Are your fitbits accurate?

I find mine is spot on. I got the Blaze in May and have lost about 12 pounds, using the sync to MFP.

FYI, I've been on MFP for years and using the fitbit has kept me on track and eating enough that I don't feel like I'm starving myself. I'm really happy with it.

Mine seems to over estimate my TDEE. I've been eating at around a 100-350 calorie deficit (negative adjustments enabled, estimated TDEE about 2000-2400 depending on how much I walk), and I ended up maintaining my weight between endo appointments (a 6 month period). I did lose at least one dress size though, so I could have lost a little bit that was hidden by water weight.

I have the fitbit hr charge. I had to alter stride length and set to dominant hand and adjust again and again until I got it the closest to being accurate. I use the HR to track my circuit training once to check the calorie burn compared to mfp estimate and it was 100 cal less than mfp so I log it that way now. The sleep feature is neat, however, it logged me as asleep during a tattoo session. Most of the features I just use as a guide and fun motivation to keep myself moving and getting outside. I work 3rd shift so this has helped me come out of my daytime hermit mode.

Mine seems to be accurate. I have the alta. I have it synced to mfp and absolutely eat the calories it gives me. I am still losing a pound weekly and never feel hungry.

A very interesting question and a large number of issues are at play too!

TL;dr: accuracy will vary and a number of factors come into play... but will be relatively close for most. Eat your Fitbit calories and evaluate how badly off they appear to be. As long as you're generally heading in the right direction don't agonise too much over how fast you are achieving that: too much could be in play.

A standard way to estimate caloric consumption is to apply an activity multiplier to BMR calories.
Fitbit (just like MFP) uses the Mifflin StJeor equation to estimate your BMR calories.

MFP then uses BMR x 1.25 as an estimate for the calories of a sedentary individual, 1.4x, 1.6x, and 1.8x for lightly active, active, and very active individuals plus deliberate exercise added on top.

Fitbit assigns you just BMR x 1 calories while you're inactive. This is an expected underestimation. Most sitting/standing/keyboarding/watching TV activities come in at BMR x 1.2 or 1.3, or more.

But, Fitbit then detects your activity and assigns you more calories while you're active. You either tell it, or it guesses, at the type of activity you are doing.

I suspect that when you tell it that you did a non step related activity, it just uses the standard compendium of physical activity tables, just like MFP and many other apps do, to estimate your burn.

But when the Fitbit has hrm and accelerometer data of its own it for the time period in question, it probably goes with that!

I believe it averages your burn over a 5 minute period and then issues calories for the 5 minutes based on that average. This average appears to me to be a slight over-estimated for most people.

But most people are only active for part of the day. And when you combine the active and inactive parts of the day... lo and behold Fitbit seems to come in quite close, while encouraging people to move a bit more

But, even assuming the Fitbit was estimating perfectly for you, would you be able to tell?

First of all your calories in would have to be counted perfectly! How do you know, were you to spot a 5% error, whether it originated from Fitbit, or from your calorie counting when you mis-judged the calories for a restaurant meal, or forgot to log the couple of beers you enjoyed at a BBQ?

Then you would have to detect your weight accurately. Because of water weight (sodium, TOM, exercise related water retention) your scale weight can change 10x faster than your underlying weight level. Are you evaluating the accuracy of your Fitbit's performance based on transient water weight? Are you using a trending weight app? is the trending weight app actually accurate?

And then you would have to know your actual body composition changes.

Because yes, we can probably assume that 3500Cal = 1lb of fat change whether up or down as fat gets stored and used up very efficiently. However, 1lb of lean mass may only yield 800Cal when burned; yet, it could take 4x to 6x more Calories in order to create 1lb of muscle mass.

In practical terms what could body composition issues mean for our results?

Well, let's take a normal weight individual who just lost 4lbs on their scale this month and is ready to compare their logged calories in to their Fitbit calories out.

4lbs... this means a 14000 Cal deficit (heck my own spreadsheet below uses this assumption because you cannot actually account for body composition most of the time!)

But if 1lb of this is a water weight error, the 14,000 was only 10,500 Cal.

And if they lost at a fat to lean mass ratio of 2:1, 1.5:1, or 1:1 (all absolutely possible outcomes when restricting calories within the normal weight range), the respective caloric deficit might have been 8600, 9680, or 10400 calories.

And that doesn't even tackle the common situation where there might be a 1.75:1 fat to lean mass loss; but, even though some lean mass is lost, you also have to account for the creation of 0.25lbs of new muscle on say the arms from doing 100 push-ups a day when in the past you did zero! because that 0.25lbs by itself may have required 4000 Cal to create!

So in the end... well, you don't really know if your 4lbs was 14,000 cal or 8600 cal

Which is where my original piece of advice came in... as long as you're heading in the right direction that's good enough... and in fact I would argue that slower is better.

And in spite of all these uncertainties, yes, I did use to use something similar to the below spreadsheet to "evaluate" my Fitbit's accuracy. Since it has been stable at an under 6% over-estimate for more than a year... I've sort of stopped looking into it and take it as a given :-)

And yes, it does count extra steps when I keyboard and it does under-count steps when I walk really fast and it is also slow to react to heart rate changes... and it still keeps coming in with a substantially less than 10% error (in fact less than 6% error in my case).

https://docs.google.com/spreadsheets/d/1VDmqNpLPu7sbQSochUJNXdp2F7AN15AGgkvS3zLw1GU/edit#gid=0