FitBit Margin of Error Calculated

keithcw_the_first

Okay, I've given the Charge HR a chance for about three months now. That's generated a decent amount of data. I've known pretty much since day one that it's over-estimating calorie burn but I hadn't known by how much.

The Method

I've been doing bodyfat measurements regularly and eating at what I think is a deficit. That's supported by my overall weight loss. I haven't magically gained pounds and pounds of muscle in three months, so the number and measurements suggest that the small amount I've lost has been fat. That gives me confidence that there aren't muscle gains offsetting the scale weight.

Based on that I say x number of pounds lost over y period of time (times 3500) gives me a total number of calories burned over that time period. I divided that by the number of days to get an apparent daily deficit.

Then I took FitBit's measured daily activity and subtracted my calories in. I've been logging as well as I can: weighing in grams, all that. By simply subtracting my intake from the reported burn, I received another higher deficit number.

Then I just compared the two; my derived deficit based on pounds lost vs. the calculated deficit from FitBit's data dump.

The Results

FitBit's numbers would have suggested an overall loss of 16 pounds over the time period. This was demonstrably false and as I said, without any increase in lean body mass I can't say "Oh well maybe I gained soooo much muscle.

Actual fat loss over the measured period was nine pounds.

Basically, my FitBit is overestimating my daily activity by about 10%. The math is ugly but I believe it's done correctly and basically to make 16 pounds 9 pounds, I have to knock 10% off of every day's reported burn.

TL:DR
One, this restores my confidence in my Charge HR a tiny bit. I always knew it couldn't be accurate but I wasn't sure by how much.

Two, this gives me a way to manipulate the food plan feature that FitBit offers. I know that if it tells me I've hit a 1,000 calorie deficit for the day, it's probably only 900 calories. If I set it to the moderate plan (750, which is where I have it set) and I nail the goal, I'm probably only running a deficit of 675 calories. And so forth.

I can't speak to everyone's devices; this is all dependent on my accuracy of food logging, plus the source of most of my active calories (chores and strength training).

Just know that yes it is overestimating, and yes you can figure out roughly how much.

heybales

That math is indeed done correctly.
lbs of change x 3500 / days in change = deficit or surplus to logged eating levels

The only caveat is if it included the first week of starting a diet, where you have increased water weight loss.

Other caveat, is if you started working out or changed in big way the types of workouts and gained water weight, which is also a part of LBM (which is everything except fat, so water, muscles, blood, bones, organs, ect).

Other caveat, is if you started Fitbit/MFP combo having already been dieting for months on end, eating even less than this current range of stats, such that body had already adapted a tad.

So I know that HRM formula has great opportunity for over-estimating calorie burn, especially on devices you can't tweak the 2 required stats - HRmax and VO2max.

Any idea in that time span how many hours of exercise you did that would have relied on the HRM formula for calorie burn, and how many calories associated with those workouts?

And did you manually log weight lifting or circuit training if done since those are known inflated calorie burn by HR?

Because indeed, not far off there.

I think it would be interesting to see the possible reasons for the difference.

My own test during a 6 week period of training for triathlon, so straight cardio, manually input workouts based on personal VO2max calorie burn formula, but otherwise daily life was all Fitbit estimated - had less than 5% difference between result based and estimated TDEE. This was Zip, but likely with all workouts manually logged, could have been any device. Oh, and height was tweaked on device so the BMR it used was matching my better estimated Katch BMR based on BF%.

keithcw_the_first

I never thought about tweaking height...

So, I don't log weight lifting at all. I just use Fitocracy to track my progress there and let the fitbit give me credit for an hour with a slightly increased heartrate. The only activity I log is the odd lunchtime walk through Runkeeper, which is plugged in to FitBit.

I had been working out in earnest for a few months prior to buying the Charge, and the data I use start about a week in since my first week of owning it included post-Christmas and New Year's Eve. I would consider that to be a significant outlier.

The period in question, about 90 days, also includes me dieting in earnest. It coincides with me starting up MFP.

How would I identify which periods of time relied solely on the HRM formula? I would imagine all of it since, like I said earlier, I rarely log exercise directly into FitBit.

It gives me 200 - 400 calories for an hour of fairly vigorous strength training; I find that's usually pretty spot on. I suspect the inflation comes from the rest of the time that I'm moving around but not actively exercising.

The difference I'm sure isn't a flat 10%. There's probably a way to weight the discrepancy by looking at my Active Calories but I'm not sure how I'd go about doing that or whether it's useful beyond Excel Nerding. Which I enjoy.

I have seen a few drastic swings in water weight but typically no more than a few pounds. I'm hoping that those are smoothed enough over the sample size to not really matter.

heybales

Yeah, the water weight fluctuations would only matter if just prior to or after a weigh-in.
Like if a totally different routine like lifting was started 3 days prior to last weigh-in, causing big water gains.
Or was sick and eating lower than normal sodium or had huge cardio workout day before first weigh-in.
Something massive like that would probably only be 5 lb max difference, so over that long time hopefully not significant.

The HRM calorie burn is used when you start an activity record by hitting the button, and your steps stay high enough for it to know this isn't just a stress elevated HR by itself.
That will auto happen too if HR and steps goes up past a certain point (not heard what it is, probably based on HR zones which are based on age).
During those times, HR is read per sec, instead of per 5 sec. And calorie burn based on the HR instead of steps like other time.

So to pull that, you'd have to have activity records for all workouts - if not done, not an easy way I'd suggest doing it now, not worth it really, and may not matter much.

But that lifting actually sounds like decent estimate. By METS database, heavy lifting is about 3.5 METS, or 3.5 x resting calorie burn, which is shown in your daily burn graph per 5 min of non-moving time.

If you know your start and end time of last workout, you could go create manual activity record and see what it had as stats during that time, including HR, and if per sec readings, you know what it used.
Then create a manual workout log same start time and correct duration, select Weights, and see what the database would have said for that time. Don't need to save it.
That won't replace the record stats, just the daily stats.

What I actually did with the height, was make Fitbit using a sleeping calorie burn BMR that is actually equal to the average of my Katch BMR and Cunningham RMR based on decent bodyfat %.
I figured since 2/3 or longer of my day was NOT sleeping and probably half of that time was sitting burning RMR level calories, I'd just take the average of the 2 figures. Because all moving time is based on resting calorie burn and other math.
Currently about 6 ft 7 in - though I really don't feel like it, and my basketball game hasn't improved at all because of it.