Attempting to Calculate my TDEE, Math heavy
Aaron_K123
Posts: 7,122 Member
So I have been logging my weight and my calories burned for close to 10 weeks now. I don't think this is truly enough data to get a good handle on my weight loss per week nor my TDEE but I thought I'd give it a try anyways just to see what it looked like.
So first my weight loss by day. I try to measure my weight everyday but not always at home with my scale. Here is my weight over time.
Of course as people note on these forums weight fluctuates up and down and so even for 10 weeks of logging the R-square is an abysmal 0.65. Can clean that up a bit by simply turning it into a moving average whereby each point is the average of a week along with its standard deviation.
That brings the R-square to 0.83 which is still not great but might be workable. The trend value establishes a loss of 0.68 pounds per week.
Now I take my data for caloric intake for that same time period, use the trend data from my weight to estimate the amount of weight I have truly lost over that period of time (try to unmask it from the fluctuations). Taking the amount of weight I lost multiplied by 3500 calories and adding that to my total calorie intake for that time should give my maintenance calories which divided by the total days would yield my TDEE.
Giving a value of about 2200 calories per day which given my activity and size seems a bit small.
I have noticed that when I started tracking my weight there seems to be two trends. In the first 5 weeks I just stall at the same weight but then in the last 5 weeks a new trend has established with a much faster weight loss. Of course I may just "stall" again but in case this new trend is more accurate I tried the same process for the most recent month.
Moving average:
So trend gives a much different 1.1 pounds per week loss with a very nice R-square of 0.93
Calculating TDEE:
That gives what I would consider a much more reasonable value of around 2600 calories per day.
Overall I think it is too early to tell. Perhaps after I have logged for about 20 weeks I will be able to lock this down. This is important to do I feel because once I lose the weight I will want to move into maintenance and knowing what my TDEE is, not based on online calculators but based on my own body, will be important for that.
Just seems like it takes a lot of data to get a feel for it.
So first my weight loss by day. I try to measure my weight everyday but not always at home with my scale. Here is my weight over time.
Of course as people note on these forums weight fluctuates up and down and so even for 10 weeks of logging the R-square is an abysmal 0.65. Can clean that up a bit by simply turning it into a moving average whereby each point is the average of a week along with its standard deviation.
That brings the R-square to 0.83 which is still not great but might be workable. The trend value establishes a loss of 0.68 pounds per week.
Now I take my data for caloric intake for that same time period, use the trend data from my weight to estimate the amount of weight I have truly lost over that period of time (try to unmask it from the fluctuations). Taking the amount of weight I lost multiplied by 3500 calories and adding that to my total calorie intake for that time should give my maintenance calories which divided by the total days would yield my TDEE.
Giving a value of about 2200 calories per day which given my activity and size seems a bit small.
I have noticed that when I started tracking my weight there seems to be two trends. In the first 5 weeks I just stall at the same weight but then in the last 5 weeks a new trend has established with a much faster weight loss. Of course I may just "stall" again but in case this new trend is more accurate I tried the same process for the most recent month.
Moving average:
So trend gives a much different 1.1 pounds per week loss with a very nice R-square of 0.93
Calculating TDEE:
That gives what I would consider a much more reasonable value of around 2600 calories per day.
Overall I think it is too early to tell. Perhaps after I have logged for about 20 weeks I will be able to lock this down. This is important to do I feel because once I lose the weight I will want to move into maintenance and knowing what my TDEE is, not based on online calculators but based on my own body, will be important for that.
Just seems like it takes a lot of data to get a feel for it.
0
Replies
-
I calculate my TDEE using 3 weeks worth of data...
It has worked for me not sure why you feel you need extra time in there...if it's for accuracy..eh..
Total calories consumed+(pounds lost x 3500)/21 should give you a good TDEE number...
no need for moving averages or standard deviations etc.
too complicated for this blond head anyway...
ETA: I did this on Sept 8th and it worked out to 1995, then I ate on average TDEE-15% and lost on average 3/4lb a week..and that has worked up until now and as you know I haven't plateaued, stalled or had any issues what so ever.
