23 Jun
2005
23 Jun
'05
4:58 p.m.
Dear all,
I'm using Zelig for rare events logistic regression. In the past, I've
run it with succes on a Linux server. However, now I would try to get it
to work on my laptop (Win XP), so I can already do some analysis during
fieldwork periods.
I however receive the following error message : "Warning message:
when cond = TRUE, fn is coerced to NULL in: setx.default(z.out1, data =
test1, cond = TRUE) " (see below for more details)
I would like to know if someone could point out what it means? And above
all: can it be solved? Or will I have to keep working on the server?
The consequence of this error message is apparently that the intercept
estimate is too high, so something must go wrong in the rare events
correction I suppose?
Regards,
Tom
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
z.out1<-zelig(depressies~archeodist + aspect1 +
aspect2 + aspect3 +
B_1 + B_2 + B_3 + B_4 + B_5 +
+ B_6 + B_7 + G_fb + G_zbr + G_zle + gully + roadgully + helling, data =
test1 ,
+ model = "relogit" , tau = 71/22788 , bias.correct = TRUE)
x.out1<-setx(z.out1, data=test1, cond=TRUE)
Warning message:
when cond = TRUE, fn is coerced to NULL in: setx.default(z.out1, data =
test1, cond = TRUE)
traceback
function (x = NULL)
{
if (is.null(x) && (exists(".Traceback", env = .GlobalEnv)))
x <- get(".Traceback", env = .GlobalEnv)
if (is.null(x) || length(x) == 0)
cat(gettext("No traceback available"), "\n")
else {
n <- length(x)
for (i in 1:n) {
label <- paste(n - i + 1, ": ", sep = "")
if ((m <- length(x[[i]])) > 1)
label <- c(label, rep(substr(" ", 1,
nchar(label, type = "w")), m - 1))
cat(paste(label, x[[i]], sep = ""), sep = "\n")
}
}
invisible()
}
<environment: namespace:base>
traceback
--
Dr. Tom Vanwalleghem
Physical and Regional Geography
K.U.Leuven
Redingenstraat 16
B-3000 LEUVEN
BELGIUM
tel +32(0)16/32.64.14
fax +32(0)16/32.64.00
new email! tom.vanwalleghem(a)kuleuven.be
23 Jun
23 Jun
7:51 p.m.
Dear Tom,
This warning message does NOT affect your results. When you are making
*conditional* predictions, we are calculating quantities of interest for
each observation of the data (i.e., in-sample inference). Therefore, we
need to set fn = NULL within setx. This warning comes up if you forget
this, but setx() automatically assigns fn = NULL. For details, see
http://gking.harvard.edu/zelig/docs/_TT_setx_TT__Setting_E.html
Best,
Kosuke
P.S. When you use traceback(), you need to have "()" in order to get
meaningful messages.
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://www.princeton.edu/~kimai
---------------------------------------------------------
On Thu, 23 Jun 2005, Tom Vanwalleghem wrote:
Dear all,
I'm using Zelig for rare events logistic regression. In the past, I've
run it with succes on a Linux server. However, now I would try to get it
to work on my laptop (Win XP), so I can already do some analysis during
fieldwork periods.
I however receive the following error message : "Warning message:
when cond = TRUE, fn is coerced to NULL in: setx.default(z.out1, data =
test1, cond = TRUE) " (see below for more details)
I would like to know if someone could point out what it means? And above
all: can it be solved? Or will I have to keep working on the server?
The consequence of this error message is apparently that the intercept
estimate is too high, so something must go wrong in the rare events
correction I suppose?
