Well, the evil of the holidays (too much food) is finally over and now I can try to get back on trendline. So what did the holidays cost me? (Let’s look at the “money” graph).

The graph above shows the projected trendline to today and the actual two points from the holiday period. Obviously the point (“consensus value, explained in previous posts) I’ve labeled as “the glitch” is the villain, but the latest (today’s) point I can appropriately label as “the fix” This graph gives the longer-term view but isn’t much good for looking close up at the last two weeks so let’s blow this up with more detail:

The graph above is all the weigh-in samples for both the “glitch” week (last week, following xmas pigouts) and the “fix” week. The regression line thus reveals an interesting number, a weight loss of 4.55lbs (4.6 in the “consensus” data) which is substantially higher than my previous weekly loss (up through week 10, 2.7267lbs/week, determined from slope of trendline for first 10 weeks of the money graph).

Now, the interesting part the statistics don’t tell is the unusual sequence of last week, where my weekly loss was 166.9% of the longer-term average. What caused this?

It turns out, a false alarm, as several other people tried to reassure me. I made the “mistake” of a mid-week weigh-in and right after a guilty pleasure (some guacamole and chips (but no beer) while watching the football games). The first number was shocking, but as I’ve previously commented there is a lot of noise (as Nate would definitely define it) in scale sample numbers. So I got more samples assuming the value was an anomaly. But it wasn’t! So it appeared, mid-week, that I had gained another 2.5lbs. Now I had drastically increased exercise to compensate for my expected New Years pigouts, so when I did the math it appeared that my food consumption had to be an average of 6000 calories/day for first three days of the week!! Total panic. Total depression. I thought I’d hit a plateau I’ve hit before which is where weight loss stops despite best efforts.

So, in a panic I tried for maximum possible change. I dropped my daily intake (for the most recent four days) to around 800 calories (my best estimates). I even had the depressing event of going to a birthday party at a bar and (almost) completely abstaining from any refreshments (did skip all booze, not even a taste). Meanwhile I went bonkers with exercise, basically exercising most of the day, including using the new xmas present recumbent bike. As a consequence I did an average of 1600 calories of exercise/day (nearly athletic training levels. So for four days that should have produced a deficit of (1600-800)*4 = 3200 calories, that is, actually burning more calories in exercise than I consumed. And according to the calculators I need about 2200 calories/day just for sedentary existence. So nominally that means ((2200*4)+3200) should be my weight loss, which works out to 3.33 pounds, about 0.6 more than my previous long-term trend (amazing how small that is, given the starving I did + much more exercise). That would have undone my midweek blowup by about a pound and thus put me around 223. But my measured was even better (lower) than that.

So my panic, over the anomalous midweek weight triggered my extreme loss regimen and thus partial compensation for “the glitch” (not just back on trendline, but actual recovery).

So when all is said and done, here’s the consequence of the holidays (the pigouts + the extreme loss):

So there it is, the 1.2lbs net result of the two weeks of holidays and hopefully I’ll now be back on regular trendline.

Now here is where I run into some questions about the analysis,

## need some help here, Nate!

*(I guess Nate is really busy because he hasn’t gotten back to me yet on my previous questions, but I know he’s out there and it’s just a matter of time)*

how do I handle “the glitch” going forward because it’s distorting my trendline:

The trendline up through week 10 shows a weekly change (its slope) of 2.7267 vs the trendline through week 12 shows weekly change of 2.5871, or, 0.1396, which translates to one less pound in about 7 weeks. Now the trendlines show a different result (nearly two pound difference four weeks in the future) because their y-intercept also changed a bit in the regression calculations. So here’s three difference possibly projects for 4 weeks into future:

- (old trendline): 208.14
- (new trendline): 209.81
- (latest consensus point, but old trendline slope): 209.29

So, Nate, which one should I use (or perhaps a better question, what would be the Bayesian approach to this problem instead of the frequentist?)

Now my simple approach might just be to average the three predictions and get: 209.1.

But I also know some factors will influence the next four weeks:

- I can’t sustain the calorie burn of exercise I’ve been doing lately, especially as I plan to switch more to bike which is easier on joints but lower METs.
- I will probably been last freaky about food intake (maybe even move in direction of the “sustainable” diet, which means a bit more consumption)
- I’ll be using the BodyMedia “Fit” Wireless Link armbit and its Activity Manager (which promises 3X better weight loss, unlikely for me given what I’m already doing) and use this to fine-tune my notion of calorie burning profiles and thus hope to get some improvement
- All the formula indicate two adverse conditions: a) as weight decreases basal metabolism calorie requirements drop, and, b) as weight decreases exercise creates less burn, so a double whammy that cuts my burn-requirement (calorie deficit) thus making the “old” trendline weekly loss more difficult.

So these are all the factors to consider on top of the purely statistical record. Plus I figure there has to be some noise and I’ll assume that goes in adverse direction, so when I combine all of this I’m going to set a “goal” (which I believe I can met/beat) of

# 210.2

in four more weeks. So stay tuned to see how good my projections go. And hopefully I won’t hit any more panics like I had this week.

