R help - Jun 2008 - How to force two regression coefficients to be equal but opposite in sign?

Is there a way to set up a regression in R that forces two coefficients

to be equal but opposite in sign?

 

I'm trying to setup a model where a subject appears in a pair of

environments where a measurement X is made.  There are a total of 5

environments, one of which is a baseline.  But each observation is for

a subject in only two of them, and not all subjects will appear in

each environment.

 

Each of the environments has an effect on the variable X.  I want to

measure the relative effects of each environment E on X with a model.

 

Xj = Xi * Ei / Ej

 

Ei of the baseline model is set equal to 1.

 

With a log transform, a linear-looking regression can be written as:

 

log(Xj) = log(Xi) + log(Ei) - log(Ej)

 

My data looks like:

 

#    E1   X1   E2    X2

1    A   .20   B    .25   

 

What I've tried in R:

 

env <-
c("A","B","C","D","E")

 

# Note: data is made up just for this example

 

df <- data.frame(

                X1
c(.20,.10,.40,.05,.10,.24,.30,.70,.48,.22,.87,.29,.24,.19,.92),

                X2
c(.25,.12,.45,.01,.19,.50,.30,.40,.50,.40,.68,.30,.16,.02,.70),

                E1
c("A","A","A","B","B","B","C","C","C","D","D","D","E","E","E"),

                E2
c("B","C","D","A","D","E","A","B","E","B","C","E","A","B","C")

)

 

model <- lm(log(X2) ~ log(X1) + E1 + E2, data = df)

 

summary(model)

 

Call:

lm(formula = log(X2) ~ log(X1) + E1 + E2, data = df)

 

Residuals:

      1       2       3       4       5       6       7       8       9
10      11      12      13      14      15 

 0.3240  0.2621 -0.5861 -1.0283  0.5861  0.4422  0.3831 -0.2608 -0.1222
0.9002 -0.5802 -0.3200  0.6452 -0.9634  0.3182 

 

Coefficients:

            Estimate Std. Error t value Pr(>|t|)  

(Intercept)  0.54563    1.71558   0.318    0.763  

log(X1)      1.29745    0.57295   2.265    0.073 .

E1B         -0.23571    0.95738  -0.246    0.815  

E1C         -0.57057    1.20490  -0.474    0.656  

E1D         -0.22988    0.98274  -0.234    0.824  

E1E         -1.17181    1.02918  -1.139    0.306  

E2B         -0.16775    0.87803  -0.191    0.856  

E2C          0.05952    1.12779   0.053    0.960  

E2D          0.43077    1.19485   0.361    0.733  

E2E          0.40633    0.98289   0.413    0.696  

---

Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05
'.' 0.1 ' ' 1

 

Residual standard error: 1.004 on 5 degrees of freedom

Multiple R-squared: 0.7622,     Adjusted R-squared: 0.3343 

F-statistic: 1.781 on 9 and 5 DF,  p-value: 0.2721 

 

----

 

What I need to do is force the corresponding environment coefficients

to be equal in absolute value, but opposite in sign.  That is:

 

E1B = -E2B

E1C = -E3C

E1D = -E3D

E1E = -E1E

 

In essence, E1 and E2 are the "same" variable, but can play two

different roles in the model depending on whether it's the first part

of the observation or the second part.

 

I searched the archive, and the closest thing I found to my situation

was:

 

http://tolstoy.newcastle.edu.au/R/e4/help/08/03/6773.html 

 

But the response to that thread didn't seem to be applicable to my

situation.

 

Any pointers would be appreciated.

 

Thanks,

Keith

 


	[[alternative HTML version deleted]]

One simple way is to do something like:
> fit <- lm(y ~ I(x1-x2) + x3, data=mydata)
The first coeficient (after the intercept) will be the slope for x1, the slope
for x2 will be the negative of that.  This model is nested in the fuller model
with x1 and x2 fit seperately and you can therefore test for differences.

