6:16 p.m.
Dear,
sample restriction does not seem to work well in panel data...
I have recently installed gretl 1.9.13 on my laptop.
Here are the issues:
1) when you restrict the sample, gretl continues to put the indication
"Full range" in the status bar.
Example:
open greene14_1.gdt
smpl year > 1970 --restrict --replace
"Full range" in the status bar does not disappear.
2) the option time-dummies in the panel command does not work as
expected when the sample is restricted and it also messes all the time
dummies.
Example:
open greene14_1.gdt
genr timedum
panel C 0 Q PF LF dt_*
panel C 0 Q PF LF --time-dummies
The previous commands work as expected and the last two estimates are
equal as they should be.
Instead, if you restrict the sample, the two estimates are different:
smpl year > 1972 --restrict --replace
panel C 0 Q PF LF dt_*
panel C 0 Q PF LF --time-dummies
The former is correct, whereas the latter includes less time dummies...
Moreover, it rewrites the time dummies and completely messes them...
Giuseppe
8:26 p.m.
Hi,
With panels I'm almost always using
--balanced together with --restrict and w or w/o --replace
Perhaps this should be the standard in panel data sets and the user has to add
--unbalanced instead?
Btw, it's very helpful that
smpl --res --rep --bal works!
Leon
Am 18.11.2013 um 19:16 schrieb Giuseppe Vittucci
<ignatiusreilly75(a)yahoo.it>:
Dear,
sample restriction does not seem to work well in panel data...
I have recently installed gretl 1.9.13 on my laptop.
Here are the issues:
1) when you restrict the sample, gretl continues to put the indication
"Full range" in the status bar.
Example:
open greene14_1.gdt
smpl year > 1970 --restrict --replace
"Full range" in the status bar does not disappear.
2) the option time-dummies in the panel command does not work as
expected when the sample is restricted and it also messes all the time
dummies.
Example:
open greene14_1.gdt
genr timedum
panel C 0 Q PF LF dt_*
panel C 0 Q PF LF --time-dummies
The previous commands work as expected and the last two estimates are
equal as they should be.
Instead, if you restrict the sample, the two estimates are different:
smpl year > 1972 --restrict --replace
panel C 0 Q PF LF dt_*
panel C 0 Q PF LF --time-dummies
The former is correct, whereas the latter includes less time dummies...
Moreover, it rewrites the time dummies and completely messes them...
Giuseppe
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
9:16 p.m.
In my case adding the option "balanced" does not solve the issue:
open greene14_1.gdt
genr timedum
smpl year > 1975 --restrict --balanced
panel C 0 Q PF LF dt_*
panel C 0 Q PF LF --time-dummies
The result is:
gretl version 1.9.13
Current session: 2013-11-18 22:13
? smpl year > 1975 --restrict --balanced
Full data set: 90 observations
Current sample: 1:1 - 6:9 (n = 54)
? panel C 0 Q PF LF dt_*
Model 1: Fixed-effects, using 54 observations
Included 6 cross-sectional units
Time-series length = 9
Dependent variable: C
Omitted due to exact collinearity: dt_15
coefficient std. error t-ratio p-value
----------------------------------------------------------------
const 2.91847e+06 2.87409e+06 1.015 0.3165
Q 3.40796e+06 466556 7.305 1.12e-08 ***
PF 1.35105 3.12512 0.4323 0.6680
LF -7.84170e+06 1.74805e+06 -4.486 6.82e-05 ***
dt_7 206516 1.63077e+06 0.1266 0.8999
dt_8 176081 1.48931e+06 0.1182 0.9065
dt_9 362840 1.37686e+06 0.2635 0.7936
dt_10 172625 847157 0.2038 0.8397
dt_11 -106171 182687 -0.5812 0.5647
dt_12 -73608.8 590531 -0.1246 0.9015
dt_13 -65699.8 399724 -0.1644 0.8703
dt_14 -69504.2 183768 -0.3782 0.7074
Mean dependent var 1512112 S.D. dependent var 1348282
Sum squared resid 2.07e+12 S.E. of regression 236406.3
R-squared 0.978537 Adjusted R-squared 0.969256
F(16, 37) 105.4330 P-value(F) 6.67e-26
Log-likelihood -734.5735 Akaike criterion 1503.147
Schwarz criterion 1536.960 Hannan-Quinn 1516.187
rho 0.560452 Durbin-Watson 0.853544
Test for differing group intercepts -
Null hypothesis: The groups have a common intercept
Test statistic: F(5, 37) = 5.4162
with p-value = P(F(5, 37) > 5.4162) = 0.000777149
? panel C 0 Q PF LF --time-dummies
Model 2: Fixed-effects, using 54 observations
Included 6 cross-sectional units
