Asymmetrical Information

You should send that link to Kevin at Calpundit.

Those kinds of statistics seem to be his speciality.

You can also to your list that Cliton benefitted from inheriting a growing economy. He came into office shortly after the unemployment situation took a turn for the better.

"Reagan was the unlucky bastard whose Fed chief had to shut down the party by raising short term interest rates to 20% to get the inflation under control, which sadly hurt his employment numbers."

Actually, the unlucky bastard was Carter, who appointed Volcker, and who saw not his unemployment numbers but his re-election chances disappear.

Bernard: interest rates topped out during Carter's presidency, but Reagan got the recession because of the lag. Carter wasn't doomed by Volcker, but by miserable poll ratings that far predated the 20% fed funds rate.

Hmmm...."Democratic power coincided with the long post-war economic boom."

Let's see, postwar we have 8 years of Truman, then 8 years of Eisenhower, then 8 years of Kennedy/Johnson, then 8 years of Nixon/Ford. I'm having a hard time seeing Democratic dominance there.

Actually, it turns out that even if you do this analysis with a lag time built in (you can pick whatever lag you want), the Dems still do better. But really, wouldn't it be enough to just make the case that neither side is much better than the other, instead of this rather tortured explanation of how all the Republicans were just unlucky because something really bad happened on their watches? I'm sure I could come up with a similar list for the Democrats if I thought it meant anything.

And just a historical note: FDR wasn't a Keynesian. He spent a lot of money, but not really out of any theoretical belief in pump priming, I think.

Spin Spin Spin...

Now, would you rather be looking for work with Roosevelt at the helm, or Bush? Don't know, What was Roosevelt's average number? Bush's has been pretty bad, though certainly not amazing by historical standards. It seems clear that he's doing nothing at all to improve the situation.

since even uber-partisan Democratic economist Paul Krugman will tell you the boom was the result of rising productivity.
Which in turn was a consequence of good government. Or are you now pretending that governement policy has no affect on productivity?

Also, Krugman is hardly a partisan democrat. He's an Anti-Bush partisan which is nowhere near the same thing. Anti-Bush partisans include most of the well-informed sane people as well as democrats.

17% was the low point, Bones. He started at 25%.

There you go, he started with awful and made things better.

And Kevin, the boom ended in 1970. Nixon presided over the very tail end.

I wasn't trying to argue, as I pointed out in the piece, that either party was particularly brilliant on unemployment. I was pointing out that there was a major exogenous variable -- the stellar productivity growth of 1940-1970 -- that makes such comparisons nearly meaningless, as well as the fact that the Keynsian policies that contributed to lowering unemployment have awfully unpleasant side effects that are missed by the table. I wasn't attempting to use the data to prove anything other than the fact that a single datum usually isn't a good measuring stick.

"Or are you now pretending that governement (sic)policy has no affect on productivity?"

Stipulating that government can affect productivity growth, it certainly has been only one of many factors, and far from the strongest. Key innovations like the semiconductor and changes in business model (like Wal-Mart's vendor and inventory management) account for substantial changes in productivity.

Of course, it was the non de-regulation de-regulation of the S and L industry that made the crisis so bad. I think that did happen during Reagan's first term, although I don't recall what input the Democrats had on the issue. Anybody who has more than $10,000 to invest, and wants the guarantee of the taxpayers, should be forced to buy Treasury securities, with the rates and liquidity such purchases entail.

Anybody who has more than $10,000 to invest, and wants the guarantee of the taxpayers, should be forced to buy Treasury securities, with the rates and liquidity such purchases entail.

WTF? This sounds like something I might have written as a lark, after eight or so beers.

FDR started WWII?

Actually, it turns out that even if you do this analysis with a lag time built in (you can pick whatever lag you want), the Dems still do better.

Oh brother, not this again.

"17% was the low point, Bones. He started at 25%."

"There you go, he started with awful and made things better"

He started with awful and kept them that way longer than necessary, delaying the recovery, by following the policy of setting up monopolies, cartels, artificial price supports and other trade restrictions.

The New Deal people had no coherent understanding of the economic causes of the Depression, but they strongly associated it with "falling prices" so they did what they could to rig prices upward as a counter. But monopolies, cartels, trade restrictions and price rigging were no better for growth in the 1930s than they are today.

There's an interesting Fed paper on how much various causes contributed to the Depression at http://minneapolisfed.org/research/qr/qr2311.html

Anybody who has more than $10,000 to invest, and wants the guarantee of the taxpayers, should be forced to buy Treasury securities, with the rates and liquidity such purchases entail.

WTF? This sounds like something I might have written as a lark, after eight or so beers.

Nah. Makes perfect sense. It's getting government out of the bank insurance business. The FDIC encourages high-risk lending, because the bank's depositors will get bailed out, no matter what.

I think they call this a "moral hazard."

GBI got the S&L crisis [...] None of these things had anything to do with Republican policy;

First of all, the S&L crisis had everything to do with Republican policy because it was fueled by a deregulatory fever that is the raison d'etre of the GOP.

Second, one of the more egregious bank collapses occurred at Silverado, under the stewardship of Neil Bush, costing U.S. taxpayers upwards of $1 billion dollars. If G.H.W.B. should not bear the blame for bringing his worthless incompetent son into this world then who should?

Third, the job creation numbers look even worse for the GOP if we factor in Herbert Hoover (cleverly omitted from the original post). I suppose we are to assume that the economy was just chugging along until F.D.R. took over?

Fourth, it is too easy to let G.W.B. off the hook. He has terrible job creation numbers because his administration has completely failed to undertake fiscal policy initiatives that are likely to lead to job growth. Sure, the tail end of the bubble bursting was going to be a drag, but we all knew that. And yet Bush continues to swear by tax cuts that no credible economist thinks is going to lead to significant job creation. Sorry, got to place blame where blame is due.

"There you go, he started with awful and made things better."

Sure, after eight years he found a very effective way to end the Depression--get involved in WWII. The right decision, of course, but FDR could hardly claim that it was a stroke of economic genius to involve the nation in a war of survival to get the economy out of the crapper--that was just a fortuitous side effect.

"Third, the job creation numbers look even worse for the GOP if we factor in Herbert Hoover (cleverly omitted from the original post). I suppose we are to assume that the economy was just chugging along until F.D.R. took over?"

And it omits the eight years before Hoover, when the economy was booming and Republicans were also in power--what's your point?

The very concept of job creation is very nebulous. Are these jobs of true value or a waste of valuable resources? Did you say that you desire to create more jobs? If so, you should make it illegal for a business owner to become more efficient. A construction business, for instance, should be forced to hire workers who use only tea spoons for digging instead of mechanical devices.

Am I being facetious? Not in the least! There is no sense engaging in this sort of discussion unless you consciously admit that job creation per se is not always a good thing. The destruction of jobs in a macroeconomic sense is often what we seek to accomplish.

Space:

While a lack of regulator oversight certainly contributed to the S&L problem, the primary cause was the mismatch between the duration of S&L deposits and loans. The S&Ls had 3% mortgages at 30 years, and savings depositors demanding interest rates double or triple that. Some of the S&L people used their banks as private hog wallows to exploit, but most of them simply made stupid investments. The reason for the deregulation was, after all, that the S&L's were going bankrupt without it, because they were forced to pay depositors much more than their lenders were paying them.

As for the rest, it's either irrelevant or wrong. Whatever Neil Bush did, he did not contribute a statistically significant amount to the crisis. I used the starting period provided by the author of the chart, but of course if you want to push it back to the beginning of the century, we can make the Republicans look much, much better. And since I don't believe it's possible to stimulate your way out of a recession, I'm afraid your last point falls on deaf ears.

hi mse,
"And it omits the eight years before Hoover, when the economy was booming and Republicans were also in power--what's your point?"

actually, the republicans in the twenties saw three booms and three busts before the great one. info as follows:

There were bull markets, notably in the periods:
July 1921- May 1923 (22 months),
July 1924- October 1926 (27 months),
November 1927 to August 1929 (21 months).

But there were also three sharp recessions in:
Feb 1920-July 21 (18 months)- A depression as well,
May 1923- July 1924 (14 months),
October 1926- October 1927 (13 months)

hi jg,
loved this:

"Looking at that, one would think that Roosevelt was the paradigm of job creating wizardry, while George Bush was a job-hating moron."

love it...

