[Openspace] Moran's I, lm lag and lm error

[Ines Koller]

Hello,

Sorry for my unreadable e-mail (the formatting of my mail went wrong).
Hopefully my next mail is readable:
 
I have three questions regarding the Moran's I, the lm lag and the lm error
tests.
My model consists of six independent variables, dummies for the months and
years (minus one period) and a serially lagged dependent variable (panel
data).
I have done the Moran's I, the lm lag and the lm error tests for two models.
First I have done the Moran's I, the lm lag and the lm error test with all
my independent variables including the serially lagged dependent variable
and dummies for all cross-sections (minus one cross-section). Then I have
done the same tests for a model without the dummies for cross-sections.
Below are the results of the tests:
 
1. Model without dummies for cross-sections
Test | Value | Probability
Moran's I (error) | 0.0062 | 0.0016  
Lagrange Multiplier (lag) | 45.3780 | 1.6245e-11
Robust LM (lag) | 43.3461 | 4.5864e-11
Lagrange Multiplier (error) | 5.2439 | 0.0220
Robust LM (error) | 3.2121 | 0.0731

2. Model with dummies for cross-sections
Test | Value | Probability
Moran's I (error) | -2.2496e-13 | 0.9954
Lagrange Multiplier (lag) | 26.5398 | 2.5816e-07
Robust LM (lag) | 26.8671 | 2.1794e-07
Lagrange Multiplier (error) | 1.3826e-22 | 1.0000
Robust LM (error) | 0.3273 | 0.5673

My questions are:
1. Why are the results for the two models so different? Why have the dummies
(= fixed effects) such a big influence on the test statistics?
2. For the model with dummies the Moran's I shows no correlation, but the lm
lag test shows high spatial correlation. How is this possible?
3. The probability of the lm error test is higher than the probability of
the robust lm error test (same for lm lag). Shouldn't be the probability of
the robust lm error higher than the probability of the lm error test?
 

Thanks
Ines

________________________________
Ines Koller
Technische Universitaet Muenchen
TUM Business School
Environmental Economics & Agricultural Policy Group 
Alte Akademie 14 
85354 Freising, Germany  

phone: +49 8161 71-3576 
fax: +49 8161 71-3408 
email: ines.koller at wzw.tum.de