[R-sig-Geo] Error object 'can.sim2' not found in spatianl panel data - splm with spdep

Tue Oct 9 17:18:34 CEST 2012

On Tue, 9 Oct 2012, Patrick Meyfroidt wrote:

> OK, now it appears that probably my problem was due to poor understanding
> of how I had to create the listw object with spdep.
>
> Starting from the nearest neighbor identification, what I was doing was:
> ruskn5 <- knearneigh(coords, k=5, longlat = NULL)
> ruskn5.nb <- knn2nb(ruskn5, row.names=rus$index, sym = FALSE)
> wrus.knb5 <- nb2listw(ruskn5.nb, glist=NULL, style="W", zero.policy=NULL)
>
> which I used before to deal with simple cross-section data.
>
> Re-reading the spdep manual, I changed to:
> ruskn5 <- knearneigh(coords, k=5, longlat = NULL)
> ruskn5.nb <- knn2nb(ruskn5, row.names=rus$index, sym = FALSE)
> wrus.knb5.mat <- nb2mat(ruskn5.nb, glist=NULL, style="W", zero.policy=NULL)
> wrus.knb5.lw <- mat2listw(wrus.knb5.mat, row.names =rus$index, style="M")
>
> Using that "wrus.knb5.lw" object created with mat2listw, and indeed
> explicitly indicating that object in both listw and listw2, I get a result
> to the spml function.
>
> Now I just have to figure out whether the result is meaningful.

The actual underlying problem may be a missing assignment in R/spfeml.R of 
can.sim2 to env. The local environment env is used in maximum likelihood 
methods in splm and spdep to find the log-det (Jacobian), and should be 
set in the calling function, see ?do_ldet in spdep. The first error 
message was probably the important one. I have added the splm maintainer 
to the CCs.

Because the log likelihood function is called often, it is much more 
efficient to put all the objects needed by the function in an environment, 
and only pass a pointer to the environment to the function, than to pass 
the objects separately at each call; this way of handling optimisation 
function efficiency was recommended among others by Doug Bates.