Cas,

"[T]he republicans in the twenties saw three booms and three busts before [Oct 1929]... there were also three sharp recessions in:

Feb 1920-July 21 (18 months)- A depression as well,
May 1923- July 1924 (14 months),
October 1926- October 1927 (13 months)"

A nit: IIRC, Harding was inagurated in early March 1921, so it seems unlikely that the last four months of a post-WWI/post-influenza-epidemic/post-Wilson-stroke depression had much to do with Harding's policies.

The more interesting question for the New Dealers in the audience--why did Oct 1929 lead to a decade of depression, when all the previous panics and depressions, from the Jackson administration to the 1890s and 1920s, blew over in a year or two?

Cas . . . a decline in a bull market is not the same thing as a recession.

"Sure, after eight years he found a very effective way to end the Depression--get involved in WWII. The right decision, of course, but FDR could hardly claim that it was a stroke of economic genius to involve the nation in a war of survival to get the economy out of the crapper--that was just a fortuitous side effect."

The 17% was in 1939, the last 'non-war' year of his presidency, as jg said in her original post. So WWII had nothing to do with the decline from 25% to 17%. Nitpicking, I know.

Jane,

Unemployment was 7.5% when Carter took office. It dropped to around 6% and stayed there, or slightly under until Jan. 1980, when it began to rise, reaching 7.5% by the fall of 1980 - election season. (It went over 10% in the early part of the Reagan administration) This corresponded rougly, unsurprisingly, with sharp increases in interest rates created by Volcker. As expected, the economy also experienced a contraction during the first half of 1980.

Poll numbers don't fall from the sky. They are a consequence of other things. If Reagan was "unlucky" in terms of this table, that bad luck was more than compensated for by his good luck in having Carter appoint Volcker, whose policies, while sensible, predictably contributed to Carter's defeat and ultimately to the prosperity of the 1980's.

Actually, it did have something to do with it, wallster, as demand was already ticking up during that period due to armament.

hi jg,
"Cas . . . a decline in a bull market is not the same thing as a recession."

you have lost me, could you please explain?

cas-

Depressions and recessions usually refer to falls in GNP. In other words, the output (goods and services produced) of the country get smaller.

A decline in a bull market (a clumsy phrase) is simply an event or time period during which the value of stocks goes down in the midst of a longer, sustained period where stock prices are rising. Since "bull market" is a broad term, I suppose you could also apply that phrase to any market (bonds, real estate, etc...)

Bob:

I guess the point of confusion is why someone who wants taxpayer supports ought to be forced to do anything at all. Expectations need not be realized, after all.

Expecting that to be the case can frequently be the result of having had a few too many. Hence my comment.

Cas, the measurement error in GNP is such that it's not really appropriate to call 1926-7 a recession; GDP was roughly flat, but didn't really decline, and unemployment, at 3.3%, was right around trend. You really don't want FDR going toe-to-toe with his Republican predecessors on any metric; GNP grew by nearly 40% over the 1920's, and unemployment ended the decade way down. But FDR had the Great Depression, you will say, and you're absolutely right, which is why it's stupid to make charts like this -- exogenous variables skew the set.

And here I thought that the S&L crisis was caused by TEFRA 86 which did away with the tax deduction for passive loss investments in real estate, which caused the stock market crash of '87 and greatly reduced the value of real estate in the country. (I've since seen a theory that the Asian markets also contributed to the stock market crash of '87.) Because S&L's were/are permitted by law to invest only in real estate, and because the banking regulations required portfolios to be marked to market, there were many many good performing loans which instantly became "undercollateralized." By definition, then, many S&L's became failing banks, and when the regulators went in to take a look, perhaps up to 10% of them had underlying fraud which had been unobservable due to their previously good performance. Then FIRREA was passed at the worst possible time (as viewed strictly from a "timing" point of view) and exacerbated the situation. A simpler solution would have been to temporarily change the banking regs so that performing loans could continue to be carried and the collateral valued at maturity instead of at market. (The underlying collateral, real estate, came back up to its pre-1987 prices by around 1997-98 in the New York metro area.) The 5-20 billion that the S&L bailout cost the taxpayers could have been greatly reduced if the taxpayers only paid for non-performing loans and not for performing loans that suddenly became undercollateralized.

BTW, a performing loan is one which is being paid off on time. A non-performing loan is one on which payments are not being made. Valuing at market means valuing the asset at current market prices, while valuing at maturity means valuing the asset when the asset reaches maturity. Think of a $10,000 bond which matures in 30 years and pays 6% interest a year. The market will establish a price for that bond based on whether the 6% interest is higher or lower than what an investor could get on a similar grade investment today. So, if available interest rates are, say, 12%, the face value of the bond ($10,000) would be discounted to compensate for the fact that it only pays 6%. Nevertheless, the bond will still pay $10,000 at maturity no matter what you pay for it today. So, the bond has a maturity value and a market value.

I think we're talking about two different things. The underlying rules, as I understand it, were changed in the early 80's in order to allow the S&L's to invest in real estate and related assets directly, because they were losing so much money on mortgages that without the rule change, they would have become insolvent. They used the new rules to not only invest in real estate, but also invest in mortgage bond deals that were facially ludicrous -- sell a bunch of mortgages to an investment bank to package, and then buy a similarly valued package of mortgage bonds, paying a fee to the bank in the process -- but camoflaged by rising prices. Etc. You're discussing the denouement; I'm discussing the original deregulation. But perhaps my understanding is flawed?

That the political cycle is loosely tied to the business cycle explains it pretty well.

For what it's worth, Michael Lewis described the S&L crisis (in Liar's Poker) as the logical outcome of "let the S&Ls gamble their way back to solvency."

If Reagan was "unlucky" in terms of this table, that bad luck was more than compensated for by his good luck in having Carter appoint Volcker, whose policies, while sensible, predictably contributed to Carter's defeat and ultimately to the prosperity of the 1980's.
~~~

A good part of Reagan's "luck" in having Volcker create the prosperity of the later 80s for him -- according to Volcker, in his book -- was that Reagan shielded the Fed from political interference even as unemployment went over 10%, the worst since the Depression, just before the 1982 Congressional election, with the Republicans taking a tremendous pounding and Reagan's own poll numbers plunging to their all-time low as a result. Volcker pointed out that this was not typical behavior for US presidents and he greatly admired it.

Reagan's biographers say this period of the 10+% unemployment and the pounding he took for it was the most difficult time for him of his entire political career, but that he trusted that having the Fed end inflation was the best thing to do, and had faith that in the end the nation would be better off for it and he would be too. And he was right.

Megan,
As I understood it at the time, CMO's backed by Ginnies, Freddies, and Fannies were considered to be equivalent to real estate and thus S&L's could invest in them. I don't know if a rule change was necessary or not; I never heard that one was required. S&L's from their inception (they are creatures of Congress, not the marketplace) could only invest in real estate. I was working at Texas Commerce Trust Company from 1987-1990, which was the agent bank for Texas Commerce Bank (which was bought by Chemical Bank) who was the Corporate Trustee for the vast majority of CMO's produced at the time. CMO's were co-developed by Solomon Bros. and TCB, so we ended up being the transfer agent and holding all the collateral for CMO's issued by Solomon, Merrill Lynch, Drexel Burnham Lambert, etc. Eventually Freddie and Fannie started issuing their own CMO's. I remember reviewing the legal paperwork and signing the bonds as transfer agent for what was packaged as a residual CMO bond, i.e., whatever was "left over" after taking the interest and principal streams of the collateral, forcasting repayment from refinancings and defaults, and developing CMO classes of bonds. The purchasers were banks such as Silverado and Columbia, which were some of the more spectacular failures. The residuals carried some high risk, but potentially high reward, and since they were considered "real estate", the S&L's could invest in them. Not too many did, but some did.

Another point on the S&L crisis that I don't think has been mentioned is that they were only half-deregulated -- the deposit side remained protected and the "reform" even increased the amount of deposit insurance, IIRC. That gave the S&L operators the incentive to attract great amounts of money for use in very risky investments while marketing them as "risk-free insured" and playing a game of "heads I win, tails somebody else loses".

Moreover, the S&Ls were hopelessly inept at managing competitive investments after 50 years of being sheltered from the market as regulated as 3-6-3ers (borrow at 3%, lend at 6%, on the golf course at 3pm). In "Liars Poker" Lewis has stories of commercial bankers who were, um, grossly taking advantage of the S&Lers to the point where they actually felt guilty about it and pauseed to ask "does the fact that they are *asking* us to do this to them make it ethical"? [Well, they used much more graphic language but I don't want to offend.]

Even Krugman in his book that covered this said the S&L crisis was a result of Congressional mal-regulation rather than deregulation, and refused to blame it on Republicans -- in fact giving credit to Bush I's people for being willing to recognize and deal with the problem to their political cost, after Congress and their executive branch predecessors had covered up the problems and let them fester for years to make things worse. (Or course, that was the pre-NY Times Krugman).

"That the political cycle is loosely tied to the business cycle explains it pretty well."
~~

Hoover was like Eisenhower in that in his time he was extremely popular and viewed as skilled rather than politically partisan, and both parties wanted him to run for them. Keynes wrote practically an Ode to Hoover based on Hoover's post-World War I public performance (while JMK was notably blasting everyone else).

If Hoover had won as a Democrat in 1928 then after the Depression hit we'd have had Republicans in power for the next 30 years.

Jim, that is what I mean by "non-deregulation deregulation". It is my non-scientific opinion that when politicians tell us that they are going to "deregulate" an industry they are just as likely to malregulate as deregulate. For every success (for consumers, if not stockholders) like the airline industry, we can probably find a debacle like the S and L industry, or the California electricity fiasco.

... that is what I mean by "non-deregulation deregulation".
~~~

