Hi Scott,
LS will be fine. Poisson with fixed-effect is a non-linear estimator that
needs more attention. I was wonder that fip variable changes over-time but
not too much, that was the reason why I asked you for a descriptive. I am
guessing that there is something with this low time variability of flip and
other variable in the model (probably the fixed-effect). If flip has very
low time-variability (almost time-invariant) you are not able to identify
the coefficients in a fixed-effects framework. Finally, your dependent
variable seems to have some outlier events... the mean is around 3,
meanwhile the minimum is 0 and the maximum 99. It would be better for your
analysis to see the description of the data under the model that you want.
You can run the LS and take a look of the variables using -xtsum if
e(sample)-
I hope this helps you
Rodrigo.
----- Original Message -----
From: "Scott Cunningham" <[email protected]>
To: <[email protected]>
Sent: Wednesday, May 24, 2006 12:06 PM
Subject: Re: st: Re: poisson will not converge
On May 24, 2006, at 11:54 AM, Rodrigo A. Alfaro wrote:
> Scott,
>
> What is "fip" variable? state? where did you get 55? It seems this
> variable
> is highly correlated with others in the model. A quick inspection
> could be
> helpful: -xtsum rp srfm age2 hgc hhd1 fip-. Moreover, is this panel
> balanced?
Rodrigo,
Thank you for your reply. I'll answer what I can.
The panel is balanced (except for hhd1, as I hadn't noticed something
before doing the xtsum [see below]). "Fip" is a state indicator
variable, and while the number goes to 55, it skips some numbers for
reasons only the Census knows so that it still totals to either 50 or
51 (I forget if DC is included separately). You say fip is highly
correlated with others in the model. With other variables in the
model? You may be right, as when I estimate the model without the -
fip- variables, it does converge fine. Also, when I re-estimate the
model with fixed effects with just year and fip, it has trouble
converging. And this is perhaps due to the fact that there is some
migration across time and state within my sample, but not much, as
the individuals in the sample are relatively young and most live in
the same state. Do you have any suggestions as to what I can do to
overcome this? Also, OLS does work fine, making me wonder if I
should be suspicious in light of what you said of those estimates.
Variable | Mean Std. Dev. Min Max |
Observations
-----------------+--------------------------------------------
+----------------
rp overall | 3.113009 8.069344 0 99 | N
= 2283
between | 5.431478 0 44.66667 |
n = 761
within | 5.969858 -38.88699 67.11301 |
T = 3
| |
srfm overall | 100.1544 11.52157 35.3187 243.9197 | N
= 2283
between | 8.116986 53.33359 172.0538 |
n = 761
within | 8.180396 46.42766 184.2222 |
T = 3
| |
age2 overall | 308.8257 74.43039 169 529 | N
= 2283
between | 47.65699 227.6667 443.6667 |
n = 761
within | 57.1899 226.159 394.159 |
T = 3
| |
hgc overall | 10.58826 1.852228 4 16 | N
= 2283
between | 1.33929 5.666667 14.66667 |
n = 761
within | 1.280086 7.254928 14.25493 |
T = 3
| |
hhd1 overall | .2415459 .4281144 0 1 | N
= 2277
between | .3970831 0 1 |
n = 761
within | .1596349 -.4251208 .9082126 | T-
bar = 2.99212
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
