My Philosophy on Diet and Weight Loss

@Can_Do_Gal I love Libra too. In libra you can adjust how the trend is calculated to make it track closer or further from your data points. I like it at around 10 days, but you can fiddle with it if you like to play with the numbers. The essential trade off is a tighter trendline will give you bad news when it isn't warranted, but a slower trendline will take longer to warn you of danger if you actually have made a change. For me, 10 days is a happy medium that gets rid of any weekend adventures...without making me feel too guilty.

@SLLRunner the weight chart is definitely just a guesstimation to make it understandable, but close enough for a lot of people. I just took the lowest number off the large frame column of a chart similar to what you linked. It's quite interesting to note how it's different - the chart I was using allows more weight at taller heights and less at shorter heights for "large frame".

Your TDEE is excellent, you must be super fit. I'd guess you have a lean body mass advantage vs many women at your height.

Osric

yarwell wrote: »

By overdamped I mean the signal is filtered and attenuated too much - it takes too long to fully show a change.

There is obviously a tradeoff between how long the moving average is and how soon it will reflect the data being fed into it, which is why it'd be nice if trendweight let you control that the way Libra does.

However in order to make the argument that it is too filtered, we'd have to show that a real signal appears that the moving average filters out. Since the daily fluctuation in actual solid mass is about 0.1 - 0.3 lbs for most women (0.3 would be > 2lb/week) and the water fluctuation for most women is probably in the range of 0.5-2.5 lbs, the moving average needs to be pretty long to enable you to actually extract the signal from the noise.

Also, TOM plays a role. In this post, we see that kommodevaran has a 200 kcal/day uptrend followed by a 200kcal/day downtrend that is purely her cycle - so arguably the trendline is too short in her case, causing her to adjust her diet needlessly every month along with her cycle when in fact she's maintaining and could have a longer trendline helping her to stay at goal. Of course the risk of that would be that if things actually changed, it'd take longer to react - but the fact remains that for there to be a real signal from the trendline in her case, the moving average needs to be longer, or she has to be content with adjusting her diet every week due to water retention fluctuations.

However, I think the most convincing argument I can make is this chart:



In this spreadsheet, I have modeled someone at 200 pounds on a diet for 40 days. Their daily deficit is 1000 calories, less a random mislogging of 0 - 400 calories (average 200 calories randomly eaten per day; sometimes 0, sometimes 400). Their water wieght fluctuates by 0 - 3 lb retained water per day.

First of all, notice how little the random logging error actually affects the red line of actual weight. Even though each day a little less or a little more is lost, actual weight loss when your deficit is between 600-1000 calories (or for that matter, 300-500 calories, I modeled it that way too, it makes no difference) is amazingly straight line perfect.

Water weight is modeled as a random amount of retention (0-2lb) over the actual body weight. So the jagged dark blue line is what comes out of the scale.

Now, at exactly the halfway point of the diet our hypothetical dieter loses hope and starts overeating by 600-1000 calories - the exact same rate and randomness that they were losing by initially. The question is, what level of dampening lets them react to the bottom if that was unintentional? I have done 5 days, 10 days, and 20 days. The surprising thing is that basically any of those smoothing values works about the same for finding the bottom. On the zoomed in chart I can see that indeed the 5 day average gives a warning about 2-3 days sooner than the 20 day one, but really 2-3 days is insignificant if you're making any effort at all to log your food.

The bottom line is that for the purposes of reacting to the real signal from the scale, almost any moving average will do. So it is sensible to choose one that allows you to average out your weekends (i.e. > 7 days) so that you're not constantly adjusting. Or for someone maintaining, a very long average (> 30 days) will help fight against adjusting your intake because of cyclical variations, and at most cost you a week or so of gain before you realize there's actually a problem.

The way to work with the trend is to use it to estimate actual deficit, and to see "is it going down, up, or staying about the same". The exact position of the trend line relative to your weight datapoints is not as important as the trend it is indicating. And the trend it is indicating is more significant the longer the moving average is, because it is more likely to reflect the true situation.

Osric

yarwell wrote: »

In your example above the measured weight is always higher than the actual weight - should it not be either side by the random fluctuations being considered ?

It's just a matter of how I modeled it; I modeled water mass as a positive number that varied from 0 (true weight) to 2lbs above true weight. The picture would look the same in the ways that matter if I'd modeled water as varying from -1lb of dehydration to 1lb of overhydration ... the thing that matters being whether the trend conceals the turnaround point in real weight or not by more than a few days. Even the 20 day moving average has clearly turned up by 4 days after the actual turning point. All the moving averages level off almost immediately. If your goal is to lose 1lb/week, you'll quickly know that something is going wrong using trendweight or any of the trending graphers.

@kommodevaran's graph is more problematic - because she is in fact maintaining, it'd be better if her chart looked like this:


[Image credit: The Hacker's Diet, John Walker]

In this case we see that a man's fluctuations are a lot tamer than a woman's so that even with a very short term moving average the gentle bumping up and down of the weight stays in a very narrow band. For women who are at maintenance, I think a very long moving average - perhaps 30 or even 40 days - would present a more accurate picture of what's going on with fat and still provide enough time to recover. She does in fact analyze her own data and maybe will chime in with a comment.

Osric

tiffanyschadow wrote: »

I need a WIFI scale, never heard of that! I will try this TrendWeight....mind = blown!

I love my scale! As far as I know there are two popular choices, the withings (that I have) and the fitbit aria (that I'm sure you can find other MFP'ers that have). The convenience just can't be beat.

Osric