[R] Interval pls

[Spencer Graves]

      RSiteSearch("interval pls") produced 7 hits for me just now.  I 
didn't study all of them carefully, but I doubt if any one of them have 
exactly what you want.  One could also search for "pls with ordered 
factors", but I didn't try that. 

      However, I doubt if it would be excessively difficult to write a 
wrapper for appropriate functions in the 'pls' packages to do what you 
want.  With the 'interval' variables as a potentially explanatory 
variable rather than a response variable, one could start by replacing 
ordered factors by contrasts, then adjusting the coding for intermediate 
levels iteratively until either (a) intermediate levels were collapsed 
because the model would otherwise reverse them or (b) the coefficients 
of nonlinear terms were all zero.  Something similar could be done if an 
'interval' variable was a response variable, with the goal of maximizing 
R2. 

      Something similar could be done using the 'sem' package, but the 
'sem' function seems to require a moment or covariance matrix as an 
input, which might make the wrapper somewhat more complex, I think.  
With 'sem' one could adjust the levels for an ordered response variable 
to minimize BIC, rather than maximizing R2.  This latter seems to me to 
be somewhat more satisfying and solid theoretically, but others may not 
agree. 

      Of course, whatever I did in this regard, I'd want to also 
accompany the effort with a more careful literature search. 

      Hope this helps. 
      Spencer Graves

Dirk De Becker wrote:
> Hey R-helpers,
>
> Does anybody know of an implementation of interval PLS in R?
>
> Thx,
>
> Dirk
>
>