| Library: | metrics |
| See also: | redun makedesign |
| Quantlet: | dpls | |
| Description: | calculates latent variables, weights, loadings and path coefficients with dynamic partial least squares algorithm |
| Usage: | {wg,b,sk,sdv,skl,slv,lk,iter} = dpls(w,d,dy,dl,y,lag,precis) | |
| Input: | ||
| w | l x k matrix, start weights | |
| d | k x k matrix, inner unlagged designs (0 or 1), no diagonal values are allowed | |
| dy | l x k matrix, outer designs (0 or 1) rows are counting manifest variables | |
| dl | k x k matrix, inner lagged designs (0 or 1), the diagonal elements are showing autoregression | |
| y | n x l matrix, manifest variables (indicators) | |
| lag | scalar, lag order | |
| precis | scalar, cancelling criterion | |
| Output: | ||
| wg | l x k matrix, weights | |
| b | l x k matrix, loadings | |
| sk | k x k matrix, path coefficients | |
| sdv | k x k matrix, standard deviations of path coefficients | |
| skl | k x k matrix, lagged path coefficients with the same dimension as d and ordered like designed | |
| slv | k x k matrix, the standard deviations of the path coefficients | |
| lk | n x k matrix, latent variables | |
| iter | scalar, the number of iterations used | |
randomize(13409)
library("metrics")
b1=0.3
c1=0.6
s=500
n1=normal(s+1)
n1lag=n1[1:s,]
n1=n1[2:rows(n1),] ;innermodel
n2=b1*n1+c1*n1lag+normal(rows(n1))/5
n=n1~n2
nn=n./sqrt(var(n)) ;loadingsmatrix
p=(1|2|3|4|0|0|0)~(0|0|0|0|5|6|7)
y=nn*p'+normal(rows(n),rows(p))/8
d=(0|1)~(0|0)
dl=(0|1)~(0|0)
w=(1|1|1|1|0|0|0)~(0|0|0|0|1|1|1)
myfit=dpls(w,d,w,dl,y,1,3)
myfit.wg
myfit.b
myfit.sk
myfit.skl
myfit.iter
Contents of wg [1,] 0.032821 0 [2,] 0.066113 0 [3,] 0.099959 0 [4,] 0.13301 0 [5,] 0 0.045413 [6,] 0 0.054268 [7,] 0 0.063904 Contents of b [1,] 0.9967 0 [2,] 2.0131 0 [3,] 3.005 0 [4,] 4.0134 0 [5,] 0 5.0093 [6,] 0 5.9997 [7,] 0 6.9937 Contents of sk [1,] 0 0 [2,] 0.423 0 Contents of skl [1,] 0 0 [2,] 0.856 0 Contents of iter [1,] 2