Let t be the trend and let t-zero be yesterday's trend.
The trend adjusts itself by taking one tenth of the difference between today's measurement and the existing trend. So if you are 2 lbs lighter on the scale then trend, the trend will adjust downward by 0.2lbs.
Why divide by 10? Taking a tenth of the difference as your true signal provides a nice, smoothing effect. Let's take a look at how the smoothing effect works. Here's a contrived example: You weigh 177 at the start of the month and immediately lose two pounds so that your true weight is 175. Of course you vary a little from day to day (let's say alternating between 174 and 176). Here's what the trend line looks like:
The red line is the trend line. It appears to take the entire month to finally match your true weight of 175! Is this a good thing or a bad thing?
Its bad in the sense that for three weeks or more you are operating under the assumption that you are losing weight when you are not. But its good in the sense that you are carrying forward your credit so you can coast for a while and still appear to lose weight!
Using a larger multiplier, like say 0.3 (dividing by 3.33) gives a little bit of smoothing, but not much. As a result it becomes much more "accurate" in that it achieves you true weight in only 10 days or so. Here's what that looks like:
In the end, I must say it probably comes down to some level of personal preference. I find the 0.1 multiplier to be not just smoothing, but soothing.
Why divide by 10? Taking a tenth of the difference as your true signal provides a nice, smoothing effect. Let's take a look at how the smoothing effect works. Here's a contrived example: You weigh 177 at the start of the month and immediately lose two pounds so that your true weight is 175. Of course you vary a little from day to day (let's say alternating between 174 and 176). Here's what the trend line looks like:
The red line is the trend line. It appears to take the entire month to finally match your true weight of 175! Is this a good thing or a bad thing?Its bad in the sense that for three weeks or more you are operating under the assumption that you are losing weight when you are not. But its good in the sense that you are carrying forward your credit so you can coast for a while and still appear to lose weight!
Using a larger multiplier, like say 0.3 (dividing by 3.33) gives a little bit of smoothing, but not much. As a result it becomes much more "accurate" in that it achieves you true weight in only 10 days or so. Here's what that looks like:
In the end, I must say it probably comes down to some level of personal preference. I find the 0.1 multiplier to be not just smoothing, but soothing.
1 comment:
Agreed ! The standard hacker diet line is very slow, it took me a while to figure out why people posted trend lines where practically all of the data points were below the line. Less smoothing, less soothing but perhaps more accurate with 0.3 or similar.
Post a Comment