00:01
So here we estimated a regression, right? we got some data.
00:04
We put it into our computer, and we got the predicted value for y is equal to a very large number of significant figures.
00:14
Depends on x, right? so there's a positive intercept and a negative slope, right? so if i was going to plot my predicted line between, let's say, x and y, up here, you've got, say, the 29 point.
00:31
8 ,000 and you've got a negative slope.
00:35
So line for y hat looks something like this, right? again, that is the slope.
00:41
So we also know, again, a whole bunch of things that were given.
00:45
We're told that r squared is equal to 0 .2596.
00:50
We're told that r is equal to minus 0 .5096, right? note that if you square little r, you get r squared naturally.
00:59
And the standard error of the estimate is 100 ,093 .9156.
01:06
So again, we get this interpretation here that i totally agree with.
01:12
25 .96 % of the variation in the number of full -time university intakes is explained by the variation in polytechnic graduates.
01:20
I totally agree with that.
01:21
That's exactly what our square means.
01:24
There is a negative and weak relationship between the number of full -time university intakes and the number of full -time polytechnicates and the number of full -time polytechnic graduates.
01:33
I don't know if i fully agree with that, right? so that would be my first comment, right? there's the claim here that the relationship is weak, right? this is unclear to me, right? we don't know what the standard error is.
01:56
We're given the standard error of the the estimate, but not of this, right? what is the standard error here, right? of the coefficient.
02:12
That's something that immediately jumps to mind to me.
02:15
I would be asking that not clear.
02:19
In general, you might want to compute a p value here, right? that's usually how you would interpret this story, right? i would suggest that you go back to the regression data set, you get the standard error of the coefficient of the slope, right? because to me, this is a pretty strong relationship, right? every time someone graduates a polytechnic, the number of full -time university students decreases by like a third.
02:46
So it's like a one to three.
02:48
That's a fairly strong relationship from my perspective, at least, right? but, and yes, the average dispersion of the data points from the regression line is 1 ,093, absolutely.
02:59
So as someone who does a lot of regressions in their day -to -day life, the biggest issue here is probably omitted variable bias, right? variable bias, right? we do not have causality, right? this is the problem.
03:25
We're interpreting this slope coefficient as the relationship between x and y, but we don't.
03:32
Know whether that is causing x and y's right so for example maybe the state of the economy affects both of these things right so the state of the economy affects polytechnic grads and it also affects university grads right but these things do not affect each other right that's a distinct problem here right you have no guarantee whatsoever that it's the polytechnic graduates causing the number of university students.
04:09
Oh, sorry, this is not graduates, this is students, the number of university students to change, right? there's, you don't know that.
04:18
It could be that some lurking omitted variable is causing both of these things to change, right? at the same time and so there may not be any correlation here, right? and if this is true, right? b only represents the simple correlation, not behavior, right, not truth.
04:48
When you read that regression equation, you want to say that each time someone graduates from a polytechnic means that the number of university students is going to go down, but that's not true, right? you don't know that here, right? that you simply are completely unaware because you don't have that that sort of regression design that lets you to make that seem true, right? so that's the biggest most fundamental issue, right? i hope that's clear...