Dear Statalist Community,
I really could use your help, as I am confused on what model to choose, considering the data I have. My dependent variable (voteshare) consists on the vote-share in a party in a given election at the municipality level (8,108 observations).
However, because my independent variables -- decline and growth -- consist of historical data portraying the lower economic growth (decline) and higher economic growth (growth) during a particular historical period precceding the election, this data is only available at the provincial level, and there are only 50 provinces.
Should I:
1. Ignore this, and run a normal:
reg voteshare decline growth [controls], r
2. Should I treat my DV as a "small N, large T", being N the number of provinces (N=50) and the T the number of municipalities within each province, with:
xtset province
and run a xtgls or xtregar ?
3. I tried using a DV N=50 with the voteshare of each province instead of municipalities, but the results look really strange... Although the scatterplots clearly show that the provinces that experienced a higher decline period have a lower voteshare, and the provinces that exprienced more growth have a higher voteshare (which is in accordance with my theoretical expectations) (Y axis=voteshare):
However, my regression does not portray these relationships, even without controls:
. reg voteshare growth decline, r
Linear regression Number of obs = 50
F(2, 47) = 7.21
Prob > F = 0.0019
R-squared = 0.2791
Root MSE = .13951
------------------------------------------------------------------------------
| Robust
voteshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
growth | 1.004445 .689592 1.46 0.152 -.3828348 2.391726
decline | -1.007512 .820127 -1.23 0.225 -2.657395 .6423708
_cons | .4703362 .0223719 21.02 0.000 .4253298 .5153427
------------------------------------------------------------------------------
I don't know why this might be and I am not sure this N=50 province-level analysis is the best solution.
4. Bayesian model? However I don't know how to go about the priors.
I really could use your advice!
Thank you very much!
I really could use your help, as I am confused on what model to choose, considering the data I have. My dependent variable (voteshare) consists on the vote-share in a party in a given election at the municipality level (8,108 observations).
However, because my independent variables -- decline and growth -- consist of historical data portraying the lower economic growth (decline) and higher economic growth (growth) during a particular historical period precceding the election, this data is only available at the provincial level, and there are only 50 provinces.
Should I:
1. Ignore this, and run a normal:
reg voteshare decline growth [controls], r
2. Should I treat my DV as a "small N, large T", being N the number of provinces (N=50) and the T the number of municipalities within each province, with:
xtset province
and run a xtgls or xtregar ?
3. I tried using a DV N=50 with the voteshare of each province instead of municipalities, but the results look really strange... Although the scatterplots clearly show that the provinces that experienced a higher decline period have a lower voteshare, and the provinces that exprienced more growth have a higher voteshare (which is in accordance with my theoretical expectations) (Y axis=voteshare):
However, my regression does not portray these relationships, even without controls:
. reg voteshare growth decline, r
Linear regression Number of obs = 50
F(2, 47) = 7.21
Prob > F = 0.0019
R-squared = 0.2791
Root MSE = .13951
------------------------------------------------------------------------------
| Robust
voteshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
growth | 1.004445 .689592 1.46 0.152 -.3828348 2.391726
decline | -1.007512 .820127 -1.23 0.225 -2.657395 .6423708
_cons | .4703362 .0223719 21.02 0.000 .4253298 .5153427
------------------------------------------------------------------------------
I don't know why this might be and I am not sure this N=50 province-level analysis is the best solution.
4. Bayesian model? However I don't know how to go about the priors.
I really could use your advice!
Thank you very much!
