Regression Discontinuity in WFM
Regression Discontinuity in WFM (RD) is a quasi-experimental method that estimates a treatment effect when treatment is assigned by whether a continuous "running variable" crosses a fixed cutoff. Units just below and just above the threshold are nearly identical except that one group got the treatment and the other did not, so the jump in the outcome at the cutoff is a credible estimate of the effect for units near it. RD applies wherever workforce management policies are triggered by sharp thresholds.[1]
How it works
A running variable (also called a forcing variable) determines treatment: units on one side of a cutoff receive it, units on the other do not. RD compares outcomes in a narrow band around the threshold, fitting a trend on each side and measuring the gap at the cutoff. The logic is that whatever else varies with the running variable does so smoothly, while the treatment switches discontinuously — so a discontinuity in the outcome at exactly the cutoff is attributable to the treatment, not to the running variable. The design dates to Thistlethwaite and Campbell's 1960 study of scholarship awards based on a test-score threshold.[2] In the sharp design treatment is fully determined by the cutoff; in the fuzzy design crossing the cutoff only changes the probability of treatment, and RD is combined with instrumental-variables estimation.[3]
WFM examples
- Tenure or score thresholds. A policy that unlocks at a tenure cutoff (eligibility for self-scheduling, a bonus, or a program) lets RD estimate the policy's effect on attrition or performance by comparing agents just above and just below the cutoff.
- Performance-band triggers. Coaching or QA interventions triggered when a score falls below a fixed threshold: compare agents narrowly above and below to estimate the intervention's effect — and to avoid the regression-to-the-mean trap that plagues naive "coach the bottom" evaluations.
- Staffing or routing rules. Volume or wait-time thresholds that trigger overflow, callbacks, or added staff allow RD estimates of those rules' effects.
Cautions
- Manipulation of the running variable. If units can sort themselves across the cutoff (e.g., scores nudged to clear a threshold), comparability breaks; a density check at the cutoff helps detect this.
- Locality. RD estimates the effect only near the cutoff; it says little about units far from the threshold.
- Functional form. Mis-modeling the trend on either side can masquerade as a jump; estimates should be robust to the bandwidth and the curve fitted.
- Other thresholds at the same cutoff. If several policies switch at the same value, RD captures their combined effect, not the one of interest.
Maturity Model Position
In the WFM Labs Maturity Model™, RD is the natural method wherever the operation runs threshold-based rules.
- Level 1–2 (Emerging / Foundational) — threshold policies are evaluated by comparing all who got the treatment to all who did not, ignoring that the groups differ on the running variable.
- Level 3 (Progressive) — analysts exploit cutoffs, compare units near the threshold, and check for manipulation and bandwidth sensitivity.
- Level 4–5 (Advanced / Pioneering) — RD (sharp and fuzzy) is a standard evaluation design for threshold rules, with density and robustness checks built into the analysis.
See also
- Causal Inference in Workforce Management
- Difference-in-Differences in WFM
- Instrumental Variables in WFM
- Potential Outcomes Framework
- Regression to the Mean in WFM
- A/B Testing for WFM Experiments
References
- ↑ Lee, D. S., & Lemieux, T. (2010). "Regression Discontinuity Designs in Economics". Journal of Economic Literature, 48(2), 281–355. doi:10.1257/jel.48.2.281.
- ↑ Thistlethwaite, D. L., & Campbell, D. T. (1960). "Regression-Discontinuity Analysis: An Alternative to the Ex Post Facto Experiment". Journal of Educational Psychology, 51(6), 309–317.
- ↑ Imbens, G. W., & Lemieux, T. (2008). "Regression Discontinuity Designs: A Guide to Practice". Journal of Econometrics, 142(2), 615–635.