Regards,
Tom
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
z.out1<-zelig(depressies~archeodist + aspect1
+ aspect2 + aspect3 +
B_1 + B_2 + B_3 + B_4 + B_5 +
+ B_6 + B_7 + G_fb + G_zbr + G_zle + gully + roadgully + helling, data =
test1 ,
+ model = "relogit" , tau = 71/22788 , bias.correct = TRUE)
x.out1<-setx(z.out1, data=test1, cond=TRUE)
Warning message:
when cond = TRUE, fn is coerced to NULL in: setx.default(z.out1, data =
test1, cond = TRUE)
traceback
function (x = NULL)
{
if (is.null(x) && (exists(".Traceback", env = .GlobalEnv)))
x <- get(".Traceback", env = .GlobalEnv)
if (is.null(x) || length(x) == 0)
cat(gettext("No traceback available"), "\n")
else {
n <- length(x)
for (i in 1:n) {
label <- paste(n - i + 1, ": ", sep = "")
if ((m <- length(x[[i]])) > 1)
label <- c(label, rep(substr(" ", 1,
nchar(label, type = "w")), m - 1))
cat(paste(label, x[[i]], sep = ""), sep = "\n")
}
}
invisible()
}
<environment: namespace:base>
traceback
24 Jun
24 Jun
3:38 a.m.
Dear Prof. Kosuke Imai,
Really appreciating your quick response and tips.
Any idea what could be causing the difference then between linux
server/windows pc (difference in intercept estimate of -3.73 vs +0.67) ?
Or is there something I'm overseeing?
I've attached all necessary information in case you would have time to
have a look.
Thanks in advance,
Tom
Kosuke Imai wrote:
Dear Tom,
This warning message does NOT affect your results. When you are making
*conditional* predictions, we are calculating quantities of interest for
each observation of the data (i.e., in-sample inference). Therefore, we
need to set fn = NULL within setx. This warning comes up if you forget
this, but setx() automatically assigns fn = NULL. For details, see
http://gking.harvard.edu/zelig/docs/_TT_setx_TT__Setting_E.html
Best,
Kosuke
P.S. When you use traceback(), you need to have "()" in order to get
meaningful messages.
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://www.princeton.edu/~kimai
---------------------------------------------------------
On Thu, 23 Jun 2005, Tom Vanwalleghem wrote:
--
Dr. ir. Tom Vanwalleghem
Physical and Regional Geography
K.U.Leuven
Redingenstraat 16
B-3000 LEUVEN
BELGIUM
tel +32(0)16/32.64.14
fax +32(0)16/32.64.00
new email! tom.vanwalleghem(a)kuleuven.be
Dear all,
I'm using Zelig for rare events logistic regression. In the past, I've
run it with succes on a Linux server. However, now I would try to get it
to work on my laptop (Win XP), so I can already do some analysis during
fieldwork periods.
I however receive the following error message : "Warning message:
when cond = TRUE, fn is coerced to NULL in: setx.default(z.out1, data =
test1, cond = TRUE) " (see below for more details)
I would like to know if someone could point out what it means? And above
all: can it be solved? Or will I have to keep working on the server?
The consequence of this error message is apparently that the intercept
estimate is too high, so something must go wrong in the rare events
correction I suppose?
Regards,
Tom
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
z.out1<-zelig(depressies~archeodist + aspect1
+ aspect2 + aspect3 +
B_1 + B_2 + B_3 + B_4 + B_5 +
+ B_6 + B_7 + G_fb + G_zbr + G_zle + gully + roadgully + helling, data =
test1 ,
+ model = "relogit" , tau = 71/22788 , bias.correct = TRUE)
x.out1<-setx(z.out1, data=test1, cond=TRUE)
Warning message:
when cond = TRUE, fn is coerced to NULL in: setx.default(z.out1, data =
test1, cond = TRUE)
traceback
