Data Science x Public Health

Model averaging is often presented as a more careful and uncertainty-aware alternative to choosing one model specification. It is supposed to reduce overconfidence and make analysis more robust. But what if all the models being averaged share the same blind spots from the start? In this episode, we break down why model averaging often overpromises, how shared structural weaknesses survive the averaging process, and why uncertainty cannot be handled simply by blending similar models.&nbs...

Censoring is one of the most common assumptions in epidemiology and survival analysis. It is often treated as a routine technical step for handling people who leave observation before the study ends. But what if leaving the study is not random noise—and is actually part of the outcome process itself? In this episode, we break down why censoring assumptions often fail, how loss to follow-up can distort longitudinal research, and why disappearing from the dataset is not the same thing as ...

Resource allocation models are supposed to help public health systems distribute scarce resources more intelligently. They promise better targeting, more efficient deployment, and stronger impact under constraint. But what if the model is optimizing inside a system whose deepest constraints should never have been treated as fixed? In this episode, we break down why resource allocation models often fail in practice, how optimization can normalize structural scarcity, and why better public he...

External validation is often presented as the gold standard for proving that a predictive model works beyond its original dataset. It is supposed to show that the model can generalize to the real world. But what if one external dataset is still far too small a test of the outside world? In this episode, we break down why external validation often overpromises, how “different” datasets can still be too similar, and why transportability is a much harder claim than validation language sugg...

Competing risks methods are often presented as a more realistic way to analyze time-to-event data in epidemiology and public health. They promise to handle situations where other events prevent the outcome of interest from ever occurring. But what if the method becomes more sophisticated while the interpretation becomes less clear? In this episode, we break down why competing risks analyses are often overtrusted, how the choice of estimand quietly changes what the result means, and why ...