I am estimating a "quasi-dated" water crop production function. My dataset
contains variables describing the average amount of water applied per
irrigation, the length of the irrigation season in days (IrrS), the average
number of days between irrigation events (IrrD) and dummy variables for 4
climatic zones (with each respondent belonging to one climate zone only).
The start of the crop's irrigation season is determined by the end of the
local wet season. The length of the irrigation season is also determined by
the length of the dry season. It is logical to assume that IrrS is
endogenous to local climate conditions. IrrD could also be endogenous.
I have modelled a 2SLS using ivreg2 defining Clim2 Clim3 Clim4 as
instruments for IrrS and lnIrrS. It's my first time estimating such models.
The Durbin-Wu-Hausman chi-sq test is signficant and the Sargan statistic is
insignificant. I have done the same with (IrrD lnIrrD = Clim2 Clim3 Clim4)
and get a significant DWH and insignificant Sargan result. The results
suggest I should treat all 4 irrigation timing variables as endogenous, but
I cant do this because of underidentification. Can anyone suggest a way to
get around this problem? I can't get anymore data and I don't want to have
to drop IrrD lnIrrD from the model as these are key variables. I am also
wondering if there is any problem with using dummy variables as instruments?
I haven't found any literature on this and just want to confirm what I am
doing is OK.
Many thanks
*
* 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/
