I’m doing a good job of demonstrating how bad my predictions are. Many people like to claim they predicted (notice the past tense) something when in reality the only “prediction” to evaluate is those made inĀ indelible form that can be compared to reality. So here’s mine:

## 195.5 (prediction) vs 196.2 (reality)

That might not seem like a bad miss but the prediction last week was already conservative and still I blew it. This is the smallest loss in 23 weeks and that puts the miss in better perspective. So let’s look at the “money graph”:

I had to put the third order polynomial fit back in there since the linear trendline is a large miss. But 23 weeks and 60lbs range is getting hard to see so let’s scale the graph for a better look:

Now statistics and graphs don’t “explain” anything. Noticing that week 21 and 22 are below both trendlines and week 23 is above the linear (a lot) and the polynomial (a little) which are the anomalous points? I think it’s a little bit of both. Week 21 and 22 were based on extreme dieting and week 23 had three bad days, especially Friday (anniversary) where it seemed appropriate to enjoy some food. But all this just points out how useless the BodyMedia app is for prediction. My average weekly calorie deficit (over 10 weeks, counting week 23) is 9428 (or 1347/day) which is pretty good (except being wrong) so that compares unfavorably to this week’s 7006 deficit (which should translate into 2.0 lbs loss instead of 0.7). So how is this ending up so far off? Over 10 weeks where I have the data my calorie deficit produces an actual weight loss of only 77.9% of predicted. IOW, the deficit overstates the expected loss by 28.3%. Now I may be under-reporting calorie consumption a little (and rarely over-reporting) but I don’t believe it’s by 28%. And the calorie burn, as measured by BodyMedia is under-reporting (for instance, yesterday I did 1800 calories on bicycle (60 miles) and essentially none of that is recorded since BodyFit is just a glorified pedometer and bicycling doesn’t record “steps”). So let’s just break this out, calorie deficit vs actual weight loss:

Notice the r^2 is awfully low and thus this correlation isn’t very meaningful, but note also the two anomalous points, which are paired (one, the negative, immediately precedes the most positive which sorta makes up for the negative), so let’s kick out this two outliers and see what we get.

Now Nate would probably be distressed at my deletion of three outlier point just to “cook” the graph and r^2 to look better, but I do think this more accurately portrays the situation. But the conclusion it shows is very disappointing. My weight “plan” (2lbs/week loss) requires 7000 calorie deficit per week, but this graph shows that the first 5800 portion of the deficit just gets me to weight maintenance.

This means I’ll have to continue fairly extreme exercise (now averaging 1104/day which is not sustainable) and fairly extreme diet (even the two days I “pigged out” were modest consumption compared to what I traditionally did).

So most of my “mysteries” of weight loss are fairly well documented and analyzed but now my my main mystery is why I didn’t weight 3000lbs before I started all this. It’s very clear to me I had a calorie surplus all the time (way less exercise, way more consumption) yet my weight was fairly stable (maybe 2lbs/year gain, vs probably 2lbs/month calorie surplus). Some “setpoint” prevented absorption of all those excess calories (as it does for most people). But as I understand it my adipose cells now have a reduced lipid capsule but the number of cells is still the same so therefore bloating back to my previous weight would be really easy (and probably faster than my loss has been). So it looks like maintenance is really going to be an ordeal.

Now one more graph, the updated daily results:

This graph, unfortunately, shows how bad last week was and my challenge going forward. Last week 5 of the 7 days (even including today) are way above the linear trendline which is then shown by looking at the third order polynomial (really curving up / flattening out). This looks especially bad after the low I hit on Monday. Translating this to real events is simple: in the previous two weeks I was extremely aggressive about diet (not so much on exercise) and this week I was bad on both (missed two days of exercise, “pigged out” three days). And why that is so bad is that my vacation is about 10 days is going to be worse than this week (even less exercise, even more eating).

So if magically I can be really good this coming week and hold to 1300 calories/day consumed + 1000 calories/day of exercise, I might knock off 2.4lbs like my long-term trendline and thus crack 195lbs before almost starting the vacation. But I suspect that’s optimistic so I think I’ll just make a conservative prediction of:

# 195.0

and now I really need to make that, or better yet, beat it.

