I tried doing a simple linear regression using RapidMiner but some of the output values (std error, std coefficient, t-stats) are greatly different from SPSS and Excel.
Why are the outputs different?
I've run into similar issues before. If I run an analysis on a laptop and analyze my data multiple times, being on battery vs plugged in produces different results. Using 32-bit vs 64-bit Windows 7 makes a difference. Using Windows 7 vs Windows 8 makes a difference. One of my professors told us to run our stats software through MS-DOS ONLY. He got different results like this too.
It is also possible that somewhere in Rapid Miner, there is a little box that is checked that changes how weights are used.
Thank you all for you comments.
1) Andrew: I'm surprised that even being plugged vs battery could also contribute to such a difference. Good to know about it. I'll take note of it the next time I run my analysis. :)
2) Alwynne: I've tried before a few statistical software and seen the differences in some parameters but since those differences were very small and I decided it to be negligible. Maybe because, just like you said, of how some values could be rounded during computation. But this is the first I find such a big variation.
I agree that it's important to specify the software, operating system, etc. as a way to overcome this problem. But it seems difficult to find papers with such details these days (personal experience).
3) Soutrik: I'm new to using data mining packages (previously only statistical packages) and rapidminer is my first. I will try to use Weka and see how it performs.
- Badges
- Science topic
- Similar topics
- Computer Science
- Data Mining