The aftermath

by Russ Roberts on August 1, 2011

in Stimulus

One of the complaints this morning after the debt deal is that we won’t be able to “stimulate” the economy any more. Yet, there is no evidence that the stimulus worked. When I tweeted something similar this morning (follow me as EconTalk on Twitter here, Cafe Hayek tweets are here), someone accused me of confirmation bias, noting that the CBO had found that the stimulus created 3 million jobs. Actually, the CBO has never estimated the job effects of the stimulus other than to assume a particular relationship between government spending and job creation. That is simply assuming the results that you are trying to discover. More here if you missed it the first time.

By the way, I was careful to say that there is no evidence that the stimulus worked. We don’t have very good evidence that it didn’t work. But we have no evidence that it did.

Be Sociable, Share!



33 comments    Share Share    Print    Email


Slappy McFee August 1, 2011 at 12:56 pm

A question I have always had. Does the CBO release the methodology they use to ‘score’?

Daniel Kuehn August 1, 2011 at 12:58 pm

Don’t we extrapolate what we know about engineering and past experience with building bridges when we walk across a brand new bridge constructed on those principles?

There is obviously considerably more uncertainty around macro estimates than there is around bridge engineering, but the principle of what the CBO is doing is exactly the same. I don’t think it’s fair to say “there is no evidence that the stimulus worked”. Or put it this way – if one can say this it’s with the same degree of relevance that someone might say “nobody has walked across this newly constructed bridge before – there is no evidence that this bridge will hold me up”.

Don Boudreaux August 1, 2011 at 1:01 pm

What evidence are you extrapolating in order to support the burden of persuading us that the ‘stimulus’ worked?

Justin P August 1, 2011 at 3:04 pm

I hear crickets.

Daniel Kuehn August 1, 2011 at 3:09 pm

All the empirical work that I’m aware of produces fairly large multipliers in downturns and small multipliers in periods of growth. Work that averages out many years produce lower multipliers, suggesting the same thing. These empirical findings are supported by theoretical findings in a variety of frameworks (which is good because you want to make sure something tenuous like this isn’t too sensitive to your assumptions).

Economic Freedom August 2, 2011 at 3:22 am

All the empirical work that Iā€™m aware of produces fairly large multipliers in downturns and small multipliers in periods of growth.


“All the empirical work done by those who already believe in such a thing as a ‘multiplier’ tends to confirm the further effects of what they already believe in.”

The fallacy is called reification, with an added dash of question-begging. Better known as “Keynesianism.”

Here’s Uncle Milton on the multiplier:

In cooperation with some of my students, I have done some fairly extensive empirical work, for the U.S. and other countries, to get some more satisfactory evidence [of the Keynesian multiplier effect].* The results are striking. They strongly suggest that the actual outcome will be closer to the quantity theory extreme than to the Keynesian [i.e., $100 spent by government either reduces private spending by $100 somewhere else or the additional $100 raises prices]. The judgment that seems justified on the basis of this evidence is that the assumed $100 increase in government expenditures can on the average be expected to add just about $100 to income, sometimes less, sometimes more. This means that a rise in government expenditures relative to income is not expansionary in any relevant sense. It may add to money income but all of this addition is absorbed by government expenditures. Private expenditures are unchanged. Since prices are likely to rise in the process, or fall less than they otherwise would, the effect is to leave private expenditures smaller in real terms. Converse propositions hold for a decline in government expenditures . . . Whether the views so widely accepted about the effects of fiscal policy be right or wrong, they are contradicted by at least one extensive body of evidence. I know of no other coherent or organized body of evidence justifying them. They are part of economic mythology, not the demonstrated conclusions of economic analysis or quantitative studies. Yet they have wielded immense influence in securing widespread public backing for far-reaching governmental interference in economic life.

*Some of the results are contained in Milton Friedman and David Meiselman, The Relative Stability of the Investment Multiplier and Monetary Velocity in the United States, 1896-1958 (forthcoming publication of Commission on Money and Credit).

[Excerpt from "Capitalism and Freedom" 40th Anniversary Edition, page 84.]

Daniel Kuehn August 2, 2011 at 5:55 am

So what Friedman calls the “Keynesian extreme” presumption at the time was something like a multiplier of 4. After permanent income hypothesis and a few other things people realized it was way to high, and what do you have Keynesians saying now?

Exactly what Friedman said: during periods of growth there is crowding out. These small multipliers get bigger during downturns.

Now – do you actually have concerns with the empirical literature or are you satisfied with “Keynesians did those studies” as a counterargument?

