[R-sig-Geo] logLik value of the spml spatial panel model

[Jeremie Juste]

Hello,

Indeed the Loglik was not returned here by the function sarpanelerror
here. I have provided a patch (attached). It's an ugly one but better
than nothing. So instead of running spml you could run test..spml after
sourcing the file to obtain the fespaterr$logLik. 

fespaterr <- test..spml(fm, data = Produc, listw = mat2listw(usaww),
                   model="within", spatial.error="b", Hess = FALSE)


The only modification I've made is in the function test..(sperrorlm)
where I've added ll=LL in the list returned.


Hope it helps,

Jeremie




-------------- next part --------------
A non-text attachment was scrubbed...
Name: test.R
Type: application/octet-stream
Size: 33277 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-geo/attachments/20181025/a28b6639/attachment.obj>

-------------- next part --------------


> Dear Jeremie:
>
> Thank you for the response. I actually only copied the scripts from
> the spml help page, reproduced here:
>
> fm <- log(gsp) ~ log(pcap) + log(pc) + log(emp) + unemp
> ## the two standard specifications (SEM and SAR) one with FE
> ## and the other with RE:
> ## fixed effects panel with spatial errors
> fespaterr <- spml(fm, data = Produc, listw = mat2listw(usaww),
>                    model="within", spatial.error="b", Hess = FALSE) #
> doesn't matter if I set Hess = T or F here, and doesn't matter if I
> set spatial.error = "b" or "kkp"
>
> And when I want to check the logLik value:
>
> fespaterr$logLik
>
> I got the "NULL" value
>
> I checked this with some other data sets, and found that if I am using
> the spatial lag specification, I usually got a value for the logLik
> parameter. If, however, I use the spatial error specification, I
> usually got the "NULL" value for the logLik parameter. I was not able
> to get to the bottom of the issue by digging into the codes a bit. Any
> leads would be greatly appreciated.
>
> Best,
>
> Danlin
>
> On 2018/10/24 15:30, Jeremie Juste wrote:
>> Danlin Yu <yud using mail.montclair.edu> writes:
>>
>>
>> Hello,
>>
>> Could you post the exact command you have used with all the relevant
>> options? If you dug into the code then you've noticed quite a lot of
>> nested control structures so a clear idea of your commands might be
>> useful here.
>>
>> for instance
>>
>>      fm <- logc ~ logp+ logpn + logy
>>      sarpool <-  suppressWarnings(splm::spml(fm,cigar,
>>                                              listw=lwcig,
>>                                              model="pooling",
>>      spatial.error="none",lag=TRUE,index=c("region","time")))
>>
>> Will produce the logLik
>>      sarpool$logLik
>>
>>
>>  From what I've understood at some point the function <nlminb> is always
>> called so a likelihood value is always produced although not always
>> returned.
>>
>>
>> Best regards,
>>
>> Jeremie
>>
>>
>>
>>> Dear List:
>>>
>>> In a recent attempt to compare spatial panel models using splm
>>> package, I intend to use the logLik value that was produced by the
>>> spml() routine. When I ran the model, I realize that the logLik value
>>> is not always produced, especially when the spatial.error is set to
>>> "b", a NULL value is returned. I know I must have missed something
>>> here, but can anyone point out why the logLik value is not returned
>>> when the spatial.error value is set to "b" (a spatial panel lag model
>>> always return a logLik value). I tried a bit to dig into the codes,
>>> but didn't find much that can answer my question.
>>>
>>> Thank you.
>>>
>>> Best,
>>>
>>> Danlin