[R-sig-Geo] Question about LM test for residual autocorrelation in R

[Hodgess, Erin]

Hi Dongwoo:

I tried the following example:

> erin1 <-     summary(COL.mix.eig, correlation=TRUE, Nagelkerke=TRUE)
> names(erin1)
 [1] "type"            "rho"             "coefficients"    "rest.se"        
 [5] "LL"              "s2"              "SSE"             "parameters"     
 [9] "logLik_lm.model" "AIC_lm.model"    "method"          "call"           
[13] "residuals"       "opt"             "tarX"            "tary"           
[17] "y"               "X"               "fitted.values"   "se.fit"         
[21] "similar"         "ase"             "rho.se"          "LMtest"         
[25] "resvar"          "zero.policy"     "aliased"         "listw_style"    
[29] "interval"        "fdHess"          "optimHess"       "insert"         
[33] "trs"             "LLNullLlm"       "timings"         "f_calls"        
[37] "hf_calls"        "intern_classic"  "coeftitle"       "Coef"           
[41] "NK"              "Wald1"           "correlation"     "correltext"     
[45] "LR1"            
> erin1$LMtest
          [,1]
[1,] 0.2891926
> 

and it does indeed have the LMtest result.

Or were you looking for the formula, please?

Thanks,
Erin

________________________________________
From: r-sig-geo-bounces at r-project.org [r-sig-geo-bounces at r-project.org] on behalf of Dongwoo Kang [dwkang1982 at gmail.com]
Sent: Tuesday, July 09, 2013 3:47 PM
To: r-sig-geo at r-project.org
Subject: [R-sig-Geo] Question about LM test for residual autocorrelation in R

Dear list,

Hello, I am Dongwoo Kang. I am studying Spatial Econometric modeling.
I've faced one question while using *spdep* package in R.
 I want to ask your help for my qeustion.

While estimating my empirical models,
I want to test whether residuals of my spatial regression models (SEM, SAR,
SARAR, SDM estimated by maximum likelihood) still have spatial
autocorrelation pattern.

I think I have two options,
1) Moran's I test using *"moran.mc"* function in R,
2) Lagrange multiplier diagnostics with LMerr option using
*"lm.LMtests"* function
in R.

But I also find that for SAR, SDM, *"summary.sarlm"* function returns "LM
test for residual autocorrelation" by default.
However, this LM test is not given for SER and SARAR.

At first, I thought that "Lagrange multiplier diagnostics" and "LM test for
residual autocorrelation" in "*summary.sarlm*" function are same tests.
But in my empirical results, they give me different statistics (please see
below example).

-----< example
>---------------------------------------------------------------------------
> summary.sarlm(sar2, Nagelkerke=T)
...
Log likelihood: -3533.378 for lag model
ML residual variance (sigma squared): 224.88, (sigma: 14.996)
Nagelkerke pseudo-R-squared: 0.76166
Number of observations: 853
Number of parameters estimated: 29
AIC: 7124.8, (AIC for lm: 7393.3)
LM test for residual autocorrelation
test value: 6.8391, p-value: 0.0089184

>
> lm.LMtests(sar2$residuals, listw=w100.listw, test=c("LMerr"))

Lagrange multiplier diagnostics for spatial dependence

data:
residuals: sar2$residuals
weights: w100.listw

LMErr = 3.7108, df = 1, p-value = 0.05406
-------------------------------------------------------------------------------------------------------

I try to find formulation of "LM test for residual autocorrelation" given
by *"summary.sarlm"* function but I couldn't.

Would you tell me where I can get some documents or explanations about "LM
test for residual autocorrelation" given by *"summary.sarlm"*?

I also want to know why "LM test for residual autocorrelation" is not
provided in SER and SARAR models.

Thank you very much for your time.

Best regards,

Dongwoo Kang

        [[alternative HTML version deleted]]

_______________________________________________
R-sig-Geo mailing list
R-sig-Geo at r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-geo