[R-sig-Geo] R: R: R: Spatial Durbin Model with regimes in a cross section framework
Roger Bivand
Roger.Bivand at nhh.no
Wed Oct 21 15:25:33 CEST 2015
On Wed, 21 Oct 2015, Loredana Mirra wrote:
> Thanks Roger,
> At least you managed to have an output even if it contains some issues which
> I could address changing variables.
> Instead sorry but I am stuck. How can I get that?
library(spdep)
sessionInfo() # spdep >=0.5-88
load("chow_error.RData")
ONE <- rep(1, nrow(highways))
NOC_Nord <- as.integer(!highways$C_Nord)
durb<-lagsarlm(GR5101m~ 0+ (ONE+LVAPC51):(C_Nord+NOC_Nord),
data=highways, listw, type="mixed")
summary(durb)
works for me (I'm using the development version of spdep, which seems to
be the reason for the difference). There was an indexing error in
create_WX() that is corrected in the development version but not in 0.5-88
on CRAN. Which OS are you using - there are reasons for not releasing the
development version yet?
Roger
> Loredana
>
>
> -----Messaggio originale-----
> Da: Roger Bivand [mailto:Roger.Bivand at nhh.no]
> Inviato: mercoledì 21 ottobre 2015 14:19
> A: Loredana Mirra
> Cc: r-sig-geo at r-project.org
> Oggetto: Re: R: R: [R-sig-Geo] Spatial Durbin Model with regimes in a cross
> section framework
>
> On Wed, 21 Oct 2015, Loredana Mirra wrote:
>
>> Here the traceback()
>> 4: stop("0 (non-NA) cases")
>> 3: lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...)
>> 2: lm(y ~ x - 1)
>> 1: lagsarlm(GR5101m ~ 0 + (ONE + LVAPC51):(C_Nord + NOC_Nord), data =
>> highwaysreg,
>> listw, type = "mixed")
>>
>> No, problem. Data are not private. I tried to use the lmSLX and I got
>> the same error
>
> OK, thanks. I cannot reproduce your error as such - I'm constructing:
>
> ONE <- rep(1, nrow(highways))
> NOC_Nord <- as.integer(!highways$C_Nord)
>
> which may not be correct. I'm seeing:
>
> Warning message:
> In lagsarlm(GR5101m ~ 0 + I(factor(C_Nord))/(LVAPC51), data = highways, :
> inversion of asymptotic covariance matrix failed for tol.solve = 1e-10
> reciprocal condition number = 6.6223e-12 - using numerical Hessian.
>
> and an issue in an interaction term between the NOC_Nord term and the
> intercept:
>
> Coefficients: (numerical Hessian approximate standard errors)
> (1 not defined because of singularities)
> Estimate Std. Error z value Pr(>|z|)
> ONE:C_Nord 0.1385889 0.1310782 1.0573 0.2904
> ONE:NOC_Nord 0.1493507 0.1322529 1.1293 0.2588
> LVAPC51:C_Nord -0.0170657 0.0010275 -16.6083 <2e-16
> LVAPC51:NOC_Nord -0.0192602 0.0018370 -10.4845 <2e-16
> lag.ONE:C_Nord 0.0195317 0.1159969 0.1684 0.8663
> lag.ONE:NOC_Nord NA NA NA NA
> lag.LVAPC51:C_Nord -0.0034108 0.0086611 -0.3938 0.6937
> lag.LVAPC51:NOC_Nord -0.0032861 0.0223143 -0.1473 0.8829
>
> Rho: 0.57171, LR test value: 1.8528, p-value: 0.17346 Approximate (numerical
> Hessian) standard error: 0.34917
> z-value: 1.6374, p-value: 0.10156
> Wald statistic: 2.6809, p-value: 0.10156
>
> Roger
>
>
>> This is the link to the file with errors
>> https://onedrive.live.com/redir?resid=4484D82DD190F118!2202&authkey=!A
>> EXvKmB
>> o2FVgrQI&ithint=file%2cRData
>>
>> Thank you
>>
>> Loredana
>>
>>
>>
>> -----Messaggio originale-----
>> Da: Roger Bivand [mailto:Roger.Bivand at nhh.no]
>> Inviato: mercoledì 21 ottobre 2015 13:26
>> A: Loredana Mirra
>> Cc: r-sig-geo at r-project.org
>> Oggetto: Re: R: [R-sig-Geo] Spatial Durbin Model with regimes in a
>> cross section framework
>>
>> On Wed, 21 Oct 2015, Loredana Mirra wrote:
>>
>>> Hi, and thank you again for your help. Following Anselin (2007)
>>> Spatial regression analysis in R I managed to test for the presence
>>> of spatial regimes (also performing a Chow test) in R.
