I have two groups of countries (0 and 1), and am trying to explain the difference as due to a variable MTP. This is a continuous variable, that ranges between approx 0-12.
I would like to be able to say: countries at the higher end of MTP (e.g., 90th percentile) are x% more likely to be in group 1 than countries with low MTP (e.g., 10th percentile)
I am using xtlogit, and the coefficient for MTP is 2.66. My understanding is that this means that an increase of 1 in MTP, the odds ratio of being 0 or 1 increases by exp(2.66)=14.3.
If i take into account that MTP regularly differs between approx 1 and 8 (10th vs 90th percentile), than the odds ratio of being 0 or 1 is 14.3 to the power of 7, equals 122,279,108.
That seems extreme, or just silly. I would only expect that if countries were nearly perfectly distributed as 0 for ones with low MTP, and 1 for ones with high MTP, but that is not the case at all (see boxplot at end).
I am using 'xtlogit, re' by the way, because the variable MTP is constant for each country.
The question then is: 1) what is wrong with my explanation, and 2) how do I obtain a result that does accurately indicate the odds ratio of being 0 or 1, depending on the value of MTP?
Array
I would like to be able to say: countries at the higher end of MTP (e.g., 90th percentile) are x% more likely to be in group 1 than countries with low MTP (e.g., 10th percentile)
I am using xtlogit, and the coefficient for MTP is 2.66. My understanding is that this means that an increase of 1 in MTP, the odds ratio of being 0 or 1 increases by exp(2.66)=14.3.
If i take into account that MTP regularly differs between approx 1 and 8 (10th vs 90th percentile), than the odds ratio of being 0 or 1 is 14.3 to the power of 7, equals 122,279,108.
That seems extreme, or just silly. I would only expect that if countries were nearly perfectly distributed as 0 for ones with low MTP, and 1 for ones with high MTP, but that is not the case at all (see boxplot at end).
I am using 'xtlogit, re' by the way, because the variable MTP is constant for each country.
The question then is: 1) what is wrong with my explanation, and 2) how do I obtain a result that does accurately indicate the odds ratio of being 0 or 1, depending on the value of MTP?
Code:
. xtlogit windex100 MTP_Wind_Os_GWh_sqkm EconGen03ln Wind12ln PowAll01ln PowAll03 PowAll05ln EconGen05 Coal03 Hydro03 Nucl03 Gas03 Oi
> l03 BioGeo03 PV04 Pollute04ln Pollute01atln EnerPri06 EnerPri08 EnerPri10 PoliWind01 PoliWind02 PoliWind03 PoliWind04 PoliWind05 Pol
> iWind06 PoliWind07 PoliWind08 PoliWind09, re nolog iterate(250)
note: PoliWind07 != 0 predicts success perfectly
PoliWind07 dropped and 19 obs not used
Random-effects logistic regression Number of obs = 3996
Group variable: ctryidnr Number of groups = 132
Random effects u_i ~ Gaussian Obs per group: min = 2
avg = 30.3
max = 35
Integration method: mvaghermite Integration points = 12
Wald chi2(27) = 767.46
Log likelihood = -183.47199 Prob > chi2 = 0.0000
--------------------------------------------------------------------------------------
windex100 | Coef. Std. Err. z P>|z| [95% Conf. Interval]
---------------------+----------------------------------------------------------------
MTP | 2.665071 .4818044 5.53 0.000 1.720752 3.60939
EconGen03ln | 8.246404 1.654326 4.98 0.000 5.003985 11.48882
Wind12ln | 8.126688 .8110303 10.02 0.000 6.537098 9.716279
PowAll01ln | 11.38055 .9115233 12.49 0.000 9.593997 13.1671
PowAll03 | -.159103 .0714346 -2.23 0.026 -.2991123 -.0190937
PowAll05ln | 5.232509 1.864259 2.81 0.005 1.578629 8.88639
EconGen05 | .038857 .0117137 3.32 0.001 .0158986 .0618154
Coal03 | .0506792 .0301829 1.68 0.093 -.0084782 .1098365
Hydro03 | -.0465293 .0391585 -1.19 0.235 -.1232786 .03022
Nucl03 | -.1067515 .0664984 -1.61 0.108 -.237086 .023583
Gas03 | -.0171653 .0107564 -1.60 0.111 -.0382474 .0039169
Oil03 | -.0306124 .0454107 -0.67 0.500 -.1196158 .0583909
BioGeo03 | -.5167287 .4115646 -1.26 0.209 -1.32338 .2899231
PV04 | 6.220702 2.15526 2.89 0.004 1.99647 10.44493
Pollute04ln | -12.32647 2.188002 -5.63 0.000 -16.61488 -8.038064
Pollute01atln | -2.133058 1.602303 -1.33 0.183 -5.273514 1.007399
EnerPri06 | -.0107482 .0185987 -0.58 0.563 -.047201 .0257046
EnerPri08 | .0893871 .0566184 1.58 0.114 -.0215829 .2003571
EnerPri10 | -.5275449 .3214577 -1.64 0.101 -1.15759 .1025006
PoliWind01 | 26.57017 21.69011 1.22 0.221 -15.94167 69.08201
PoliWind02 | -3.424483 2.499803 -1.37 0.171 -8.324008 1.475041
PoliWind03 | 16.37729 3.28862 4.98 0.000 9.93171 22.82286
PoliWind04 | -24.31482 8.150193 -2.98 0.003 -40.2889 -8.340731
PoliWind05 | -8.727784 13.54992 -0.64 0.519 -35.28514 17.82957
PoliWind06 | -1.551728 8.474649 -0.18 0.855 -18.16173 15.05828
PoliWind07 | 0 (omitted)
PoliWind08 | -7.17042 4.932269 -1.45 0.146 -16.83749 2.496649
PoliWind09 | 7.813706 4.775382 1.64 0.102 -1.54587 17.17328
_cons | -388.7335 35.53703 -10.94 0.000 -458.3848 -319.0823
---------------------+----------------------------------------------------------------
/lnsig2u | 5.677371 .2015317 5.282376 6.072366
---------------------+----------------------------------------------------------------
sigma_u | 17.09328 1.722419 14.02986 20.8256
rho | .9888656 .0022189 .9835611 .9924716
--------------------------------------------------------------------------------------
Likelihood-ratio test of rho=0: chibar2(01) = 626.45 Prob >= chibar2 = 0.000
Array