Regression Discontinuity Lab: Gaokao

Learn why RD only works at the true discontinuity. Elite college admission occurs at 580; try a placebo cutoff elsewhere and see how the estimated “effect” disappears or flips.

Parameters

True Elite Cutoff: 580 (Fixed)
Students with Gaokao ≥ 580 attend elite colleges
520630

Try 580 for valid RD, other values for placebo cutoff

015

Income benefit of elite college (10k RMB)

1560
312
015

Students manipulate scores above 580

Try These

RD Analysis at Cutoff 580 (True Elite Cutoff: 580)

Analysis: 580Gaokao ScoreFuture Income (10k RMB)Regular CollegeElite CollegeRD = 7.6

Results

Valid RD Analysis!
You're analyzing at 580, close to the true cutoff (580). RD estimate should detect the true effect of 8.
7.6
RD Estimate
Detected effect
8
True Effect
Actual premium
0.4
Error
|Estimate - True|
804
Sample Size
In bandwidth

Analysis Details

Analysis Cutoff:580
True Cutoff:580
Below Analysis Cutoff:410 students
Above Analysis Cutoff:394 students

Performance

RD Estimate:7.58
True Effect:8.00
Accuracy:Good
Cutoff Distance:0

Key RD Lessons

Why Cutoff Matters

RD only works when you analyze at the TRUE discontinuity. Analyzing at placebo cutoffs produces weak/no effects because there's no actual policy change there.

Real-World Application

In practice, researchers must know institutional details to identify true cutoffs. Fishing for effects at arbitrary thresholds invalidates the design.

Experiment: Try setting the analysis cutoff to 560 or 600. Notice how RD finds little effect even though elite colleges have a real impact!

Stata Implementation

Master All Causal Inference Methods

Now you understand how RD depends on knowing the true discontinuity. Compare this with other quasi-experimental methods that identify causal effects.