Right. The California electricity market was a prime example of "deregulation" actually being a new set of bonehead regulations put together by various interest groups through extensive lobbying to protect themselves.(No long term contracts ... brilliant!)

Smarter deregulation may be able to eliminate that problem, but the S&Ls and California electricity market both demonstrated another problem that may be unavoidable: that when everybody in an industry has been sheltered from unregulated competition for their entire careers, many of them just aren't going to have the skills and ability to handle it if they are suddenly thrown into it. There's no way around that. You've got to expect some carnage before the ones who can handle it emerge.

"A good part of Reagan's "luck" in having Volcker create the prosperity of the later 80s for him -- according to Volcker, in his book -- was that Reagan shielded the Fed from political interference even as unemployment went over 10%,"

Then Reagan behaved admirably, as did Carter before him.

http://www.pla.blogspot.com/ has a must-read response to Jane's argument.

While I do not disagree with Jane’s larger point that it is difficult to draw meaningful conclusions from small data sets, what is striking about the job growth data is how clean it is. Job growth was higher under every Democratic President than under any of the seven Republican Presidents.

How large is the possibility that such a dramatic separation would occur through random chance?

He does the math and concludes that the chance is about 1 in 2000.

There's lots more there about other measures of performance as well.

The patsies who put together the new California electricity regs, under the helpful instruction of the sharks who put the patsies to sleep, actually thought they could speculate in electricity effectively, rendering long-term contracts unnecessary. Not to be envious, but why is it so seldom that I find these people at a card table?

The S&L crisis had its roots in the rise of the money market fund and the high interest rates of the late 70's. The rates S&L's paid depositors were limited, so they lost deposits to the money funds.

When this regulation was removed they were able to compete for funds, but, as others have pointed out, two problems arose: Deposits were newly insured to $100,000, up from (I think) $40,000; and lots of the managers didn't really know what they were doing.

The insurance situation created moral hazard - you could always attract new depositors, and the incompetence (along with a fair amount of fraud) meant you needed to, because lots of bad investments were made.

"Bernard: interest rates topped out during Carter's presidency, but Reagan got the recession because of the lag."

Wrong. Real GDP dropped -1% in 1980, recovered in 1981, and then dropped -2.2% in 1982. 20% interest rates are what happens when you have the adminstration trying a stimulative fiscal policy while the Fed is running a contractionary monetary policy.

Jim Glass wroteL "He started with awful and kept them that way longer than necessary, delaying the recovery,"

If by "delaying the recovery" you mean "maintained a 7.4% mean GDP growth from 1933-1941" that is.

Care to name just *one* year where a GOP president managed to beat that mean growth? Go on, try.

Sorry to bother: has anyone crunched these numbers to verify them? Apparently the original link is the NY Times (see graphic; http://www.nytimes.com/2003/07/03/business/03JOBS.html). They say the sources were BLS and Economic Policy Institute. I can't find the data on the EPI site. I was looking at BLS numbers (annual averages, from ftp://ftp.bls.gov/pub/special.requests/lf/aat1.txt (link from P.L.A) or the monthly data here http://data.bls.gov/cgi-bin/surveymost?bls) and I get that under Clinton civilian employment rose (base year 1993 through 2000) by 16.6 million not 22.71 million, and Reagan (base year 1981 through 1988) by 14.6 million, not 16.1 million (these descrepencies are 37% and 10%, respectively, and Clinton didn't increase the *non*civilian workforce right? I don't see how they can be this different) What am I doing wrong? Is there something more to this than arithmetic? If you then take the percentage growth from the base year (how are they doing it?), it is 14.5% and 13.8% for Reagan and Clinton, respectively (R then C, NOT annualized). Sorry if I am missing something obvious. Thanks.

Bones, PLA is not quite the statistical wizard he thinks he is. The probability that this particular outcome would occur if it were a coin toss is indeed 1 in 2000. However, in a multivariable world, such calculations are just silly.

If by "delaying the recovery" you mean "maintained a 7.4% mean GDP growth from 1933-1941" that is.

