I only know very basic statistics.

I have data that is calculated by two different codes. The results should ideally be identical but never are.

I need a way to calculate how closely the two sets of data match. They share the same domain of independant values.

Am I aiming for R-squared? Mean square error? Std dev? I don't even know what to google.

10/6/2007 9:09:42 PM

What codes? What are you talking about?

10/6/2007 10:36:09 PM

Can't talk about the codes other than to say they perform simulations. One of the codes calculates 'predicted data.' The other code uses real-world results to back-calculate and produce 'measured data.'

I need to verify that both sets of results agree with each other to within a certain amount (let's say, within 90%).

The datasets cannot be fit via any method as they are sporadic in shape.

My initial thought was the following:
Dataset A (measured data)
Dataset B (predicted data)
% difference = (A - B) / B

Would that be an adequate way to verify agreement?

10/6/2007 10:45:25 PM

I think you want a correlation.

10/6/2007 11:03:08 PM

what is it you are trying to compare? Means, variance, etc. If you are looking to compare means, you probably need something like a two-sample t-test or something along those lines. That way you can see if they differ statistically significant from one another. A correlation measures the linear relationship between two variables and r-squared tells you how much variation is explained by your model so I'm not sure if either of those is what you are looking for.

10/6/2007 11:10:19 PM

Hmm.. Problem is I'm too ignorant to even know what I actually want other than to say I'd like a single number which would be an indicator of how close the two sets match. Much like how R-squared is a 0 - 1 value, 1 being the best, that tells you how well a dataset fits its own mean curve. 'Cept I'm dealing with two datasets that cannot be fit.

10/6/2007 11:30:16 PM

If you're looking for a single number, what's wrong with just averaging the data and comparing the average?

Also the t test would result in a single number.

10/6/2007 11:36:06 PM

Because the data is all time dependant. The averages mean nothing because the y values are changing constantly.

Will check out t test.

[Edited on October 6, 2007 at 11:37 PM. Reason : ]

10/6/2007 11:36:52 PM

Hope the results of this analysis have no impact on public safety.

10/7/2007 12:16:05 AM

^ who knows

Quote :

"Major : Nuclear engineering"

10/7/2007 12:28:12 AM

Of course it doesn't. I'm posting on TWW. This is a personal side project.

Additionally, nuclear technology today does not pose a threat to public safety.

[Edited on October 7, 2007 at 12:36 AM. Reason : ]

10/7/2007 12:35:25 AM

You want a correlation.

10/7/2007 12:55:20 AM