After reviewing the Stata data, the current political climate, I was interested to identify a relationship (if any) between federal spending on border security and vote choice, the premise being that republicans and democrats view border security differently.
Run the initial linear probability model regression:
reg who2012 fedspend_bordercongress_therm male fedspend_terrorismfedspend_schools independent republican, r
|
Linear regression
|
|
Number of obs
|
= 719
|
|
|
F(7, 711)
|
= 231.88
|
|
|
|
Prob > F
|
= 0.0000
|
|
|
|
R-squared
|
= 0.6108
|
|
|
|
Root MSE
|
= .29704
|
|
|
|
|
|
|
|
|
Robust
|
|
|
|
who2012
|
Coef.
|
Std. Err.
|
t P>t
|
[95% Conf. Interval]
|
|
|
|
|
|
|
|
fedspend_border
|
.0147129
|
.0074025
|
1.99 0.047
|
.0001795 .0292464
|
|
congress_therm
|
.0021285
|
.0005851
|
3.64 0.000
|
.0009798 .0032773
|
|
male
|
.0471352
|
.0231848
|
2.03 0.042
|
.0016162 .0926541
|
|
fedspend_terrorism
|
.0188195
|
.0062908
|
2.99 0.003
|
.0064687 .0311702
|
|
fedspend_schools
|
-.0334393
|
.0093583
|
-3.57 0.000
|
-.0518125 -.015066
|
|
independent
|
-.2529824
|
.0633581
|
-3.99 0.000
|
-.3773738 -.128591
|
|
republican
|
-.7149579
|
.033041
|
-21.64 0.000
|
-.7798275 -.6500883
|
|
_cons
|
.7257504
|
.0561219
|
12.93 0.000
|
.615566 .8359348
|
|
|
|
|
|
|
Based on these results, it looks like there is a positive increase in the probability of voting for Obama (democrat) as the value of fedspend_border increases, which actually corresponds to a decrease in spending. However, the magnitude is small.
To test an LPM predicted value, we'll use the following characteristics:
Male, independent, rates Congress at 50, believes terrorism spending and school spending should remain constant, and believes border security spending should be increased a great deal
To find the z value:
scalar z = _b[_cons] + _b[fedspend_border]*1 + _b[congress_therm]*50 + _b[male]*1 + _b[fedspend_terrorism]*4 + _b[fedspend_schools]*4 + _b[independent]*1 + _b[republican]*0
display z
This particular example gives us the percentage probability that the voter will select Obama.
Now we can find the probability of someone with similar characteristics, BUT a change in his opinion of border spending, in this case, someone who believes spending should remain the same (fedspend_border = 4):
scalar z = _b[_cons] + _b[fedspend_border]*4 + _b[congress_therm]*50 + _b[male]*1 + _b[fedspend_terrorism]*4 + _b[fedspend_schools]*4 + _b[independent]*1 + _b[republican]*0
display z
Now we can find the probability of someone with similar characteristics, BUT a change in his opinion of border spending, in this case, someone who believes spending should decrease a great deal (fedspend_border = 7):
scalar z = _b[_cons] + _b[fedspend_border]*7 + _b[congress_therm]*50 + _b[male]*1 + _b[fedspend_terrorism]*4 + _b[fedspend_schools]*4 + _b[independent]*1 + _b[republican]*0
display z
These are all of our Linear Probability Model values.
Now, let's determine the probit model results:
probit who2012 fedspend_bordercongress_therm male fedspend_terrorismfedspend_schools independent republican, r
|
Iteration 0: log pseudolikelihood = -460.59874
|
|
Iteration 1: log pseudolikelihood = -216.29088
|
|
Iteration 2: log pseudolikelihood = -212.61059
|
|
Iteration 3: log pseudolikelihood = -212.59664
|
|
Iteration 4: log pseudolikelihood = -212.59663
|
|
Probit regression Number of obs
|
= 719
|
|
Wald chi2(7)
|
= 313.68
|
|
Prob > chi2
|
= 0.0000
|
|
Log pseudolikelihood = -212.59663 Pseudo R2
|
= 0.5384
|
|
|
|
|
Robust
|
|
|
who2012 Coef. Std. Err. z P>z
|
[95% Conf. Interval]
|
|
|
|
|
fedspend_border .0765934 .0458889 1.67 0.095
|
-.0133472 .1665339
|
|
congress_therm .0111291 .0035811 3.11 0.002
|
.0041103 .018148
|
|
male .2450341 .1430008 1.71 0.087
|
-.0352423 .5253105
|
|
fedspend_terrorism .1208896 .0412771 2.93 0.003
|
.039988 .2017912
|
|
fedspend_schools -.1820228 .0476661 -3.82 0.000
|
-.2754467 -.0885988
|
|
independent -.9743602 .20477 -4.76 0.000
|
-1.375702 -.5730184
|
|
republican -2.39355 .1491906 -16.04 0.000
|
-2.685958 -2.101142
|
|
_cons .4506491 .3076945 1.46 0.143
|
-.1524211 1.053719
|
|
|
|
We can now use the same Stata scalar command, however, to determine the probit value, we must use:
display normprob(z)
How do we determine the rest of the relevant results and what does the final table look like?