Dear List-Members,
I have a problem with the function lqs() from package MASS. In some cases it
produces different results for the same settings and needs a random seed to be
set, in other cases not.
I really cannot understand, why this happens. As well I do not understand what
exactly you need the random seed for. Is it a starting point for iterations? Or
do different results occur because of the estimation doesn't converge?
I tried data "phones" from package MASS. You find this example as well
in the MASS-book on page 160.
> lqs(calls~year, data=phones, method="lms")
Call:
lqs.formula(formula = calls ~ year, data = phones, method = "lms")
Coefficients:
(Intercept) year
-55.947 1.155
Scale estimates 0.9377 0.9095
>
> lqs(calls~year, data=phones, method="S")
Call:
lqs.formula(formula = calls ~ year, data = phones, method = "S")
Coefficients:
(Intercept) year
-52.5 1.1
Scale estimates 2.129
You can do it over and over again and get the same coefficients. In contrast, if
u use other data like cats from MASS or simulated data, u get different outputs
every time u start the code if not electing a random.seed.
> lqs(Hwt~Bwt, data=cats, method="S")
Call:
lqs.formula(formula = Hwt ~ Bwt, data = cats, method = "S")
Coefficients:
(Intercept) Bwt
0.2625 3.6250
Scale estimates 1.474
> lqs(Hwt~Bwt, data=cats, method="S")
Call:
lqs.formula(formula = Hwt ~ Bwt, data = cats, method = "S")
Coefficients:
(Intercept) Bwt
0.4714 3.5714
Scale estimates 1.474
Example with simulated data:
> b0<--1
> b1<-6
> b2<-0.8
> b3<--0.5
>
> x1<-runif(200,-3,3)
> x2<-runif(200,20,40)
> x3<-rbinom(200, 1, 0.7)
> e<-rnorm(200,0,1)
> y<-b0+b1*x1+b2*x2+b3*x3+e
> lqs(y~x1, method="lms")
Call:
lqs.formula(formula = y ~ x1, method = "lms")
Coefficients:
(Intercept) x1
22.239 4.964
Scale estimates 5.379 4.891
>
> lqs(y~x1, method="S")
Call:
lqs.formula(formula = y ~ x1, method = "S")
Coefficients:
(Intercept) x1
23.193 5.743
Scale estimates 5.642
> lqs(y~x1, method="lms")
Call:
lqs.formula(formula = y ~ x1, method = "lms")
Coefficients:
(Intercept) x1
21.176 5.255
Scale estimates 5.383 5.023
>
> lqs(y~x1, method="S")
Call:
lqs.formula(formula = y ~ x1, method = "S")
Coefficients:
(Intercept) x1
22.55 5.46
Scale estimates 5.642
Thanks for your help,
I appreciate it!
K. Schmidt
________________________________________________________________
Neu: WEB.DE Doppel-FLAT mit Internet-Flatrate + Telefon-Flatrate
f?r nur 19,99 Euro/mtl.!* http://produkte.web.de/go/02/
