Brian's solution:
The Grunfeld dataset that accompanies Gujarati's text is actually a panel
of 4 firms over 20 years. In the example on p. 641 he runs a pooled
regression, stacking the data for the four firms on top of each other.
To get Stata to compute the Durbin-Watson statistic, we need to trick it
into thinking this is time series data.
Here is how I did that:
insheet using guj.txt // From the CD-ROM accompanying the book
drop v5 v6
rename v2 I
rename v3 lagF
rename v4 lagC
gen year = real(substr(v1, length(v1)-3, .))
gen str company = substr(v1, 2, 2)
sort company year
gen t = _n
tsset t
regress I lagF lagC
estat dwatson
Durbin-Watson d-statistic( 3, 80) = .3097946
fails to take into account the gaps in the data. You should not be taking the differences between different firms' errors! The enclosed does not match Gujarati exactly, but it comes close:
webuse grunfeld,clear
keep if company<4 | company == 8
set obs 81
replace company = 2 in 81
replace year = 1934 in 81
set obs 82
replace company = 3 in 82
replace year = 1934 in 82
set obs 83
replace company = 8 in 83
replace year = 1934 in 83
sort company year
gen t=_n
tsset t
l
regress invest mvalue kstock
dwstat
Kit Baum, Boston College Economics
http://ideas.repec.org/e/pba1.html
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
