In academia calorie burn gets measured about the produced heat in a closed system. How accurate is the Fitbit Charge 2 when compared against that gold standard?
The calorie burn estimates that Fitbit provides takes into account the person's basal metabolic rate based on gender, age, height, and weight as well as the step activity recorded and any activities they log manually.
My impression is that the calorie burn reported by Fitbit trackers is really just a ballpark estimate and those in the Fitbit community often question the accuracy and consistency of the calculations. An article in Medicine & Science in Sports & Exercise stated that a study by Ball State University in Muncie, Indiana showed Fitbit and other trackers were highly inaccurate regarding calories to burn especially on non-step based activity, as much as 50% off.
To me, this seems close to being an X-Y question. The medical professionals (and to some extent the laboratory equipment manufacturers) are pushing back against these low-cost, continuous measurment sensors, because they're not accurate - but failing to appreciate that there are probably other gains to be made from replacing high precision with high volume data (both in time, and across populations).
In common with lots of embedded applications, the Fitbit doesn't measure calorie burn directly. What it does it take some more tractable measurements and user data to provide an estimate based on identified modes of activity (for example -sleeping, jogging, walking). Across a group of individuals, there is potential for significant variation between the observed parameters and actual calorie burn.
Where the data can be expected to be more accurate is in comparing relative calorie burn with different days for a specific individual, and the obvious advantage is the trivial instrumentation overhead compared with attempting any precision calorimitry in an open-air activity.
Its important to appreciate that the calorie number is an estimate, but when does anyone care about anything precise with calories? Even with blood pressure, the value is being able to identify changes over time in the majority of even medically monitored applications (rather than one-off diagnostic measurements).
Some effort does need to be made to formalise the value of herd measurements like this, educate professionals to work with low quality data, and educate users to appreciate this cheap data is not always individually informative.