Hi all, I try to test a linear slope using offset. I have:> m2 <- glm(Y~X*V) > summary(m2)Call: glm(formula = Y ~ X * V) Deviance Residuals: Min 1Q Median 3Q Max -2.01688 -0.56028 0.05224 0.53213 3.60216 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 1.3673 0.8476 1.613 0.119788 X 4.0235 0.1366 29.453 < 2e-16 *** Vn2 0.9683 1.1987 0.808 0.427131 Vn3 4.6043 1.1987 3.841 0.000787 *** X:Vn2 4.1108 0.1932 21.279 < 2e-16 *** X:Vn3 -4.0069 0.1932 -20.740 < 2e-16 *** --- Signif. codes: 0 `***' 0.001 `**' 0.01 `*' 0.05 `.' 0.1 ` ' 1 (Dispersion parameter for gaussian family taken to be 1.53955) Null deviance: 15303.977 on 29 degrees of freedom Residual deviance: 36.949 on 24 degrees of freedom AIC: 105.39 Number of Fisher Scoring iterations: 2 It is clear that X slope is diferent of Zero. X 4.0235 0.1366 29.453 < 2e-16 *** It is too clear that X:Vn2's slope is diferent of X's slope and diferent of Zero, because is greater than X'slope. X:Vn2 4.1108 0.1932 21.279 < 2e-16 *** But, the X:Vn3' slope is different of X'slope, but not necessarily different of Zero. How I make to introduce this parameter in a new model for test? An offset only with Vn3 slope??? I try: m2 <- glm(Y~V*offset(0*X)) m2 <- glm(Y~X*V+V*offset(0*X)) m2 <- glm(Y~V:X+V*offset(0*X)) but neither work :(( Thanks for all. Ronaldo -- If you live in a country run by committee, be on the committee. -- Graham Summer -- | //|\\ [*****************************][*******************] || ( ? ? ) [Ronaldo Reis J?nior ][PentiumIII-600 ] | V [ESALQ/USP-Entomologia, CP-09 ][HD: 30 + 10 Gb ] || / l \ [13418-900 Piracicaba - SP ][RAM: 128 Mb ] | /(lin)\ [Fone: 19-429-4199 r.229 ][Video: SiS620-8Mb ] ||/(linux)\ [chrysopa at insecta.ufv.br ][Modem: Pctel-onboar] |/ (linux) \[ICQ#: 5692561 ][Kernel: 2.4.18 ] || ( x ) [*****************************][*******************] ||| _/ \_Powered by Gnu/Debian Woody D+:) | Lxuser#: 205366
Hi,
in order to avoid the parameter of x:v3 to be estimated, you can
re-formulate the model.
You fitted
y~1+x+v2+v3+x:v2+x:v3
with 6 df
Now you would fit
y~0+v1+v2+v3+x:v1+x:v2)
with 5 df as x:v3 is not included in the model, i.e. the parameter x:v3 is
constrained to be zero.
best,
vito
A toy example is following:
x<-1:100 #explanatory continuous variable
f<-factor(rep(1:3,length=100)) #categorical variable
f<-model.matrix(~f-1)
xx<-f*x
X<-cbind(f,xx) #design matrix
colnames(X)[4:6]<-c("xf1","xf2","xf3")
b<-c(2,5,-3,-2,3,0) #parameters
y<-X%*%b+rnorm(100,0,1) #response
m0<-lm(y~0+X) #full model
m1<-lm(y~0+X[,-6]) #reduced model
anova(m0,m1) #test for nonnull slope
----- Original Message -----
From: "Ronaldo Reis Jr." <chrysopa at insecta.ufv.br>
To: "R-Help" <r-help at stat.math.ethz.ch>
Sent: Tuesday, February 04, 2003 8:50 PM
Subject: [R] testing slope
Hi all,
I try to test a linear slope using offset.
I have:
> m2 <- glm(Y~X*V)
> summary(m2)
Call:
glm(formula = Y ~ X * V)
Deviance Residuals:
Min 1Q Median 3Q Max
-2.01688 -0.56028 0.05224 0.53213 3.60216
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 1.3673 0.8476 1.613 0.119788
X 4.0235 0.1366 29.453 < 2e-16 ***
Vn2 0.9683 1.1987 0.808 0.427131
Vn3 4.6043 1.1987 3.841 0.000787 ***
X:Vn2 4.1108 0.1932 21.279 < 2e-16 ***
X:Vn3 -4.0069 0.1932 -20.740 < 2e-16 ***
---
Signif. codes: 0 `***' 0.001 `**' 0.01 `*' 0.05 `.' 0.1 ` '
1
(Dispersion parameter for gaussian family taken to be 1.53955)
Null deviance: 15303.977 on 29 degrees of freedom
Residual deviance: 36.949 on 24 degrees of freedom
AIC: 105.39
Number of Fisher Scoring iterations: 2
It is clear that X slope is diferent of Zero.
X 4.0235 0.1366 29.453 < 2e-16 ***
It is too clear that X:Vn2's slope is diferent of X's slope and diferent
of
Zero, because is greater than X'slope.
