An Estimate With No Error Bar Is a Guess in a Suit
Adaptive maintenance calories come with an interval showing how far the estimate could be off. A narrow one means your scale data is consistent, not that the number is right.
- The estimate is only as reliable as the weight trend underneath it.
- The interval comes from the standard error of the fitted trend, widened for small samples with the t distribution.
- Gaps in your food logging add a second margin, and logging every day of the window removes it.
- Under-reported intake is the largest error in practice, and no interval can measure it.
Same number, different certainty
Adaptive maintenance calories are worked out by reconciling what you logged against how your weight actually moved. That makes the figure only as reliable as the weight trend underneath it, and a trend fitted through eight tightly clustered weigh-ins is a very different object from one fitted through four scattered ones, even when both produce the same headline number.
How the interval is built
Kriterion reports the second thing as well: the standard error of the fitted trend, which is how far the slope could plausibly be off given how scattered the points are, converted into calories and scaled by how much the estimate actually leans on that trend rather than on the formula. With only two weigh-ins it reports no interval at all, because a line drawn through two points fits them perfectly and an interval of zero would claim certainty rather than admit ignorance. Small samples are handled with the t distribution, the version of the usual bell curve that widens as the sample shrinks, rather than the familiar 1.96. Four weigh-ins carry a critical value above four, and rounding that down would advertise confidence exactly where there is least of it.
Gaps in your logging
The weight trend is not the only input with something left over, though, so the interval carries a second term: your average intake is a mean over whichever days you logged, and where the window has gaps that average has its own margin, which is added in alongside the trend's. Log every day of the window and that term disappears entirely, which is the correct answer rather than a convenient one: the days you logged are not a sample of some larger population, they are all of the days there were.
What it leaves out
What the interval deliberately leaves out is under-reported intake, which is the largest error in practice by some distance. It is systematic rather than random, so it shifts the estimate instead of widening it, and nothing in the logged data can measure it. A narrow interval here means your scale data is consistent. It does not mean the number is correct.
Kriterion prints that interval next to your maintenance figure, so you can see whether it is a number to plan a week around or one to keep logging into.
References
- Hall KD, Sacks G, Chandramohan D, et al. Quantification of the effect of energy imbalance on bodyweight. The Lancet, 2011;378(9793):826-837.
- Thomas DM, Schoeller DA, Redman LA, et al. A computational model to determine energy intake during weight loss. American Journal of Clinical Nutrition, 2010;92(6):1326-1331.
- Dhurandhar NV, Schoeller D, Brown AW, et al. Energy balance measurement: when something is not better than nothing. International Journal of Obesity, 2015;39(7):1109-1113.
- Draper NR, Smith H. Applied Regression Analysis. 3rd ed. Wiley, 1998 (standard error of the slope, weighted least squares).
Read next
Kriterion is the app this reasoning is built into
It measures your maintenance calories from what you actually ate against what the scale actually did, tells you how firmly the number is pinned down, and checks its own past predictions against your results. Free logging, UK food data, and no ads.
Join the list and you get the founding offer: the launch price, held for as long as you stay subscribed. It is capped at the first 1,000 subscribers or 90 days from launch, whichever comes first, and it will not be extended. One confirmation now, one email on launch day. Not a newsletter.