The U.S. General Social Survey, a random sample of adults, asked respondents the following question in 2006 and 2010, a couple of years before and a couple of years after the crisis:

The first issue is federal income taxes. Some people say that reducing income taxes almost always helps the economy grow. Others say that reducing income taxes has very little long-term impact on economic growth but could cause a big increase in the deficit. Please look at Card B10. (On a scale of 1 to 5, where 1 means “Very Well” and 5 means “Not at All”), how well do [economists] understand the likely effects of reducing federal income taxes?

As far as I can tell responses have not been analyzed in the literature. This is not a particularly good question to proxy beliefs about the state of economic science: it’s not well-posed. I’m not sure what the distribution of answers would look like if we randomly sampled economics professors, but I’d be really surprised if they were piled up on “5 Very Well.” Yet it doesn’t seem unreasonable to interpret is as a rough index of beliefs over the credibility of economic science. The distribution of answers (n=1,287) is displayed below, stratified by year:

Economists enjoy more popular support than I would’ve guessed. Most people think economists understand the effect of taxes on growth at least moderately well. Overall, just under a third think we understand “very well,” although that proportion declines between 2006 and 2010 from 34% to 24% (p<0.01). About 15% of respondents think we understand the effect "Not at all" or just a bit better (responses four or five on the five point scale) in 2006, rising modestly to 18% in 2008.

Who tends to believe economists know what they're doing? Who became more cynical of economic understanding between 2006 and 2010? Regression time. I choose some basic demographics as explanatory variables: age, sex, ethnicity, education, and I also consider a measure of religious behavior (does the respondent attend church at least once per week) and of political ideology: dummies for "conservative" or "liberal," relative to "moderate," generated by lumping all conservative or moderate responses on a seven point scale together.

The dependent variable is an ordered response taking integer values from one through five. Consider linear probability estimates (OLS) of the probability the respondent chooses category 1, economists understand "Very well," and those of categories 4 or 5, "Not at all." I choose OLS since I want to include interactions between a 2010 dummy and all covariates to check, and test, how things change between 2006 and 2010, and interaction terms in nonlinear models are a pain in the ass. Marginal effects from probit models were similar.

A set of regression estimates can be viewed by clicking here. Older people are somewhat less likely to think economists know what they’re doing. In 2010 but not 2006 there is some evidence that people who attend church are more likely to believe economists know what they’re doing and less likely to believe economists know nothing at all. Men appear to be a little more likely to believe economists know nothing at all. Political ideology, perhaps surprisingly, has no traction in explaining credence in economists.

The big story is education: people with little education are much more likely to report that economists don’t know anything at all, and much less likely to report that economists understand things very well, than people with better education. The unconditional probabilities of people who report economists know “nothing at all” are displayed below:

and the regression estimates on educational categories (omitted: dropouts) are as follows:

More education is highly and statistically significantly associated with lower probability of reporting economists understand little, and with higher probability of reporting economists understand a great deal.

Perhaps because the sample here is relatively small, none of the differences between coefficients in 2006 and 2010 are statistically significant in explaining either high or low credence in economists, and moreover they are not jointly significant. Ordered probit and semiparametric ordered response models (not displayed) lead to similar conclusions, although there is some evidence that people with graduate educations revised their beliefs downwards more so than people with less education.

REPLICATION FILES.

If you wish to replicate or extend these results, download the public use GSS dataset from NORC here. Download the Stata .do file which generated the results above here (change the extension from .doc to .do, it’s not really a Word file). Please let me know should you find an error in the code.

In The Sideways Effect: A test of changes in demand for merlot and pinot noir wines Cuellar et al provide some econometric evidence. They look for post-Sideways price and quantity changes in merlot and pinot noir relative to the “control” varietal cabernet sauvignon using scan data from the U.S.

First consider quantities:

“Promoted” here means the product was advertised. The red vertical line is at 2004, when Sideways was released. Quantity of pinot increases markedly in 2005 relative to merlot or cab. The story for price is less clear:

Promoted pinots increase in price a bit over 10% in 2000, and then trend up another few percentage points post 2004. Merlot prices, promoted or not, fall relative to the other varietals over this period, but the decline appears to begin in 2002.

These results are what we might expect if it’s not very hard to switch production from one varietal to another, and the demand shock manifests as quantity changes rather than prices changes. However, since current prices depend on last (and previous) years’ production and planting, it is somewhat surprising that we don’t see a relative price spike for pinot in 2004. The authors present some regression analysis further detailing these results (which is a little strange in my opinion—instead of a difference in difference approach, they assume a quadratic time trend, then look for a structural break in the linear component only in 2004).

My casual empiricism agrees with Bryan’s. What does actual empiricism say? As a crude first pass, I counted the number of articles in Jstor economics journals that contain the word “regression.” Of course not all articles that contain that word are empirical, nor do all empirical papers contain that word, but it seems a reasonable signal of empirical content. The proportion over time (one observation per decade) looks like
which shows that in the sample of journals tracked by jstor (which changes over time, muddying interpretation) the proportion of empirical papers, or at least papers which for some reason mention regressions, has been steadily increasing, although the rate of increase seems to have been lower 1970-2000 than prior to 1970 or after 2000.

