Hi,
I'd like to do a multinomial glm with nested and random Effects
The "multinom" function from (nnet) give me this:
> multinom(mc$c ~ (1|mc$date)+mc$lma + mc$poid+(mc$pop) +mc$male %in%
> mc$pop,family="multinomial" )
# weights: 60 (44 variable)
initial value 6801.484618
iter 10 value 5704.912781
iter 20 value 5373.764816
iter 30 value 4977.685313
iter 40 value 4746.075251
iter 50 value 4740.904959
final value 4740.900034
converged
Call:
multinom(formula = mc$c ~ (1 | mc$date) + mc$lma + mc$poid +
mc$pop + mc$male %in% mc$pop, family = "multinomial")
Coefficients:
(Intercept) 1 | mc$dateTRUE mc$lma mc$poid mc$pop2 mc$pop3
2 -3.3725062 -3.3725062 1.0578694 -0.2416550 -0.1112114 1.96731873
3 2.2583341 2.2583341 -0.1744330 -0.2552745 -0.5112047 -0.05992067
4 0.4713678 0.4713678 0.3032407 -0.3910529 -0.3587365 -0.31611663
5 0.1176258 0.1176258 0.4085705 -0.3465320 -0.4342574 -0.24052282
mc$pop4 mc$pop1:mc$male mc$pop2:mc$male mc$pop3:mc$male mc$pop4:mc$male
2 -0.6584972 0.03764097 0.042722762 -0.04506099 0.03963405
3 -0.8932160 -0.03943989 -0.005425975 -0.03670574 -0.01614433
4 -0.6258679 -0.02901482 -0.007115797 -0.01195997 -0.00537634
5 -1.2768885 -0.02291026 -0.005762747 -0.02240182 0.01579168
Residual Deviance: 9481.8
AIC: 9561.8
So, multinom, seems me work with nested and random effect....
Notice that a "TRUE" is added behind the random effect
> multinom(mc$c ~ mc$mom %in% (1|mc$date) +(1|mc$date)+mc$lma + mc$poid+mc$pop
> +mc$male %in% mc$pop,family="multinomial" )
# weights: 70 (52 variable)
initial value 6801.484618
iter 10 value 5208.112867
iter 20 value 5084.345631
iter 30 value 4655.378383
iter 40 value 4521.895062
iter 50 value 4518.961455
final value 4518.961019
converged
Call:
multinom(formula = mc$c ~ mc$mom %in% (1 | mc$date) + (1 | mc$date) +
mc$lma + mc$poid + (mc$pop) + mc$male %in% (mc$pop), family = "multinomial")
Coefficients:
(Intercept) 1 | mc$dateTRUE mc$lma mc$poid mc$pop2 mc$pop3
2 -3.54938862 -3.54938862 1.0472658 -0.2389110 -0.1126589 1.97228680
3 1.98030198 1.98030198 -0.1879820 -0.2524073 -0.5180808 -0.05418255
4 1.18772479 1.18772479 0.3435563 -0.4005161 -0.3395149 -0.33326040
5 0.09882458 0.09882458 0.4066034 -0.3458763 -0.4339492 -0.23909966
mc$pop4 mc$mom:1 | mc$dateFALSE mc$mom:1 | mc$dateTRUE mc$pop1:mc$male
2 -0.6607706 0 0.19949190 0.03740662
3 -0.8997561 0 0.30402706 -0.03987009
4 -0.6022370 0 -0.98117482 -0.02774351
5 -1.2764239 0 0.02313478 -0.02291646
mc$pop2:mc$male mc$pop3:mc$male mc$pop4:mc$male
2 0.042692605 -0.045517538 0.039313888
3 -0.005367900 -0.037430186 -0.016624565
4 -0.007190628 -0.009714267 -0.004076609
5 -0.005761103 -0.022468428 0.015746601
Residual Deviance: 9037.922
AIC: 9125.922
Here, the "TRUE" is added at the same place, and "FALSE" is added for the
interaction with the random factor, coefficients are 0
multinom(formula = mc$c ~(1 | mc$mom %in% mc$date) + (1 | mc$date) + mc$lma
+ mc$poid + (mc$pop) + mc$male %in% (mc$pop), family = "multinomial")
work well, but is it true ?
hugo mathé hubertEmail : hug...@hotmail.fr
approximatively, my experimental design is :
some individuals (lizard) have a behaviour (a dependent unordered categorical
variable).
lizards come from several populations, for each lizard, behaviour is registered
at morning , at noon, at evening, during several days.
So, individuals are nested in populations and moments of the day is nested in
the day. (isn't it ??)
I'd
like to identifies some type of individual so, (among other things) I need to
the interaction between individuals and moment of the day
which are tow nested variable...
it's wrong to take the ( individuals : moment of the day) nested in (population
: day)
(mc$male : mc$mom ) %in% ( mc$pop : mc$date)
(note
that, I have only ONE value (behaviour) by moment of the day by day by
individual ; but days are time series so each days aren't totally
independent...)
_________________________________________________________________
Hotmail: Trusted email with MicrosoftÂ’s powerful SPAM protection.
[[alternative HTML version deleted]]
______________________________________________
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.
