"p-Value of zero from linear regression"

dysprosium

Just a quick question - I'm getting a p-Value of 0 for one of the attributes In the output from the linear regression operator. Is this a genuine zero, or an extremely small value? If the latter, what's the threshold below which Rapid Miner returns zero for a p-Value?

David_A

It can be both.
For very small numbers RapidMiner just shows a 0 in the result view, but the actual value is used for further calculations (for example you can sort according to the p-values).
But for very small values it can happen that the p-value becomes a genuine 0, this depends on the underlying distribution functions and their parameters, so a fixed threshold cannot be given.
For all practical concerns the differences between a genuine 0 and a very small, non-zero, p-value should not matter.

dysprosium

I was getting 0 for the p-Value of one attribute, and 0.000 for another attribute, and wondering how they were different. I've just sorted the results by the p-Value, and that shows 0 to be smaller than 0.000. That's probably precise enough for my purpose (although it would be interesting to know if it's a genuine 0). Thanks for the explanation.