TSLS: R^2 extraction and autocorrelation and heterokedasticity tests

On Thu, 14 May 2009, spencerg wrote:

What code did you use to produce "output"?

He probably has used tsls() from "sem". ivreg() in "AER" also provides
two-stage least squares but along with a few more methods, such as 
terms(), model.matrix() etc.

This will be sufficient to call bptest() etc. from "lmtest" without 
error. However, this might be misleading. Internally, many "lmtest" 
functions re-fit the linear model and it depends on the test function 
and how the ivreg model was fitted whether what happens. Some 
functions internally re-fit the right model, others might re-fit the 
usual OLS model. I have never thought about using the "lmtest" tests 
with "ivreg" objects, I'll try to incorporate that in future updates.

In the meantime, I recommend that you re-fit the second stage of the 
2SLS "by hand" and then call the lmtest functionality. That way you 
can be sure that the second stage model is really used. In pseudo-code:
  fm_2sls <- ivreg(y ~ x1 + x2 | z1 + z2 + z3, data = mydata)
  fm_aux <- lm(model.response(model.frame(fm_2sls)) ~
    model.matrix(fm_2sls)[,-1])
  bptest(fm_aux)

etc.
Z

Have you tried "str(output)" and "names(output)"?


If this is not enough to help you answer your question, you might try 
rephrasing your question using commented, minimal, self-contained, 
reproducible code, as suggested in the posting guide 
"http://www.R-project.org/posting-guide.html".


Spencer Graves


jos? maria Rodriguez wrote:

Hi,

I'm actually I?m performing a TSLS linear multiple regression on 
annually
data which go from 1971 to 1997. After performing the TSLS 
regression, I
tried to extract the R squared value using ?output$r.squared? 
function and
to perform autocorrelation (Durbin Watson and Breush Godfrey) and
heterokedasticity tests (Breush-pagan and Goldfeld Quandt)  but I have
errors messages. More specifically, this is function that I write to 
R and
below its response :
for R^2 :

output$r.squared

NULL
for heterokedasticity tests :

bptest(reg1)

Error in terms.default(formula) : no terms component
and for autocorrelation test, when I try :
durbin.watson(reg1$residuals, max.lag=10)
 [1] 1.509 2.520 2.247 2.001 1.743 1.092 1.392 1.439 1.468 1.035
this give me only the durbin watson value and not the probabilities
(p-value)
When performing these tests on lm object I have no problem. So my 
question
is how to extract R^2 from a tsls regression (object) and how to 
perform
autocorrelation and heterokedasticity tests on tsls regression. I 
looked at
the sem package but I found no answer to my questions. So please is 
there
any person who can help me.

Think you in advance

    [[alternative HTML version deleted]]

  
------------------------------------------------------------------------