using lme4 to model regression with non-independent(nested?) data
On Fri, 17 Jan 2014, Louis wrote:
Dear lme4 list, I am having trouble modelling regression with seemingly non-independent, nested observations and I think a mixed model approach (using *lme4* for example) is required. I am examining the effect of genetic relatedness of males on the proportion of offspring sired in cases of multiple paternity (multiple males contributing to a brood of eggs from one female). I have estimates of the genetic relatedness of each male and the female it mated with, as well as the proportion of offspring that male sired in the female's brood of eggs. There are 3-5 males contributing to a brood, and 9 separate broods sampled for paternity. *My questions are: *1)Should observations be nested within a female, 2)and/or should female be treated as a random effect (9 broods from 9 females) or both? 3) should relatedness remain a fixed effect? Again, I am concerned that I have not adequately dealt with the non-independence of male proportion values. ID Female Relat. Prop. 1 A .12 .3 2 A .03 .02 4 A .23 .68
Heavens! I don't think lme4 is the right tool, though I think you could get an approximately right model (negative covariance between males within each clutch). The female only comes into the multinomial bit, setting total clutch size, as I see it, so I don't see the point of a RE there. Personally, I would permute males within clutches, with a correlation coefficient as the measure of association. Female Male N_Sired relatedness A A1 3 0.12 A A2 1 0.03 A A3 6 0.68 B B1 2 0.05 ... | David Duffy (MBBS PhD) | email: David.Duffy at qimrberghofer.edu.au ph: INT+61+7+3362-0217 fax: -0101 | Genetic Epidemiology, QIMR Berghofer Institute of Medical Research | 300 Herston Rd, Brisbane, Queensland 4006, Australia GPG 4D0B994A