4
library(nlme)
fm1 <- nlme(height ~ SSasymp(age, Asym, R0, lrc),
            data = Loblolly,
            fixed = Asym + R0 + lrc ~ 1,
            random = Asym ~ 1,
            start = c(Asym = 103, R0 = -8.5, lrc = -3.3))
summary(fm1)
fm2 <- update(fm1, random = pdDiag(Asym + lrc ~ 1))
summary(fm2)
anova(fm1, fm2)

    Model df      AIC      BIC    logLik   Test  L.Ratio p-value
fm1     1  5 239.4856 251.6397 -114.7428                        
fm2     2  6 238.9662 253.5511 -113.4831 1 vs 2 2.519406  0.1125

I'm trying to see which model is preferred using an LRT. Here the p-value is > 0.05, so I fail to reject the null hypothesis and conclude that the smaller model fm1 is sufficient. And if the p-value were < 0.05, then I would conclude that the larger model fm2 is preferred. However, from reading this post, the top answer states that "the model with the least df gets preferred in case of a significant difference." So...does this mean that if p-value < 0.05, fm1 would be preferred?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Adrian
  • 1,665
  • 3
  • 22
  • 42
  • If you have a significant p-value you should use the model with the smaller AIC. – Roland Feb 14 '17 at 15:10
  • @Roland, so I should use the model with the smaller AIC instead of the smaller df? Could you elaborate? – Adrian Feb 14 '17 at 15:41
  • non-significant p-value --> use model with smaller df; significant p-value --> use model with smaller AIC – Roland Feb 14 '17 at 15:45
  • @Roland, thanks for the response. In that case what you proposed (non-significant p-value --> use model with smaller df) is contradictory to the top answer in the post I referred to (the model with the least df gets preferred in case of a significant difference)? – Adrian Feb 14 '17 at 15:52
  • DF tells you which model is more complex/simpler. The p-value tells you if the likelihoods can be considered different. None of these tell you which model fit can be considered better. The values of the likelihood (and AIC or BIC which are derived from it) can tell you that. – Roland Feb 14 '17 at 16:00

0 Answers0