00:01
So here we've got a regression model, and our regression model is score is equal to some constant, beta 0, plus beta 1 times hours, plus an error term.
00:15
Right.
00:16
This is our single variable linear regression model.
00:20
The first one here is a, right? the way that i would say this is that beta 1ā2 will not equal the true.
00:31
True beta 1, right? we want to actually figure out some sort of true beta 1, what the actual effect of ours is.
00:40
But if i try to estimate this with data, it's not going to work because we have an endogeneity problem.
00:47
Now, sometimes people call this endogenity.
00:50
Some people would call this a missing variable problem.
00:53
If you're in the medical field, you might call this a confounding problem.
01:00
The idea here is that this does not completely capture what goes into score.
01:10
Right.
01:10
So suppose, let's say, iq matters, right? iq matters for score.
01:21
And let's suppose that there is a correlation between iq and ours.
01:27
Right.
01:30
So now, when you're we plug this into the data, we don't observe iq.
01:35
We just think it matters.
01:37
But the model, just the pure arithmetic, is going to misrepresent the true beta 1 because beta 1 is going to be set to minimize the sum of square errors.
01:48
If beta 1, sorry, if ours is positively or negatively correlated with iq in the error term, we're going to have a problem, right? and i can illustrate.
02:01
That like this.
02:02
Suppose that the truth is actually beta 0 plus beta 1 hours plus in the inside the error that the real model is really beta 2 times iq plus some other error term, the true error term...