the result:

71.4637 + 912*.0587639=125.0537

With a regression coefficient of .999294.

But is this any good? (hinting at n=3)

2. ## Re: Can I trust this regression analysis?

Welcome to MHF, iBurger!

A regression coefficient of .999294 is pretty good.
It means your data points are pretty accurately on a straight line.

n=3 measurements is not very much though as you probably already surmised.
It leaves a pretty big chance on systematic mistakes that a regression analysis does not cover.

Do you also have the t-test significance value of the slope coefficient?
At the very least that one should be below 0.05 to indicate that the chance that you wrongfully conclude you have a linear relationship, is less than 5%.

3. ## Re: Can I trust this regression analysis?

Thank you! And thanks for running through the numbers!

I'm still amazed that I actually managed to find something in the data. Have to get much better at statistics though and develop some good intuition.

I read up on the suggested t-test and I get its purpose. [h0: b1=0 vs. h1: b1=/0]. I tried to calculate it by hand and found mad numbers, far under the 0.001 p-level however, I'll need to install Excel and rerun that this test because I don't trust my own calculations. I'll keep you posted, probably next week I'll have more time.

So! big thanks! really fun experience!