Economic Freedom August 2, 2011 at 8:07 am

Now ā€“ do you actually have concerns with the empirical literature

Like most of the others here, my only concern is that you’re a zombie. If you actually had “empirical studies” that were worth a darn, you’d be posting them. You don’t have anything. (Nice save, though.)

Economic Freedom August 2, 2011 at 8:28 am

Now ā€“ do you actually have concerns with the empirical literature

(Old Marxist joke in former USSR):

“Father, what exactly is ‘dialectical materialism’?”

“I will teach you. Suppose, son, that a chimney-sweep has just cleaned a chimney. Which of his hands has soot on it?”

“I, uh, don’t know for sure. His right?”

“Wrong, my son. His left. His right hand was holding a bucket and wore a glove. Once more: if a chimney sweep has just cleaned a chimney, which hand has soot on it?”

“Um, the left!”

“Wrong, my son. The right. You see, the left hand must have been holding the bucket and wore a glove. Once again: if a chimney-sweep has just cleaned a chimney, which hand has soot on it?”

“I’ve got it! The right hand!”

“I’m sorry, my son, that’s incorrect. Neither hand had soot on it. Obviously, he was wearing gloves on both hands. Once again, and this time pay close attention: if a chimney sweep has just cleaned a chimney, which hand had soot on it?”

[long pause]

“Father, could it be . . . neither hand?”

“You’re a dolt, son. Can’t you see? Obviously, both hands must have had soot on them — what were you thinking . . . that he could afford gloves?”

“But father! This is unfair. You’re simply changing things to make the answer come out whichever way you want!”

“I’m proud of you, son! Now you understand what ‘dialectical materialism’ is.”

Substitute “empirical studies of the multiplier done by Keynesians” for “dialectical materialism” and you’ll get the moral of this story.

Campbell F August 1, 2011 at 1:54 pm

As an engineer, I can tell you that bridges behave based on principles that are modeled (successfully) using Gaussian distributions for various components, architectures, environments, etc. and then given a safety factor to compensate for the tails of the Gaussian distributions. These things can be tested in a lab environment changing one variable at a time and measuring the range of results.

There is a massive difference between creating models based on these experiments and the “experiments” done by economists (or worse by econometricians) which can only hope to possibly measure correlations.

Daniel Kuehn August 1, 2011 at 2:01 pm

How much econometrics experience do you have, Campbell?

Quasi-experimental work obviously has its liabilities and the precision of these estimates are obviously different. But it’s not a causal vs. corrollary difference. This is a common mistake for people who primarily work with experimental empirics and don’t really understand exactly what standard quasi-experimental specifications are held to.

Josh S August 1, 2011 at 3:02 pm

A model cannot be used to validate itself. Econmetrics does not play by special rules where you get to say, “See, f(X) = Y in real life because I have a model that says f(X) = Y, and when I put in X, I got Y out!” Your science has to play by the same rules everyone else’s does.

Daniel Kuehn August 1, 2011 at 3:09 pm

Can you be more specific about who has done something like that?

brotio August 2, 2011 at 3:44 pm


In your hypersensitivity, you accused Justin of calling you names in the thread below. You’ve been posting since Justin advised you to reread the post that got your panties in a bunch. Do you plan to do so?

Justin P August 1, 2011 at 3:04 pm

If your going to use that line against Campbell. Then you really don’t have any leg to stand on when your talking about engineering do you?

So from now on your not allowed to use an engineering metaphor, unless you have engineering experience. Do you?

I’m not trying to be an a$$ but just holding you to your own standard.

Daniel Kuehn August 1, 2011 at 3:12 pm

If he challenges my understanding of engineering I’m happy to defer.

Please don’t call me an ass. I’ve never insulted you like that. Campbell is welcome to correct my understanding of engineering but there doesn’t seem to be a big problem with that. He seems to misunderstand exactly how econometricians draw their conclusions, and I’m certainly going to bring that up.

Daniel Kuehn August 1, 2011 at 3:15 pm

Experimental scientists have a very straightforward identification strategy. Non-experimental scientists don’t. Experimental scientists often don’t understand the intricacies of the identification problem precisely because it’s not a problem they spend a lot of time working on. So they see no experiment and they assume we’re only dealing with correlations – it’s precisely what Campbell claims here. It’s also inaccurate. What could possibly be wrong with me pointing that out?

Justin P August 1, 2011 at 3:23 pm

Dan, reread what I said. I said I’m not trying to be an a$$. I didn’t call you any names. So I expect an apology.

mcwop August 1, 2011 at 3:52 pm

I certainly would not use an economic model to build a bridge. That would not be a very predictable driving experience. If you spend 12 minutes trying to accurately predict the economy, you’ve probably wasted 12.

