One Curve Fit to Rule Them All

Collaborative Efforts to Improve Environmental Monitoring
Oral Presentation

Prepared by R. Burrows
Retired, 5625 Youngfield St, Arvada, CO, 80002, United States


Contact Information: [email protected]; 303-885-3348

ABSTRACT

Back in the bad old days, many assessors and data validators would opine that regressions for calibration had to be linear and that weighting, if not illegal, was certainly immoral and not to be countenanced. People working in the lab complained that this requirement had nothing to do with data quality. Instead, this simple regression was the only one the assessors could get to work in their Excel spreadsheets. (This may or may not have been true.)
Since then, things have loosened up a bit and one can see all kinds of weighting and creative use of curvature. A European-built instrument I saw recently had options for cubic spline, Logit-Log 4 and Logit Log 5, amongst other exotic offerings.
What if we could take a page from the old days, and allow only one type of regression, but instead of choosing the worst possible option (linear unweighted) chose the best possible? We could, and that curve fit is:
Going to be revealed in this presentation, along with the rationale and plenty of examples.