Your weight consists of two parts: what actually changes, and what happens to be added on any given day. In signal processing the first is called signal and the second noise. The scale always shows you both together, and at one to two kilos the noise is considerably larger than the weekly change you actually want to see.
A moving average separates the two. That is the whole trick, and it is an old one: Happy Scale, Libra and John Walker's Hacker's Diet have worked on the same principle for years.
How the 14-day window works
The 14-day average is one of the methods Trendcurve offers: the average of all daily values from the last 14 days, today included. Every day the window moves one day forward: the new value comes in, the oldest one drops out.
An example with heavily rounded numbers. Suppose your last 14 daily values range between 83.4 and 85.6 kg and add up to 1176.0 kg. Divided by 14 that is a trend weight of 84.0 kg. If an outlier of 85.9 kg arrives the next day while an old value of 84.5 kg drops out, the trend shifts by (85.9 − 84.5) / 14 = 0.1 kg.
That is the decisive point: a day that shoots up by 1.4 kg moves the trend by one tenth of a kilo. The noise is not gone, it is compressed to one fourteenth.
If a day has no measurement, Trendcurve simply calculates with the values present in the window. Gaps therefore do not break the trend, they only make it a little jumpier because fewer values carry the average.
Which method does the app use?
Since version 1.1 you pick the method yourself. The default is exponential smoothing: it gives the most recent daily value the largest weight and older ones steadily less, instead of treating fourteen values alike. The result is similarly calm but reacts faster to real change. Next to it are the 7 and the 14 day average and double exponential smoothing, which also carries the direction along and bridges gaps.
Everything said here about smoothing, lag and the 7-day change holds for all four methods. They only differ in how much calm they trade for responsiveness. Under "Settings & more → Trend algorithm" every method comes with its pros and cons and today's trend value, so you can see the difference on your own data.
The 7-day change
The trend weight alone tells you where you stand. Direction comes from the 7-day change: trend weight today minus trend weight seven days ago.
This number is more robust than any direct comparison of two daily values, because both sides of the subtraction are already smoothed. Mathematically it equals the average daily change over a week, which is exactly what people mean colloquially when they say "I am losing half a kilo a week".
| What you see | What it means |
|---|---|
| Trend weight 84.0 kg | Your smoothed position today |
| 7-day change −0.45 kg | Your trend sits 450 g below the one from a week ago |
| Daily value 85.3 kg | Today's measurement, noise included |
Why the trend "lags behind"
This is the most common irritation, and it is not a flaw but the direct consequence of averaging. An average of the last 14 days naturally describes the middle of that period rather than its end.
So while you are losing weight, the trend weight sits above today's reading, because heavier days are still in the window. While you are gaining it sits below. That lag amounts to roughly one week.
It can only be shortened at the price of more noise. A 7-day window reacts faster but jumps around considerably more; a 30-day window sits almost motionless and shows an improvement only when it is already two weeks old. Fourteen days is the compromise that tolerates one or two outliers a week and still responds to real change within a few days.
From trend to forecast
Once the trend line is calm enough, you can fit a straight line through it. Its slope is your weekly rate, and with it the history turns into a prediction: with a goal of 78 kg, a trend of 84 kg and 0.45 kg per week, 6 kg remain, so roughly 13 weeks.
With the raw daily value that calculation would be worthless. The slope through noisy points depends almost entirely on which day you happen to catch as the start and which as the end. Only smoothing makes the rate stable enough to derive a date from it.
That is also why Trendcurve says nothing when the basis is insufficient: if the trend is barely moving, if it is running away from the goal, if there are too few measurements or if the last one is more than three days old, you get the reason instead of a date. An invented target date would be more comfortable and worthless.
How to read the trend day to day
Look at the daily value but do not judge it. It is the raw measurement, nothing more. It is allowed to be high.
Make decisions based on the 7-day change. Whether the calories are right, whether a change is working, whether a break is needed: that is decided by the smoothed week, not by Tuesday morning.
Give every change two weeks. Because the trend lags by about a week, a dietary change can only be judged fairly after ten to fourteen days. Adjusting after three days means correcting noise.
Expect plateaus. Even with a perfectly maintained deficit the trend can run sideways for two to three weeks, for instance because retained water is masking fat loss. What makes sense then is covered in How fast should you lose weight?.
Why the noise is so large in the first place and where it comes from is explained in Why your weight fluctuates every day.