One of my favorite new tools is L1 regression, which you use the same way as L2, but because you are just taking absolute values of the residual rather than squaring it, does not weight the outliers so highly and therefore they don't affect the results as much. It's a little like using a median rather than a mean. The hard part is finding software that supports L1 regression, and so far I've had to write my own in MATLAB (it's not hard: there is a short algorithm called "ADMM" that you program). But it makes much nicer results than least squares in situations where you have a few outliers, and then you don't have to justify with various excuses why you chose to throw away part of your data...
>I find this fascinating because of the ubiquity of the “He’s really shooting up!” reaction. I even shared the sense that he’s been growing rapidly of late, though maybe not as strongly as many of the other adults who have remarked on it.
I'm wondering how much of this impression is an angular change thing? If he grows at the same rate each year, then as he's getting closer to an adult height, the angle you have to tilt your head at becomes less and less pretty quickly, compared to all the time he spend between say, 1' and 4' when adults would basically be looking straight down.
Angular change is, in fact, a model I was toying with, and I think there's probably something to that picture. Trying to figure it out involves inverse trig functions, though, so a quick pass got messy and I had to put it aside to deal with something else.
My daughter, who topped out at exactly 183cm, has a very interesting perspective on the "I'm six feet tall" claims of men.
Zeno becomes more and more terrified every time the difference between The Pip's height and his own is halved.
One of my favorite new tools is L1 regression, which you use the same way as L2, but because you are just taking absolute values of the residual rather than squaring it, does not weight the outliers so highly and therefore they don't affect the results as much. It's a little like using a median rather than a mean. The hard part is finding software that supports L1 regression, and so far I've had to write my own in MATLAB (it's not hard: there is a short algorithm called "ADMM" that you program). But it makes much nicer results than least squares in situations where you have a few outliers, and then you don't have to justify with various excuses why you chose to throw away part of your data...
>I find this fascinating because of the ubiquity of the “He’s really shooting up!” reaction. I even shared the sense that he’s been growing rapidly of late, though maybe not as strongly as many of the other adults who have remarked on it.
I'm wondering how much of this impression is an angular change thing? If he grows at the same rate each year, then as he's getting closer to an adult height, the angle you have to tilt your head at becomes less and less pretty quickly, compared to all the time he spend between say, 1' and 4' when adults would basically be looking straight down.
Angular change is, in fact, a model I was toying with, and I think there's probably something to that picture. Trying to figure it out involves inverse trig functions, though, so a quick pass got messy and I had to put it aside to deal with something else.
Plot the derivative
It's noisy, but pretty constant.