Care to name just *one* year where a GOP president managed to beat that mean growth? Go on, try.
~~~

1983Q1 to 1984Q1, 8.6% working with the handicap of following up a much, much milder recesssion.

http://research.stlouisfed.org/fred/data/gdp/gnpc96

For more references on how the New Deal slowed the pace of recovery from the Depression in the US, if you don't like the Fed paper I linked to, you can refer to such good liberal Keynesian economic historians as Peter Temin of MIT and his "Lessons from the Great Depression" Lionel Robbins lectures, in paperback even. He goes into some detail on how the New Deal and NRA slowed the recovery in the US into being the slowest anywhere in the world.

BTW, you seem to think monopolies, cartels and higher legislated prices of the New Deal were growth boosters in the 1930s, when enacted by Democrats. Would you support them as growth boosters if enacted today by Republicans?

"1983Q1 to 1984Q1, 8.6% working with the handicap of following up a much, much milder recesssion.

http://research.stlouisfed.org/fred/data/gdp/gnpc96"

Using annual numbers (not quarterly; the numbers I was using for FDR were annual), the BEA gives the Real GDP growth in 1984 as 7.3%, and the growth in 1983 as 4.3%. Sorry.

Yeah, I don't like the Fed paper you gave, because it has one howler at the start: it says output remained below 1929 levels in 1939. This is (barely) correct in nominal terms, but neglects the fact that you had a 25% deflation from 1929-1932; so output (in real terms) surpassed the 1929 level by 1936.

"BTW, you seem to think monopolies, cartels and higher legislated prices of the New Deal were growth boosters in the 1930s, when enacted by Democrats. "

Did I say that, or did you just want me to?

"However, in a multivariable world, such calculations are just silly."

Why, exactly, are they silly?

Because the odds of an outcome in a single variable experiment has little to do with the odds of a particular outcome in a multi-variable experiment. I may as well quote the odds of experiencing 12 humid days in a row as if it were merely a coin-toss, and report the results as having relevance, regardless of whether I lived in Seattle or Phoenix. Wouldn't that be a just a tad silly?

Lots of factors affect humidity. Suppose I gave you average humidity readings for two unnamed cities over a period of years in the past and one was consistently higher than the other.

Would that arouse any suspicions as to which was Phoenix and which Seattle?

Bones, PLA is not quite the statistical wizard he thinks he is. The probability that this particular outcome would occur if it were a coin toss is indeed 1 in 2000. However, in a multivariable world, such calculations are just silly.
No, its not silly if what you are trying to determine is whether or not the party of the President IS one of the variables.

If it's not one of the variables, or dependant on one of the variables, then it should be expected to be random with respect to the outcome.

PLA's back of the envelope math suggests strongly that it is not. Now, this doesn't tell us whether this is a (partial) cause, or another effect of the underlying causes. But that's really another question.

It pretty much kills the argument that Republican presidents are better for the economy in any case. If the party of the president is a causal factor, then it's best to choose Democrat. And if the party of the president is NOT a causal factor, best not to use the economy as a deciding factor. Either way, it's stupid to vote for a republican because you believe it will be better for the economy.

Yes, of course, but that doesn't reduce the silliness of stating that the odds of having 12 consecutively more humid days in Seattle, as opposed to Phoenix, are xxxx-1, as if it were a simple coin toss. The odds favor that Seattle will have 12 consecutively more humid days, simply becasue it lies on the ocean, and Phoenix in the desert, and our simple and silly calculation actually misleads us. Imagine this statement: "The chance of Seattle being more humid 12 days in a row is xxxx-1, so it must be meaningful that Seattle lies at a more northernly latitude". One must attempt to account for all variables before one can draw meaningful conclusions. If it ain't a coin toss, it is pointless to calculate the odds as if it were, and then draw conclusions regarding a single variable's importance.

No, its not silly if what you are trying to determine is whether or not the party of the President IS one of the variables.

Yes it is silly...no make it stupid. What Dwight is saying is it is analogous to a coin flip.

Further, what this says is that if Gore was President we'd not have a recession. Never mind that the data in 2000 was pointing to a coming recession. Gore must have had a Secret Plan, that would have avoided all this unpleasantness (GDP would be higher, unemployment and inflation lower).

Now doesn't that just sound dumb?

If it's not one of the variables, or dependant on one of the variables, then it should be expected to be random with respect to the outcome.

That is just it. Nobody isn't saying it isn't one of the variables, just that it isn't THE ONLY GODDAMNED VARIABLE. Is that simple and clear enough?

As Jane pointed out, Volker raised interest rates. What could Reagan do about that? Fire Volker, he can't. Volker was appointed 2 years before Reagan got into office. Pointing to that and saying it is all Reagan is just silly.

So Jane is not saying that it is purely random (like a coin toss). Dwight's interpreting it that way is either ignorance or dishonesty.

Just to point out that when you see a little statistic like this, there's usually a bigger picture you're missing. And especially with presidents, where the meaningful data set is far too small to separate out the effects of chance from the effects of policy, it's very hard to draw meaningful conclusions.

What this means is that the statistical results could be result of random chance. Precision is related to sample size, the smaller the sample the less percision you tend to have. If I said I had a sample of 3 would you take me seriously?

Basically there is stuff that happens in the economy that is beyond the control of the President. So factoring in these "exogenous" effects into the analysis should be done. Failing to do so leads to highly questionable statistics.

Bones, if I knew nothing about the geography of the earth, and wanted to live in a city in the northern hemisphere with less humidity, and was supplied with statistics regarding latitude and humidity of Phoenix and Seattle, and then employed the reasoning in your post above, I might say, "It pretty much kills the argument that living closer to the equator provides more humidity, since the odds of having every day in Phoenix less humid than Seattle, if it were a coin toss, are xxxxxxx-1. This reasoning would be incorrect, since living closer to the equator would, on average, given the extreme lack of humidity of the interior northern continental masses in winter, actually likely be more humid. We simply don't have enough data to make that conclusion.

"Gore must have had a Secret Plan, that would have avoided all this unpleasantness (GDP would be higher, unemployment and inflation lower)."

"And especially with presidents, where the meaningful data set is far too small to separate out the effects of chance from the effects of policy,"

We're talking 50-odd years of data here. If you can't draw conclusions from that, better tell the entire econometrics profession to start flipping burgers.

"What this means is that the statistical results could be result of random chance. Precision is related to sample size, the smaller the sample the less percision you tend to have."

Err, no. In this case, we're talking about descriptive statistics, not predictive ones; we *have* the entire population of post-1948 GDP annual growth rates.
Growth rates under Dem v. GOP presidents *are* different, particularly in Years 2 & 3 of an administration; the differences are *not* a result of sampling of a portion of a population of data.

(Now of course, assigning those differences to differences in policy variables is a non-trivial matter. But, given the way the data breaks down GOPers have the uphill battle to fight).

I have it On good authority that PLA does not consider himself a statistical wizard in any way. He can barely count to 13 as his bridge partnets can attest.

I was just trying to calculate the chances of the rankings having all Democrats above all Republicans IF the rankings had been determined by random selection. At no time did I suggest that the rankings WERE random.

I happen to think that it is NOT random. To see if I was right, I tested the hypothesis that it was random and found it to be unlikely. Is that really so hard to understand?

Lots of you folks are trying to explain WHY the rankings came out as they did. I think that is a worthy exercise but it is not the exercise in which I was engaged. I was trying to determine WHAT happened not WHY.

WHAT happened is that over a whole host of measures, performance was better under Democratic presidents than under Republican presidents.

Explanations of WHY that happened are interesting but do not change the fact that it DID happen.

"Yeah, I don't like the Fed paper you gave, because it has one howler at the start: it says output remained below 1929 levels in 1939. This is (barely) correct in nominal terms, but neglects the fact that you had a 25% deflation from 1929-1932; so output (in real terms) surpassed the 1929 level by 1936."
~~~

Watch your howlers when accusing professionals of making howlers.

*Real* GDP (1996 dollars)...
1929: $822.2 billion
1936: $822.5 billion

http://www.bea.doc.gov/bea/ARTICLES/2002/08August/0802GDP_&Other_Major_NIPAs..pdf

That is *no* real growth, leaving the economy 20% to 25% under trend -- with unemployment to match, we will remember.

By your claim, the economy should have been 25% larger than in 1929 -- i.e., back at trend level with *full* employment. Which *self evidently* it wasn't, speaking of howlers.

"Did I say that, or did you just want me to?"

Well, you said the New Deal's policies didn't retard recovery. I was just wonderding if those same NRA policies -- upping prices by means of monopolies, cartels, legislated minimum prices, etc. -- wouldn't retard recovery today. With nothing else working maybe we should try them.

BTW, this whole bizarre discussion about "my presidents had higher growth rates than your presidents" might be put in perspective by noting that the Red Chinese have had higher growth rates even than FDR ever since Deng.

We should have Communists running the economy!!

Agreed,dwight, but I certainly perceived a desire on the part of some to draw conclusions about causation.

No Will, you projected a desire to draw conclusions about causation.

Dwight said nothing at all about causation, while I pointed out that IF you presume causation, it's not good news for Republicans, and if you presume NO causation, it's still not good news for Republicans (but for a different reason).

It's always easier to win an argument if you presume that your opponent is saying something stupid.

During the tax-cut debates, you couldn't throw a brick without hitting some liberal blogger who was going on about the wonderful things Eisenhower's 91% top marginal rate did for the economy. Now all of a sudden Ike's just a Republican again.

We're talking 50-odd years of data here. If you can't draw conclusions from that, better tell the entire econometrics profession to start flipping burgers.

If this were a time series analysis of say GDP or unemployment or something else along those lines, then yes, there'd be lots of data. But we aren't. We are looking at the question, does the political party of the President have an impact on things like GDP. Now a careful analysis might look like

GDPt = a + b1*x1,t + ... + bn*xn,t + g1*d1,t + ... + gk*dk,t + et

Where the xi,t are explantory variables, the dj,t are dummy variables. One candidate for dummies could be Democrat (1) Republican (0). (Yes I know this is not the only formulation for testing this. There are more sophisticated ways of going about it, but this would at least get going in the direction of doing more detailed analysis.)

But we aren't doing a careful analysis are we? No. We are simply calaculating some averages by political party over time, ignoring all other possible effects then pronouncing one party better or worse.

Dwight said nothing at all about causation, while I pointed out that IF you presume causation, it's not good news for Republicans, and if you presume NO causation, it's still not good news for Republicans (but for a different reason).

Sorry, but when you have things like this from Dwight

The performance under Democratic presidents was superior to the performance under Republicans in each of those measures.

I find it hard to believe that there wasn't at least a strong implication of causality there. And Will is certainly correct that others are drawing this conclusion.

So the discrepancy I noted above seems to come from the difference between using CES and CPS data. Using CPS data, the order of job growth is adjusted slightly, with Reagan and Nixon (barely) nudging Clinton (1.96%, 1.89%, 1.87%) Carter then Kennedy/Johnson (I started with Kennedy) were still on top in annualized employment growth (you know where Bush-I was).

Comparing the two data series for the Reagan and Clinton years shows something strange: while the difference between the data sets was about constant during the Reagan years (about 10 million with a total variation of about 1 million), it trended down and became about 4.3 million smaller in the Clinton years. This accounts for most of the discrepancy I noted above, and seems important if you are comparing different eras. The CES data (also from the BLS site I linked to above) for the 1980's has a (c) for corrected. How it was corrected may affect whether or not it is fair to use these numbers to compare different times. The CPS may have similar issues, although I didn't see any notes to that effect on the BLS site. If you look at the graphs of both, they seem to be identical as far as trends go, but the CPS is just offset higher (except in the 90's when the CES catches up).

