Table 5. Hypothetical Bias Probit Model
|
Explanatory Variables |
RDD Sample |
NPF Sample | ||||||
|
Mean |
Coefficient P -Value |
ME |
Mean |
Coefficient |
P -Value |
ME | ||
|
Hypothetical |
0.082 |
1.362*** |
0.001 |
0.136 |
0.039 |
0.348** |
0.034 |
0.126 |
|
All Revenue |
0.628 |
-0.177 |
0.395 |
-0.007 |
0.636 |
0.030 |
0.376 |
0.012 |
|
Low Visits |
0.803 |
-1.329*** |
0.002 |
-0.131 |
0.453 |
-0.409*** |
0.000 |
-0.157 |
|
Young |
0.149 |
0.580 |
0.163 |
0.038 |
0.138 |
-0.055 |
0.346 |
-0.021 |
|
Old |
0.242 |
-0.400 |
0.206 |
-0.014 |
0.300 |
0.177** |
0.043 |
0.067 |
|
Male |
0.433 |
-0.517 |
0.106 |
-0.022 |
0.477 |
0.062 |
0.252 |
0.024 |
|
BS degree |
0.357 |
0.537 |
0.140 |
0.026 |
0.394 |
0.093 |
0.210 |
0.036 |
|
Professional degree |
0.236 |
0.068 |
0.452 |
0.003 |
0.353 |
0.129 |
0.144 |
0.049 |
|
Low Income |
0.312 |
-0.400 |
0.204 |
-0.019 |
0.255 |
0.054 |
0.361 |
0.020 |
|
White |
0.846 |
-0.577 |
0.156 |
-0.038 |
0.877 |
0.224* |
0.080 |
0.087 |
|
NE |
0.048 |
0.168 |
0.438 |
0.008 |
0.079 |
-0.475*** |
0.008 |
-0.187 |
|
ENC |
0.113 |
0.178 |
0.383 |
0.008 |
0.117 |
-0.297** |
0.038 |
-0.117 |
|
WNC |
0.160 |
0.117 |
0.427 |
0.005 |
0.155 |
-0.375** |
0.012 |
-0.147 |
|
GP |
0.071 |
0.122 |
0.433 |
0.005 |
0.082 |
-0.730*** |
0.000 |
-0.285 |
|
SE |
0.165 |
-0.187 |
0.383 |
-0.007 |
0.177 |
-0.307** |
0.018 |
-0.120 |
|
WSC |
0.065 |
-3.832 |
0.461 |
-0.032 |
0.025 |
-0.471** |
0.046 |
-0.186 |
|
SW |
0.067 |
-1.256 |
0.084 |
-0.020 |
0.071 |
-0.353** |
0.039 |
-0.139 |
|
RM |
0.097 |
-1.114 |
0.133 |
-0.022 |
0.109 |
-0.171 |
0.147 |
-0.067 |
|
Constant |
1.000 |
-0.639 |
0.235 |
-- |
1.000 |
0.435** |
0.037 |
-- |
|
Sample Size |
180 |
830 | ||||||
|
Hypothetical Bias Scaling Factor (eδ) |
1.43 |
1.30 | ||||||
Notes. (***), (**), and (*) refer to statistical significance at the 1, 5 and 10 percent levels. The estimation was carried out
using the Constrained Maximum Likelihood (CML 2.0) package in Gauss version 3.5. The nonlinear optimization routine
was Newton-Raphson with a convergence criterion of 1×10-5 for the gradient of the coefficients. The estimates for “don’t
know” and “missing” dummy variables are not shown. ME = Marginal Effect. 26 and 17 protest households (those that
would not be interested in the NRP even if it were offered free of charge) were removed from the RDD and NPF samples.
30