Regression on Regression: Mapping Data-Driven Binary Black Hole Merger Rate Fits to Progenitor Histories

Abstract

The binary black hole (BBH) merger rate is governed by the progenitor formation rate and the distribution of delay-times between formation and merger, but these functions remain poorly constrained. We introduce a framework that maps the parameters of physics-driven models directly onto existing data-driven fits of the BBH merger rate. This ``regression on regression'' approach enables physical interpretation of flexible population models without the computational burden of reanalyzing the underlying gravitational-wave event data. Applying this method to the \textsc{B-Spline} merger-rate posteriors from the Fourth Gravitational-Wave Transient Catalog, we fit the minimum delay time ($τ_{\text{min}}$), delay-time power-law index ($α$), and progenitor formation parameters controlling the normalization ($\mathcal{A}$), early-time growth ($γ$), and late-time decay ($δ$). Increasing the number of anchoring redshift points from two to four reduces the median sum-squared error (SSE) by a factor of $\approx 4.5$. However, residuals reveal that the physical model does not pass through all four anchors, exposing model misspecification and demonstrating a key strength of the framework: unlike standard inference methods, which preferentially weight compatible curves and mask underlying tensions, our approach exposes BBH posteriors irreconcilable with the model. Despite uncertainties at $z\gtrsim1$, the shape of the progenitor formation rate at low-$z$ is robust and evolves more steeply than the global star formation rate (SFR), supporting a preference for low metallicity environments. Specifically, the log-space slope of the progenitor rate is $\approx 5.3$ times steeper than the SFR between $z=0.1$ and $z=1.0$. Ultimately, a more complex phenomenological model is required to match the \textsc{B-Spline} merger rates.