I redid the calculation recently because I am down 10lbs since the last time and it workout out to 2250 (7 months of lifting has apparently been good to me..) so that is my maitenance number...I am at 1800 atm just upped another 100 this week.0 -
I calculate my TDEE using 3 weeks worth of data...
It has worked for me not sure why you feel you need extra time in there...if it's for accuracy..eh..
Total calories consumed+(pounds lost x 3500)/21 should give you a good TDEE number...
no need for moving averages or standard deviations etc.
too complicated for this blond head anyway...
ETA: I did this on Sept 8th and it worked out to 1995, then I ate on average TDEE-15% and lost on average 3/4lb a week..and that has worked up until now and as you know I haven't plateaued, stalled or had any issues what so ever.
I redid the calculation recently because I am down 10lbs since the last time and it workout out to 2250 (7 months of lifting has apparently been good to me..) so that is my maitenance number...I am at 1800 atm just upped another 100 this week.
I feel I need more data because if you chose any random 21 day period within the total amount I have logged you will get very different values for TDEE because of the weight fluctuations. I'm trying to get enough data to get the trend into a good r-square (high confidence) while having sufficient amount of points over a sufficient amount of time. Don't know about you by my weight can vary as much as 4 pounds randomly day to day so if I just pick some arbitrary 21 day window I might be very inaccurate in my TDEE estimate. This error is evident in my calculations as well, I pick the full period I get a TDEE of 2200...I pick the last month I get a TDEE of 2600. That is very different.
Think of it this way. I weight myself on day 0 and I'm 188 pounds. Am I retaining a lot of water that day? Am I dehydrated? Who knows. Then 21 days later I weigh myself and I am 185 pounds. Am I retaining a lot of water that day? Am I dehydrated? Who knows. If I use those numbers it'll tell me I've lost 3 pounds and that's 1 pound per week but I've only used 2 data points, one on Day 0 and one on Day 21, which makes the error HUGE. After all perhaps I have put on 1 pound of fat while losing 4 pounds of water when I weighed on day 21, can't tell. How can you ever tell? Well the more data you include in an average to establish a trend the lower the error. With 10 weeks I still have pretty high error so I'm thinking 20 weeks might do it.
As another example right now my ticker says I have lost 9 pounds. Yesterday it said I had lost 6 pounds. This is because my weight fluctuates. By taking this moving average I can get a feel for the idea that my "genuine" weight loss over this time has been closer to 6.5 pounds which tells me that likely when I weighed this morning I was a bit dehydrated. But I don't have enough data to say with high confidence that I have lost 6.5 pounds of fat yet, error is still probably plus or minus 2.0 -
Realize I posted a big wall of text there so the TL:DR version:
"Total calories consumed+(pounds lost x 3500)/21 should give you a good TDEE number..."
This formula is correct but it is only as accurate as your measure of pounds lost. As we all know your weight fluctuates quite a lot so it can take a long time to establish the trend to determine how much weight you have actually lost as opposed to a random dip or a random spike. Once you've lost like 50 pounds okay that 3 pound fluctuation doesn't matter much but if you have lost 6 pounds that 3 pound fluctuation is huge. If you base that pounds lost on just 2 datapoints (Day 0 and Day 21) you will have a huge error. After all your weight fluctuates by pounds per day and over 21 days you lost something like 2 pounds so the error bigger than the measurement. If you base it instead on the trendline fit to all collected datapoints during that time you have less error. If you then extend that time further and include more data your error will be even less. Longer you log, more data you have, the more accurate that "pounds lost" will be and the more accurate your TDEE estimate will be. That's all.
All I mean for myself is having lost only 7 pounds and logged for only 10 weeks I think the error in my "pounds lost" estimate is still quite large so I can't really pin down my TDEE yet.0 -
Yah I just don't want to over complicate it...
If I take all my data, holidays and vacations included (where i couldn't track but guesstimated) I have a TDEE of 2100...
but that also doesn't take into account the fact as I have lost weight I have become more active...