Hope this helps,

--
Gregory (Greg) L. Snow Ph.D.
Statistical Data Center
Intermountain Healthcare
greg.snow at imail.org
(801) 408-8111


> -----Original Message-----
> From: r-help-bounces at r-project.org
> [mailto:r-help-bounces at r-project.org] On Behalf Of Woolner, Keith
> Sent: Friday, June 06, 2008 10:07 AM
> To: r-help at r-project.org
> Subject: [R] How to force two regression coefficients to be
> equal but opposite in sign?
>
> Is there a way to set up a regression in R that forces two
> coefficients
>
> to be equal but opposite in sign?
>
>
>
> I'm trying to setup a model where a subject appears in a pair of
>
> environments where a measurement X is made.  There are a total of 5
>
> environments, one of which is a baseline.  But each observation is for
>
> a subject in only two of them, and not all subjects will appear in
>
> each environment.
>
>
>
> Each of the environments has an effect on the variable X.  I want to
>
> measure the relative effects of each environment E on X with a model.
>
>
>
> Xj = Xi * Ei / Ej
>
>
>
> Ei of the baseline model is set equal to 1.
>
>
>
> With a log transform, a linear-looking regression can be written as:
>
>
>
> log(Xj) = log(Xi) + log(Ei) - log(Ej)
>
>
>
> My data looks like:
>
>
>
> #    E1   X1   E2    X2
>
> 1    A   .20   B    .25
>
>
>
> What I've tried in R:
>
>
>
> env <-
c("A","B","C","D","E")
>
>
>
> # Note: data is made up just for this example
>
>
>
> df <- data.frame(
>
>                 X1 >
c(.20,.10,.40,.05,.10,.24,.30,.70,.48,.22,.87,.29,.24,.19,.92),
>
>                 X2 >
c(.25,.12,.45,.01,.19,.50,.30,.40,.50,.40,.68,.30,.16,.02,.70),
>
>                 E1 >
c("A","A","A","B","B","B","C","C","C","D","D","D","E","E","E"),
>
>                 E2 >
c("B","C","D","A","D","E","A","B","E","B","C","E","A","B","C")
>
> )
>
>
>
> model <- lm(log(X2) ~ log(X1) + E1 + E2, data = df)
>
>
>
> summary(model)
>
>
>
> Call:
>
> lm(formula = log(X2) ~ log(X1) + E1 + E2, data = df)
>
>
>
> Residuals:
>
>       1       2       3       4       5       6       7
> 8       9
> 10      11      12      13      14      15
>
>  0.3240  0.2621 -0.5861 -1.0283  0.5861  0.4422  0.3831
> -0.2608 -0.1222
> 0.9002 -0.5802 -0.3200  0.6452 -0.9634  0.3182
>
>
>
> Coefficients:
>
>             Estimate Std. Error t value Pr(>|t|)
>
> (Intercept)  0.54563    1.71558   0.318    0.763
>
> log(X1)      1.29745    0.57295   2.265    0.073 .
>
> E1B         -0.23571    0.95738  -0.246    0.815
>
> E1C         -0.57057    1.20490  -0.474    0.656
>
> E1D         -0.22988    0.98274  -0.234    0.824
>
> E1E         -1.17181    1.02918  -1.139    0.306
>
> E2B         -0.16775    0.87803  -0.191    0.856
>
> E2C          0.05952    1.12779   0.053    0.960
>
> E2D          0.43077    1.19485   0.361    0.733
>
> E2E          0.40633    0.98289   0.413    0.696
>
> ---
>
> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05
'.' 0.1 ' ' 1
>
>
>
> Residual standard error: 1.004 on 5 degrees of freedom
>
> Multiple R-squared: 0.7622,     Adjusted R-squared: 0.3343
>
> F-statistic: 1.781 on 9 and 5 DF,  p-value: 0.2721
>
>
>
> ----
>
>
>
> What I need to do is force the corresponding environment coefficients
>
> to be equal in absolute value, but opposite in sign.  That is:
>
>
>
> E1B = -E2B
>
> E1C = -E3C
>
> E1D = -E3D
>
> E1E = -E1E
>
>
>
> In essence, E1 and E2 are the "same" variable, but can play two
>
> different roles in the model depending on whether it's the first part
>
> of the observation or the second part.
>
>
>
> I searched the archive, and the closest thing I found to my situation
>
> was:
>
>
>
> http://tolstoy.newcastle.edu.au/R/e4/help/08/03/6773.html
>
>
>
> But the response to that thread didn't seem to be applicable to my
>
> situation.
>
>
>
> Any pointers would be appreciated.
>
>
>
> Thanks,
>
> Keith
>
>
>
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>