I need help with an iv regression in stat

[goofball]

I'm getting super funky results. Just making sure I have the damn syntax/order right would be awesome. I'll coach you at poker in return!

[lippy]

I can't help you, but happy birthday. You're a good poster, so cheers.

[edtost]

ivreg depvar [varlist1] (varlist2 = varlist_iv) [if] [in] [weight] [,options]

http://www.stata.com/help.cgi?ivreg

[ QUOTE ]
ivreg fits a linear regression model using instrumental variables (or two-stage least squares) of depvar on varlist1 and varlist2, using varlist_iv (along with varlist1) as instruments for varlist2.

In the language of two-stage least squares, varlist1 and varlist_iv are the exogenous variables, and varlist2 are the endogenous variables.

[/ QUOTE ]

my understanding is that an example like

ivreg y1 x1 x2 (y2 = z1 z2 z3) x3

is equivilant to

regress y2 z1 z2 z3
predict yhat
regress y1 x1 x2 yhat x3

in other words, that the order of (varlist2 = varlist_iv) in [varlist1] doesn't matter, but my knowledge of stata isn't all that great.

[Jdanz]

[ QUOTE ]
ivreg depvar [varlist1] (varlist2 = varlist_iv) [if] [in] [weight] [,options]

http://www.stata.com/help.cgi?ivreg

[ QUOTE ]
ivreg fits a linear regression model using instrumental variables (or two-stage least squares) of depvar on varlist1 and varlist2, using varlist_iv (along with varlist1) as instruments for varlist2.

In the language of two-stage least squares, varlist1 and varlist_iv are the exogenous variables, and varlist2 are the endogenous variables.

[/ QUOTE ]

my understanding is that an example like

ivreg y1 x1 x2 (y2 = z1 z2 z3) x3

is equivilant to

regress y2 z1 z2 z3
predict yhat
regress y1 x1 x2 yhat x3

in other words, that the order of (varlist2 = varlist_iv) in [varlist1] doesn't matter, but my knowledge of stata isn't all that great.

[/ QUOTE ]

I'm not completely sure this is correct, I'm running some tonight so I'll try to let you know.

edit: I know for a fact the order in which you but your instruments inside the parens doesn't matter at all.

Also not sure how familiar you are with IV, but the results tend to simply be super funky unless you have like perfect instruments or a ridiculously large sample.

[edtost]

also, ivreg as opposed to reg predict reg should give the same coefficients, but std dev's and other statistics will be completely different.

[Jdanz]

depends on the sample, ivreg will automatically exclude subjects for which there is incomplete data, whereas reg reg will use all the data it can for each individual regression.

I spend a ridiculous number of hours confused by this.