When I started lifting lets say I was only lifting 3x a week that was it...since the new year I have added in HIIT 2x a week and now that it's summer I am not doing as much HIIT but walking a couple miles on my non lifting days and will be adding in biking and swimming in the summer...
there are too many variables over a course of that large of a period of time to get an accurate read...but that is why even using our own data there has to be room for error.
My actual calculation over the past 3 weeks is 2250...an increase of 255 to my TDEE seems a bit high so I am going to aim for 2000 as maintenance and see what happens...if I lose I will up it by 100...and see how I do for 1 month...you know the drill...
and it's a moving target all the time anyway...activity goes up in my summer and fall, could go down in the winter (if I let it) etc...
I think this way gives one of the best estimates I have seen esp considering how well it has worked for me over the last 7 months.0 -
Fair enough, I mean you are right that if you change your activity level or your metabolism over time then things will change and if you take all the data that won't show you the changes that have been happening only in the last few months. I just mean that the longer I log the more data I will have to work with the better feel I will get for my TDEE. I think right now that 1 pound a week loss and a TDEE of 2600 is probably relatively accurate for me, I'm just going to keep on doing what I'm doing because it seems to be working and the longer I do it for the better idea I will have of "how" it is working so I can adjust later if need be.0
-
Yup me too..working my way up to 2000...stay there for a bit and if I continue to lose over that month another 100 gets added...and I will just keep tabs on my BF%, weight and measurments...
I personally am looking forward to it...although it's a bit scary at times...lifting at maitenance, possible recomp to get to 20% BF..and who knows a bulk at that time...
Good luck to you...
PS this post is so math heavy it crashed my internet....:laugh: :bigsmile: :laugh:0 -
This is why I use a trend line. I use the old formula from the Hacker's Diet, for computing it on my spreadsheet. That formula is [(today's_weight - previous_day_trend) * 0.10 + previous_day_trend]. By using the weighted trend-line, daily fluctuations are flattened out. Even a recent 3kg scale jump (water weight from a night drinking) only caused my trend-line to bump up 0.21kg... basically nothing, especially when looking at 30 day periods.
What is nice about this? It gives me a much more accurate picture of rate of change over time. For example, using rolling 30 day periods and just my scale weight, I get r^2 values of:
Max: 0.92105
Min: 0.30499
Average: 0.67470
Basically, change is weight over 30 days is pretty much useless. But, over those same 30 days, my trend gives me r^2 values of:
Max: 0.99568
Min: 0.90567
Average: 0.96692
All of these tell me that using the trend is going to give me a fairly accurate picture of actual weight lost. That means my TDEE is going to be better. I originally was doing 10-day rolling average (I still have them) but they are less sure. The r^2 ranges from 0.69464 - 0.99189 {average: 0.89957}. That's still better than scale weight, but it's nothing compared to the trend-line.
Anyway, if you use this sort of trend, 10 weeks is plenty of data to get a reasonably accurate estimate. As your number of days increases, the estimate will improve (if you want it to). I prefer to limit my calculations to the last 30 days, so that I can see if my TDEE changes with weight loss. But, I could go back and use every data point (it actually gives me essentially the same number (2,648.7 calories) as my typical 30-day (today is 2,633 calories for example)).
That's not to say my calculated TDEE doesn't vary. For various reasons, my TDEE does vary around 2,650 +/- 150 calories. This is related to temporary stalls on the scale or possibly increases/decreases in average amount of physical activity over the 30 days being compared. But, in any case, that estimate is pretty solid.0 -
This is why I use a trend line. I use the old formula from the Hacker's Diet, for computing it on my spreadsheet. That formula is [(today's_weight - previous_day_trend) * 0.10 + previous_day_trend]. By using the weighted trend-line, daily fluctuations are flattened out. Even a recent 3kg scale jump (water weight from a night drinking) only caused my trend-line to bump up 0.21kg... basically nothing, especially when looking at 30 day periods.
