[R] Creating a correlation matrix from a vector

Peter Ehlers ehlers at ucalgary.ca
Fri Oct 26 19:32:35 CEST 2012


On 2012-10-26 08:58, David Winsemius wrote:
>
> On Oct 26, 2012, at 6:52 AM, F_Smithers wrote:
>
>> I'm looking to create a correlation matrix, but I have already obtained the
>> correlations, which are stored in a vector. (Basically, I'm running a
>> simulation which requires a correlation matrix, but I am simulating the
>> various correlations.)
>>
>> My aim is to create a function that can take the vector, and fit the values
>> into their respective locations in a correlation matrix. (The correlations
>> are ordered as if working along the upper triangle of the correlation matrix
>> row-wise.)
>>
>> The initial step in the function was to create a diagonal matrix of length
>> n, (n being the number of factors) and then add the correlations at each
>> level using a for command.
>
>> mat<- matrix(1:16, 4,4)
>> mat[ upper.tri(mat) ]
> [1]  5  9 10 13 14 15
>> newmat <- matrix(NA, 4,4)
>> newmat[ upper.tri(newmat) ] <- mat[ upper.tri(mat) ]
>> newmat
>       [,1] [,2] [,3] [,4]
> [1,]   NA    5    9   13
> [2,]   NA   NA   10   14
> [3,]   NA   NA   NA   15
> [4,]   NA   NA   NA   NA
>> newmat[ lower.tri(newmat) ] <- t(newmat[upper.tri(newmat)])
>> newmat
>       [,1] [,2] [,3] [,4]
> [1,]   NA    5    9   13
> [2,]    5   NA   10   14
> [3,]    9   13   NA   15
> [4,]   10   14   15   NA
>> diag(newmat) <- 1
>> newmat
>       [,1] [,2] [,3] [,4]
> [1,]    1    5    9   13
> [2,]    5    1   10   14
> [3,]    9   13    1   15
> [4,]   10   14   15    1
>

This doesn't quite work; the resulting matrix is not symmetrical.
The culprit is the t(newmat[....]) line. I find it easiest to
start with a matrix of zeros, fill in the upper.tri part as you
have done, then just add newmat and t(newmat), then fix the diagonal.

Peter Ehlers




More information about the R-help mailing list