Anyway, one interesting thing to look at in the CPS data is the growth of the population over 16 (i.e. the total potential labor force) during the presidential terms. The number of jobs created during the Reagan years was greater (barely) than the increase in the 16+ population, that is the only time it happened. In the Clinton years, there were 1 million fewer jobs created than the 16+ population growth. Also, Nixon/Ford got trounced with the largest (8 year term) 16+ population growth with 21 million! That may have affected their unemployment numbers a bit.

Dwight, you're still completely not understanding: you are calculating the probability incorrectly for even the simple thing you are trying to calculate.

For one thing, the kind of calculation you are using is used only for discrete, independent events such as a coin toss, where the outcome is not only purely dependent on chance, but also independent of all the other variables. I think even you will agree that the election of a president of a particular party is not independent of all the other variables.

For another, you are measuring presidents instead of presidential terms, which is both simply wrong for that sort of calculation (you need events of equal duration) and has the effect of overweighting the boom presidents and severely overweighting FDR, who did very little with the economy for eight years and then was rescued by an exogenous war which he certainly didn't get us into in order to improve the economy.

For a third, a simple calculation is inadequate to measure a conditional probability, such as the probability that, given that the economy was good, the president would be a democrat. That is what you need to measure in order to demonstrate that the effect was unlikely to be random. It would still be meaningless, as such probability would still have to be a single, binary value with a probability that remains static and a value that does not change, from event to event, but at least you would be closer.

For a fourth, the probability that a Democrat would be elected changed dramatically over your time series, and was highest during the long post-WWII boom that was driven, almost all economists agree, by a productivity boom for which the president was not responsible. Yet your sample treats the probability as unitary, which it was not. It is only appropriate to do a simple probability calculation when the probability stays the same from period-to-period or event to event.

Those are just off the top of my head. You don't know what you're doing, yet you've managed to create a new factoid that I've now seen on several web sites. And you are clinging to your factoid despite your cheerful admission that you have no idea what you're doing.

Sorry Bones, the following is an ignorant, if not stupid, thing to say:

"It pretty much kills the argument that Republican presidents are better for the economy in any case. If the party of the president is a causal factor, then it's best to choose Democrat. And if the party of the president is NOT a causal factor, best not to use the economy as a deciding factor. Either way, it's stupid to vote for a republican because you believe it will be better for the economy."

It's ignorant, becasue it presumes the 2000-1 odds mentioned are relevant to the question at hand. One can believe that which party holds the Presidency is important, and still believe that the Republicans are better, because other variables have produced the results. To say that other variables have produced the results in the past is NOT synonymous with saying that Presidential selection is unimportant in the future, because those other variables may be more benign in future outcomes. All things are not equal, nor do they remain constant. It ain't a coin flip, so to calculate odds as if it were tells us exactly nothing.

I saw this about 55 comments back, and noticed that nobody corrected it...

First of all, the S&L crisis had everything to do with Republican policy because it was fueled by a deregulatory fever that is the raison d'etre of the GOP.

The point man for the deregulation of the S&L industry was Fernand St. Germain (D-RI), who was pretty obviously bought and paid for by the S&L industry.

Who led the charge for deregulation of the airline industry? Ted Kennedy (D-MA).

And who was the architecht of the California power "deregulation" project? Steve Peace (D-El Cajon).

The GOP may have a predispostion towards deregulation, but all three plans were sheparded through by Demcrats, in legislative bodies that had strong Democratic majorities at the time of passage.

Don't try to pin any of the resulting fiascos on the GOP, because that dog doesn't hunt.

"It would still be meaningless, as such probability would still have to be a single, binary value with a probability that remains static and a value that does not change, from event to event, but at least you would be closer."

Wow. Talk about meaningless.

Maybe Republicans agree with Thomsen that job destruction is often a good thing, whereas Democrats don't. When Reagan bit the bullet in 1980 or so, wasn't that part of the rationale? Tom Friedman has praised Reagan for his anti-labor policies, since it made the U.S. more efficient.

Democratic electoral politics depends heavily on organized labor.

So now we have a rough statistic, a motive for the Dems, and a motive for the Repubs.

A long time ago a socialist author (Domhoff??) concluded that the Repubs were the tight-money anti-labor party and the Dems the easy-money full-employment party. Or something like that. Both establishment capitalist parties, but with different capitalist strategies.

Straw-in-the-wind statistics as univocal as Dwight's don't prove anything, but they do give you an idea of what to look for.

Jane:

You are certainly right about one thing. What I was calulating is quite simple and you are trying to make it very difficult.

I was trying to determine if random chance could explain the rankings. I concluded that it could not. I do not think random chance can explain the rankings and you do not think so.

The reason I used a model (you call it a coin flip, I prefer to call it a lottery) based on pure random chance is because the hypothesis I was testing was whether or not pure random chance could explain the rankings. Why would I use a model based on something other than randon chance to test a hypothesis about random chance?

Steve: I made no assertions as to causation. In the statement you quote ("The performance under Democratic presidents was superior to the performance under Republicans in each of those measures."), the performance referred to is that of the economy and not that of the Presidents.

As such, it is statement of empirical fact. That fact may have resulted from productivity booms, business cycles, external events (such as oil shocks, non-random but hard to predict events (such as natural disasters), monetary policy by the Fed, fiscal policy of the various administrations, political trends and any number of other factors. It seems quite likely to me that all of the above (and more) had a role.

To say that those factors likely had a role is the easy part. Defining exactly what role each played is beyond my skill. That is exactly why I expressed no opinion as to causation.

Dwight, you're trying to figure out if random chance can explain a two-dependent-variable system using a one-independent-variable mathematical operation. Your result is meaningless. Please, consult someone other than myself who understands statistics, such as Mark Kleiman, Kieran Healy, or Iain Murray and have them explain it to you if you refuse to respect my explanation. Your result is simply incorrect. When a statistician calculates that "the probability of x occurring by chance is less than y", he does not use math that looks anything whatsoever like yours. Yet you are passing off your number as the same as his.

As I say, I have no idea how you would even begin to calculate the possibility of this occurring by random chance. But I do know that your math is wildly wrong, both because it's wrong, and because no reputable statistician would make any such statement when using a sample as small as eighteen -- the confidence interval would be on the order of 50% or less, which, even if your math were right, would lead to a statement like "there is a less than fifty percent chance that there is a one in two-thousand chance that this occurred at random."

Jane,

What are the two dependent variables you're talking about? There's no "explanation" here, no regressions, just a simple piece of combinatorial probability.

You have an urn with six black balls and seven white ones. You draw out six balls What is the chance that all six are black? 1/1716, as any statistician will tell you. Indeed, as anyone who passed Probability 101 will tell you.

Does this give you any reason to suspect there's something funny going on, that maybe for some reason the draws really weren't random? Yes.

And if the draws are not just balls taken out of an urn, but measurements of a more complex process, could there be factors that argue that the draw was random? Also yes.

But make your case on that basis, not by throwing around a lot of statistical gibberish.

Jane,

I have to echo Bernard here. What two dependent variables are you talking about? Dwight has not made any causation analysis, which I think is what you are focusing on. In fact his example seems taken straight out of the first pages of a Probability book, where you study what are the chances of a certain ordering happening. It can be blue balls or presidencies. Same thing.

I also don’t understand why you talk about confidence intervals and samples. There are no samples here. We have the whole universe. All talk about sample size is wrong and misleading. If I throw a dice six times and get a 6 every time and then try to calculate what the likelihood of that happening is it would be ridiculous to talk about sample size.

I agree that just based on Dwight’s calculations you can’t really reach any conclusions about policy or outcomes. But then again he’s made that clear himself. All this is telling you is that it is not likely to have happened purely by chance. Why it happened requires further research.

"You have an urn with six black balls and seven white ones. You draw out six balls What is the chance that all six are black? ...
Does this give you any reason to suspect there's something funny going on, that maybe for some reason the draws really weren't random? Yes." -Bernard

Actually, no.

"If I throw a dice six times and get a 6 every time and then try to calculate what the likelihood of that happening is it would be ridiculous to talk about sample size."-GT

If you roll a presumed-fair die 6 times (6 samples) and get a 6 each time the sample size is too small to gain insight into whether or not the die is, in fact, fair. What you (both) seem to be arguing is like saying that the chance of getting a Yahtzee on your first roll is so small that you can't get one randomly (it is even less likely than pulling out 6 black balls from the urn, 1 in 7776 if I calculated correctly). Low likelihood of an outcome does not imply nonrandomness of the space generating the outcome. If you roll a die a billion times and you get a 6 each time, now you may want to look at the presumed-fairness (ie the randomness) of the die (or course, technically speaking, this is a possible outcome of a *random* experiment, and so is one 6 followed by 999,999,999 5s, or three 6s followed by 999,999,994 5s and three 4s, etc, and all of these have the *same* probability of happening given a fair die).

