Question: What do you think mean zero measurement error in the measure of class attendance will do to your estimate of the returns to class attendance and why (please derive this)? Gist Mean zero measurement error in leads to attenuation bias in the regression coefficient estimate on the x1 parameter - As the as the variance of the error term increases (as people are more incorrect as to their estimate of x1) the coefficient estimate gets biased down.
"Proof" Where , where is the observed value (observed with error, is the true value, and is the mean zero error term, with 

Now, appears in four places. We now need to consider x_1 given its error. Let's take these four occurrences of x_1 one by one.
1) First occurrence of x_1: as usual.
2) Second:

Going term by term of the five terms above, in the probability limit 






Third Location: 
Through a similar process to 





Fourth Occurrence of x_1: 




Also unchanged: 





So after all that, let's plug in what we found for these probability limits 





Okay! So the bias in our estimate for the depends on
Note that if the term on the right goes to one, and we have no bias.
If (and since it's a squared term, it'll be positive if it ain't zero), then will decrease our estimate of Just like in the inivariate case. | |
|
|