I’m really having trouble wrapping my mind around the concept of heritability.
These numbers are obviously wrong, but let’s just use them for convenience:
Let’s pretend that scientists say that 80% of height is due to my genetics. If I’m 100cm tall, I understand that the scientist is NOT saying that 80 of my cm is due to my genetics.
I also understand that it doesn’t mean that if there are 100 people in my population, it doesn’t mean that 80% of them are only as tall as they are because of their genetics.
Does it mean that 80% of the reason I am as tall as I am is because of my genetics?
I started wondering about this when someone mentioned something to the effect of, “If your dad was a drinker, it runs in the family.” and also things like, “you’ll probably go bald because your mom did”. I understand that heritability is a term used for population studies, but I’m not sure why it doesn’t apply to individuals as well.
And if you read this and think “hoooo boy, OP’s gotta go ALLL the way back to square 1”, I’m more than happy to click a link and read!
Thanks a lot for your help!
In: 2
Usually the numbers you see are percent of the variance in the trait (in your case, height) explained by the gene. In more technical terms, it’s the [coefficient of determination](https://en.wikipedia.org/wiki/Coefficient_of_determination), which you’ve probably seen as r^2 in a basic statistics class but has more general applications.
Given a collection of things, the variance (which is the square of the more familiar standard deviation) describes how spread out that data is. Higher variance means more spread out, lower variance means less spread out. Height, for example, might have a variance of (say) 100 cm^2 (so a standard deviation of 10 cm), though that’s probably low for real data.
If you have some prediction, like a best-fit line, you can instead measure how spread out the real values are *from your predicted values* instead of from one another. This value measures how much your model *isn’t* explaining. This results in a lower number than the original variance. Say it’s 40 cm^(2) in a model predicting height. Then you’d say that 40% of the variance is *not* explained, i.e., 60% of it *is* explained.
(The reason this value is always lower than the original variance is that the original variance is basically just this with a model of “assume everyone’s average”, and best-fit models never do any worse than that.)
Latest Answers