What is nice about this? It gives me a much more accurate picture of rate of change over time. For example, using rolling 30 day periods and just my scale weight, I get r^2 values of:
Max: 0.92105
Min: 0.30499
Average: 0.67470
Basically, change is weight over 30 days is pretty much useless. But, over those same 30 days, my trend gives me r^2 values of:
Max: 0.99568
Min: 0.90567
Average: 0.96692
All of these tell me that using the trend is going to give me a fairly accurate picture of actual weight lost. That means my TDEE is going to be better. I originally was doing 10-day rolling average (I still have them) but they are less sure. The r^2 ranges from 0.69464 - 0.99189 {average: 0.89957}. That's still better than scale weight, but it's nothing compared to the trend-line.
Anyway, if you use this sort of trend, 10 weeks is plenty of data to get a reasonably accurate estimate. As your number of days increases, the estimate will improve (if you want it to). I prefer to limit my calculations to the last 30 days, so that I can see if my TDEE changes with weight loss. But, I could go back and use every data point (it actually gives me essentially the same number (2,648.7 calories) as my typical 30-day (today is 2,633 calories for example)).
That's not to say my calculated TDEE doesn't vary. For various reasons, my TDEE does vary around 2,650 +/- 150 calories. This is related to temporary stalls on the scale or possibly increases/decreases in average amount of physical activity over the 30 days being compared. But, in any case, that estimate is pretty solid.
Yeah I think what you are describing is a moving average where you take a time-period window, like 21 days, and shift it by 1 day logging each 21 day value for each day and plotting that. I sort of did that using just the week long averages with standard deviation and trend fitting to that. If this specific method is giving you r-squares that high though that's awesome. I think I did something similar to what you described with my weekly average over the last month and that's where I got my r-square value to 0.93. I'll have to check out the Hackers Diet...is that a website?0 -
This is an excellent conversation, thank you for this! Knowing my personal TDEE is really going to help me in the next couple of months!0
-
oh math- I have a stats course, its the last one of my masters studies and it starts in july and i am dreading it, i hate hate hate math!
The TDEE stressed me out big time- the people that say "eat at ____ calories for a few weeks and ive you have gained, eat less, if you stayed the same, bingo you found your maintenance..." bug me, cause like you- my weight fluctuates MASSIVE amounts over the course of 3-5 days, there can be a difference of 11lbs
I have type 1 diabetes and I am a distance runner, I often deplete my glycogen stores and that can leave with me a 5lb loss easily
The i could be dehydrated from the run and really weigh a lot less...
Or, I could have had a large carb loaded pre race day meal, high sodium treats, done heavy weights, and be up 9lbs lol
Averages and me dont mix, my body is anything but average.
I got a fitbit- I still dont believe the TDEE on it- but I think its a good estimate and its better than any "average" these online calculators give me!0 -
Yeah I think what you are describing is a moving average where you take a time-period window, like 21 days, and shift it by 1 day logging each 21 day value for each day and plotting that. I sort of did that using just the week long averages with standard deviation and trend fitting to that. If this specific method is giving you r-squares that high though that's awesome. I think I did something similar to what you described with my weekly average over the last month and that's where I got my r-square value to 0.93. I'll have to check out the Hackers Diet...is that a website?
Not quite a simple moving average. The only thing that is moving is the window I am looking at. The calculations themselves incorporate all past values (to some degree). This is a form of a weighted moving average. Instead of adding up the last 21 days and dividing by 21. We are weighting each day. Today is 10%. Yesterday is 9%. Two days ago is 8.1%, Three days ago is 7.29%. Four days ago is 6.561%. Five days ago is 5.9049% ... and so on... all the way back through every day we've ever logged. This means our weight 190 days ago is still part of this average. Sure it's 0.0000002% of the sum ... but it's in there.
Of course, the last 21 days account for nearly 90% of our current number. So, it's fair to compare it to a 21 day average.
This gets into the math: http://www.fourmilab.ch/hackdiet/e4/signalnoise.html0 -
oh math- I have a stats course, its the last one of my masters studies and it starts in july and i am dreading it, i hate hate hate math!
The TDEE stressed me out big time- the people that say "eat at ____ calories for a few weeks and ive you have gained, eat less, if you stayed the same, bingo you found your maintenance..." bug me, cause like you- my weight fluctuates MASSIVE amounts over the course of 3-5 days, there can be a difference of 11lbs
I have type 1 diabetes and I am a distance runner, I often deplete my glycogen stores and that can leave with me a 5lb loss easily
The i could be dehydrated from the run and really weigh a lot less...
