Applying a function to a group of a group (tapply in ddply) - r

Question

My dataset looks like this:
d = data.frame(year=rep(2000:2002,each=40),month=rep(c(rep(1:12,3),5,6,7,8),3),species=rep(c(rep(letters[1:12],3),"a","b","g","l"),3),species_group=NA,kg=round(rnorm(120,15,6),digits=2))
d$species_group=ifelse(d$species %in% letters[1:5],"A","B")
I would like to have per year and per species group (so excluding the levels of month and species) the mean weight and the number of species included. This works fine with ddply. However, I also would like to include a value of the “quality” of my data. That is, if the number of species per month is balanced or there are for example more species included during the summer months. Therefore I thought I might simply calculate the yearly standard deviation of the number of unique species per month.
I tried doing this with tapply in ddply as follows:
s=ddply(d,c("year","species_group"),function(x) cbind(n_species=length(unique(x$species)),
quality=tapply(x,x$month,sd(length(unique(x$species)))),
kg=sum(x$kg,na.rm=T)))
but this gives me an error
Error in match.fun(FUN) : 'sd(length(unique(x$species)))' is not a function, character or symbol
What I would like to obtain is something like this:
output=data.frame(year=rep(2000:2002,each=2),species_group=rep(c("A","B"),3),n_species=rep(c(7,9),3),quality=round(rnorm(6,2,0.3),digits=2),kg=round(rnorm(6,15,6),digits=2))
I cannot first use ddply by month, year and species group because this means I couldn’t know anymore the number of unique species per year.
I suppose I could also calculate n_species and quality separately and put them together afterwards, but this would be a cumbersome approach.
How can I make my function work, or how can I do this more properly?
ANSWER:
The easiest solution came from shadow, who noted my mistake in the use of tapply. Furthermore, a standard error should be more appropriate than standard deviation, giving the following formula:
s=ddply(d,c("year","species_group"),function(x) cbind(n_species=length(unique(x$species)),
quality=sd(tapply(x$species,x$month, function(y) length(unique(y))))/sqrt(length(tapply(x$species,x$month, function(y) length(unique(y))))),
kg=sum(x$kg,na.rm=T)))

Answer 1

Not clear how do you define your quality criteria. So How I would do this.
First I define my quality criteria in a separate function. Note that your function should retuen a single value not a vector (in your solution you are using tapply which return a vector).
## returns the mean of sd variation per month
get_quality <-
function(species,month)
mean(tapply(species,month,
FUN=function(s)sd(as.integer(s))),
na.rm=TRUE)
Then I use it within ddply . To simplify code I am also creating a function to be applied by group.
ff <-
function(x) {
cbind(n_species=length(unique(x$species)),
quality= get_quality(x$species,x$month),
kg=sum(x$kg,na.rm=TRUE))
}
library(plyr)
s=ddply(d,.(year,species_group),ff)
year species_group n_species quality kg
1 2000 A 5 0.4000000 259.68
2 2000 B 7 0.2857143 318.24
3 2001 A 5 0.4000000 285.07
4 2001 B 7 0.2857143 351.54
5 2002 A 5 0.4000000 272.46
6 2002 B 7 0.2857143 331.45