Campbell F August 1, 2011 at 5:57 pm

I actually have a little bit mostly on microeconomic issues, only amateur stuff, playing around in MATLAB and reading books to further my understanding. I just reject the idea that one can use the same tools from (relatively simple) engineering issues to model complex things such as human activity or financial systems. The furthest I would extend econometric models would be to an individual level, generously a firm level.

Daniel Kuehn August 1, 2011 at 2:02 pm

What standard their held to – or even simply how they are done. If you think econometricians are just running around confusing correlation for causality you probably don’t have a very good sense of what econometricians do.

Bruce August 1, 2011 at 3:49 pm


With all due respect it seems to me that there is a flaw in your analogy. In engineering, one can perform physical tests to prove or disprove one’s hypothesis using controls and variables. With repect to the stimulus, one cannot test the counterfactual. There is no way to establish a control group. Lacking a control, the hypothesis is reduced to a self validating mathematical formula. Put another way, the Mythbusters could tell you with a high degree of certainty whether it’s safe to cross the bridge, they’ll never be able to tell you how many jobs government spending will “create”.

Daniel Kuehn August 2, 2011 at 5:57 am

There are other options besides experimental empirical work and theoretical work. Namely, non-experimental empirical work.

Treibs August 1, 2011 at 1:24 pm

It seems like the bridge is a lot easier to test.

Emil August 1, 2011 at 2:00 pm

Fair point, but that should be something the politicians should think about before pissing the tax payers money up the wall

Ben Hughes August 1, 2011 at 1:26 pm

Even if your commenter’s point on jobs was not fallacious from the start, why is the presence of additional jobs taken as evidence of success for the stimulus? As we all know, you could pay 3 million people to dig ditches and fill them back up. In such a case – by that reasoning – the stimulus worked, yet prosperity declines.

Ben Hughes August 1, 2011 at 1:28 pm

Me guess is that many people presume ex-ante that more employment necessarily means an increase in prosperity. But that’s committing the same “simply assuming the results that you are trying to discover” Russ already pointed out, no?

Jim August 1, 2011 at 3:17 pm

Let’s step back from the regressions for a moment and get real here. No sane organization would spend 7% of its revenue and feel satisfied by statisticians who need to hunt in the data to figure out what the benefit was.

I say if you spend $1 Trillion and double government spending in less than 10 years, it should be obvious to everyone you have accomplished something or you shouldn’t be doing it. The elites take a sledge hammer to the economy and then measure it with a microscope. Are we really all this silly?

Brent Buckner August 1, 2011 at 3:21 pm

Depends upon what you mean by “evidence” and “worked” I suppose, but Feyrer and Sacerdote took a swing at it.

Andrew_M_Garland August 1, 2011 at 10:27 pm

Economic News of the Future:

Keynesian economists I.M.Whacknut and M.T.Mindset have released a breakthrough paper. By simplifying their observation protocol, they have identified the direct operation of the Keynesian Multiplier. “We are surprised that this simple system shows such a clear signal”, reports Mindset.

They set out to observe money in action “in the wild”, in a local administrative office. Whacknut: “We thought that there would be a better paper trail in an office”. They were rewarded almost immediately.

The company president Perry took $9 from corporate cash (stimulus spending) to buy 12 donuts for a morning meeting. He presented these donuts to the staff. It took Whacknut and Mindset only a moment to realize what they had witnessed.

12 donuts of value came into possession of Perry, and then 12 donuts of value appeared in the meeting system, where unfortunately they were consumed. The total of 12 + 12 = 24 donuts is a direct representation of the Keynesian multiplier of 2.0 in this case.

Whacknut: “These transaction are usually complex, fragmented, poorly documented, and devilishly hard to measure. The tiny scale of these transactions nevertheless reveals the essential nature of the Multiplier.”

Whacknut and Mindset are working on a more sophisticated experiment, where Perry gives the donuts to secretary Sally, who then presents the donuts to the staff. Whacknut: “We believe that we can achieve a Multiplier of 3 in this case, and even higher values in the future”.

Herman August 2, 2011 at 1:19 am


muirgeo August 1, 2011 at 11:01 pm

When the data doesn’t support your contentions question the data….

Sean August 2, 2011 at 2:59 pm

The problem with the debt deal is that it is just putting off the inevitable. The U.S. needs to control spending, create tax incetives which spur economic and job growth, cut wasteful social spending, which creates the spiral of social stagnation, crime, and poverty. We need a clensing of both sides of the aisle of Congress and get rid of the, “business as usual,” attitude.

Previous post:

Next post: