Exuse me, I don't claim R^2 can't be negative. What I say if I get R^2 negative then the data are useless. I know, that what Thomas said is true in general case. But in my special case of data, using nonzero intercept is nonsense, and to get R^2 less than 0.985 is considered poor job (standard R^2>0.995). (R^2 given by R^2 = 1 - Sum(R[i]^2) / Sum((y[i])^2) )
Because lm() uses two differrent formulas for computing R^2, it is confusing to get R^2 closer to 1 when linear model with zero intercept y=a*x (a = slope) is used, rather than in case with model y=a*x+b (a=slope, b= nonzero intercept). I think R^2 is only measure of good fit for least squares optimization and it doesn't matter which formula is used: (R^2 = 1 - Sum(R[i]^2) / Sum((y[i])^2) or R^2 = 1 - Sum(R[i]^2) / Sum((y[i])^2-y*)), but using both is confusing. So I would like to know why two different formulas for R^2 are used? -- View this message in context: http://r.789695.n4.nabble.com/Strange-R-squared-possible-error-tp3382818p3383992.html Sent from the R help mailing list archive at Nabble.com. ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
