[R] stats on transitions from one state to another

[Murali.Menon at avivainvestors.com]

David: Brilliant! Thanks very much.

As Berend pointed out, I was not precise in the query, sorry. Please note in the example that we have a run of three state 2 after 0, and later another run of two state 2 after 0.

0 2 2 2 .... 0 0 2 2

Whenever the state moves from 0 to 2, I want to compute the mean of the run of values in state 2. 

I tried to modify David's example by doing the 'interaction' on states and runs:

y <- rle(sss[,"State"])
yy <- cbind(y$lengths, y$values, c(NA, y$values[-length(y$values)]))
colnames(yy) <- c("RunLength","State","PrevState")

tapply(yy[,"RunLength"], INDEX=interaction(yy[,"PrevState"], yy[,"State"]), c)

This gives me the list of runs for each transition-pair.

Now I need to match these lists of runs against the corresponding rows in sss of values V1, V2, and compute stats on them.

Is there an easier way?

Thanks,

Murali

-----Original Message-----
From: David Winsemius [mailto:dwinsemius at comcast.net] 
Sent: 20 February 2012 15:31
To: Menon Murali
Cc: r-help at r-project.org
Subject: Re: [R] stats on transitions from one state to another


On Feb 20, 2012, at 10:11 AM, <Murali.Menon at avivainvestors.com> wrote:

> Folks,
>
> I'm trying to get stats from a matrix for each transition from one  
> state to another.
>
> I have a matrix x as below.
>
> structure(c(0, 2, 2, 2, 0, 0, 0, 1, 1, 1, 1, 2, 2, 1, 1, 1, 0,
> 0, 2, 2, 0.21, -0.57, -0.59, 0.16, -1.62, 0.18, -0.81, -0.19,
> -0.76, 0.74, -1.51, 2.79, 0.41, 1.63, -0.86, -0.81, 0.39, -1.38,
> 0.06, 0.84, 0.51, -1, -1.29, 2.15, 0.39, 0.78, 0.85, 1.18, 1.66,
> 0.9, -0.94, -1.29, -0.23, -0.92, -0.21, 1.02, -0.77, -0.68, -0.33,
> 0.04), .Dim = c(20L, 3L), .Dimnames = list(NULL, c("State", "V1",
> "V2")))
>
> Is it possible to get, say, mean values of each variable in state 1  
> when the previous state was 0, in state 2 when the previous state  
> was 0, and so on with all available transitions between states 0, 1,  
> 2?
>
> In the above case, mean of V1 in state 2 when previous state was 0  
> would be
>
> mean(c(-0.57, -0.59, 0.16, 0.06, 0.84)) = -0.02
>
> while the mean of V1 in state 0 when previous state was 2 would be:
>
> mean(c(1.62, 0.18, -0.81)) = 0.33
>
> If I try something like
>
> by(x[, 2:3], x[, 1], FUN = colMeans)
>
> I get the means for each state. I'm not sure how to get the split by  
> transition?

Add an extra column of previous states: and tabulate:

 > sss <-cbind(sss,  c(NA, sss[,"State"][-nrow(sss)]) )

 > table(sss[,"State"], sss[,4])

     0 1 2
   0 3 1 1
   1 1 5 1
   2 2 1 4

The requested means for "V1" by transition types:
 > tapply(sss[,"V1"], INDEX=interaction(sss[,"State"], sss[,4]), mean)
    0.0    1.0    2.0    0.1    1.1    2.1    0.2    1.2    2.2
-0.670 -0.190 -0.255  0.390 -0.640  2.790 -1.620  1.630  0.205

The counts on which those means are based:
 > tapply(sss[,"V1"], INDEX=interaction(sss[,"State"], sss[,4]), length)
0.0 1.0 2.0 0.1 1.1 2.1 0.2 1.2 2.2
   3   1   2   1   5   1   1   1   4

-- 

David Winsemius, MD
West Hartford, CT