This paper introduces the Difference-in-Kinks (DiK) design, an econometric framework that extends the standard regression kink design to settings in which the slope of a policy rule varies over time. By combining the key features of the regression kink and difference-in-differences approaches, the DiK design identifies causal effects from variation in kink intensity over time. We formalize both sharp and fuzzy versions of the estimator and derive the identification conditions under a parallel-trends assumption. Applying DiK to Finland’s 2011 guarantee pension reform demonstrates that changes in marginal incentives significantly increased the probability of retirement, while the standard regression kink design would have obtained implausibly large estimates in the opposite direction. The DiK design thus offers a flexible framework for policy evaluation in dynamic, nonlinear environments.