There is no section tonight!!
Quoting Alexander Greenfield Liebman <liebman(a)fas.harvard.edu>:
> hey jacob,
>
> i'm wondering if there was any resolution about the section/gov party
> conflict. i think a bunch of us would like to go (especially because it would
>
> be lame if none of the first years were there -- and we'd like to see what
> the
> skit is like before we do it next year).
>
> just wondering if you were moving section or what.
>
> thanks,
> alex
>
>
I was looking at data.approval$inflation and noticed that a lot of
numbers repeat, often one right after the other. For example, the
second and third entries are both 0.7547206, later on there's a couple
of 0.7462721, and so forth. I think it'd be pretty rare to get the
exact same inflation rate to 7 decimal places so often. Does this have
something to do with the way inflation is calculated?
Also, I don't really understand the difference between inflation and
inflationMonthly, what are we supposed to be taking into account when
choosing one variable over the other?
Kai-Hua
--
~-~-~-~-~-~-~-~-~-~-~-~-
Kai-Hua Yu
364 Currier Mail Center
Cambridge, MA 02138
(617) 493-6916
Currier House, Gilbert 416
Hi Jason,
This isn't really on point for Table 5, but it makes complete sense to
put in approval at t-1 and approval at t-1 in a model. The effect of
approval t-2 is only accounted for by approval t-1 if we assume the
effect geometrically declines over time. See the document I refered
to in the email on lags I sent out earlier. But they didn't include
multiple lags for the model in Table 5. This is really an issue for
the granger causality stuff which doesn't concern us in homework #2.
Jas.
Jason Morris Lakin writes:
>
>
>
> Hey Jas: question from note 3. It says that they lagged the 'candidate
> "causal" variables' by a quarter and that they have added 2 quarters previous
> approval...shall we interpret this to mean they used both a lag and the real
> variable, or did they just use the lag? we think it might mean they used both
> the lag and the actual, so, in other words, they included for inflation:
> inflation at t and inflation at t-1, and for approval: approval at t-1 and
> approval at t-2. however, it seems a bit redundant to include both approval at
> t-2 and t-1---the effect of t-2 should be sopped up in t-1...or maybe not. Help!
>
> jason et al
>
Model 2 gives the significance of the t+1 variables as .017 -- has
anybody been able to get it that low? Am I correct in that we're
basically comparing model 2 with model 1 to get that figure?
Kai-Hua
--
~-~-~-~-~-~-~-~-~-~-~-~-
Kai-Hua Yu
364 Currier Mail Center
Cambridge, MA 02138
(617) 493-6916
Currier House, Gilbert 416
Hi Sean,
In the question, random assignment was successful so there is NO
association between vouchers and any other baseline variable (observed
or unobserved).
> (one of the benefits of experimental design as the Reiter article repeatedly
> said--randomized experimental designs trump the observational efficacy of
> regression by already controlling for a whole slew of potential errors).
Correct. The correlation between vouchers and the other variables is zero.
> but there is obviously something going
> on here involving how Y mapped onto parental ed/income and race is pushing the
> voucher beta way up on the significance scale. I thought it was interaction,
> but now I am puzzled.
There is something going on. I would suggest looking at the equation
for how the variance of vouchers (or X1 in the multiple regression) is
estimated. This tells us a lot.... Reading about multiple regression
in W&W may give you a big hint.
I like question 1 a lot--:)
Jas.
Sean L. Yom writes:
> Jas,
>
> Somebody mentioned a point that made me rethink how I answered number one--and
> thus much of my thinking about regression. How much are we to assume about the
> researcher (not the reviewer) in terms of random assignments?
>
> My initial thoughts were thought if the total effect of the simple gave
> insignificant results (I assume through p-values or F-testing) and adding other
> variables makes the voucher coefficient significant, then any significance can
> be purely attributed to the presence of other variables. My intuition tells me
> that there is interaction occurring between the variables, they are related
> (i.e., parents with poor income will also tend to be less educated, and in the
> real world they will be the ones getting vouchers). But does random assignment
> mean that the researcher does not pay attention to *any* of other variables in
> the design? I.e., the researcher doesn't pick a population with lesser edu.
> parents with lower incomes who happen to be federally disadvantaged persons of
> color, the researcher gives vouchers to a set number of people who span the
> normal distribution and have no appreciable racial or socioeconomic profile?