Roger

>
> Thank you very much Rafael,
>
> best regards,
>
> Patrick
>
>
>> Yes,
>> in that case the error I get is:
>>
>> Error in match(x, table, nomatch = 0L) : object 'wrus.knb5' not found
>>
>> Thus, no matter I specify this for listw or listw2, in both cases R cannot
>> find the object.
>>
>> Which I suppose answer my first question: it is not the "listw" function
>> which is not found, but the object I use as argument of that function,
>> right?
>>
>> But now I'm still unclear about what to do.
>>
>> Thanks anyway
>>
>>> Hi Patrick
>>>
>>> Have you tried what I suggested?
>>>
>>> spml(formula = fm, data = rusp9508.subsvars, index ="index", listw =
>>> wrus.knb5,
>>> listw2 = wrus.knb5, model = "within", lag = TRUE, spatial.error="b")
>>>
>>> Rafael
>>>
>>> On 09.10.2012, at 11:25, Patrick Meyfroidt wrote:
>>>
>>>> Hi Rafael,
>>>>
>>>> thank you very much for the answer.
>>>>
>>>> I read the page you referred to, and it was indeed very interesting,
>>>> but
>>>> I
>>>> haven't yet figured out how to use that in my problem.
>>>>
>>>> When I do:
>>>>  ls( envir = .GlobalEnv )
>>>> I see that the object I use for listw, which is "wrus.knb5", is indeed
>>>> located in .GlobalEnv
>>>>
>>>> I tried - not sure it was necessary or useful - to force spml to be
>>>> enclosed in .GlobalEnv by doing:
>>>>  environment( spml ) = .GlobalEnv
>>>> But still the object listw is not found.
>>>>
>>>> I also tried to specify where to look for wrus.knb5, by using:
>>>>  spml( ... listw = .GlobalEnv:::wrus.knb5 ...)
>>>> but, as can be expected, it does not work because .GlobalEnv is not a
>>>> package.
>>>>
>>>> So, two questions:
>>>> - What is R looking in the wrong place? Is it the function listw, or
>>>> the
>>>> object wrus.knb5 that I use as input for listw ?
>>>> - Based on that, how can I help R to look at the right place?
>>>>
>>>> Thanks,
>>>>
>>>> Patrick
>>>>
>>>>
>>>>> Hi Patrick
>>>>>
>>>>> I can not judge whether it makes any sense, but if you want listw and
>>>>> listw2 to be the same, try:
>>>>>
>>>>> spml(formula = fm, data = rusp9508.subsvars, index ="index", listw =
>>>>> wrus.knb5, listw2 = wrus.knb5, model = "within", lag = TRUE,
>>>>> spatial.error
>>>>> =
>>>>> "b")
>>>>>
>>>>> Otherwise, I guess, R is looking for listw in the wrong place. Look at
>>>>> http://obeautifulcode.com/R/How-R-Searches-And-Finds-Stuff/
>>>>> for explanations.
>>>>>
>>>>> HTH,
>>>>> Rafael
>>>>>
>>>>> On 08.10.2012, at 13:47, Patrick Meyfroidt wrote:
>>>>>
>>>>>> OK, I may have some update, sorry for the disturbance.
>>>>>>
>>>>>> By explicitly indicating that listw2 is the same as listw, like that:
>>>>>>
>>>>>> spml(formula = fm, data = rusp9508.subsvars, index ="index", listw =
>>>>>> wrus.knb5, listw2 = listw, model = "within", lag = TRUE,
>>>>>> spatial.error
>>>>>> =
>>>>>> "b")
>>>>>>
>>>>>> I now get the following error message:
>>>>>>
>>>>>> Error in match(x, table, nomatch = 0L) : object 'listw' not found
>>>>>>
>>>>>> So apparently the listw object cannot be read.
>>>>>>
>>>>>> Is it because the weights were calculated once for each country, but
>>>>>> the
>>>>>> dataset is a panel one?
>>>>>>
>>>>>> Does someone have a clue?
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Patrick Meyfroidt
>>>>>>
>>>>>>
>>>>>>> Dear list,
>>>>>>>
>>>>>>> I have spatial panel data (regions over years), that I subsetted to
>>>>>>> make
>>>>>>> sure that the panel is balanced.
>>>>>>>
>>>>>>> Using spdep, I calculated weights using nearest neighbors on these
>>>>>>> regions, with the following code:
>>>>>>>
>>>>>>> --
>>>>>>> RUSCRS <- CRS("+proj=lcc +lat_1=15 +lat_2=65 +lat_0=30 +lon_0=95
>>>>>>> +x_0=0
>>>>>>> +y_0=0 +units=m +datum=WGS84")
>>>>>>> rus <- readShapeSpatial([filename], IDvar = "index", proj4string =
>>>>>>> RUSCRS)
>>>>>>> coords <- coordinates(rus)
>>>>>>> ruskn5 <- knearneigh(coords, k=5, longlat = NULL)
>>>>>>> ruskn5.nb <- knn2nb(ruskn5, row.names=rus$index, sym = FALSE)
>>>>>>> wrus.knb5 <- nb2listw(ruskn5.nb, glist=NULL, style="W",
>>>>>>> zero.policy=NULL)
>>>>>>> --
>>>>>>>
>>>>>>> Everything looks fine, i.e. I can print out the results at each
>>>>>>> step,
>>>>>>> and
>>>>>>> each unit apparently has indeed 5 neighbors with 0.2 weight for
>>>>>>> each.
>>>>>>>
>>>>>>> Then I use splm package to perform a spatial panel regression, using
>>>>>>> the
>>>>>>> code:
>>>>>>>
>>>>>>> --
>>>>>>> fm <- rnabd ~ rlex + pbpl
>>>>>>> spml(formula = fm, data = rusp9508.subsvars, index ="index", listw =
>>>>>>> wrus.knb5, model = "within", lag = TRUE, spatial.error = "b")
>>>>>>> --
>>>>>>>
>>>>>>> But I end up with the following error message:
>>>>>>>
>>>>>>> "Error in get("can.sim2", envir = env) : object 'can.sim2' not
>>>>>>> found"
>>>>>>>
>>>>>>>
>>>>>>> Can someone explain what is this "can.sim2" object and possible
>>>>>>> reasons
>>>>>>> for the error message?
>>>>>>>
>>>>>>> Thank you very much for any input,
>>>>>>>
>>>>>>> Patrick Meyfroidt
>>>>>>> Université catholique de Louvain - Belgium
>>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> R-sig-Geo mailing list
>>>>>> R-sig-Geo at r-project.org
>>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Rafael Wüest
>>>>> Swiss Federal Research Institute WSL
>>>>> Zürcherstrasse 111
>>>>> 8903 Birmensdorf
>>>>> Switzerland
>>>>>
>>>>> +41 44 7392126
>>>>> rafael.wueest at wsl.ch
>>>>> http://www.wsl.ch/info/mitarbeitende/wueest/index_EN
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>
>>>
>>>
>>> --
>>> Rafael Wüest
>>> Swiss Federal Research Institute WSL
>>> Zürcherstrasse 111
>>> 8903 Birmensdorf
>>> Switzerland
>>>
>>> +41 44 7392126
>>> rafael.wueest at wsl.ch
>>> http://www.wsl.ch/info/mitarbeitende/wueest/index_EN
>>>
>>>
>>>
>>>
>>
>>
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo

-- 
Roger Bivand
Department of Economics, NHH Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no