Or, I could have had a large carb loaded pre race day meal, high sodium treats, done heavy weights, and be up 9lbs lol
Averages and me dont mix, my body is anything but average.
I got a fitbit- I still dont believe the TDEE on it- but I think its a good estimate and its better than any "average" these online calculators give me!
I've had massive scale swings as well. That is why I use the trend line. It smooths them out. If you're having that many large swings, you may want to use averages from month to month. Like weigh yourself every day for 60 days... compare the average of the first 30 with the average of the second 30 to get an idea of how much you lose over 30 days (it would compare day 15.5 with day 45.5... if you averaged the days 1-30 and 31-60).0 -
Yeah I think what you are describing is a moving average where you take a time-period window, like 21 days, and shift it by 1 day logging each 21 day value for each day and plotting that. I sort of did that using just the week long averages with standard deviation and trend fitting to that. If this specific method is giving you r-squares that high though that's awesome. I think I did something similar to what you described with my weekly average over the last month and that's where I got my r-square value to 0.93. I'll have to check out the Hackers Diet...is that a website?
Not quite a simple moving average. The only thing that is moving is the window I am looking at. The calculations themselves incorporate all past values (to some degree). This is a form of a weighted moving average. Instead of adding up the last 21 days and dividing by 21. We are weighting each day. Today is 10%. Yesterday is 9%. Two days ago is 8.1%, Three days ago is 7.29%. Four days ago is 6.561%. Five days ago is 5.9049% ... and so on... all the way back through every day we've ever logged. This means our weight 190 days ago is still part of this average. Sure it's 0.0000002% of the sum ... but it's in there.
Of course, the last 21 days account for nearly 90% of our current number. So, it's fair to compare it to a 21 day average.
This gets into the math: http://www.fourmilab.ch/hackdiet/e4/signalnoise.html
Huh interesting, I'll have to take a look at that. So you assign a weight to each term based on its perceived influence on your current TDEE with more recent data having a higher weight than older data. I guess that is a valid assumption but I wonder if it is based on anything specific. I'll check out the link later, thanks.0 -
Yeah, my current "mass" is calculated using today's mass and previous masses. The more recent the scale reading, the higher weight it is given. Sure, I could add up the last 7 days. But, why should my mass last Friday be as important as my mass today? If we're talking 21 day moving averages (which would remove most of the variations we're trying to get rid of), why would I consider my mass three Fridays ago at the same weight as I do today? I could have lost almost 2kg in that time.
As it is, the formula looks confusing. But, that's only when it's expanded. We just take the previous day's trend at 90% weight and today's scale measurement at 10% weight. If my trend yesterday was 75kg, and my scale today read 73kg, I would add 75*.9 + 73*.1 = (67.5 + 7.3) = 74.8kg.
The nice thing about this is that it is very easy to compute, it gives you a value every day, and it provides the same (relatively speaking) smoothness as using a running 21 day average.
* replaced "weight" with mass when I was talking about the number that showed up on the scale to reduce confusion from also using weight to refer to the amount of credence I give to each particular value.0
This discussion has been closed.
Categories
- All Categories
- 1.4M Health, Wellness and Goals
- 395.4K Introduce Yourself
- 44.1K Getting Started
- 260.7K Health and Weight Loss
- 176.2K Food and Nutrition
- 47.5K Recipes
- 232.7K Fitness and Exercise
- 445 Sleep, Mindfulness and Overall Wellness
- 6.5K Goal: Maintaining Weight
- 8.6K Goal: Gaining Weight and Body Building
- 153.2K Motivation and Support
- 8.2K Challenges
- 1.3K Debate Club
- 96.4K Chit-Chat
- 2.5K Fun and Games
- 4.2K MyFitnessPal Information
- 16 News and Announcements
- 1.3K Feature Suggestions and Ideas
- 2.9K MyFitnessPal Tech Support Questions