function (x = NULL)
{
if (is.null(x) && (exists(".Traceback", env = .GlobalEnv)))
x <- get(".Traceback", env = .GlobalEnv)
if (is.null(x) || length(x) == 0)
cat(gettext("No traceback available"), "\n")
else {
n <- length(x)
for (i in 1:n) {
label <- paste(n - i + 1, ": ", sep = "")
if ((m <- length(x[[i]])) > 1)
label <- c(label, rep(substr(" ", 1,
nchar(label, type = "w")), m - 1))
cat(paste(label, x[[i]], sep = ""), sep = "\n")
}
}
invisible()
}
<environment: namespace:base>
traceback
z.out1<-zelig(depressies~archeodist + aspect1 +
aspect2 + aspect3 + B_1 + B_2 + B_3 + B_4 + B_5 +
+ B_6 + B_7 + G_fb + G_zbr +
G_zle + gully + roadgully + helling, data = test1 ,
+ model = "relogit" , tau = 71/22788 , bias.correct = TRUE)
summary(z.out1)
Call: zelig(formula = depressies ~ archeodist + aspect1 + aspect2 + aspect3 + B_1 +
B_2 + B_3 + B_4 + B_5 + B_6 + B_7 + G_fb + G_zbr + G_zle + gully + roadgully +
helling, model = "relogit", data = test1, tau = 71/22788, bias.correct =
TRUE)
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 0.6715 0.6289 1.0677 0.2866
archeodist -1.2674 0.5087 -2.4916 0.0133
aspect1 -0.3278 0.5352 -0.6124 0.5408
aspect2 0.8988 0.4045 2.2223 0.0271
aspect3 -0.0967 0.4801 -0.2014 0.8405
B_1 -0.2294 0.5677 -0.4041 0.6865
B_2 -15.0894 808.0271 -0.0187 0.9851
B_3 0.8899 0.4247 2.0955 0.0371
B_4 0.0234 0.5695 0.0412 0.9672
B_5 -0.4487 0.6823 -0.6577 0.5113
B_6 -1.1223 0.7798 -1.4392 0.1513
B_7 -15.8236 1552.4077 -0.0102 0.9919
G_fb 3.3175 1.4523 2.2843 0.0232
G_zbr -0.4735 0.3663 -1.2926 0.1973
G_zle -0.8981 0.6614 -1.3580 0.1756
gully -1.6922 0.8754 -1.9330 0.0543
roadgully -1.5486 0.9249 -1.6744 0.0953
helling -0.0756 0.0405 -1.8677 0.0629
Intercepts with prior correction for:
tau = 0.0031
-3.73
Degrees of Freedom: 279 Total (i.e. Null); 262 Residual
Null Deviance: 283
Residual Deviance: 220 AIC: 256
Rare events bias correction performed.
z.out1<-zelig(depressies~archeodist + aspect1 +
aspect2 + aspect3 + B_1 + B_2 + B_3 + B_4 + B_5 +
+ B_6 + B_7 + G_fb + G_zbr +
G_zle + gully + roadgully + helling, data = test1 ,
model = "relogit" , tau = 71/22788 , bias.correct = TRUE)+
summary(z.out1)
Call: zelig(formula = depressies ~ archeodist + aspect1 + aspect2 + aspect3 + B_1 +
B_2 + B_3 + B_4 + B_5 + B_6 + B_7 + G_fb + G_zbr + G_zle + gully + roadgully +
helling, model = "relogit", data = test1, tau = 71/22788, bias.correct =
TRUE)
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -3.733e+00 6.289e-01 -5.935e+00 9.294e-09
archeodist -1.267e+00 5.087e-01 -2.492e+00 1.334e-02
aspect1 -3.278e-01 5.352e-01 -6.124e-01 5.408e-01
aspect2 8.988e-01 4.045e-01 2.222e+00 2.711e-02
aspect3 -9.672e-02 4.801e-01 -2.014e-01 8.405e-01
B_1 -2.294e-01 5.677e-01 -4.041e-01 6.865e-01
B_2 -1.509e+01 8.080e+02 -1.867e-02 9.851e-01
B_3 8.899e-01 4.247e-01 2.095e+00 3.709e-02
B_4 2.343e-02 5.695e-01 4.115e-02 9.672e-01
B_5 -4.487e-01 6.823e-01 -6.577e-01 5.113e-01
B_6 -1.122e+00 7.798e-01 -1.439e+00 1.513e-01
B_7 -1.582e+01 1.552e+03 -1.019e-02 9.919e-01
G_fb 3.317e+00 1.452e+00 2.284e+00 2.315e-02
G_zbr -4.735e-01 3.663e-01 -1.293e+00 1.973e-01
G_zle -8.981e-01 6.614e-01 -1.358e+00 1.756e-01
gully -1.692e+00 8.754e-01 -1.933e+00 5.431e-02
roadgully -1.549e+00 9.249e-01 -1.674e+00 9.525e-02
helling -7.563e-02 4.050e-02 -1.868e+00 6.292e-02
Degrees of Freedom: 279 Total (i.e. Null); 262 Residual
Null Deviance: 283
Residual Deviance: 219.9 AIC: 255.9
Prior correction is made with tau = 0.00312
Rare events bias correction was made.
