WHY WE HAVE PEER REVIEW. Ohio University's Richard Vedder and some of his colleagues have set up something called the Center for College Affordability and Productivity and commenced publishing a weblog. Professor Vedder was a member of the Spellings Commission on Higher Education and the content of the weblog suggests Center members will use the site as an extended minority report on the Commission's statement.

One post caught my eye as it offers me an opportunity to get around to a book review I have been putting off. Professor Vedder has a frank exchange of views with Gary Becker.

Two events over the weekend demonstrated once again something that I already knew -- that the Higher Education Establishment in America is in a state of denial about the reality that America's colleges and universities are increasingly costly and inefficient -- and probably, at the broader social level, not a sector on which we should be devoting huge amounts of new resources without some fundamental changes.
In Boston on Saturday, I got into a bit of a civilized contretemps with a man I greatly admire, Professor Gary Becker of the University of Chicago, a Nobel laureate in economics and truly a great economist. I had suggested that empirical work I had done shows that perhaps new government investments in higher education have low rates of return, since state government appropriations and state economic growth are negatively correlated -- more spending, lower growth. He replied that we as a nation underestimate the true rate of return on higher education. College graduates commit fewer crimes, are less likely to be unemployed, have fewer low birth weight babies, etc. (Some other data from the estimable postsecondary.org even suggests that college educated drunk drivers are more likely to wear seat belts than their less educated drunken colleagues).
Like most good scholars, Becker is usually very fastidious about taking into account other factors that might explain the phenomenon he is investigating. But not here. I would speculate that if the individuals who graduated from college had gone into construction as semi-skilled or unskilled laborers, they likely would have been more responsible citizens than average Americans. College students are, on average, brighter, more motivated, more disciplined, etc., than non-college graduates. College admissions is a screening process, and the truly non-bright, poorly motivated, undisciplined individuals either do not apply for admission or are rejected. Many of the positive virtues attributed to college education may have been largely (or even completely) developed elsewhere -- in the home, through church attendance, etc. Becker seemingly dismisses or ignores some evidence that might suggest higher education's role in determining the quality of our lives may be less unambigously positive.

The negative correlation argument appears in Going Broke by Degree: Why College Costs too Much, which I read long ago and now offer Book Review No. 23. On one hand, I'm likely to sympathize with a man who can note (p. 213)

The corruption of college athletics, the occasional scandals over college admissions, the growing politicalization of the academy, lax standards, the construction of extremely luxurious facilities, and excessive student party-going threaten public trust in our institutions of higher education, along with the immense subsidies that allow universities to operate as they do.

On the other hand, if he's going to write about "undisciplined individuals [who] either do not apply for admission or are rejected" today, has he rethought "Although students are probably no more accomplished academically and arguably less so than their parents" (p. 112) or his case for performance-based scholarships to attenuate the moral hazard of subsidies that he suggests "would lower somewhat the scandalously high attrition rate" (p. 225) or his suggestion that universities outsource their remediation to Sylvan Learning (p. 179?) (He does not suggest that Sylvan send the bill to the high schools that dropped the ball in the first place.) One thesis of Going Broke by Degree is that third-party payments attenuate the incentive for the universities (the state-supported ones thanks to taxpayers, the privates thanks to endowments and student loans) to contain their costs, which is logical enough, but the book also maintains that those third-party payments induce students and parents to make inefficient choices of college. (He leaves unexplored the possibility that the low tuitions (relative to equilibrium) name colleges charge (to boost their selectivity rating, see p. 19) also induce inefficiently many applications and admissions. (Regular readers know where I stand on this point.)

But it is to that asserted negative correlation between college spending and state economic growth rates I must turn. Professor Vedder reports two such regressions. In the first, (p.135)

(Change in?) Personal Income Per Capita (1977 to 2002) = 35.837 (1.172) - 11.933 (Level of?) Higher Education Spending (3.570) - 0.004 1977 Income Per Capita (1.957) + 1.402 Taxes as Percent of Income, 1977 (1.070) + 0.091 Changes in Taxes as Percent of Income, 1977-1999 (0.064) - 0.476 Percentage Workforce in Unions (1.582) + 0.164 Age of State (4.362) + 0.152 Percent Days the Sun Shines (0.831) + 0.912 Percent Population over 65, 1981 (1.177). The adjusted R² is 0.641 and the equation F-statistic (what are the degrees of freedom?) 12.170. The t-statistics are in parentheses and I have bolded the estimates with t-statistics greater than 2.

Going Broke by Degree is a publication of the American Enterprise Institute and as such might be susceptible to editing in such a way as to highlight favorable findings. The editors did not require Professor Vedder to provide much by way of description of his sample (are we looking at 51 observations, showing the average annual growth rate in fifty states and the District of Columbia over a 25 year interval, or are we looking at a panel of 51x25 observations?) That makes a difference for the equation F-statistic as well as for the interpretation of some of the t-statistics close to 2, such as the coefficient on 1977 income per capita. And a referee for Review of Economics and Statistics or Regional Science and Urban Economics would raise a number of questions about potential autocorrelation in the case of a panel. There's also an obligation for a researcher to say something more about specification. We're not looking at estimates of an indirect utility function, which has some basis in theory, or the Becker-Mincer kitchen sink, which is a generally accepted specification. By way of explanation we get, "I included seven variables used in table 7-1 for control purposes (to approximate more closely the usual assumption of 'holding everything else constant')." At least these seven variables occur in levels, preempting the gripe I raise about people using "controlling for" rather than "dummying out," a frequent feature of the kitchen sink regression. But why these seven? Why not the won-lost record of the nearest pro football team or width of temperature range, January to July or square miles of waterfront?

A few pages later comes another regression, which is even more troubling.

(Change in?) Personal Income Per Capita (1977 to 2002) = 44.570 (1.760) - 13.002 Higher Education Spending, 1980 (4.431) - 6.583 Change in Higher Education Spending, 1980 - 2000 (1.533) + 1.186 Percent Population 18-24 Years Old, 1981 (0.561) - 0.004 Income Per Capita, 1977 (2.397) + 2.320 Average Tax Burden (1.875) - 0.705 Average Percentage of Workforce in Unions, 1973, 1984 (2.214) + 0.130 Age of State (4.213) - 0.232 Energy Production, 1977 (2.649). The adjusted R² is 0.690 and the equation F-statistic is 14.925. In this regression Professor Vedder reports that his sample is 48 observations.

This result is autistic number-crunching. (That's my single biggest complaint with the book, as there are other examples in other chapters, most of those, however, not concealed behind the phony precision of a regression estimate.)

We're supposed to believe that 25 years of change in personal income per capita can be captured in a single set of regressors with a different set of control variables "To deal with a possible problem of omitted-variable bias" (p. 139.) Possible problem?? How about a very real problem of specification bias (there is no theoretical framework in the book or in the footnotes leading to the choice of this data base and set of regressors) or an even greater problem of aggregation bias (the college-age population in 1981 will be prime-age workers in 2002, but that workforce is being augmented by subsequent graduates, and where have the retirees gone?) I fear that Going Broke by Degree will convince people who are already convinced, but it won't do much to convince the unconvinced, particularly those with a working knowledge of statistical inference.