economic statistic
HhaNn
S0159.zipS0159.do
use "G:\comp_ia_bootstrap4_RCT.dta", clear * 1 tabstat newreg busy age female vote00 state comp_mi comp_ia vote98, by(treat_real) logistic treat_real newreg logistic treat_real busy logistic treat_real age logistic treat_real female logistic treat_real vote00 logistic treat_real state logistic treat_real comp_mi logistic treat_real comp_ia logistic treat_real vote98 * 3 tab vote02 treat_real, ch col * 4 logistic vote02 treat_real, coef logistic vote02 treat_real newreg, coef logistic vote02 treat_real busy, coef logistic vote02 treat_real age, coef logistic vote02 treat_real female, coef logistic vote02 treat_real vote00, coef logistic vote02 treat_real state, coef logistic vote02 treat_real comp_mi, coef logistic vote02 treat_real comp_ia, coef logistic vote02 treat_real vote98, coef
S0159.docx
1.
The means of the two groups, the differences and the p-values are shown in the following table (some p-values are not calculable):
|
|
treatment |
control |
diff |
p-value |
|
newreg |
0.052 |
0.049 |
0.003 |
0.140 |
|
busy |
0.032 |
0.000 |
0.032 |
- |
|
age |
56.056 |
55.768 |
0.288 |
0.086 |
|
female |
0.554 |
0.564 |
-0.009 |
0.037 |
|
vote00 |
0.739 |
0.736 |
0.003 |
0.441 |
|
state |
1.000 |
1.000 |
0.000 |
- |
|
comp_mi |
0.000 |
0.000 |
0.000 |
- |
|
comp_ia |
1.000 |
1.000 |
0.000 |
- |
|
vote98 |
0.581 |
0.572 |
0.009 |
0.035 |
2.
The table shows the randomization may not work well here since some of the differences between treatment and control are statistically significant, though the absolute difference values are all very small.
3.
From the Stata output, we find the voting probability is 59.55% in the control group and 61.27% in the treatment group. The difference is 1.72%, and statistically significant (p-value < 0.001); however, it is not large in a practical sense.
4.
The following table shows the coefficient based on the logistic regression, where the first column only include treat_real and the following columns add one more covariate.
|
|
Treat_real |
+newreg |
+busy |
+age |
+female |
+vote00 |
+state |
+comp_mi |
+comp_ia |
+vote98 |
|
B0 |
0.387 |
0.450 |
0.387 |
-0.986 |
0.458 |
-1.491 |
0.387 |
0.387 |
0.387 |
-0.668 |
|
B1 |
0.072 |
0.077 |
0.072 |
0.069 |
0.072 |
0.088 |
0.072 |
0.072 |
0.072 |
0.069 |
|
B2 |
- |
-1.292 |
0.020 |
0.025 |
-0.012 |
2.555 |
0.000 |
0.000 |
0.000 |
1.977 |
5.
We find some of the covariates (e.g. vote00) have big effects on the treatment effect. This phenomenon may be caused by the imbalance between the two groups, suggesting some of the covariates also have effects on the outcome.
No, it will not get us an unbiased estimate of the causal effect. From the previous results, we see some of the covariates also have big effects on the outcome, so we have to control this before establishing the causal relationship between the treatment and the outcome.
Pearson chi2(1) = 15.8502 Pr = 0.000
100.00 100.00 100.00
Total 85,628 14,990 100,618
59.55 61.27 59.80
1 50,988 9,185 60,173
40.45 38.73 40.20
0 34,640 5,805 40,445
vote02 0 1 Total
treat_real
column percentage
frequency
Key
