Mathematics graduate working on statistical inference in epidemic models — specifically on how much of a published conclusion survives the choices its author never wrote down.

Independent researcher, Lahore, Pakistan. Currently applying for funded Master's study in applied / computational mathematics.

dengue-multiverse — Does the evidence for climate-driven dengue transmission survive the analyst?

221 dengue outbreaks across 33 countries, each fitted under all 144 combinations of six routine analysis choices — 33,152 model fits. Roughly three-quarters of the variation in whether climate forcing is endorsed lies between analyses of the same outbreak, not between outbreaks. Most of that is an outbreak × analysis interaction that no standardised convention can remove.

The paper does not stop at the diagnosis: combining eight analyses by Rubin's rules turns a 95% confidence interval that covers 71% of the time into one that covers 97%, for a few minutes of computation.

Python · SciPy · compartmental models · maximum likelihood · multiverse analysis

I keep a dated analysis log in every project recording the formulations that failed, the hypotheses that were tested and rejected, and the bugs — including the ones that produced plausible wrong numbers rather than crashing. A result is easier to trust when the route to it is visible, including the wrong turns.

Where a paper quotes a number, a test checks that number against the pipeline that produced it. If the manuscript drifts, the test suite fails.

Background: BS Mathematics (University of the Punjab) · 3.5 years teaching mathematics at college level · data analyst

Interests: mathematical epidemiology · numerical methods for differential equations · model selection and identifiability · reproducible research

Corrections and objections to anything here are welcome — open an issue.