Skip to content

R^2 in linear regression

3 messages · Peter Dalgaard, Terias

Terias
# 
Hello,

I was doing a linear regression with the following formula:
lm(y~x+0), so it passes through the origin. But when I called the summary of
the regression i saw that R squared is abnormally high (it's a lot lower in
other programs such as SigmaPlot and MS Excel).The manual explained the
cause of the difference (because of the different computing method), but
what should I do to get the same R^2 in excel and R?

WITHOUT PASSING THROUGH THE ORIGIN:
R^2: Multiple R-squared: 0.9711,     Adjusted R-squared: 0.9654
In MS Excel: 0,9711

So it's OK.


WITH PASSING THROUGH THE ORIGIN:
Multiple R-squared: 0.9848,     Adjusted R-squared: 0.9822
In MS Excel: 0,8907

So almost 10% difference.

Thank you for your help.

Csanad Bertok, Hungary
Peter Dalgaard
#
Terias wrote:
If you insist, I think you can get what Excel does like this:

 > x <- 1:10
 > y <- rnorm(10)

## Residual sums of squares
 > ss1 <- anova(lm(y~1))[1,2]
 > ss2 <- anova(lm(y~x+0))[2,2]

## Relative reduction in sum of squares
 > (ss1-ss2)/ss1
[1] -0.08576713

Now if you dislike the fact that R^2 can come out negative, that's your 
problem....