variance-covariance matrix for GMerrorsar

Dear list,

I want to use bootstrapping to derive confidence intervals for marginal
wtp after GMerrorsar command.

It works for stsls since covariance matrix is directly available. However,
I cannot find covariance matrix for GMerrorsar.

For example, the following code works for stsls:

model1.beta <- coef(model1)

model1.vcov <- summary(model1)$var

model1.sim <- rmultnorm(10000, mu = model1.beta, vmat = model1.vcov)

model1.mwtp <- model1.sim * Pbar

model1.ci <- apply(model1.mwtp, 2, quantile, c(0.025, 0.975))

when I apply the same code for GMerrorsar:


model2.beta <- coef(model2)

model2.vcov <- summary(model2)$var

model2.vcov

NULL


How can I obtain covariance matrix for GMerrorsar?

Chelsea

	[[alternative HTML version deleted]]

On Tue, 11 Apr 2017, Qiuhua Ma wrote:

Dear list,

I want to use bootstrapping to derive confidence intervals for marginal
wtp after GMerrorsar command.

It works for stsls since covariance matrix is directly available. However,
I cannot find covariance matrix for GMerrorsar.

For example, the following code works for stsls:

model1.beta <- coef(model1)

model1.vcov <- summary(model1)$var

model1.sim <- rmultnorm(10000, mu = model1.beta, vmat = model1.vcov)

model1.mwtp <- model1.sim * Pbar

model1.ci <- apply(model1.mwtp, 2, quantile, c(0.025, 0.975))

The DGP for this model is (I - \rho W)^{-1} (X \beta + e), so I'm in geat
doubt about whether your proposal is correct (model1.vcov is has one more
row and column than X has columns, so including \rho); the first element of
model1.beta is \rho.

when I apply the same code for GMerrorsar:


model2.beta <- coef(model2)

model2.vcov <- summary(model2)$var


model2.vcov

NULL


How can I obtain covariance matrix for GMerrorsar?

Reading the code, you'll see where the matrices occur. Running under

debug, you can assign the outside the environment of the function if you
like (use <<- ). I've added a vcov component in the returned object (source
on R-Forge, I can send a source package or a Windows binary package).

You should also look at sphet::spreg, which does return a var component.
Please note that you should think of the DGP first and foremost, the coef
and var may return the values for what you are treating as nuisance parts
of the model. Getting the distribution of the willingess to pay also
probably involves them and their variability.

Have you considered getting the WTP marginal from a Bayesian approach?

Hope this helps,

Roger

Chelsea

        [[alternative HTML version deleted]]

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; e-mail: Roger.Bivand at nhh.no
Editor-in-Chief of The R Journal, https://journal.r-project.org/index.html
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

Thanks for your quick reply. You are right. Marginal wtp should take into
account rho for spatial lag model.

I still would like to use GMerrorsar. Can you please send me the source
package?

Best,

Chelsea

On Tue, Apr 11, 2017 at 7:54 AM, Roger Bivand <Roger.Bivand at nhh.no> wrote:

On Tue, 11 Apr 2017, Qiuhua Ma wrote:

Dear list,

I want to use bootstrapping to derive confidence intervals for marginal
wtp after GMerrorsar command.

It works for stsls since covariance matrix is directly available.
However,
I cannot find covariance matrix for GMerrorsar.

For example, the following code works for stsls:

model1.beta <- coef(model1)

model1.vcov <- summary(model1)$var

model1.sim <- rmultnorm(10000, mu = model1.beta, vmat = model1.vcov)

model1.mwtp <- model1.sim * Pbar

model1.ci <- apply(model1.mwtp, 2, quantile, c(0.025, 0.975))

The DGP for this model is (I - \rho W)^{-1} (X \beta + e), so I'm in geat
doubt about whether your proposal is correct (model1.vcov is has one more
row and column than X has columns, so including \rho); the first element of
model1.beta is \rho.

when I apply the same code for GMerrorsar:


model2.beta <- coef(model2)

model2.vcov <- summary(model2)$var


model2.vcov

NULL


How can I obtain covariance matrix for GMerrorsar?

Reading the code, you'll see where the matrices occur. Running under

debug, you can assign the outside the environment of the function if you
like (use <<- ). I've added a vcov component in the returned object (source
on R-Forge, I can send a source package or a Windows binary package).

You should also look at sphet::spreg, which does return a var component.
Please note that you should think of the DGP first and foremost, the coef
and var may return the values for what you are treating as nuisance parts
of the model. Getting the distribution of the willingess to pay also
probably involves them and their variability.

Have you considered getting the WTP marginal from a Bayesian approach?

Hope this helps,

Roger

Chelsea

        [[alternative HTML version deleted]]

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; e-mail: Roger.Bivand at nhh.no
Editor-in-Chief of The R Journal, https://journal.r-project.org/
index.html
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

Is this the link to the package?

https://r-forge.r-project.org/R/?group_id=182

Chelsea

On Tue, Apr 11, 2017 at 10:59 PM, Qiuhua Ma <qiuhuanihao at gmail.com> wrote:

Thanks for your quick reply. You are right. Marginal wtp should take into
account rho for spatial lag model.

I still would like to use GMerrorsar. Can you please send me the source
package?

Best,

Chelsea

On Tue, Apr 11, 2017 at 7:54 AM, Roger Bivand <Roger.Bivand at nhh.no> wrote:

On Tue, 11 Apr 2017, Qiuhua Ma wrote:

Dear list,

I want to use bootstrapping to derive confidence intervals for marginal
wtp after GMerrorsar command.

It works for stsls since covariance matrix is directly available.
However,
I cannot find covariance matrix for GMerrorsar.

For example, the following code works for stsls:

model1.beta <- coef(model1)

model1.vcov <- summary(model1)$var

model1.sim <- rmultnorm(10000, mu = model1.beta, vmat = model1.vcov)

model1.mwtp <- model1.sim * Pbar

model1.ci <- apply(model1.mwtp, 2, quantile, c(0.025, 0.975))

The DGP for this model is (I - \rho W)^{-1} (X \beta + e), so I'm in geat
doubt about whether your proposal is correct (model1.vcov is has one more
row and column than X has columns, so including \rho); the first element of
model1.beta is \rho.

when I apply the same code for GMerrorsar:


model2.beta <- coef(model2)

model2.vcov <- summary(model2)$var


model2.vcov

NULL


How can I obtain covariance matrix for GMerrorsar?

Reading the code, you'll see where the matrices occur. Running under

debug, you can assign the outside the environment of the function if you
like (use <<- ). I've added a vcov component in the returned object (source
on R-Forge, I can send a source package or a Windows binary package).

You should also look at sphet::spreg, which does return a var component.
Please note that you should think of the DGP first and foremost, the coef
and var may return the values for what you are treating as nuisance parts
of the model. Getting the distribution of the willingess to pay also
probably involves them and their variability.

Have you considered getting the WTP marginal from a Bayesian approach?

Hope this helps,

Roger

Chelsea

        [[alternative HTML version deleted]]

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; e-mail: Roger.Bivand at nhh.no
Editor-in-Chief of The R Journal, https://journal.r-project.org/
index.html
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en