>
> If so, then I have some trouble conceptualizing the problem. Random assignment
> of vouchers with *no* knowledge of other variables means the researcher can say
> that there is close to zero correlation between vouchers and observed variables
> (one of the benefits of experimental design as the Reiter article repeatedly
> said--randomized experimental designs trump the observational efficacy of
> regression by already controlling for a whole slew of potential errors). If
> that is the case, then can there be any interaction effects between variables
> which have no correlation by design? So the significance gained by adding
> other variables cannot be attributed to interaction effects? More data means
> less residual error, that much I know; but there is obviously something going
> on here involving how Y mapped onto parental ed/income and race is pushing the
> voucher beta way up on the significance scale. I thought it was interaction,
> but now I am puzzled.
>
> Sorry for long-windedness, but I may have misconceptualized a major issue in
> multiple-variable regression and experimental design, and want to nip the
> problem in the bud before I go out in the Middle East and start spouting
> significance tests...
>
>
> Thanks in advance,
>
> Sean
>
-----Original Message-----
From: Jacob Kline [mailto:jkline@fas.harvard.edu]
Sent: Wednesday, December 10, 2003 12:19 PM
To: 'Kai-Hua Yu'
Subject: RE: [gov1000-list] Significance of t+1 variables
Yes and No -- you are comparing model 2 with a model that is a
restriction of model 2 -- removing the t+1 variables. In this case, the
restricted version of model 2 is the same as the unrestricted model 1.
-----Original Message-----
From: gov1000-list-admin(a)fas.harvard.edu
[mailto:gov1000-list-admin@fas.harvard.edu] On Behalf Of Kai-Hua Yu
Sent: Wednesday, December 10, 2003 6:40 AM
To: gov1000-list(a)fas.harvard.edu
Subject: [gov1000-list] Significance of t+1 variables
Model 2 gives the significance of the t+1 variables as .017 -- has
anybody been able to get it that low? Am I correct in that we're
basically comparing model 2 with model 1 to get that figure?
Kai-Hua
--
~-~-~-~-~-~-~-~-~-~-~-~-
Kai-Hua Yu
364 Currier Mail Center
Cambridge, MA 02138
(617) 493-6916
Currier House, Gilbert 416
_______________________________________________
gov1000-list mailing list
gov1000-list(a)fas.harvard.edu
http://www.fas.harvard.edu/mailman/listinfo/gov1000-list
Note 5 on page 609 ends with these sentences:
"The event 'effects' in the Approval equations are in the range of 6
points and highly significant. In every case, we have been careful to
exclude information from the previous presidential administration from
the Approval prediction equations."
Can somebody explain these two sentences to me? I'm thinking the first
sentence has something to do with their events series, but I don't know
what they mean by the 6-point range (I thought their events were either
+2, +1, or -1). For the second sentence it seems like they are throwing
out some variables (lag variables, maybe?) when another administration
begins. Can somebody unravel this?
Kai-Hua
--
~-~-~-~-~-~-~-~-~-~-~-~-
Kai-Hua Yu
364 Currier Mail Center
Cambridge, MA 02138
(617) 493-6916
Currier House, Gilbert 416
Hi,
A number of you have asked for substantive readings on the issues the
homework is about. If you want to readup on this stuff (in a
substantive way) you may want to take a look at the following books.
THESE ARE PURELY FOR YOUR INTEREST (they are in not required in any
way):
Alesina, Alberto, and Howard Rosenthal. 1995. \textit{Partisan
Politics, Divided Government, and the Economy.} New York: Cambridge
University Press. ISBN: 0521436206.
*The Alesina and Rosenthal book is extremely good work and probably the
*best statement to date of the issues we're talking about.
Hibbs, Douglas. 1987. \textit{The American Political Economy:
Electoral Policy and Macroeconomics in Contemporary America.}
Cambridge, MA: Harvard University Press. ISBN: 0674027361.
Blanchflower, David G., and Andrew J. Oswald. 1994. \textit{The Wage
Curve.} Cambridge, MA: MIT Press. ISBN: 026202375X.
The last one is just about economics, but is relevant. The first two
are about politics AND economics. The Hibbs book is a bit dated but
an easier read than the Alesina and Rosenthal book. If you just focus
on the data parts of the Alesina and Rosenthal book, it is pretty
readable (ignore the formal models).
Since we are talking about MacKuen et al., you may also want to take a
look at:
Erikson, Mackuen, Stimson. The Macro Polity. Cambridge University
Press. January 2002.
Cheers,
Jas.
Hello all,
If you're publishing an article using regression, is it kosher to include
interaction effects in your model without mentioning that you did in the body
of the article, the charts, or the footnotes?
Thanks.
Alex