Time-series length = 9
Dependent variable: C
coefficient std. error t-ratio p-value
-----------------------------------------------------------------
const 3.24479e+06 1.19361e+06 2.718 0.0098 ***
Q 3.41244e+06 459635 7.424 6.61e-09 ***
PF 1.00610 2.15612 0.4666 0.6434
LF -7.88439e+06 1.70371e+06 -4.628 4.22e-05 ***
dt_2 192948 191771 1.006 0.3207
dt_3 62244.2 512074 0.1216 0.9039
dt_4 -114577 1.13902e+06 -0.1006 0.9204
dt_5 -30165.8 1.46139e+06 -0.02064 0.9836
dt_6 -43973.9 1.32486e+06 -0.03319 0.9737
dt_7 -74924.6 1.15304e+06 -0.06498 0.9485
dt_8 -18568.0 1.07066e+06 -0.01734 0.9863
Mean dependent var 1512112 S.D. dependent var 1348282
Sum squared resid 2.07e+12 S.E. of regression 233350.1
R-squared 0.978524 Adjusted R-squared 0.970046
F(15, 38) 115.4254 P-value(F) 6.14e-27
Log-likelihood -734.5909 Akaike criterion 1501.182
Schwarz criterion 1533.005 Hannan-Quinn 1513.455
rho 0.560989 Durbin-Watson 0.851294
Test for differing group intercepts -
Null hypothesis: The groups have a common intercept
Test statistic: F(5, 38) = 5.63826
with p-value = P(F(5, 38) > 5.63826) = 0.000554147
Wald test for joint significance of time dummies
Asymptotic test statistic: Chi-square(7) = 2.79884
with p-value = 0.902966
Giuseppe
On Mon, 2013-11-18 at 21:26 +0100, Pindar wrote:
Hi,
With panels I'm almost always using
--balanced together with --restrict and w or w/o --replace
Perhaps this should be the standard in panel data sets and the user has to add
--unbalanced instead?
Btw, it's very helpful that
smpl --res --rep --bal works!
Leon
> Am 18.11.2013 um 19:16 schrieb Giuseppe Vittucci <ignatiusreilly75(a)yahoo.it>:
>
> Dear,
>
> sample restriction does not seem to work well in panel data...
> I have recently installed gretl 1.9.13 on my laptop.
>
> Here are the issues:
>
> 1) when you restrict the sample, gretl continues to put the indication
> "Full range" in the status bar.
> Example:
> open greene14_1.gdt
> smpl year > 1970 --restrict --replace
>
> "Full range" in the status bar does not disappear.
>
> 2) the option time-dummies in the panel command does not work as
> expected when the sample is restricted and it also messes all the time
> dummies.
>
> Example:
> open greene14_1.gdt
> genr timedum
> panel C 0 Q PF LF dt_*
> panel C 0 Q PF LF --time-dummies
>
> The previous commands work as expected and the last two estimates are
> equal as they should be.
>
> Instead, if you restrict the sample, the two estimates are different:
>
> smpl year > 1972 --restrict --replace
> panel C 0 Q PF LF dt_*
> panel C 0 Q PF LF --time-dummies
>
> The former is correct, whereas the latter includes less time dummies...
> Moreover, it rewrites the time dummies and completely messes them...
>
> Giuseppe
>
> _______________________________________________
> Gretl-users mailing list
> Gretl-users(a)lists.wfu.edu
> http://lists.wfu.edu/mailman/listinfo/gretl-users
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
11:06 p.m.
On Mon, 18 Nov 2013, Giuseppe Vittucci wrote:
Dear,
sample restriction does not seem to work well in panel data...
I have recently installed gretl 1.9.13 on my laptop.
Here are the issues:
1) when you restrict the sample, gretl continues to put the indication
"Full range" in the status bar.
Example:
open greene14_1.gdt
smpl year > 1970 --restrict --replace
"Full range" in the status bar does not disappear.
OK, I see that here, but it's just cosmetic. It'll be dealt with before
long.
2) the option time-dummies in the panel command does not work as
expected when the sample is restricted and it also messes all the time
dummies.
Example:
open greene14_1.gdt
genr timedum
panel C 0 Q PF LF dt_*
panel C 0 Q PF LF --time-dummies
The previous commands work as expected and the last two estimates are
equal as they should be.
Instead, if you restrict the sample, the two estimates are different:
smpl year > 1972 --restrict --replace
panel C 0 Q PF LF dt_*
panel C 0 Q PF LF --time-dummies
Yes, this was messed up. Thanks for spotting the problem. It's now fixed
in CVS and the fix will be in gretl 1.9.14, to be released shortly.
Allin Cottrell
4051
days inactive
4051
days old
3 comments
3 participants
participants (3)
-
Allin Cottrell -
Giuseppe Vittucci -
Pindar