Answers are grounded in this paper and cite their sources.
Endpoint behavior and weak identification complicate the analysis of bounded data. We study a random-stage Coxian framework that connects sequential stopping with distributions on the unit interval. The unit member admits a random-length beta-product representation, and stable distributional formulas remain available when rates coincide. A specialization of the phase-type small-time expansion shows that, at the upper endpoint, the hazard ratio relative to the baseline tends to the minimum admissible stage count. For an ordered Coxian subfamily, we distinguish nonidentification when continuation is absent from singularity at equal rates and establish likelihood-ratio limits under fixed compact rate-ratio restrictions. A separately defined dimensionless profile statistic has a nuisance-free law under the no-continuation null, allowing finite-sample Monte Carlo level control in exact arithmetic. Direct bootstrap experiments assess equal-rate inference, and comparisons with the corresponding fixed-stage submodel identify both predictive gains and the cost of unnecessary flexibility in the studied settings. Profile and Wald diagnostics document the effects of weak identification and boundary fits on interval precision and availability. Applications to electricity shares and piped-water coverage distinguish relative fit from goodness of fit, with data screening and dependence limitations reported explicitly. These results connect endpoint structure to inference while keeping theoretical likelihood suprema distinct from numerical search results.
No takes yet. Share an insight, caveat, or question.
Gomaa et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: