6
$\begingroup$

From p. 277 of R Cookbook:

Let's say I have a R model lm(formula = y ~ u + v + w) and the Summary() shows:

Multiple R-Squared: 0.4981, Adjust R-Squared: 0.4402 F-statistic:
8.603 on 3 and 26 DF, p-value: 0.0003915

Using Adjusted r-Squared I can say that my model explains 44.02% of the variance of y with the remaining 55.98 unexplained.

Question: Does the associated F-statistic (with the p-value being < .05) tell me:

  1. the model, in general, is significant (not taking into account other values from Summary)
  2. the model is significant in explaining the 44.02% variance (adjusted r-squared)
$\endgroup$
  • $\begingroup$ How did you arrive at 3, and 26 d.f.? $\endgroup$ – Subhash C. Davar Dec 3 '17 at 12:43
13
$\begingroup$

The F-statistics tells you if the model fits the data better than the mean. Or, in other words, if $H_0:\;R^2=0$ should be rejected.

See: Wikipedia

To illustrate that the formula given in the link is indeed used by summary.lm:

x1 <- 1:10
set.seed(42)
x2 <- rnorm(10)
y <- x1+2*x2+rnorm(10)

fit0 <- lm(y~1)
fit1 <- lm(y~x1+x2)

summary(fit1)
#F-statistic:  14.1 on 2 and 7 DF,  p-value: 0.003507 

RSS0 <- sum(residuals(fit0)^2)
RSS1 <- sum(residuals(fit1)^2)

Fvalue <- (RSS0-RSS1)/(3-1)/RSS1*(10-3)
#14.10014
pf(Fvalue,2,7,lower.tail=FALSE)
#0.00350697
| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ The question concerns about the chisquared test of variability and does not ask for the significance of mean. $\endgroup$ – Subhash C. Davar Jun 28 '19 at 23:06
  • 1
    $\begingroup$ My answer does not say anything about significance of the mean. $\endgroup$ – Roland Jun 29 '19 at 11:17
  • $\begingroup$ The two formulations are different. I think, one (Ist) imbibes - so called random effects and second is based on on (true )(fi xed- effects model.itmaybe noted that the two shoud yield same results. To my knowlege, first one model acts on assumption of random effects assumption and second one under fixed effeccts assumption. Therefore, I am not conviced your structure and proof. $\endgroup$ – Subhash C. Davar Jun 30 '19 at 0:55
  • $\begingroup$ Sorry, but both OP's model and the so-called Null model y ~ 1 are pure fixed effects models. $\endgroup$ – Roland Jun 30 '19 at 4:19
  • $\begingroup$ fit1 <- lm(y~x1+x2) please explain briefly. thanks for your valuable response. I am a halfbaked statistics worker. $\endgroup$ – Subhash C. Davar Jun 30 '19 at 10:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.