Microbial source tracking — a censored-data pipeline
Host-associated Bacteroidales qPCR — HF183, DG3, Rum2Bac, GFD, CGOF — resolves fecal contamination to its source where the regulatory indicator cannot. The analytical difficulty is that the resulting data are heavily left-censored, and the common workaround is to discard the non-detects.
A reproducible pipeline implements EPA Method 1696 and Helsel censored-data conventions, propagating censoring flags through every downstream computation rather than dropping them: half-LOQ substitution recovered 27.7% of observations that a detected-only convention would have thrown away, concentrated in the low-detection markers where power is weakest. Continuous antecedent-precipitation classification, censored Kendall τ in place of Pearson r on substituted data, 10,000-iteration bootstrap intervals and permutation nulls, and watershed-scale land-cover covariates. Raw qPCR workbooks in, stakeholder deliverables out.
MPH practicum · findings not public