7 Aug
2007
7 Aug
'07
3:24 p.m.
Hi,
I've run across a problem in the two stage least squares model in
Zelig, which may be due to my code or to a bug. I want to use 2SLS
for an instrumental variable regression. However, whenever I attempt
to do so, I get an error:
Error in checkNrReq(modelNumEqn, userNumEqn, modelPar) :
The parameter "mu" requires between 2
and Inf equation(s). You have provided 1 See model doc.
for more details
I've provided some simulated data and the code that produces the
error below. Note that I used the same syntax as published in the
Zelig manual. Any thoughts on where I/zelig might be going wrong?
Doing the procedure with lm() seems to recover the correct parameters.
Many thanks,
Andy
nn<-1000 #sample size
V<-rnorm(nn) #confounding omitted variable
W<-rnorm(nn) #instrument
X<-rnorm(nn) #exogenous regressor
Z<-rep(NA, nn) #instrumental var. holder
for (i in 1:nn){
+ Z[i]<-rnorm(1, mean=V[i]+W[i]) #instrumental var as a
function of
instrument and omitted variable.
+ }
Y<-rep(NA, nn)
for (i in 1:nn){
+ Y[i]<-rnorm(1, mean=V[i]+0.25*Z[i]) #outcome as func of
omitted
var. and Z.
+ }
data<-as.data.frame(cbind(X,W,Z,Y))
names(data)<-c("X","W", "Z", "Y")
### using the 2sls syntax in section 12.48 of the zelig manual
fml <- list ("mu" = Y ~ X + Z,
+ "inst" = Z ~ W + X)
z.out <- zelig(formula = fml, model =
"twosls", data=data)
Error in checkNrReq(modelNumEqn, userNumEqn,
modelPar) :
The parameter "mu" requires between 2
and Inf equation(s). You have provided 1 See model doc.
for more details
### 2sls using lm()
lm.out1<-lm(Z~W+X, data=data)
lm.out2<-lm(Y ~ X + lm.out1$fitted, data=data)
summary(lm.out1)
Call:
lm(formula = Z ~ W + X, data = data)
Residuals:
Min 1Q Median 3Q Max
-3.7840 -0.9809 0.0200 0.9883 4.2447
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.00241 0.04329 -0.06 0.96
W 0.99100 0.04236 23.39 <2e-16 ***
X 0.02788 0.04414 0.63 0.53
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 1.37 on 997 degrees of freedom
Multiple R-Squared: 0.355, Adjusted R-squared: 0.353
F-statistic: 274 on 2 and 997 DF, p-value: <2e-16
summary(lm.out2)
Call:
lm(formula = Y ~ X + lm.out1$fitted, data = data)
Residuals:
Min 1Q Median 3Q Max
-4.975 -1.122 -0.055 1.105 5.361
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.00901 0.05170 -0.17 0.86164
X -0.04195 0.05274 -0.80 0.42653
lm.out1$fitted 0.18591 0.05104 3.64 0.00028 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 1.63 on 997 degrees of freedom
Multiple R-Squared: 0.0135, Adjusted R-squared: 0.0116
F-statistic: 6.85 on 2 and 997 DF, p-value: 0.00111
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Andy Harris
Ph.D. Candidate
Department of Government
Harvard University
8 Aug
8 Aug
3:53 p.m.
Hi,
The two-stage least squares procedure we have in Zelig is based on
the systemfit package which takes a system of more than one equation
along with instrumental variables as inputs. What you have in mind is
slightly different from that procedure is identical to tsls() in sem
package. Unfortunately, we don't have that included in Zelig yet.
Maybe, you would like to contribute?
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics Fax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
---------------------------------------------------------
On Aug 7, 2007, at 3:24 PM, Andy Harris wrote:
Hi,
I've run across a problem in the two stage least squares model in
Zelig, which may be due to my code or to a bug. I want to use 2SLS
for an instrumental variable regression. However, whenever I
attempt to do so, I get an error:
Error in checkNrReq(modelNumEqn, userNumEqn, modelPar) :
The parameter "mu" requires between 2
and Inf equation(s). You have provided 1 See model doc.
for more details
I've provided some simulated data and the code that produces the
error below. Note that I used the same syntax as published in the
Zelig manual. Any thoughts on where I/zelig might be going wrong?
Doing the procedure with lm() seems to recover the correct parameters.
Many thanks,
Andy
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)lists.gking.harvard.edu
[un]subscribe Options: http://lists.gking.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
nn<-1000 #sample size
V<-rnorm(nn) #confounding omitted variable
W<-rnorm(nn) #instrument
X<-rnorm(nn) #exogenous regressor
Z<-rep(NA, nn) #instrumental var. holder
for (i in 1:nn){
+ Z[i]<-rnorm(1, mean=V[i]+W[i]) #instrumental var as a
function of
instrument and omitted variable.
+ }
Y<-rep(NA, nn)
for (i in 1:nn){
+ Y[i]<-rnorm(1, mean=V[i]+0.25*Z[i]) #outcome as func of
omitted
var. and Z.