10:53 a.m.
Dear Tom,
Assuming "data1" in the iamge file is the same as "test1" in your
code,
on my linux box, I get the same results as your Windows results. I'm not
exactly sure what is going on here, but you might want to try reinstalling
Zelig from CRAN via
install.packges("Zelig")
and see if the problem persists. For your infomation, you can check the
version of the software by packageDescription("Zelig"). Make sure you get
version 2.1-4 which is the most current version.
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://www.princeton.edu/~kimai
---------------------------------------------------------
On Fri, 24 Jun 2005, Tom Vanwalleghem wrote:
Dear Prof. Kosuke Imai,
Really appreciating your quick response and tips.
Any idea what could be causing the difference then between linux
server/windows pc (difference in intercept estimate of -3.73 vs +0.67) ?
Or is there something I'm overseeing?
I've attached all necessary information in case you would have time to
have a look.
Thanks in advance,
Tom
Kosuke Imai wrote:
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Tom,
This warning message does NOT affect your results. When you are making
*conditional* predictions, we are calculating quantities of interest for
each observation of the data (i.e., in-sample inference). Therefore, we
need to set fn = NULL within setx. This warning comes up if you forget
this, but setx() automatically assigns fn = NULL. For details, see
http://gking.harvard.edu/zelig/docs/_TT_setx_TT__Setting_E.html
Best,
Kosuke
P.S. When you use traceback(), you need to have "()" in order to get
meaningful messages.
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://www.princeton.edu/~kimai
---------------------------------------------------------
On Thu, 23 Jun 2005, Tom Vanwalleghem wrote:
Dear all,
I'm using Zelig for rare events logistic regression. In the past, I've
run it with succes on a Linux server. However, now I would try to get it
to work on my laptop (Win XP), so I can already do some analysis during
fieldwork periods.
I however receive the following error message : "Warning message:
when cond = TRUE, fn is coerced to NULL in: setx.default(z.out1, data =
test1, cond = TRUE) " (see below for more details)
I would like to know if someone could point out what it means? And above
all: can it be solved? Or will I have to keep working on the server?
The consequence of this error message is apparently that the intercept
estimate is too high, so something must go wrong in the rare events
correction I suppose?
Regards,
Tom
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
z.out1<-zelig(depressies~archeodist + aspect1
+ aspect2 + aspect3 +
B_1 + B_2 + B_3 + B_4 + B_5 +
+ B_6 + B_7 + G_fb + G_zbr + G_zle + gully + roadgully + helling, data =
test1 ,
+ model = "relogit" , tau = 71/22788 , bias.correct = TRUE)
x.out1<-setx(z.out1, data=test1, cond=TRUE)
Warning message:
when cond = TRUE, fn is coerced to NULL in: setx.default(z.out1, data =
test1, cond = TRUE)
traceback
function (x = NULL)
{
if (is.null(x) && (exists(".Traceback", env = .GlobalEnv)))
x <- get(".Traceback", env = .GlobalEnv)
if (is.null(x) || length(x) == 0)
cat(gettext("No traceback available"), "\n")
else {
n <- length(x)
for (i in 1:n) {
label <- paste(n - i + 1, ": ", sep = "")
if ((m <- length(x[[i]])) > 1)
label <- c(label, rep(substr(" ", 1,
nchar(label, type = "w")), m - 1))
cat(paste(label, x[[i]], sep = ""), sep = "\n")
}
}
invisible()
}
<environment: namespace:base>
traceback
11:42 a.m.
Dear Prof. Kosuke Imai,
Sorry to bother again - many thanks for your advice.
Indeed, data1=test1 (some leftover code from some trial-and-error)
I have checked your suggestions, but strangely enough the old one is on
the linux machine.
Windows machine:
Package: Zelig
Version: 2.1-4
Date: 2005-5-22
Linux machine:
Package: Zelig
Version: 1.0-7
Date: 2004-05-26
Just the same as it surprises me that you say that the results I'm
getting on my Windows machine are correct? ?
This is very strange to me since the value for the intercept I'm
getting on my Windows machine (with the latest Zelig-version) is the same
as with a standard logistic regression in SAS (no rare events
correction)? I've attached the SAS output and the Zelig-Windows output...
