Fun With My Trending App

AnnPT77
AnnPT77 Posts: 31,730 Member
Warning: There's no Big Point to this post. I'm just chatting about something I find fun/interesting, in maintenance, but NBD.

I'm a long-term Libra user. One of the nice things about it is that I can change the number of days it uses for smoothing and projections. Generally, in maintenance, I like using longer periods. Over the last few months, I've been working on slowly drifting weight downward, maybe half a pound a week or less, and recently went back to the default settings. (Changing the settings doesn't change one's weight data, just changes the calculations & displays done with that data, so switching back and forth is fine.)

With Libra's default settings, 7 days smoothing and 7 days projection, my recent months look like this:

xcs5revluor3.png

This could mislead: There's that bump in July, which looks like regain, and I admit even I worried there briefly, even though nothing had changed on the calorie side (in or out) to make gain happen. As I expected - when not giving into my emotions 😆), it was about the couple of pounds of water weight gain I usually see when I restart strength training after a hiatus. I restarted around the end of June/beginning of July. Now, the continuing slow loss is starting to show up, emerging from the noise of the water weight. (Note that this set of values says I'll reach 120lbs on October 11.)

Like I said, in regular maintenance, when not trying to gain/lose, I like longer setting, 30 days smoothing, 60 days projection. With the exact same daily weight data & time period as the chart above, that would look like this:

g1q678v60d7f.png

Looks like steady loss, eh? The July pseudo-gain is totally missing, but it says I won't reach 120 until October 31.

For general maintenance, I like the longer smoothing/projection, because I don't get meaningless gain/lose signals from minor fluctuations. That misleading stuff didn't happen as much during weight loss, because the fat loss going on in the background happened at a fast-ish rate. If my fat levels are pretty steady (if I'm actually maintaining) a perfectly fine high carb/sodium/calorie day can pull the 7-day trend upward (and annoy me, even though I know it's fake news 😉).

While I'm going this slow-loss thing, though, the longer smoothing/projection can lull me into a false sense of security, because the tolerances are pretty tight, trying to keep losing, but at a very (very) slow rate. I do tend to eat unevenly - bank calories most days, indulge sometimes - so I tend to see wider daily swings with that strategy than if I had a more strict daily consistency. (I'm a bit more consistent in the pandemic, mostly eating my own cooking, than I was when restaurant/social eating & drinking was more frequent.)

TL; DR. Weight trending: It's a statistical projection, not a magic crystal ball. But it's fun to play with. 🙂

Replies

  • Jthanmyfitnesspal
    Jthanmyfitnesspal Posts: 3,515 Member
    Well, @AnnPT77 , you are an inspiration! I lost 5lbs in Jan/Feb in support of the triathlon season. Then the quarantine hit and I gained it back. I'm wanting to lose it again, but with no real progress thus far.

    It all looks very useful-- much more so than just the daily numbers, which bounce around all over the place.

    There's something funny about the second trend. It appears that ALL the "outliers" are below the smoothed line. That's not possible. The first one shows the daily data bouncing around the smoothed line, which looks very reasonable. Any idea why?
  • MadisonMolly2017
    MadisonMolly2017 Posts: 10,875 Member
    Yes, I love this post & I also manipulate how I have data presented to me based upon my goal at the time.

    I just realized you can divide your avg calorie deficit by 10 to get a pretty good estimate of your yearly loss at that deficit. Multiply by 0.1043 for more exact 🤣

    I’m pondering losing 3.6 lbs to see if I can break through the 157’s (in maintenance). Last time I tried, I got hungry. I think I’m more experienced now at preventing that.

    So if I have a 78 cal deficit, I’ll be there Jan 1, my 5 year anniversary from when I began my journey. If I’m too hungry, I’ll decrease the deficit until I’m not & it will just take longer!

    Waves to @SummerSkier Im always impressed by folks who can maintain without weighing daily. I weigh daily as part of my “Keeping My Head in the Game” plan. Every other time I lost a lot of weight, I stopped paying attention for a “good reason” & regained.

    Not this time!
  • AnnPT77
    AnnPT77 Posts: 31,730 Member
    It just goes to show how data can be manipulated. You are making good progress though regardless of how you smooth it. Interesting post. Thank you for putting that together. I am just closing out year three of maintaining and I actually don’t weigh myself anymore. I do like to log my food in order to see what I am eating calorie wise and also macro wise. Scale Fluctuations mess with my head way too much. I go with clothes fit and energy. Of course working from home I have to make sure to put on jeans at least once or twice a week. 😃

    In this context, I'd say "used to one's advantage" rather than "manipulated". 😉

    It's extremely useful, IMO, to develop a little - truly a little - understanding of basic statistics and how to use them. There's a great little book "How to Lie with Statistics" that's fun to read, and really informative - useful to at least avoid the main ways of being hoodwinked.

    https://en.wikipedia.org/wiki/How_to_Lie_with_Statistics

    It's common around here to see what are IMO misleading understandings of simple math: For example, the idea that EPOC ("afterburn") for strength exercise or HIIT is twice as much as EPOC for moderate steady state cardio. It is, in the sense that HIIT EPOC is estimated (usually) around 14%, and SS EPOC more like 7%, but that's as a percentage of calorie burn during the exercise, so usually numerically trivial at best, in calories - possibly irrelevant, since HIIT duration is self-limiting, and strength training tends to burn low base calories. (Similar to those clickbait "reduces XYZ cancer mortality by 50%!". then you learn that whatever it is reduces the mortality from 1 person per 10,000 diagnosed to 0.5 per 10,000, so 0.0001% raw percent of those diagnosed, or something underwhelming like that.)

    Your post reminded me, I should have said: I'm in year 4+ of maintaining, after losing around 50 pounds back in 2015, after preceding *decades* of obesity. I've weighed myself daily (and recorded the result) for well over a decade, even when not trying to lose weight, so I arrived at weight loss and MFP calorie counting in 2015 already very calm about fluctuations. To me, it's just potentially useful raw data, a single moment's snapshot of my body's temporary relationship with gravity. The trending app has been a useful tool, since I started using it part way through weight loss.

    To some, daily weighing can sound (or be) obsessive, but obsession is about habits of mind, not so much about practices. The question is whether one feels deeply *compelled* to do the thing, stressed if it's skipped, etc. Daily weighing can be stressful/obsessive for some, but that's not universal by any means.

    I'm a data geek, so I have data about lots of things, because I think data is fun: Records of my garden plantings, data about rowing training & performance, and who knows what-all. I come by it honestly: My dad used to measure key trees in his woods to see how many board-feet they were making. 😆
  • AnnPT77
    AnnPT77 Posts: 31,730 Member
    Well, @AnnPT77 , you are an inspiration! I lost 5lbs in Jan/Feb in support of the triathlon season. Then the quarantine hit and I gained it back. I'm wanting to lose it again, but with no real progress thus far.

    It all looks very useful-- much more so than just the daily numbers, which bounce around all over the place.

    There's something funny about the second trend. It appears that ALL the "outliers" are below the smoothed line. That's not possible. The first one shows the daily data bouncing around the smoothed line, which looks very reasonable. Any idea why?

    Sure. It's not just possible, it's kind of cooked in. Think of it the other way around - consider the trend first.

    When I use more data points (include a longer span of past days in the arithmetic), in a context where the overall direction is loss (even slow loss) the trend line is going to point down. I almost think of the numbers as pulling it down. One or two higher weight days, during 30 days or 60 days that are mostly even or downward, don't exert much upward "pull" on the trend. (Even in the smooth trend, you can see around 3 days "pulling up" the trend, i.e., the above the line values causing a slight micro-hill in the downhill slope. Moguls, maybe. 😆)

    If, instead, it's calculating based on 7 days with one or two higher weights, and a lower net loss across the whole 7-day time period, the higher weight can pull the trend line up, because it doesn't have as many lowered weights, or weights lowered as far, to pull it down.

    Consider a situation where someone is in a super-crazy-high dangerous deficit (we see one around here once in a while). They're weighing every day, and every day they see a loss (for a while 😉). All - literally all - their weights are going to be below the trend line: How could they possibly be above it? Conceptually, metaphorically, those decreasing daily weights are what's consistently "pulling down" on the trend line, *causing* it to be a downward line.

    Here's another way to look at it: Simplify it. Pretend I weigh bi-monthly only, then pretend I weigh weekly only, and I only compare the two end data points (Pretend I forget history, just look at two points, in each case).

    Over 60 days, say May 15 to July 15, the my weight went from 131.0 to 130.8. If you just connect those two points, ignoring a roller coaster in the middle, you get a down trend - admittedly a very small one, just because of the dates I randomly selected without cherry-picking the data for a good result. But still, that long term line slopes slowly downward, if calculated from just the two points.

    Over 7 days, say the 7 days leading up to July 15, my weight went from 128.8 to 129.4, a subset of that same time period**. That 7-day line slopes up. What the heck? Am I gaining?!?! Of course not. It's just not a long enough time span to get a picture of what's slowly happening over the long haul. It's a slightly extended variation on people freaking out because their weight goes up a pound overnight when they were doing everything right. Over-reacting to that individual week would just be freaking out because my weight went up over 7 days when I was doing everything right. 😆

    Because my ongoing fat loss is ssssoooooo sssslllooooowwww, the random fluctuations yank the short-term (7 day setting) trend line all over the place, but the long 30/60 setting is long enough for that down-trend to emerge from the "noise" of daily fluctuations.

