Ken Wharton and I have proposed a mechanism for quantum entanglement. The key ingredient is the familiar statistical phenomenon of collider bias, or Berkson's bias. In the language of causal models, a collider is a variable causally influenced by two or more other variables. Conditioning on a collider typically produces non-causal associations between its contributing causes, even if they are actually independent. It is easy to show that this phenomenon can produce associations analogous to Bell correlations, in suitable post-selected ensembles. It is also straightforward that such collider artefacts may become real connections, resembling causality, if a collider is 'constrained' (e.g., by a future boundary condition). We consider the time-reversed analogues of these points in the context of retrocausal models of QM. Retrocausality yields a collider at the source of an EPR-Bell particle pair, and in this case constraint of the collider is possible by normal methods of experimental preparation. It follows that connections resembling causality may be expected to emerge across such colliders, from one branch of the experiment to the other. Our hypothesis is that this constrained retrocausal collider bias is the origin of entanglement. This talk will be based on an explanation of the idea for general audiences available at https://arxiv.org/abs/2212.06986 , itself based on a suggestion we first made in arXiv:2101.05370v4.