>>>
>>> Since, I need to apply a Spatial Durbin model, I tried fitting it
>>> using the following (just adding the option "mixed") to the sar
>>> command in the cases of regimes (Centre_North and South in my case):
>>>
>>> durb<-lagsarlm(GR5101m~ 0+ (ONE+
>>> LVAPC51):(C_Nord+NOC_Nord),data=highways,listw, type="mixed")
>>> summary(durb)
>>>
>>> I always get this error
>>> error in lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...) :
>>> 0 (non-NA) cases
>>
>> Please provide the output of traceback() after the error has occurred.
>> I think that if the same model can be fitted with type="lag", but not
>> with "mixed", then the lagged X variables are possibly to blame.
>>
>> Maybe also try to use lmSLX() on the same formula, data and listw
>> objects, to see whether you also see failures in the lagged X
>> variables. The underlying create_WX() function does check for NAs, but
>> may not do it correctly - possibly for this formula.
>>
>> If this sounds hard, do:
>>
>> save(highways, listw, file="chow_error.RData")
>>
>> and make the "chow_error.RData" file available on a link (possibly
>> off-list if the data are private).
>>
>> Hope this helps,
>>
>> Roger
>>
>>>
>>> I checked if there were NA cases and tried with other variables or
>>> weight matrices. I wonder if this is a good approach or what is wrong.
>>>
>>> Thank you for your attention
>>> Loredana Mirra
>>>
>>>
>>>
>>>
>>> -----Messaggio originale-----
>>> Da: Roger Bivand [mailto:Roger.Bivand at nhh.no]
>>> Inviato: venerdì 2 ottobre 2015 10:59
>>> A: Loredana Mirra
>>> Cc: r-sig-geo at r-project.org
>>> Oggetto: Re:[R-sig-Geo] Spatial Durbin Model with regimes in a cross
>>> section framework
>>>
>>> On Fri, 2 Oct 2015, Loredana Mirra wrote:
>>>
>>>>
>>>> Thank you very much for your kind help. Yes, I should check also if,
>>>> after having estimated separated coefficients, the difference, if
>>>> any, is statistically significant (using a chow test). Implementing
>>>> such test is the major issue to me. Can you give me an hint about this?
>>>
>>> Please see ch. 4 in Kleiber and Zeilis (2008) Applied Econometrics with
> R.
>>> You may use anova() on the no-regime model and the regime model,
>>> which is equivalent to a Chow test (personal communication, Achim
>>> Zeileis,
>> 2005).
>>> It will however suffer from misspecification such as outliers,
>>> discussed in section 4.3 and the use of a Wald test with an HC
>>> covariance
>> matrix.
>>>
>>>> Thanks also for the suggestion about the interpretation of output in
> SDM.
>>>
>>> A "Chow" test might go through a Likelihood Ratio test, but should
>>> arguably be presented using the empirical distributions of the
>>> impacts for each X variable for the no-regime model and for each regime.
>>>
>>> Hope this helps,
>>>
>>> Roger
>>>
>>>> Best Regards
>>>> Loredana
>>>> -----Messaggio originale-----
>>>> Da: Roger Bivand [mailto:Roger.Bivand at nhh.no]
>>>> Inviato: giovedì 1 ottobre 2015 20:45
>>>> A: Loredana Mirra
>>>> Cc: r-sig-geo at r-project.org
>>>> Oggetto: Re: [R-sig-Geo] Spatial Durbin Model with regimes in a
>>>> cross section framework
>>>>
>>>> On Thu, 1 Oct 2015, Loredana Mirra wrote:
>>>>
>>>>> I am writing to ask you a suggestion about the availability of a R
>>>>> routine or piece of program for the purposes of my empirical analysis.
>>>>> I should revise my paper using a Spatial Durbin model. This was
>>>>> recommended in order to mitigate a problem due to the possible
>>>>> presence of
>>>> unobserved factors.
>>>>>
>>>>> Unfortunately data do not permit to perform a panel but a cross
>>>>> section analysis. I do not have troubles to perform a Spatial
>>>>> Durbin Model on the entire sample, but I need to study the presence
>>>>> of Spatial Regimes. The problem is that I cannot find a program, in
>>>>> a cross sections framework. Do you know if there is a program
>>>>> available in R
>>>> to perform this analysis?
>>>>
>>>> How do you want to handle the "regimes"? Estimate separate
>>>> coefficients for each regime (regime defined as a factor)? If so,
>>>> this is a standard formula construction, and can be used with
>>>> formula objects in model fitting
>>>> functions:
>>>>
>>>> lm(mpg ~ 0 + I(factor(am))/(disp + wt), data=mtcars)
>>>>
>>>> or similar - check in detailed discussions of formula objects.
>>>>
>>>> You need to be careful to report impacts, not coefficient values,
>>>> from the Spatial Durbin model.
>>>>
>>>> Hope this helps,
>>>>
>>>> Roger
>>>>
>>>>> Thank you very much in advance for your help.
>>>>>
>>>>> Best regards
>>>>>
>>>>> Loredana Mirra
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> -------------------------
>>>>>
>>>>>
>>>>>
>>>>> Dott.ssa Loredana Mirra
>>>>>
>>>>> Universit??? di Roma ???Tor Vergata???
>>>>>
>>>>> Dipartimento di Economia Diritto e Istituzioni
>>>>>
>>>>> Via Columbia, 2
>>>>>
>>>>> 00133 Roma
>>>>>
>>>>> Italia
>>>>>
>>>>> tel. +390672595725
>>>>>
>>>>> <mailto:pozzolo at unimol.it> loredana.mirra at uniroma2.it
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> ---
>>>>> Questa e-mail ??? stata controllata per individuare virus con Avast
>>>> antivirus.
>>>>> https://www.avast.com/antivirus
>>>>>
>>>>> [[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; fax +47 55 95 91 00
>>>> e-mail: Roger.Bivand at nhh.no
>>>>
>>>>
>>>> ---
>>>> Questa e-mail è stata controllata per individuare virus con Avast
>>> antivirus.
>>>> https://www.avast.com/antivirus
>>>>
>>>>
>>>
>>> --
>>> Roger Bivand
>>> Department of Economics, Norwegian School of Economics, Helleveien
>>> 30,
>>> N-5045 Bergen, Norway.
>>> voice: +47 55 95 93 55; fax +47 55 95 91 00
>>> e-mail: Roger.Bivand at nhh.no
>>>
>>>
>>> ---
>>> Questa e-mail è stata controllata per individuare virus con Avast
>> antivirus.
>>> https://www.avast.com/antivirus
>>>
>>>
>>
>> --
>> Roger Bivand
>> Department of Economics, Norwegian School of Economics, Helleveien 30,
>> N-5045 Bergen, Norway.
>> voice: +47 55 95 93 55; fax +47 55 95 91 00
>> e-mail: Roger.Bivand at nhh.no
>>
>>
>> ---
>> Questa e-mail è stata controllata per individuare virus con Avast
> antivirus.
>> https://www.avast.com/antivirus
>>
>>
>
> --
> Roger Bivand
> Department of Economics, Norwegian School of Economics, Helleveien 30,
> N-5045 Bergen, Norway.
> voice: +47 55 95 93 55; fax +47 55 95 91 00
> e-mail: Roger.Bivand at nhh.no
>
>
> ---
> Questa e-mail è stata controllata per individuare virus con Avast antivirus.
> https://www.avast.com/antivirus
>
>
--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 91 00
e-mail: Roger.Bivand at nhh.no
More information about the R-sig-Geo
mailing list