Michael,

Once again. Nobody is making any assertions as to what will happen in the future. Nobody is saying what the causality, if any, is. We are just noting a fact.

In the example of the die throw if I buy a brand new die , throw it 10 times , and get a six every time I would have reason to wonder if there is something going on. That's all.

Same here. Given the specific ordering, where every Dem president saw more job growth than every GOP one, it is perfectly valid to calculate how likely that ordering is, if the events were independent and random.

I just learned that I may have gotten out of being drafted to fight the Viet Nam war because they screwed up the draft lottery, contrary to the specific Congressional mandate of a "random" lottery.

I.e., my draft number was 339, and my birthday is in November. But after the lottery was held it was noted that while there were 366 numbers in it, so the average for each month should have been around 183.5 with each month having an equal chance of being over or under it, in fact the first six months had averages under it and the last six had averages over it.

Critics said this could be explained by the bin holding the numbers not being adequately mixed, and a 1970 NY Times story on this reported...

"Two graduate students at the University of Wisconsin have estimated that the odds against obtaining the results of the drawings by a truly random process are 50,000 to 1. Other statisticians arrived at similar results."

http://www.dartmouth.edu/~chance/chance_news/recent_news/chance_news_6.10.html#draftlottery

Although I don't know what they were computing there, as it looks to me like the odds of what was described were 1 in 4096 (the odds of any given pattern of 12 independent binary outcomes).

In any event, in spite of Sen. Ted Kennedy's protests Nixon's people said "We're not doing this again", so maybe somebody born in February served for me. If so, thank you very much.

Michael,

Sample size is implicit in the probabulity calculation. It's not a separate issue. Why would a billion 6's make you curious, while six wouldn't? Because the probability is a lot less. So you are really talking about your own threshold of suspicion.

As far as the Yahtzee example goes, I think what you are arguing is that any result is as likely as any other, and something has to happen, after all. But even if all outcomes are equally likely, all events - sets of outcomes - are not. Here we have the event: {all six balls are black}. This is much less likely than the event {three of the balls re black and three are white}, for example. Hence the suspicions.

Oh, wait, I read that damn Times draft story backward. It was the guys in the first six months who got the break and they tried to shaft me.

Apparently, though, if called I'd have been rejected 4-F because of my reading incapacity.

The point about the probability stands though, and there are some other interesting examples and links on that web page.

GT- When did I say anything about predicting the future? I was talking about randomness, and the inference thereof from an experiment with a small sample size.

Bernard-I was using the die example to discuss GT's point. The sample size doesn't affect the *relative* probability (more on this below) of getting 1 billion 6s versus 999,999,999 6s and a 5, they are the same. But we aren't just talking about probabilities here, we are talking about inferring randomness from a small sample size by, in effect, comparing probabilities of two events (admittedly slightly more complicated than a die roll). If we are trying to infer randomness of a die toss, I will take the 1 billion experiments and you can have the 6. Whose distribution would better show that it is likely random?

Both-
Let's take your analysis to its full ends. To be specific, I calculated, using the same idea as PLA, the probability of Reps and Dems alternating in the top 6 slots. The probability was 2.04% or 35 times the probability of all Dems getting in the slots (0.058%), by this type of analysis. Alternating is pretty random, right? So let's define the alternating one (R,D,R,D,R,D) as our standard and say that, for example, that anything 30 times below that is too suspicious to be nonrandom (this is what you seem to be saying, in effect, am I misunderstanding you?).

But now, lets say that we played *one* hand of 5 card stud poker (since we are talking about percentages already, any random game will do), and I got a full house, and you got two-pair. Let's say you take two-pair as your randomness detection standard (it's what you are looking at when I grin and take all your money). Agreeing with the standard, your analysis here would force you to state that I must have been cheating (i.e. the full house is not due to randomness), because the full house is suspiciously 32.2 times less likely than two-pair. (0.15 % versus 4.83% (with jokers), yeah, I looked them up. If I got a straight (not very random-looking) and you only got a pair (pretty random-looking), the difference is 60 times!)

So by this sort of analysis, we could say: in the great game of Job-growth Politics, the parties played a single hand of 5 card stud; the Democrats (what skill!) got a full house, and the Republicans (unlucky randomness!) got a two-pair (using CES numbers, anyway). Dems won. It happens. (Ok, so the analogy isn't perfect; the Reps "would have got" a two-pair, had things been random or something, this isn't *my* analysis ok?)

Anyway, I think saying that it *couldn't* have happened by chance using your arguments is just a bit nonsensical.

You have a disease for which there are two treatments being tested.

Treatment D is tested by six physicians on their patients and treatment R is tested by seven different physicians on their patients. The six groups receiving D all do better than the the seven receiving R.

Is this information of any interest to you?

Bernard,

Now you are trying to mess with sample size. If each physician were treating a large group (you said patient*s*) perhaps (and confounding variables were taken into account), but your Dems versus Reps statements are equivalent to each treating one (in a random probabilistic way) and there being some metric of betterness (as I showed above, some metrics even flip the order a bit, but nobody cares about that apparently), and the sample size is then too small (especially for medical purposes, sheesh!). Actually, this might be a good analogy for these things, if we were talking about causation (health of patient before term, number of diseases to deal with, what was prescribed, etc.) as well as analyzing things from a statistical point of view.

But we are for whatever reason focused on the what, what the numbers mean in terms of *simple probability* (why we use this I do not know). By *your* analysis, I concluded that they don't mean much. Again, we are already talking about percentages, so we don't need any more analogies, any game will do. Changing the analogy to something more emotionally important like medicine doesn't change what your analysis says in terms of simple probability (*your* chosen method of analysis). Analogies are useful for thinking about the relative probabilites and poker is a easy game to think about. So why keep switching things around? Let's summarize the debate using an SAT-style analogy quiz:

Dems in all top 6 slots: alternating Reps and Dems

a) Full house: two-pair

b) Not due to chance: random

c) Republicans are terrible at managing the economy: or at least they can't claim to be better

d) Did I say anything about causation?: I'm just doing simple probability here.

e) b, c, and d

My answer: a (using your analysis)
Your answer: b
Combined contra-Galt group's answers: e

a is the only correct answer, given your analysis. b is nonsensical, unless you are a really unpleasant poker player. c may be true, but not given your analysis. d is a joke to give 5 selections.

So, are you going to change your answer from b to a? If not, why not? Your explanation should not require any use of sample sizes or other analogies. The two analogies are set, the sample sizes fixed, and the percentages have been calculated/looked up.

"You have a disease for which there are two treatments being tested.

"Treatment D is tested by six physicians on their patients and treatment R is tested by seven different physicians on their patients. The six groups receiving D all do better than the the seven receiving R.

"Is this information of any interest to you?"
~~~

Well, it depends.

1) Are all other factors except the treatment being held the same? Or is each patient acting differently and being subjected to a multitude of different outside forces than all the other patients? (As is the case with the Presidential comparisons.)

If the latter, than you need a *lot* more than 13 patients to get a meaningful result, in order to reduce the effect of the all the outside forces and different influences to random noise.

First rule of statistical analysis: Correlation is not causation.

2) Do we all have an agreed upon definition of "do better" before we run our trial? Or after we run it, do the six doctors who administered D go data mining through the results to select a way to say that all their patients did "better" than those who received treatement R -- while ignoring equally credible ways to conclude that some who received R were among the six who did best?

(E.g., if we look at GDP growth per four-year persidential term, a much more uniform measure than that used in the cited stats, Nixon and Reagan were in the top six performers of the 13 since WWII.)

3) Do we actually have a scientifically credible reason for believing the results we are testing for could result from the treatments being tested? Or is it actually much more plausible that the results we see are caused by some other unspecified outside factor? E.g., as if we are testing voodoo against bleeding with leeches, *without* holding other factors equal, as mentioned above in 1).

ISTM the ranking of the Presidents as given flunks all three tests, and if I ever do get a disease like you say I sure hope the medicine I receive will have been subject to much better analysis than that.

Add a fourth condition, Jim: Treatment D needs to be the same treatment each time, not a bottle of unknown pills pulled at random from the shelf with a "D" label stuck on it. Ditto for R. If the boys in the R&D (haha) lab really think that LBJ's expansion of welfare and Clinton's limiting of it were two doses of the same medicine, they should be looking for different jobs.

Wow, this topic sure does produce a lot of comments. I haven't heard what role Congress plays in this. I don't know about the founding fathers, but I think that politics has turned into a big game, a career. Those involved want to keep their jobs and/or power as long as they can.

The variables within the series are dependant on each other. The prior outcome in the series has an impact on the outcome in the next event; the fact that Carter was a Democrat had an effect on the fact that the next president elected was a Republican. So unless the effect of a prior outcome on the sequential outcome is exactly the same for the two series, position at the start of the series would considerably alter the conjunction of outcomes, even if the series were entirely independent from each other, which is what Dwight seems to mean by "at random". Interseries occurrence could be entirely independent yet still generate a high probability of one or the other parties predominating during the peaks.

As for what I mean by a sample, well, I presume that the ultimate point is to generalize from this to the larger pool of presidential candidates: you are more likely to get a good economy under candidates for x party than y party. We analyze the probabilities based on a smaller subset of presidential candidates; those who have served as president between 1932 and 2002. The question is, can one generate meaningful data from thirteen data points (actually I'd prefer eighteen -- the number of presidential terms -- but Dwight didn't do that). And any statistician would, I think, say that any data generated from a set of 18 would generate a confidence interval as to the probability that data from the set represented intrinsic properties of the population, rather than random variance, of less than 50%. That's a guess, of course, as I'm no statistician. But I highly doubt you'd catch them making claims that the probability of a conjunction occurring at random was 1 in 2000 based on only 13 events, as the probability of random variance is simply too high.

And Bernard, the FDA most certainly doesn't approve treatments based on a double-blind with six members. I believe there are legions of cases where just such random variance has made treatments seem promising that later turned out to be useless -- and you didn't find any reputable doctor claiming 99.995% confidence, either.

"The variables within the series are dependant on each other. The prior outcome in the series has an impact on the outcome in the next event; the fact that Carter was a Democrat had an effect on the fact that the next president elected was a Republican. So unless the effect of a prior outcome on the sequential outcome is exactly the same for the two series, position at the start of the series would considerably alter the conjunction of outcomes..."
~~~

Yes indeed. The collapse of 1928-32 guaranteed both that the numbers for the next many years would look much better by comparison *and* that that the next president would be from the other party.

So to the extent that the election of 1928 was probablistic, everything for at least the next 20 years could have flip-flopped, if only the other party had won in 1928.

That is, unless one thinks that Hoover personally caused the Great Depression (instead of being the guy who was unlucky enough to be in office when it hit).

And if you think that, remember that the Democrats tried to get him to run for them.

Jane,

Probabilities do not have confidence intervals. Estimates have confidence intervals. Probabilities are calculated numbers. They no more have confidence intervals than sums do.

When a poll reports that 60% of those surveyed support Senator Foghorn, with a margin of error of 3%, the 60% is not a probability, it is an estimate. Therefore sample size, implicit in the confidence interval, is important.

But when I say that, if employment growth is independent of the party in the White House, the probability that all six Democrats will outrank all seven Republicans in employment growth (annualized, by the way, according to the original post) is 1 in 1716 that number has no confidence interval attached.

Now, you want to argue that there is some sort of serial correlation in both series that happens to cause these things to match up. Could be. But you've presented no evidence whatsoever. You haven't even described the process you have in mind. If it's cyclical, as you suggest, why do the cycles match presidential terms?

Jim even goes so far as to claim that had Hoover run as a Democrat in 1928 the entire series would have been inverted. Now that's stretching.

"If Hoover had won as a Democrat in 1928 then after the Depression hit we'd have had Republicans in power for the next 30 years."

Ok, let's say Hoover ran on the Democratic ticket. He still would have advocated doing nothing to deal with the depression, and would have been driven out of office as out-of-touch and so on, replaced by the GOP candidate.

The GOP candidate would then have advocated doing nothing to deal with the depression, which would have been wildly popular and establishing a political realignment because......erm, runs out of steam there.

Realignment didn't happen because people were pissed at Hoover; it happened because people thought FDR was trying to help.

Jim,

I did not ask,

"Does this prove D is better than R?"

I asked whether it was of interest and, to be fair, I implicitly ask whether it causes you to suspect D might possibly be better than R.

In response to your three tests:

1. Of course you should look for possible sources of bias in the data. And they are more likely to be present in a small sample. But just saying "small sample size" is not enough. A probability of x is a probability of x, whether the sample is 13 or 13,000. So you have to find something pretty important here. Sample size is an issue when you cite a proportion, not a probability. If I say, “This coin is weighted. It came up heads 60% of the time,” it would matter enormously whether I had flipped it 10 times or 100 times. If I said “the probability that a fair coin would come up heads as often as this one did is only 1%,” then it wouldn’t matter.

2. True and very often, usually even, overlooked or misunderstood. But as I recall, Dwight was looking specifically at employment growth and he used employment growth figures. And referring back to the original post, they were annualized. Doesn’t seem unreasonable to use employment growth to measure employment growth. Of course if someone said this by itself demonstrates the overall superiority of Democratic policies your point would be valid.

3. I would turn this around and ask, “ do we have a scientifically credible reason for believing that there are factors that distort our results?” The calculation assumes independence. If you believe it is not valid, tell me why that’s wrong. Simply saying “correlation does not imply causation” does nothing. If we accept your broad interpretation of this then NO statistical test ever shows anything.

Paul,

You are correct with respect to the medical example, but I don’t think the argument carries over to the presidential case except in a sort of “meta” way. I might argue, for example, that Democrats are more flexible and pragmatic than Republicans, hence more likely to adopt sensible policies for dealing with existing problems and less ideologically hidebound than Republicans. I’m not saying this is necessarily so, (though W does sem to regard tax cuts as a universal nostrum) just that it is one possible line of reasoning that allows for differing policies.

OK. See everybody next time.

"He still would have advocated doing nothing to deal with the depression, and would have been driven out of office as out-of-touch and so on, replaced by the GOP candidate."

He *didn't* advocate "doing nothing" to deal with the Depression, geeze.

And FDR ran on the 1932 Democratic party platform of doing two things -- balancing the budget by cutting waste out of the government. Talk about "doing nothing".

But that's what people voted for because the guy who is *in* when something very bad happens gets the *blame*, deserved or not, which makes the guy who is out the popular alternative. So the out guy carefully avoids saying anything at all controversial or specific that could be challenged, speaks a lot of optimistic platitudes, and wins in a walk. How many times have we seen this from the local mayoral level on up? That was the 1932 election.

Returning to the GOP alternative wins in '32 scenario:

"The GOP candidate would then have advocated doing nothing to deal with the depression, which would have been wildly popular and establishing a political realignment because .... erm, runs out of steam there."

No, because economic historians today say the New Deal *slowed* the recovery from the Depression -- see my comments and links above -- so the Republicans could have done even *better* getting more credit for a faster recovery than the Democrats did.

Which points out yet *another* problem with the D versus R stuff -- there's no third control group, and no comparison to what the other party would have accomplished in the *same* time period under the same conditions.

So after 1932 there must be faster growth than before, but the Democrats *screw up* and hold it down compared to what it should be, yet people say: "Whoopie! Look how much better the Democrats did compared to before! One point for them!"

I note again, anybody who believes this logic should see how the annual growth in mainland China since Deng has surpassed US annual growth ever -- what are the odds!!??? -- and conclude we should elect a Communist as president to *really* spur the economy.

Bernard, you're making hte same mistake that Dwight is. The probability of b given a is not the same as the probability of a given b. The probability that the economic cycle is not causally related to the presidential cycle, given that Democrats scored in the top of the GDP range, is not the same as the probability that the Democrats scored in teh top of the GDP range, given that the economic cycle is not causally related to the presidential cycle. I agree that Dwight has correctly calculated the result of 6 binary events occurring in sequence, given that they are random; I do not agree that his calculation correctly describes the outcome of presidential elections, or that he has correctly calculated the probability that they are random, given that they in fact occurred as they did. Am I clear now?

"Am I clear now?"

No.

"The probability of b given a is not the same as the probability of a given b. The probability that the economic cycle is not causally related to the presidential cycle, given that Democrats scored in the top of the GDP range, is not the same as the probability that the Democrats scored in teh top of the GDP range, given that the economic cycle is not causally related to the presidential cycle. "

All true, except the numbers were employment growth, but that's irrelevant.

