----- Original Message ----- From: "JDG" <[EMAIL PROTECTED]> To: "Killer Bs Discussion" <[email protected]> Sent: Thursday, May 05, 2005 10:58 PM Subject: Re: US riches, actual and hypothetical
> At 10:24 PM 5/5/2005 -0500, Dan M. wrote: > >> Unless you and Dan have some brilliant economic theory as to why > >> Republicans tend to cause recessions and Democrats tend to produce > >> uninterrupted economic growth regardless of the business cycle, your > >> analysis is deeply flawed. > > > >This is one area where we differ. I believe that data come first, theory > >comes second. > > In Economics, the prevalence of spurious correlations makes that a > dangerous paradigm. I won't say that no serious Economists follow that > paradigm, but "data mining" is broadly looked upon with skepticism in > Economics. But, I asked a very obvious question: the effects of the implementation of two general schools of thoughts on ecconomics. It's not as thought I tried hundreds of correlations, until I got a 3 sigma one. > One reason for this is that Economics relies heavily upon time-series data, > and any two non-stationary time series will tend towards correlation over > time. > > To give an example from another case of mixing Economics and Presidential > Politics, there is a Economics professor - I believe at Yale - out there, > who on a bit of lark constructed an Economic model that predicts the > outcome of the two-way US Presidential race based upon economic factors. > By all the usual statistical tests, this model is very robust. And yet, > every four years that same model is spectacularly wrong. And so, after > each Presidential election the model is tweaked to account for the latest > observation - all to no avail. Every four years the model's future > predictions are invariably wrong. Do you know _why_ what I did and what he did are different? > So, to return to the original point, the data says that 8 out of 9 > recession have occurred under Republican Presidencies. Do you believe > that this is inherently significant? If you look at the policies that were undertaken by Democrats and Republicans, then I would expect recessions to be more likely, longer and worse when Republican economic techniques are used. In my time series, we looked at two 12 consecutive year spans when Republicans had the White House and one 20 year span when Democrats did. Try a number of different types of cyclical functions, and see how likely this type of occurrence would occur randomly. One would expect a cycle to have periodicity that one doesn't really see here. In short, I agree one has to be very careful about seeing correlations when one tries hundreds of different combinations until one sees a signal. But, when one asks the one obvious question about Republican vs. Democratic economic policy, one can use the statistics that are valid for asking 1 question, not trying hundreds of questions. It may be worthwhile to start a thread on statistics. I'd be willing to walk through the foundations of statistics and ways to check for valid vs. invalid use of statistics. As I've mentioned before; Monte Carlo techniques are very good at turning tacit assumptions into explicit assumptions, allowing one to more clearly see the question one is asking. Dan M. _______________________________________________ http://www.mccmedia.com/mailman/listinfo/brin-l