### Like this:

Like Loading...

*Related*

## About dmill96

old fat (but now getting trim and fit) guy, who used to create software in Silicon Valley (almost before it was called that), who used to go backpacking and bicycling and cross-country skiing and now geodashes, drives AWD in Wyoming, takes pictures, and writes long blog posts and does xizquvjyk.

## Applied Nate Silver – The holidays cost me 1.2lbs

Well, the evil of the holidays (too much food) is finally over and now I can try to get back on trendline. So what did the holidays cost me? (Let’s look at the “money” graph).

The graph above shows the projected trendline to today and the actual two points from the holiday period. Obviously the point (“consensus value, explained in previous posts) I’ve labeled as “the glitch” is the villain, but the latest (today’s) point I can appropriately label as “the fix” This graph gives the longer-term view but isn’t much good for looking close up at the last two weeks so let’s blow this up with more detail:

The graph above is all the weigh-in samples for both the “glitch” week (last week, following xmas pigouts) and the “fix” week. The regression line thus reveals an interesting number, a weight loss of 4.55lbs (4.6 in the “consensus” data) which is substantially higher than my previous weekly loss (up through week 10, 2.7267lbs/week, determined from slope of trendline for first 10 weeks of the money graph).

Now, the interesting part the statistics don’t tell is the unusual sequence of last week, where my weekly loss was 166.9% of the longer-term average. What caused this?

It turns out, a false alarm, as several other people tried to reassure me. I made the “mistake” of a mid-week weigh-in and right after a guilty pleasure (some guacamole and chips (but no beer) while watching the football games). The first number was shocking, but as I’ve previously commented there is a lot of noise (as Nate would definitely define it) in scale sample numbers. So I got more samples assuming the value was an anomaly. But it wasn’t! So it appeared, mid-week, that I had gained another 2.5lbs. Now I had drastically increased exercise to compensate for my expected New Years pigouts, so when I did the math it appeared that my food consumption had to be an average of 6000 calories/day for first three days of the week!! Total panic. Total depression. I thought I’d hit a plateau I’ve hit before which is where weight loss stops despite best efforts.

So, in a panic I tried for maximum possible change. I dropped my daily intake (for the most recent four days) to around 800 calories (my best estimates). I even had the depressing event of going to a birthday party at a bar and (almost) completely abstaining from any refreshments (did skip all booze, not even a taste). Meanwhile I went bonkers with exercise, basically exercising most of the day, including using the new xmas present recumbent bike. As a consequence I did an average of 1600 calories of exercise/day (nearly athletic training levels. So for four days that should have produced a deficit of (1600-800)*4 = 3200 calories, that is, actually burning more calories in exercise than I consumed. And according to the calculators I need about 2200 calories/day just for sedentary existence. So nominally that means ((2200*4)+3200) should be my weight loss, which works out to 3.33 pounds, about 0.6 more than my previous long-term trend (amazing how small that is, given the starving I did + much more exercise). That would have undone my midweek blowup by about a pound and thus put me around 223. But my measured was even better (lower) than that.

So my panic, over the anomalous midweek weight triggered my extreme loss regimen and thus partial compensation for “the glitch” (not just back on trendline, but actual recovery).

So when all is said and done, here’s the consequence of the holidays (the pigouts + the extreme loss):

So there it is, the 1.2lbs net result of the two weeks of holidays and hopefully I’ll now be back on regular trendline.

Now here is where I run into some questions about the analysis,

## need some help here, Nate!

(I guess Nate is really busy because he hasn’t gotten back to me yet on my previous questions, but I know he’s out there and it’s just a matter of time)how do I handle “the glitch” going forward because it’s distorting my trendline:

The trendline up through week 10 shows a weekly change (its slope) of 2.7267 vs the trendline through week 12 shows weekly change of 2.5871, or, 0.1396, which translates to one less pound in about 7 weeks. Now the trendlines show a different result (nearly two pound difference four weeks in the future) because their y-intercept also changed a bit in the regression calculations. So here’s three difference possibly projects for 4 weeks into future:

So, Nate, which one should I use (or perhaps a better question, what would be the Bayesian approach to this problem instead of the frequentist?)

Now my simple approach might just be to average the three predictions and get: 209.1.

But I also know some factors will influence the next four weeks:

So these are all the factors to consider on top of the purely statistical record. Plus I figure there has to be some noise and I’ll assume that goes in adverse direction, so when I combine all of this I’m going to set a “goal” (which I believe I can met/beat) of

## 210.2

in four more weeks. So stay tuned to see how good my projections go. And hopefully I won’t hit any more panics like I had this week.

## Rate this:

## Share this:

## Like this:

Related## About dmill96

old fat (but now getting trim and fit) guy, who used to create software in Silicon Valley (almost before it was called that), who used to go backpacking and bicycling and cross-country skiing and now geodashes, drives AWD in Wyoming, takes pictures, and writes long blog posts and does xizquvjyk.