This thread isn’t really much about Nate Silver any more but for consistency I’ll continue using this title for the thread. And, in fact, there is a Nate Silver element to this, how I fell victim to “confirmation bias” until I just used the data. So let’s look at the money graph (so labeled in previous posts):
The point for week 9 was expected (continuing on my regimen and thus tread line), but the point for week 10 was a (pleasant) surprise. Being the holidays I had multiple events last week where eating was higher than normal. On Thursday a quick weigh-in confirmed a reduction in rate of loss. But herein lies all my confirmation bias: a) I expected not to do as well so I cherry-picked evidence to prove it, b) despite knowing the variability of scale (previous posts) I believed a single measurement. Today’s “consensus” result (slightly pessimistic vs the statistical value) shows the effect, as easily predicted by Nate, that using the data is always more correct:
This graph shows the considerable variability in individual measurements this morning, not the widest range ever, but certainly wide. Interestingly the trendline is now almost identical between the money graph (consensus value, my “fudging” in choosing a weekly value) and this more “honest” (raw data in later weeks) graph than Nate would want to use. In fact, with four weeks of data now let’s just look at this:
The range-only (just points where I have multiple Sunday measurements) graph is just a smidgen more pessimistic, i.e. demonstrates slightly less weight loss per week. In addition to seeing this in the slope numerical value, the extrapolation backwards underestimates the actual value and the extrapolation forward predicts a slightly higher value (than the money graph, for instance). This might be a real effect (not just artifact of changing how I get weekly data) since I am against two effects: a) the same amount and intensity of exercise burns less calories (both by the formula, but also probably for valid basic reasons), and, b) my daily requirements drop with weight. This produces the double whammy that my calorie deficit per day is steadily dropping and that has to show up in reduced weight loss, which I certainly expect in just a few weeks. I’ve partially compensated for this effect with an increase in amount of exercise but I’m getting near the peak I can do.
So, much to my surprise, I managed to maintain the trend (weekly weight loss) despite some fundamental changes. This weeks has even more interference with my regimen so I do expect to see that effect next week but unless I really pig out this week I predict I’ll be able to start the new year (don’t need a resolution for this, already committed, so I’ll just re-commit on 1Jan and make a new commitment to be 20lbs on 31Dec2013, which really means I will have succeeded at “maintenance” since I should hit my target around April – lots of predictions to check in the future).