Hello, I am attempting to use a fuzzy regression discontinuity design. In the study, I am trying to assess the impact of an unconditional cash transfer program for women on their uptake of reproductive health services. The following are the main variables in this study:
pmt= proxy means test,a poverty score card used to deem eligibility for the transfer (HHs having a score of 16.17 or below are eligible)
center= the pmt score centered by subtracting the values from the cutoff
bisp= treatment status variable, 1 if the HH is receiving the cash transfer
d= binary variable, 1 if the place of delivery was in a health facility
The rdplot command yields the following graph, which does show a slight jump
However, when a non-parametric analysis is conducted, the p-value is insignificant
Mass points detected in the running variable.
And when I conduct a parametric analysis (ivreg2), to my surprise the p-value is significant. Can someone please explain, how this is possible?
.
pmt= proxy means test,a poverty score card used to deem eligibility for the transfer (HHs having a score of 16.17 or below are eligible)
center= the pmt score centered by subtracting the values from the cutoff
bisp= treatment status variable, 1 if the HH is receiving the cash transfer
d= binary variable, 1 if the place of delivery was in a health facility
The rdplot command yields the following graph, which does show a slight jump
Code:
rdplot p center, binselect(espr) graph_options(legend(pos(6) row(1)))
However, when a non-parametric analysis is conducted, the p-value is insignificant
Code:
rdrobust d center, fuzzy(bisp)
Code:
Fuzzy RD estimates using local polynomial regression. Cutoff c = 0 | Left of c Right of c Number of obs = 10400 -------------------+---------------------- BW type = mserd Number of obs | 2113 8287 Kernel = Triangular Eff. Number of obs | 1092 1316 VCE method = NN Order est. (p) | 1 1 Order bias (q) | 2 2 BW est. (h) | 5.363 5.363 BW bias (b) | 8.088 8.088 rho (h/b) | 0.663 0.663 Unique obs | 613 3400 First-stage estimates. Outcome: bisp. Running variable: center. -------------------------------------------------------------------------------- Method | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------------+------------------------------------------------------------ Conventional | -.01054 .03145 -0.3352 0.737 -.072194 .051105 Robust | - - -0.2413 0.809 -.082984 .064788 -------------------------------------------------------------------------------- Treatment effect estimates. Outcome: d. Running variable: center. Treatment Status: bisp. -------------------------------------------------------------------------------- Method | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------------+------------------------------------------------------------ Conventional | -3.0134 10.157 -0.2967 0.767 -22.9214 16.8946 Robust | - - -0.3002 0.764 -27.5515 20.2332 -------------------------------------------------------------------------------- Estimates adjusted for mass points in the running variable.
And when I conduct a parametric analysis (ivreg2), to my surprise the p-value is significant. Can someone please explain, how this is possible?
.
Code:
ivreg2 d center (bisp=eligible)
Code:
IV (2SLS) estimation -------------------- Estimates efficient for homoskedasticity only Statistics consistent for homoskedasticity only Number of obs = 10400 F( 2, 10397) = 376.88 Prob > F = 0.0000 Total (centered) SS = 2333.119615 Centered R2 = -0.0803 Total (uncentered) SS = 6866 Uncentered R2 = 0.6329 Residual SS = 2520.486281 Root MSE = .4923 ------------------------------------------------------------------------------ d | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- bisp | -.6058425 .1740404 -3.48 0.000 -.9469553 -.2647297 center | .0055215 .0008215 6.72 0.000 .0039113 .0071316 _cons | .6425867 .0288817 22.25 0.000 .5859797 .6991937 ------------------------------------------------------------------------------
Comment