Surprisingly, the small and mostly dated literature I could find on the topic leans towards the opposite conclusion:

• Oswald (1991) showed that the proportion of empirical papers in Economic Journal rose from 3% in 1959 to 20% three decades later, but almost all of that increase occurred by the mid 1970s. Oswald wondered if the discipline had been in an “equilibrium in which large numbers of researchers treat the subject as if were a kind of mathematical philosophy.”
• Coelho and McClure (2005) reported that papers top general interest journals became more complex and less data-driven over the last several decades.
• Morgan (1988) found that the proportion (about one-half) of empirical work in the mid 1980s in the AER was lower than the proportion of empirical work in political science (58%), sociology (78%), chemistry (100%), or physics (88%).
• Bridges and Haywood (2003) found that the proportion of empirical papers in health economics journals declined from 70% to 60% between 1982 and 2001.

These papers give little support to the notion of econometrics ascendent.

However, a recent paper by Oswald and Ralsmark stongly suggests that empirical work currently dominates. The authors assessed the CVs of 109 assistant professors in top 10 U.S. departments. They find 20% of these researchers predominately write pure theory, which tells us that only a small proportion of the best and brightest are pure theorists, and that departments with their choice of freshly minted PhDs are overwhelmingly picking empirical people.

It would be interesting to see an updated version of the papers that came out circa 1990 studying the proportion of theory papers, and not just in the AER or other top general interest journals, which may not be representative (Figlio 1994 found that top journals were substantially less likely than second tier outlets to publish empirical papers, but also that as of that time convergence appeared to be occurring).

More questions: What is the optimal proportion of theory? How do improvements in hardware, software, and data availability affect the optimal proportion of theory?

David Mineta, Deputy Director for Demand Reduction for the Office of National Drug Control Policy, writes in The Tennessean:

Proponents of marijuana legalization often argue it will do everything from fixing our economy to ending violent crime (“Marijuana legalization bill offers safer alternative,” Tennessee Voices, Aug. 15). Yet, the science is clear: Marijuana use is not a benign drug and it is harmful to public health and safety.

No reasonable observer believes that marijuana legalization will “fix the economy” nor, obviously, would it “end” violent crime. Charitably suppose Mr. Mineta intends to counter the common view among drug policy experts that the costs of prohibition exceed the benefits. Mr. Mineta believes that demonstrating that “marijuana use [sic] is not a benign drug” is equivalent to showing that prohibition is sound policy, which is mistaken.  Mr. Mineta believes the science supports his claims:

Decades of scientific study, including research from the prestigious National Institutes of Health, show marijuana use is associated with addiction, treatment admissions among young people, fatal drugged driving accidents, and visits to emergency rooms. Data also reveal that marijuana potency has almost tripled in the past 20 years.

The negative social consequences Mineta claims support the case for prohibition are ER visits and traffic accidents.  But he makes an error in assuming that because marijuana use causes ER visits and traffic accidents, marijuana prohibition reduces ER visits and traffic accidents.

Consider a simple model in which traffic accidents are caused by alcohol use , marijuana use , holding all other causes constant:

Suppose and , that is, assume that all else equal increases in either marijuana use or alcohol use causally increases accidents.

Now suppose marijuana is legalized and let the new consumption levels be and . We expect marijuana use to increase under legalization, so . What happens to alcohol use depends on whether alcohol and marijuana are economic substitutes or complements. If these drugs are substitutes, alcohol use falls when the (full) price of marijuana decreases with legalization, so

The left hand side is the decrease in alcohol-caused accidents induced by marijuana legalization and the right-hand side is the increase in marijuana-caused accidents induced by marijuana legalization. The sign of the effect depends on demand elasticities and on the the relative dangers of these two drugs.

The evidence (e.g., Bates and Blakely 1999) suggests that it is likely that marijuana use causes accidents but that the effect is much smaller than that of alcohol, so . It follows that even a small increase in alcohol use may lead us to the superficially counterintuitive conclusion that: (1) marijuana use causes traffic accidents and (2) marijuana legalization reduces traffic accidents.

The evidence suggests that reductions in enforcement of marijuana prohibition cause decreases in traffic accidents. Chaloupka and Laixuthai estimate that marijuana decriminalization reduces youth traffic fatalities by 5.5% (p<0.01). Mr. Mineta's claim about ER admissions fails for the same reasons: Model (1993) shows that marijuana decriminalization increases marijuana ER mentions (note that a mention does not mean the drug was the direct nor indirect cause of the ER visit) but decreases
mentions of all other illicit drugs by about 14% (p<0.01).

Mr. Mineta then confuses correlation and causation:

This is especially troubling for use among teens because the earlier a person begins to use drugs, the more likely they are to develop a more serious abuse and addiction problem later in life.

The correlation that Mineta asserts does hold, but it does not follow that early use causes heavier use later in life. Even in the case of nicotine, which is an extremely physically addictive drug, the evidence suggests that exogenously changing age of initiation (such as through policy changes) has little effect on usage patterns later in life, see for example Glied (2003), Auld (2005), and Eisenberg and Rowe (2009). Put another way: the same characteristics which tend to induce a given person to begin smoking early in life also tend to induce that person to smoke more later in life. Since marijuana is much less addictive than tobacco, it would be very surprising to find that exogenous changes in age at initiation into marijuana use cause substantial changes in use later in life.

Mr. Mineta proceeds to argue that regulating marijuana would be bad policy because prescription drugs are regulated in Tennessee but nonetheless Tennessee has higher rates of prescription drug abuse than in other states, which is a senseless argument for several reasons, perhaps most notably that those other states also regulate prescription drugs.