EPR vs LRR: which shoreline change rate should you report?

If you've pulled shoreline positions into DSAS and run the rate statistics, you've seen both EPR and LRR sitting in the same output table for the same transect, often disagreeing by a meter a year or more. One stakeholder wants "the rate." Picking which column to hand over, and why, is a judgment call baked into every shoreline change report.
What EPR actually measures
End Point Rate is the simplest math in the toolbox. Take your oldest shoreline position and your newest one at a transect, measure the distance between them, divide by the number of years. Two dates, one number.
The appeal is obvious. A county commissioner can follow it without a statistics background, and it uses dates you probably already have: an old aerial from a flood study, whatever you flew last year. The catch is that EPR throws out every shoreline position in between. If your baseline year happened to catch the beach right after a storm, or your end year caught it mid-recovery from one, that single year's noise becomes your entire 20- or 30-year trend.
What LRR actually measures
Linear Regression Rate fits a least-squares line through every shoreline position you have at that transect and reports the slope. Five dates, ten dates, it doesn't matter. Each position gets weighted into the fit, so one anomalous survey year pulls the line a little but doesn't define it.
This is why USGS and most academic shoreline studies default to LRR once there are more than two or three dates available. It's a better statistical description of long-term trend, and it comes with an r-squared you can report alongside the rate to show how well the line fits the scatter of positions. A low r-squared is worth stating plainly in the report, since it means the shoreline at that transect isn't moving in one consistent direction.
The tradeoff is data. LRR is only as good as the number and spacing of shoreline positions feeding it. Two widely spaced, noisy shorelines run through a regression will spit out a number that looks more rigorous than EPR but isn't built on any more information.
Picking between the two for a given stretch of coast
The rule most practitioners work by: with two dates, you're really doing EPR whether or not you call it that. With four or more, run LRR and report the r-squared alongside it. DSAS computes both, plus WLR and a few weighted variants, in the same batch, so there's rarely a reason not to generate all of them and let the number of usable shoreline positions decide which one goes in the final report.
The input data is where most shoreline change reports actually break down, long before the rate statistics run. LRR needs annual, or near-annual, shoreline positions, digitized the same way every time, to earn its rigor. A lot of coastal programs end up running LRR on positions pulled from three inconsistent sources: an old T-sheet, a LiDAR-derived line from one vendor, a hand-digitized line from aerial photos, each on a different vertical datum and a different waterline definition. The regression math doesn't know the inputs are apples and oranges. It just fits a line through whatever you feed it.
That's the gap a year-over-year waterline layer closes before the rate statistics ever run. A consistent extraction method applied to the same annual high-res pass, segment by segment, gives LRR the kind of evenly spaced, comparably derived shoreline positions it was designed for, instead of a patchwork of surveys stitched together after the fact.
If your next shoreline change report needs rates that hold up to that kind of scrutiny, it's worth seeing what a clean annual waterline record looks like for your stretch of coast.