+ }
data<-as.data.frame(cbind(X,W,Z,Y))
names(data)<-c("X","W", "Z", "Y")
### using the 2sls syntax in section 12.48 of the zelig manual
fml <- list ("mu" = Y ~ X + Z,
+ "inst" = Z ~ W + X)
z.out <- zelig(formula = fml, model =
"twosls", data=data)
Error in checkNrReq(modelNumEqn, userNumEqn,
modelPar) :
The parameter "mu" requires between 2
and Inf equation(s). You have provided 1 See model doc.
for more details
### 2sls using lm()
lm.out1<-lm(Z~W+X, data=data)
lm.out2<-lm(Y ~ X + lm.out1$fitted, data=data)
summary(lm.out1)
Call:
lm(formula = Z ~ W + X, data = data)
Residuals:
Min 1Q Median 3Q Max
-3.7840 -0.9809 0.0200 0.9883 4.2447
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.00241 0.04329 -0.06 0.96
W 0.99100 0.04236 23.39 <2e-16 ***
X 0.02788 0.04414 0.63 0.53
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 1.37 on 997 degrees of freedom
Multiple R-Squared: 0.355, Adjusted R-squared: 0.353
F-statistic: 274 on 2 and 997 DF, p-value: <2e-16
summary(lm.out2)
Call:
lm(formula = Y ~ X + lm.out1$fitted, data = data)
Residuals:
Min 1Q Median 3Q Max
-4.975 -1.122 -0.055 1.105 5.361
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.00901 0.05170 -0.17 0.86164
X -0.04195 0.05274 -0.80 0.42653
lm.out1$fitted 0.18591 0.05104 3.64 0.00028 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 1.63 on 997 degrees of freedom
Multiple R-Squared: 0.0135, Adjusted R-squared: 0.0116
F-statistic: 6.85 on 2 and 997 DF, p-value: 0.00111
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Andy Harris
Ph.D. Candidate
Department of Government
Harvard University
9 Aug
9 Aug
9:49 p.m.
After further investigations, it turns out that Andy's problem can
also be handled in Zelig with a different syntax. :) Suppose your
model is:
Y = X + T + e1
T = X + Z + e2
then, the formula would look like:
list("mu1" = Y ~ X + T, "mu2" = T ~ X + Z, "inst" = ~ X +
Z)
The key is that you need to include both X and Z as instruments. The
current docs are misleading and so we will fix the docs in the next
version.
Best,
Kosuke
On Aug 8, 2007, at 3:53 PM, Kosuke Imai wrote:
Hi,
The two-stage least squares procedure we have in Zelig is based
on the systemfit package which takes a system of more than one
equation along with instrumental variables as inputs. What you have
in mind is slightly different from that procedure is identical to
tsls() in sem package. Unfortunately, we don't have that included
in Zelig yet. Maybe, you would like to contribute?
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics Fax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
---------------------------------------------------------
On Aug 7, 2007, at 3:24 PM, Andy Harris wrote:
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)lists.gking.harvard.edu
[un]subscribe Options: http://lists.gking.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Hi,
I've run across a problem in the two stage least squares model in
Zelig, which may be due to my code or to a bug. I want to use 2SLS
for an instrumental variable regression. However, whenever I
attempt to do so, I get an error:
Error in checkNrReq(modelNumEqn, userNumEqn, modelPar) :
The parameter "mu" requires between 2
and Inf equation(s). You have provided 1 See model
doc. for more details
I've provided some simulated data and the code that produces the
error below. Note that I used the same syntax as published in the
Zelig manual. Any thoughts on where I/zelig might be going wrong?
Doing the procedure with lm() seems to recover the correct
parameters.
Many thanks,
Andy
nn<-1000 #sample size
V<-rnorm(nn) #confounding omitted variable
W<-rnorm(nn) #instrument
X<-rnorm(nn) #exogenous regressor
Z<-rep(NA, nn) #instrumental var. holder
for (i in 1:nn){
+ Z[i]<-rnorm(1, mean=V[i]+W[i]) #instrumental var as a
function
of instrument and omitted variable.
+ }
Y<-rep(NA, nn)
for (i in 1:nn){
+ Y[i]<-rnorm(1, mean=V[i]+0.25*Z[i]) #outcome as func of
omitted
var. and Z.
+ }
data<-as.data.frame(cbind(X,W,Z,Y))
names(data)<-c("X","W", "Z", "Y")
### using the 2sls syntax in section 12.48 of the zelig manual
fml <- list ("mu" = Y ~ X + Z,
+ "inst" = Z ~ W + X)
z.out <- zelig(formula = fml, model =
"twosls", data=data)
Error in checkNrReq(modelNumEqn, userNumEqn,
modelPar) :
The parameter "mu" requires between 2
and Inf equation(s). You have provided 1 See model
doc. for more details
### 2sls using lm()
lm.out1<-lm(Z~W+X, data=data)
lm.out2<-lm(Y ~ X + lm.out1$fitted, data=data)
summary(lm.out1)
Call:
lm(formula = Z ~ W + X, data = data)
Residuals:
Min 1Q Median 3Q Max
-3.7840 -0.9809 0.0200 0.9883 4.2447
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.00241 0.04329 -0.06 0.96
W 0.99100 0.04236 23.39 <2e-16 ***
X 0.02788 0.04414 0.63 0.53
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 1.37 on 997 degrees of freedom
Multiple R-Squared: 0.355, Adjusted R-squared: 0.353
F-statistic: 274 on 2 and 997 DF, p-value: <2e-16
summary(lm.out2)
Call:
lm(formula = Y ~ X + lm.out1$fitted, data = data)
Residuals:
Min 1Q Median 3Q Max
-4.975 -1.122 -0.055 1.105 5.361
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.00901 0.05170 -0.17 0.86164
X -0.04195 0.05274 -0.80 0.42653
lm.out1$fitted 0.18591 0.05104 3.64 0.00028 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 1.63 on 997 degrees of freedom
Multiple R-Squared: 0.0135, Adjusted R-squared: 0.0116
F-statistic: 6.85 on 2 and 997 DF, p-value: 0.00111
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Andy Harris
Ph.D. Candidate
Department of Government
Harvard University
6127
days inactive
6130
days old
2 comments
2 participants
participants (2)
-
Andy Harris -
Kosuke Imai