Thanks. Both of those expositions were helpful...
cheers,
jason
----- Original Message -----
From: "Jasjeet Singh Sekhon" <jasjeet_sekhon(a)harvard.edu>
To: "Jason Morris Lakin" <jlakin(a)fas.harvard.edu>du>;
<gov1000-list(a)fas.harvard.edu>
Cc: <jkline(a)fas.harvard.edu>
Sent: Sunday, November 23, 2003 1:54 PM
Subject: Re: [gov1000-list] RE: HW clarification
Jason,
As I noted in lecture, there are a few ways of thinking about this.
1) The most direct and intuitively useful answer is Jacob's (see
attached). Regression let's us ask what will happen to Y if X changes
by some number of units. Correlation does not directly provide this
information because it has removed both the units of X and Y. And
covariance does not directly provide this information because it
contains information about both the units of X and Y. What we want is
how changes in X map to Y.
2) Although the formula for Beta which is presented is just for simple
regression, multiple OLS regression can also be thought of as analysis
of variance (and covariance)---the equation just looks a lot more
complicated. Recall that covariance is a measure of linear
relationships so in some sense it is not surprising that *linear*
regression is also a way of analyzing variances and covariances. But
OLS can be generalized to the setting of multiple independent
variables while this cannot really be done with covariance or
correlation.
3) Under a set of assumptions, OLS obtains some nice proprieties like
unbiasedness and efficiency. Recall that under the 5 assumptions, OLS
is BLUE: the Best Linear Unbiased Estimator. If these assumptions
hold, these proprieties provide good reasons to favor the OLS beta
equation.
4) Even if the assumptions do not hold, OLS has the orthogonality
propriety mentioned in class and asked about in the homework. See the
last homework question. This property implies that the linear
information about Y contained in X has been extracted and that the
remaining residual (Y minus our predicted Y) is uncorrelated with X.
This is true even if none of the assumptions hold, and it is a nice
property for any measure of relationship to have. This propriety
comes from the fact that the OLS beta is the value of beta at which
the least squares loss function is minimized. When minimizing the
loss function we never explicitly said we want anything to do with
covariance, but it just turns out that the beta which minimizes the
loss is very closely related to covariance.
Jas.
>From: "Jacob Kline" <jkline(a)fas.harvard.edu>
>Sender: gov1000-list-admin(a)fas.harvard.edu
>To: "'Jason Morris Lakin'" <jlakin(a)fas.harvard.edu>du>,
> "'Jasjeet Singh Sekhon'" <jasjeet_sekhon(a)harvard.edu>
>Cc: <gov1000-list(a)fas.harvard.edu>
>Subject: RE: [gov1000-list] RE: HW clarification
>Date: Sun, 23 Nov 2003 11:19:05 -0500
>
>Jason,
>
>Try to think about this way: When we calculate the correlation, we want
>a number that is standardized to be between -1 and 1. We specifically
>don't care about the range of either variable. On the other hand, when
>we run a regression y~x, what we care about is how a change in x will
>correspond to a change in y, so we wouldn't want to divide out something
>like var(y) -- which you could think of as what tells us what a standard
>change in the size of y looks like.
>
>Does this help?
>
>Hope you are well,
>
>Jacob
>
>
>
>Jason Morris Lakin writes:
> > Hi Jas, others.
> >
> > Can i get some help with the intuition on the formula for the Betas?
Why
does
> > it make sense that we would standardize the
coefficient by only the
variance of
> > the independent variable? What is the
difference between the
information we get
> > from the correlation (standardized by a
version of the variance of
both
> > variables) and what we get from the B
(standardized only by our
modeled
> > independent variable)? I understand the
equations but not their
meaning...
> >
> > thanks.
> > jason
> >
> >
> > Quoting Jasjeet Singh Sekhon <jasjeet_sekhon(a)harvard.edu>du>:
> >
> > >
> > > The online homework file Homework_Assignment_6.pdf has been
corrected.
> > > See
http://www.courses.fas.harvard.edu/~gov1000/assignments/
> > >
> > > Cheers,
> > > Jas.
> > >
> > > Jasjeet Singh Sekhon writes:
> > > >
> > > > Wherever it says "employment" in an equation it should
say
> > > > "unemployment". I see that the typo is also in one
equation in
> > > > question #3. I'll correct the online file.
> > > >
> > > > Jas.
> > > >
> > > > Jason Lakin writes:
> > > > > Greetings all. In problem 2, is the independent variable in
the model
> > > correctly named, or should it say
"unemployment"?
> > > > >
> > > > > thanks.
> > > > > jason
> > > > >
> > > > > <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0
Transitional//EN">
> > > > > <HTML><HEAD>
> > > > > <META http-equiv=Content-Type content="text/html;
> > > charset=iso-8859-1">
> > > > > <META content="MSHTML 6.00.2800.1276"
name=GENERATOR>
> > > > > <STYLE></STYLE>
> > > > > </HEAD>
> > > > > <BODY bgColor=#ffffff>
> > > > > <DIV><FONT face=Arial size=2>Greetings all. In
problem 2, is
the
independent
> > variable in the model correctly named, or should it say
"unemployment"?
> > </FONT></DIV>
> > <DIV><FONT face=Arial size=2></FONT> </DIV>
> > <DIV><FONT face=Arial size=2>thanks.</FONT></DIV>
> > <DIV><FONT face=Arial size=2>jason</FONT></DIV>
> > <DIV><FONT face=Arial size=2></FONT> </DIV>
> > <DIV><FONT face=Arial
size=2></FONT> </DIV></BODY></HTML>
> _______________________________________________
> gov1000-list mailing list
> gov1000-list(a)fas.harvard.edu
>
http://www.fas.harvard.edu/mailman/listinfo/gov1000-list
_______________________________________________
gov1000-list mailing list
gov1000-list(a)fas.harvard.edu
http://www.fas.harvard.edu/mailman/listinfo/gov1000-list
_______________________________________________
gov1000-list mailing list
gov1000-list(a)fas.harvard.edu
http://www.fas.harvard.edu/mailman/listinfo/gov1000-list