X:Vn2 4.1108 0.1932 21.279 < 2e-16 ***
But, the X:Vn3' slope is different of X'slope, but not necessarily
different
of Zero.
How I make to introduce this parameter in a new model for test? An offset
only
with Vn3 slope???
I try:
m2 <- glm(Y~V*offset(0*X))
m2 <- glm(Y~X*V+V*offset(0*X))
m2 <- glm(Y~V:X+V*offset(0*X))
but neither work :((
Thanks for all.
Ronaldo
--
If you live in a country run by committee, be on the committee.
-- Graham Summer
--
| //|\\ [*****************************][*******************]
|| ( ? ? ) [Ronaldo Reis J?nior ][PentiumIII-600 ]
| V [ESALQ/USP-Entomologia, CP-09 ][HD: 30 + 10 Gb ]
|| / l \ [13418-900 Piracicaba - SP ][RAM: 128 Mb ]
| /(lin)\ [Fone: 19-429-4199 r.229 ][Video: SiS620-8Mb ]
||/(linux)\ [chrysopa at insecta.ufv.br ][Modem: Pctel-onboar]
|/ (linux) \[ICQ#: 5692561 ][Kernel: 2.4.18 ]
|| ( x ) [*****************************][*******************]
||| _/ \_Powered by Gnu/Debian Woody D+:) | Lxuser#: 205366
______________________________________________
R-help at stat.math.ethz.ch mailing list
http://www.stat.math.ethz.ch/mailman/listinfo/r-help
For testing the hypothesis
Beta[X] + Beta[Vn3] + Beta[X:Vn3] = 0
you can use the estimable function from the gregmisc package:
estimable(m2, c("Intercept"=0,
"X"=1,
"Vn2"=0,
"Vn3"=1,
"X:Vn2"=0,
"X:Vn3"=1)
)
See the estimable help page for details.
-Greg
> -----Original Message-----
> From: Ronaldo Reis Jr. [mailto:chrysopa at insecta.ufv.br]
> Sent: Tuesday, February 04, 2003 2:51 PM
> To: R-Help
> Subject: [R] testing slope
>
>
> Hi all,
>
> I try to test a linear slope using offset.
>
> I have:
>
> > m2 <- glm(Y~X*V)
> > summary(m2)
>
> Call:
> glm(formula = Y ~ X * V)
>
> Deviance Residuals:
> Min 1Q Median 3Q Max
> -2.01688 -0.56028 0.05224 0.53213 3.60216
>
> Coefficients:
> Estimate Std. Error t value Pr(>|t|)
> (Intercept) 1.3673 0.8476 1.613 0.119788
> X 4.0235 0.1366 29.453 < 2e-16 ***
> Vn2 0.9683 1.1987 0.808 0.427131
> Vn3 4.6043 1.1987 3.841 0.000787 ***
> X:Vn2 4.1108 0.1932 21.279 < 2e-16 ***
> X:Vn3 -4.0069 0.1932 -20.740 < 2e-16 ***
> ---
> Signif. codes: 0 `***' 0.001 `**' 0.01 `*' 0.05 `.' 0.1 `
' 1
>
> (Dispersion parameter for gaussian family taken to be 1.53955)
>
> Null deviance: 15303.977 on 29 degrees of freedom
> Residual deviance: 36.949 on 24 degrees of freedom
> AIC: 105.39
>
> Number of Fisher Scoring iterations: 2
>
> It is clear that X slope is diferent of Zero.
> X 4.0235 0.1366 29.453 < 2e-16 ***
>
> It is too clear that X:Vn2's slope is diferent of X's slope
> and diferent of
> Zero, because is greater than X'slope.
> X:Vn2 4.1108 0.1932 21.279 < 2e-16 ***
>
> But, the X:Vn3' slope is different of X'slope, but not
> necessarily different
> of Zero.
>
> How I make to introduce this parameter in a new model for
> test? An offset only
> with Vn3 slope???
>
> I try:
>
> m2 <- glm(Y~V*offset(0*X))
> m2 <- glm(Y~X*V+V*offset(0*X))
> m2 <- glm(Y~V:X+V*offset(0*X))
>
> but neither work :((
>
> Thanks for all.
> Ronaldo
> --
> If you live in a country run by committee, be on the committee.
> -- Graham Summer
> --
> | //|\\ [*****************************][*******************]
> || ( ? ? ) [Ronaldo Reis J?nior ][PentiumIII-600 ]
> | V [ESALQ/USP-Entomologia, CP-09 ][HD: 30 + 10 Gb ]
> || / l \ [13418-900 Piracicaba - SP ][RAM: 128 Mb ]
> | /(lin)\ [Fone: 19-429-4199 r.229 ][Video: SiS620-8Mb ]
> ||/(linux)\ [chrysopa at insecta.ufv.br ][Modem: Pctel-onboar]
> |/ (linux) \[ICQ#: 5692561 ][Kernel: 2.4.18 ]
> || ( x ) [*****************************][*******************]
> ||| _/ \_Powered by Gnu/Debian Woody D+:) | Lxuser#: 205366
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> http://www.stat.math.ethz.ch/mailman/listinfo/r-help
>
LEGAL NOTICE\ Unless expressly stated otherwise, this message is ... [[dropped]]