Best regards,
Tom
Kosuke Imai wrote:
Dear Tom,
Assuming "data1" in the iamge file is the same as "test1" in your
code,
on my linux box, I get the same results as your Windows results. I'm not
exactly sure what is going on here, but you might want to try reinstalling
Zelig from CRAN via
install.packges("Zelig")
and see if the problem persists. For your infomation, you can check the
version of the software by packageDescription("Zelig"). Make sure you get
version 2.1-4 which is the most current version.
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://www.princeton.edu/~kimai
---------------------------------------------------------
Call: zelig(formula = depressies ~ archeodist + B_3 + aspect2, model =
"relogit", data = test1, tau = 71/22788, bias.correct = TRUE)
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -1.47e+00 2.91e-01 -5.05e+00 7.93e-07
archeodist -1.01e+00 3.89e-01 -2.61e+00 9.68e-03
B_3 1.14e+00 3.44e-01 3.32e+00 1.01e-03
aspect2 1.16e+00 3.07e-01 3.78e+00 1.89e-04
1:30 p.m.
Hi Tom,
First, please update your Zelig on your Linux machine. There have been
important bug fixes since the version you have, some of which are related
to relogit.
Second, I now understand your confusion. The output of relogit with a
specified value of tau isn't very good. The corrected intercept appears at
the end of the output:
tau = 0.0031
-3.73
So, the corrected intercept is -3.73 with your tau value 0.0031. I think
we should have printed this as the intercept estimate. I will fix that in
the next version. Sorry for the confusion.
However, I checked the code again, but the corrected version of the
intercept is used in sim(). To see how specifying tau makes differences in
your results. Try the following:
load("data1.RData")
library(Zelig)
z.out1 <- zelig(depressies~archeodist + aspect1 + aspect2 + aspect3 +
B_1 + B_2 + B_3 + B_4 + B_5 + B_6 + B_7 + G_fb + G_zbr +
G_zle + gully + roadgully + helling, data = data1,
model = "relogit" , tau = 71/22788)
x.out1 <- setx(z.out1)
s.out1 <- sim(z.out1, x=x.out1)
print(summary(s.out1))
z.out2 <- zelig(depressies~archeodist + aspect1 + aspect2 + aspect3 +
B_1 + B_2 + B_3 + B_4 + B_5 + B_6 + B_7 + G_fb + G_zbr +
G_zle + gully + roadgully + helling, data = data1,
model = "logit")
x.out2 <- setx(z.out2)
s.out2 <- sim(z.out2, x=x.out2)
print(summary(s.out2))
Thanks again for your feedbacks on this.
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://www.princeton.edu/~kimai
---------------------------------------------------------
On Fri, 24 Jun 2005, Tom Vanwalleghem wrote:
Dear Prof. Kosuke Imai,
Sorry to bother again - many thanks for your advice.
Indeed, data1=test1 (some leftover code from some trial-and-error)
I have checked your suggestions, but strangely enough the old one is on
the linux machine.
Windows machine:
Package: Zelig
Version: 2.1-4
Date: 2005-5-22
Linux machine:
Package: Zelig
Version: 1.0-7
Date: 2004-05-26
Just the same as it surprises me that you say that the results I'm
getting on my Windows machine are correct? ?
This is very strange to me since the value for the intercept I'm
getting on my Windows machine (with the latest Zelig-version) is the same
as with a standard logistic regression in SAS (no rare events
correction)? I've attached the SAS output and the Zelig-Windows output...
Best regards,
Tom
Kosuke Imai wrote:
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Tom,
Assuming "data1" in the iamge file is the same as "test1" in your
code,
on my linux box, I get the same results as your Windows results. I'm not
exactly sure what is going on here, but you might want to try reinstalling
Zelig from CRAN via
install.packges("Zelig")
and see if the problem persists. For your infomation, you can check the
version of the software by packageDescription("Zelig"). Make sure you get
version 2.1-4 which is the most current version.
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://www.princeton.edu/~kimai
---------------------------------------------------------
6904
days inactive
6905
days old
5 comments
2 participants
participants (2)
-
Kosuke Imai -
Tom Vanwalleghem