generate multinomial random varibles with varying

问题:

I need to genereate multinomial random variables with varying sample size.

Let say i already generated my sample sizes as follows,

samplesize =c(50,45,40,48)

then i need to generate multinomial random variables based on this varying sample size. I tried this using a for loop and using a apply function(sapply).

Using For loop ,

p1=c(0.4,0.3,0.3)
for( i in 1:4)
{
xx1[i]=rmultinom(4, samplesize[i], p1)
} 

If my code is correct then i should get a matrix that have 4 columns and 3 rows. Where column totals should equal to the each value in sample sizes. But i am not getting that.

Using Sapply ,

sapply( samplesize ,function(x)
{
  rmultinom(10, samplesize[x], p1)
})

I am getting an error here also.

So can any one help me to figure out what went wrong ?

Thank you

回答1:

samplesize <- c(50, 45, 40, 48)
p <- c(0.4, 0.3, 0.3)

## method 1
set.seed(0)
xx1 <- matrix(0, length(p), length(samplesize))
for(i in 1:length(samplesize)) {
  xx1[, i] <- rmultinom(1, samplesize[i], p)
  }
xx1
#     [,1] [,2] [,3] [,4]
#[1,]   24   17   20   24
#[2,]   11   14    8   16
#[3,]   15   14   12    8
colSums(xx1)
#[1] 50 45 40 48

## method 2
set.seed(0)
xx2 <- sapply(samplesize, rmultinom, n = 1, prob = p)
xx2
#     [,1] [,2] [,3] [,4]
#[1,]   24   17   20   24
#[2,]   11   14    8   16
#[3,]   15   14   12    8
colSums(xx2)
#[1] 50 45 40 48

---

Note: rmultinom is not "vectorized" like other distribution functions say rnorm.

set.seed(0)
fail <- rmultinom(length(samplesize), samplesize, p)
#     [,1] [,2] [,3] [,4]
#[1,]   24   19   25   24
#[2,]   11   16   10   17
#[3,]   15   15   15    9
colSums(fail)
#[1] 50 50 50 50

So the R-level for loop or sapply loop or using sugar function Vectorize is necessary.

回答2:

You can avoid the loop with mapply if you like:

samplesize <- c(50, 45, 40, 48)
p <- c(0.4, 0.3, 0.3)

mapply(rmultinom, samplesize, MoreArgs = list(n=1, prob=p))

#     [,1] [,2] [,3] [,4]
#[1,]   15   22   14   18
#[2,]   13    9   14   12
#[3,]   22   14   12   18

回答3:

i think i can use this method using replicate function as well to solve my problem isn't it ?

r1= c(0.34,0.33,0.33)
    rep=10
    size=500
    alpha=0.05
    q=0.1

    set.seed(1)
    didnt_vote=rbinom(rep, size, q)
    replicate(n=1,rmultinom(rep,didnt_vote,r1) )