"I do not agree that his calculation correctly describes the outcome of presidential elections, or that he has correctly calculated the probability that they are random, given that they in fact occurred as they did. "

He wasn't trying to calculate the probability that elections are random. What he calculated was this:

If employment growth is independent of the party in the White House, what is the probability that the six Presidents out of thirteen whose terms experienced the highest growth in employment would be the six Democrats.

This is identical to various other examples I've cited, like drawing six balls out of an urn containing six black ones and seven white ones and getting the six black ones. It's 1/1716.

It is what you refer to as,

"the probability that the Democrats scored in teh top of the GDP range, given that the economic cycle is not causally related to the presidential cycle. "

Though you should say "is independent of" rather than "not causally related to"

What does that mean? It means that if the assumption of independence is correct we have witnessed a very low-probability event. Hence we may reasonably have doubts about that assumption.

Of course there are all kinds of issues that can be raised, and neither I nor, as I understand it, Dwight, are claiming that this is proof that Democrats are better at job creation than Republicans. A more complex analysis might well produce dramatically different results. But this number is what it is. It is not "meaningless."

Bernard, we're going around in circles.

Dwight "The chance that this occurred at random is about 1 in 2000"

Parsed: The chance that the Democrats occupied the top six slots due to random chance, rather than some relationship between the party of the president and the outcome of the economy, is 1 in 2000.

That probability is the probability of a, given b: the probability that, given that the top six slots were occupied by Democrats (which is the information we have), economic performance is unrelated to the party of the president.

But what he actually calculated is b given a: the probabability that, given that economic performance is unrelated to the party of the president, the top six slots were occupied by Democrats.

Those two probabilities are not the same thing.

For those who don't understand why the two probabilities should be different, consider this simple example: the probability that I have breast cancer, given that I am a woman, is not the same as the probability that I am a woman, given that I have breast cancer. The probability that I have breast cancer, given that I am a woman, is much less than 1%. But the probability that I am a woman, given that I have breast cancer, approaches 100%.

I misspoke when I said confidence interval; I meant simply "confidence". Brain freeze from typing too fast, no doubt. But the point stands; that's just quibbling.

How much weight did Dwight put on his stats? I think that he was just saying that six out of six was a pretty striking number and that maybe we should look into this. I do not believe that he was claiming to have proved that Republican presidents are bad for the economy. Just that -- gee, there's something sort of funny here.

The criteria you use to decide whether to investigate a question are different than the criteria you use to decide the question. I think that this question should be looked into.

And sometimes a very very unsophisticated statistical analysis can be very significant if the data are striking enough. You use sophisticated stats to tease small amounts of significance out of dirty, incomplete data.

The fact that Republicans and Democrats tend to look differently at unemployment seems highly relevant.

Yomtov, I'm bailing you out this time, but nothing has changed as far as my friends are concerned. Got that, chum?

No, Zizka, when you say that you have calculated that there is a very small chance that something occurred at random, you are stating that you have calculated that there is a very large chance that there is a causal link between the two series; the two are reciprocal. You can't then weasel and say that you didn't mean any such thing while still allowing other people to use your numbers to say that very thing you claim you didn't mean. There are people out there now claiming that Dwight's number "proves", at the very least, that Republicans aren't any better for the economy, and more likely that the Democrats are downright superior economic managers, which is indeed what Dwight implied. Incorrectly.

The reason that no one is investigating this fascinating link, other than inummerate partisans, is that the data is far too weak to provide any sort of cogent analysis, for reasons I've outlined, and I'm sure many others besides. This is why people seriously interested in finding out the effect of presidents on the economy analyze policies, not parties.

“Bernard, we're going around in circles.”

True.

“Dwight: The chance that this occurred at random is about 1 in 2000

Parsed: The chance that the Democrats occupied the top six slots due to random chance, rather than some relationship between the party of the president and the outcome of the economy, is 1 in 2000.”

True.

“That probability is the probability of a, given b: the probability that, given that the top six slots were occupied by Democrats (which is the information we have), economic performance is unrelated to the party of the president.”

False. It is the probability that, if employment growth is independent of the party of the President, employment growth under every one of the six Democrats will be higher than under every one of the seven Republicans

“But what he actually calculated is b given a: the probabability that, given that economic performance is unrelated to the party of the president, the top six slots were occupied by Democrats.”

True. And that is what he should have calculated.

This is not about conditional probabilities. The concept “employment growth is independent of the President’s party” is not an event. It has no probability. It is an assumption underlying the calculation. You can talk in terms of P(a) given (b), but (b) is an assumption, or a condition, not an event with a probability. It is not the same as the events “Jane is a woman,” or “Jane has breast cancer” in your example.

Take the example. Suppose we wanted to find out whether breast cancer affects women more than men. We look up the respective rates of incidence and find out that, say , only .001% of the victims, one in 100,000, are men. (That number is a total guess). If breast cancer incidence were truly independent of sex, then the chance of this finding is vanishingly small. So we suspect that the assumption is wrong. Here the assumption “it’s independent of sex” is again not an event. No probability.

In a (much) more rigorous setup, the notion of independence would be the null hypothesis. The alternative would be that performance is better under Democrats. We would set a significance level. Then we calculate the probability of the observed result under the null, i.e. assuming the null is true. When we got a very low probability – less than significance - we would reject the null in favor of the alternative.

Another approach is to think of yourself on a jury. The “null” is “not guilty.” But then you decide that, if the defendant is not guilty the probability of the observed events – the evidence – is 1/2000. How would you vote?

Again, the statistics presented are very very far from a rigorous analysis for lots of reasons. They are suggestive. That’s all. Zizka put his tentacles around the subject quite well. But you can’t blame Dwight, or me, or Zizka, if some people choose to overinterpret.

There are any number of ways to challenge this suggestion. (Why start with Hoover, for example – that smacks of data mining; lags? exogeneous events biasing the data?) But you need to be clear about what you’re challenging.

Bernard made most of the points I was going to make.

I'll just add that it is simply not true that no one is investigating this link. In fact there are many studies looking at the relationship between who controls the WH and different economic indicators. A quick Google search found this one for example. In this case it's the relationship with the stock market.

Dwight's point was a simple one. It had nothing to do with conditional probabilities, samples, or confidence intervals. It had nothing to do with dependent variables or causation. It was simply a test of the likelihood of a certain event happening (If presidential affiliation and job growth are unrelated what are the chances that the historical ordering we have seen could happen?).

If you want to understand the why, it will require much more analysis.

You're right, FDR ran declaring he wanted to balance the budget. He also, however, attacked Hoover at length for not doing a thing.

But whatever, you're right - the 1932-1968 political consensus relied on an accidental quirk of the business cycle. Who knew? Someone tell Reagan about 1980.....

Steve wrote:

"Watch your howlers when accusing professionals of making howlers."

I fail to see the howler, Steve. Real GDP was higher in 1936 than it was in 1929. The article you referenced was misleading when it said that output was lower in 1939 than it was in 1929. In real terms, it was >20% higher.

If you reread the article, you'll find that the authors have a statistical analysis that bundles the unexplained portions of the slower growth and asserts that this portion is due to New Deal policies. They don't make a detailed causal argument; and there are other explanations they neglect, see below for a link to one made by their colleagues at the Minneapolis Fed.

"That is *no* real growth, leaving the economy 20% to 25% under trend"

And the economy fell by 20-25% in 1929-1932, you'll note.

" -- with unemployment to match, we will remember.'

I can't see how FDR & the New Deal can be held responsible for the massive slowdown & deflation under Hoover (which, because of the decline in investment, would also have affected productivity growth). Sorry.

"By your claim, the economy should have been 25% larger than in 1929"

No, that is not what my claim was. I claimed that recovery was at an average of 7.4% from 1933 to 1941.

"Did I say that, or did you just want me to?"

Well, you said the New Deal's policies didn't retard recovery.'

No, I did not say that - I did not say that there were *no* New Deal policies that did not retard recovery. I pointed out that FDR's record of growth was very strong, compared with historical GDP growth rates in other administrations.

An alternate explanation for productivity declines during the Great Depression is found in this Minneapolis Fed paper (who seem to be having an office row over the 1930s): http://minneapolisfed.org/research/qr/qr2622.pdf, where they suggest that increasing returns to scale and disruption of industrial infrastructure led to drops in productivity.

Now, would you rather be looking for work with Roosevelt at the helm, or Bush?

_______________________________________________

Well, I would assume that in the 30's Bush would be acting like Hurbert Hoover. Would you rather be looking for work under Hoover or Roosevelt?

Comments are Closed.