Excel - Linear Regression Question

[TheRobster]

Hi All,

I've recently figured out that Excel can perform linear regression analysis on graph data points and return an equation for the line. However, I am unsure about the 'R squared' value. I know that the closer to 1 this is then the better the line fits the given equation.

But I don't know if there is an 'acceptable' value for R squared. Some of my lines have corrosponding equations where R is almost 1 and others have values of (for example) 0.8 or 0.91.

How do I know if the value for R squared is acceptable or not?

Cheers
-Rob

[osprey4]

It really depends on the data. For some correlations, for example, of drug plasma concentrations to clinical efficacy, r squared (a.k.a., coefficient of determination) may be in the 0.7-0.8 range to be considered significant. In analytical chemistry, a response factor to concentration correlation must have an r squared above 0.998.

What kind of data are you looking at?

[TheRobster]

Hi Osprey,

Well the type of data varies quite a lot - but one of the main data sets I have relates to rainfall, mainly rainfall depth vs. frequency and duration of rainfall event. The general trend is that the more frequent and longer duration storms have less depth (measured in mm per hour) than do rarer and more intense storms.

There is a definite 'trend' to the data and the R squared value ranges anywhere from between 0.8 and close to 1. I just want to know if the lower values of 0.8 still show a good enough 'fit' for me to be able to use the given equation to predict future rainfall patterns or whether the equation would give results too far from reality to be useful.

Cheers
-Rob

*edit* I just saw that your profile states you're a scientist. What's your area of interest, if you don't mind me asking?

[butch81385]

personally i would stick with .85-.95 with the ones closer to 1 weighted more heavily (for prediction purposes). but it all depends on how accurate you want to be..

[osprey4]

Originally posted by TheRobster
I just saw that your profile states you're a scientist. What's your area of interest, if you don't mind me asking?

Analytical chemistry, but of course!!

For that type of data set, I agree with butch81385. Even a weak correlation supports your point. You're not trying to prove beyond a shadow of a doubt, just showing a nice correlation.