    ** For fun, if I cherry pick, I can find 7-day chunks that are a losing trend (July 6, 129.6; July 12, 128.6) or completely flat (July 8, 130.0; July 14, 130.0). In one sense, though it isn't literally true, you could think of the bumpier first trend chart as a long sequence of these little up-y down-y single week trends, and the smooth down slope to be more like biting off a 30/60 day chunk all at once instead.

    Does that make sense?

    When loss is very fast (pound a week or more), the 7-day settings still give you enough net loss most of the time that the overall trendline drops. The fat loss "signal" is louder than the daily-fluctuation "noise". If I showed you the data from my initial faster fat loss, there are periods of a month, maybe more, where literally all the daily weights are on/below the trend line, zero above. To get that same kinda consistent signal-louder-than-noise metaphorical result with ultra-slow fat loss, I need to look at longer time periods.

    So: There's nothing funny about the 2nd trend. It's just a longer trend. Looking at the longer trend highlights some features, obscures others. Looking at the shorter trend highlights some features (different features), obscures others. If one has a trending app that allows one to fiddle with stuff like that, different settings may work better - show different things - at different times. You can get the same basic kind of thing, probably, if you use a spreadsheet, and use different-sized chunks of days as your window for a rolling average.

    Data play is fun! 🙂
  • AnnPT77
    AnnPT77 Posts: 31,730 Member
    @Jthanmyfitnesspal - An afterthought to what I wrote. Are you thinking that this should look like a regression curve-fitting kind of thing, least-squares method or something like that, where the curve tends to move through the center-ish of the display of data points, some above & some below the fitted/smoothed function line? It isn't that. It's not IMU finding a mathematical function that the data points represent (or a formula that describes the behavior of the data points - not sure how to say it best); it's a trend, a moving average kind of deal, where the data points themselves are sort of shaping the line.
  • charsuzy
    charsuzy Posts: 234 Member
    I'm going to try the app - it looks fun :)
  • Jthanmyfitnesspal
    Jthanmyfitnesspal Posts: 3,515 Member
    @AnnPT77 : I wouldn't lose any sleep on it. The 7 day looks pretty much like a moving average but the 30 day doesn't. It is weighting the high values more than the lows, for some reason. The web site probably explains it somewhere.
  • AnnPT77
    AnnPT77 Posts: 31,730 Member
    @AnnPT77 : I wouldn't lose any sleep on it. The 7 day looks pretty much like a moving average but the 30 day doesn't. It is weighting the high values more than the lows, for some reason. The web site probably explains it somewhere.

    I don't know what the specific smoothing algorithm is - the available documentation is limited.
  • AnnPT77
    AnnPT77 Posts: 31,730 Member
    edited August 2020
    AnnPT77 wrote: »
    @AnnPT77 : I wouldn't lose any sleep on it. The 7 day looks pretty much like a moving average but the 30 day doesn't. It is weighting the high values more than the lows, for some reason. The web site probably explains it somewhere.

    I don't know what the specific smoothing algorithm is - the available documentation is limited.

    Messing with it a little, it seems to be an exponential moving average algorithm rather than a simple moving average, with the smoothing days as the smoothing factor (or a component of the smoothing factors, i.e. still not sure it's simple linear exponential smoothing). The larger the smoothing days, the less effect recent changes have.
  • srk369
    srk369 Posts: 256 Member
    I tend to lose in a 5-7 day cycle...drop 1 day and then jump up for 4-6 days and then drop down again. I changed my Smoothing to 10 days to cover that cycle and the forecast to 14 days and that seems to keep my chart about .5 to .7lb average loss. When it was 7/7 I was always jumping between small loss to small gain and it was annoying!! I'm with you, I like seeing the data!!
  • AnnPT77
    AnnPT77 Posts: 31,730 Member
    Just for clarity, for folks talking about trying a trending app:

    Libra works on Androids.
    If you have an Apple device, Happy Scale is similar.

    There are also web-based weight trending apps, like Trendweight (requires a free Fitbit account, I'm told, but doesn't require a Fitbit device) or Weightgrapher.

    Likely there are others.

    Because I'm an Android/Libra person, I don't know what options any of the others have for playing with the trend lines or projections, if any.
  • PAV8888
    PAV8888 Posts: 13,399 Member
    Trendweight for sure and weightgrapher afaik don't adjust the parameters of their exponentially smoothed averages that, by design, give more weight (heh, heh) to recent measurements while still considering past values and correctly reflecting one's trend and inflection points

    Afaik a lot of the apps were inspired by the hacker's diet which was published back in the Palm pilot days by the founder of Autodesk
  • fairy_daisy
    fairy_daisy Posts: 247 Member
    This was really great to see, thanks @AnnPT77. I'm on Apple, so using Happy Scale and I'm not sure it's got so many options to tweak. I'm weighing daily and even though I know the fluctuations aren't really the story, I do find myself reacting to them a little. I've now been going three months and I'd love the option of smoothing out all the noise.