- This topic has 6 replies, 3 voices, and was last updated 8 years, 8 months ago by Anonymous.
29th March 2012 at 8:55 am #2530
I have completed an analysis in which I have performed multiple linear regression for two dependent variables, Y1 (rate of use of a service) and Y2 (variety of use of the service). These variables are driven by a set of variables X1 (perceived social value), X2 (perceived emotional value), X3 (perceived performance/quality value), X4 (perceived value for money), and other variables.
I did regressions separately for Y1 and Y2. A reviewer of my paper has pointed out that since Y1 and Y2 are determined simultaneusly by X1, X2, X3, X4,… I should have performed Two Stage Least Squares (2SLS) regression.
Can someone point me to a source which explains, in simple terms, how to perform 2SLS in SPSS?30th March 2012 at 2:17 pm #2536AnonymousInactive
depending on which version of SPSS you have, i believe the newer ones have an option under the ‘regression’ menu. i have SPSS 17.0 so if you go Analyze —> Regression at the very end you should be able to see an option for “2SLS”
the coefficients are interpreted the same as in Path Analysis where you create (pseudo)-causal diagrams..3rd April 2012 at 4:35 am #2535
Thanks a ton, Oscar.
I do have the option to do 2SLS in SPSS. Is there a reference which explains what explanatory and instrumental variable are, and how 2SLS is used for situations in which two dependent variables are simultaneously determined by some explanatory variables? I see that 2SLS regression still calls for selecting one dependent variable at a time.3rd April 2012 at 8:27 am #2534AnonymousInactive
well, most of the people who work in econometrics do a lot of 2SLS models so i guess i just know this stuff out from econ courses i took? basically an instrumental variable is anything that predicts your dependent variable indirectly. for example, say you have predictors A, B and C. if A, B and C predict X and X predicts Y but A,B and C do not necessarily predict Y, then A, B and C are instrumentals, X is your independent variable and Y is your dependent variable.
i think it’s important to keep in mind that 2SLS is still what is referred to as a “first generation” multivariate technique. it is, after all, still a regression and as such you only get one dependent variable of interest, unless you can somehow turn one of your Y’s into another predictor. to be honest i’m a little bit surprised that someone would’ve asked you to perform 2SLS since it’s kind of an ‘old school’ technique. you can always have a look at Path Analysis. you should be able to analyze a dataset like yours through Path Analysis and it should take care of the endogeneity of the variables. if you’re familair with SPSS, it has an extension called AMOS which does structural equation modeling and path analysis. that would allow you to have two dependent variables which are being estimated simultaneously by your predictors…
(now this is not to say that you can’t do this through 2SLS. there is such a thing as multivariate 2SLS which is what i guess the person who asked you to perform is expecting for you to do, but i dont think you can do multivariate 2SLS in SPSS… you can do it in R though through the tsls() function in the sem package)4th April 2012 at 3:52 am #2533
Once again, thanks a ton. I’m much clearer now.
Ashish24th April 2012 at 2:22 pm #2532Muir HoustonMember14th May 2012 at 1:06 pm #2531
First of all many thanks. I couldn’t reply earlier as I was busy with my “day” job.
I haven’t been able to understand this completely, though. If it’s possible, can you kindly explain this a bit more: (i) what were the instrumentals, independent variable and dependent variable? (ii) in what steps did you solve the two equations?
- The forum ‘Default Forum’ is closed to new topics and replies.