### 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 – a warning

I’m doing a good job of demonstrating how bad my predictions are. Many people like to claim they predicted (notice the past tense) something when in reality the only “prediction” to evaluate is those made inĀ indelible form that can be compared to reality. So here’s mine:

## 195.5 (prediction) vs 196.2 (reality)

That might not seem like a bad miss but the prediction last week was already conservative and still I blew it. This is the smallest loss in 23 weeks and that puts the miss in better perspective. So let’s look at the “money graph”:

I had to put the third order polynomial fit back in there since the linear trendline is a large miss. But 23 weeks and 60lbs range is getting hard to see so let’s scale the graph for a better look:

Now statistics and graphs don’t “explain” anything. Noticing that week 21 and 22 are below both trendlines and week 23 is above the linear (a lot) and the polynomial (a little) which are the anomalous points? I think it’s a little bit of both. Week 21 and 22 were based on extreme dieting and week 23 had three bad days, especially Friday (anniversary) where it seemed appropriate to enjoy some food. But all this just points out how useless the BodyMedia app is for prediction. My average weekly calorie deficit (over 10 weeks, counting week 23) is 9428 (or 1347/day) which is pretty good (except being wrong) so that compares unfavorably to this week’s 7006 deficit (which should translate into 2.0 lbs loss instead of 0.7). So how is this ending up so far off? Over 10 weeks where I have the data my calorie deficit produces an actual weight loss of only 77.9% of predicted. IOW, the deficit overstates the expected loss by 28.3%. Now I may be under-reporting calorie consumption a little (and rarely over-reporting) but I don’t believe it’s by 28%. And the calorie burn, as measured by BodyMedia is under-reporting (for instance, yesterday I did 1800 calories on bicycle (60 miles) and essentially none of that is recorded since BodyFit is just a glorified pedometer and bicycling doesn’t record “steps”). So let’s just break this out, calorie deficit vs actual weight loss:

Notice the r^2 is awfully low and thus this correlation isn’t very meaningful, but note also the two anomalous points, which are paired (one, the negative, immediately precedes the most positive which sorta makes up for the negative), so let’s kick out this two outliers and see what we get.

Now Nate would probably be distressed at my deletion of three outlier point just to “cook” the graph and r^2 to look better, but I do think this more accurately portrays the situation. But the conclusion it shows is very disappointing. My weight “plan” (2lbs/week loss) requires 7000 calorie deficit per week, but this graph shows that the first 5800 portion of the deficit just gets me to weight maintenance.

This means I’ll have to continue fairly extreme exercise (now averaging 1104/day which is not sustainable) and fairly extreme diet (even the two days I “pigged out” were modest consumption compared to what I traditionally did).

So most of my “mysteries” of weight loss are fairly well documented and analyzed but now my my main mystery is why I didn’t weight 3000lbs before I started all this. It’s very clear to me I had a calorie surplus all the time (way less exercise, way more consumption) yet my weight was fairly stable (maybe 2lbs/year gain, vs probably 2lbs/month calorie surplus). Some “setpoint” prevented absorption of all those excess calories (as it does for most people). But as I understand it my adipose cells now have a reduced lipid capsule but the number of cells is still the same so therefore bloating back to my previous weight would be really easy (and probably faster than my loss has been). So it looks like maintenance is really going to be an ordeal.

Now one more graph, the updated daily results:

This graph, unfortunately, shows how bad last week was and my challenge going forward. Last week 5 of the 7 days (even including today) are way above the linear trendline which is then shown by looking at the third order polynomial (really curving up / flattening out). This looks especially bad after the low I hit on Monday. Translating this to real events is simple: in the previous two weeks I was extremely aggressive about diet (not so much on exercise) and this week I was bad on both (missed two days of exercise, “pigged out” three days). And why that is so bad is that my vacation is about 10 days is going to be worse than this week (even less exercise, even more eating).

So if magically I can be really good this coming week and hold to 1300 calories/day consumed + 1000 calories/day of exercise, I might knock off 2.4lbs like my long-term trendline and thus crack 195lbs before almost starting the vacation. But I suspect that’s optimistic so I think I’ll just make a conservative prediction of:

## 195.0

and now I really need to make that, or better yet, beat it.

## 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.