[R] asypow.noncent: how does it work?

[Kjetil Brinchmann Halvorsen]

david_foreman at doctors.org.uk wrote:

>I am trying to do power calculations for the proportional odds model using the asypow library.
>
>The code
> 
>noncenta90b10<-asypow.noncent(theta.ha=a9010,info.mat=infomatrixa90b10,constraints=constrt)
>
>returns
>
>Error in max(..., na.rm = na.rm) : invalid "mode" of argument.
>
>the various arguments I've used are:
> a9010
>           [,1]
>[1,] -1.7357568
>[2,] -0.1928619
>specifying the theta.ha array as a row not a column makes no difference
>
>
>  
>
>>infomatrixa90b10
>>    
>>
>             [,1]         [,2]
>[1,]  0.967005807 -0.004699262
>[2,] -0.004699262  0.903852346
>  
>  
>
>>constrt
>>    
>>
>     [,1] [,2] [,3]               
>[1,] "1"  "a"  "-1.92861865194525"
>[2,] "1"  "b"  "0"                
>
>
>I'm probably missing something very simple, but I can't see what it is.  Can anyone help?
>
>  
>
Problem is with your constarint matrix. From the help file (which 
admittedlt could be clearer)
constrain is a 3-column matrix. First element in a row should be 1 to 
indicate a param set == to a value,
2 to indicate equality of two params. You put 1, so we assume first 
case.  Second row element should be , in this case,
index of param set to value, and cannot be "a" or "b". Lets suppose 
yours are first and second parameter, then it shoud be 1 and 2, 
respectively.
Finally, in this case, third row element should be value.

So you should use something like
constrt <- matrix( c(1,1,-1.9286, 1,2,0), 2, 3, byrow=TRUE)

and then, using your values for the other arguments,

 > asypow.noncent(a9010, infomatrix, constrt)
$w
           [,1]
[1,] 0.06993048

$df
[1] 2

Kjetil


-- 

Kjetil Halvorsen.

Peace is the most effective weapon of mass construction.
               --  Mahdi Elmandjra