Explaining Increases in Higher Education Costs
Archibald, Robert B., Feldman, David H., Journal of Higher Education
The real cost of higher education per full-time equivalent student has grown substantially over the last 75 years, and the rapid rise since the early 1980s is a cause of considerable public concern. Opinion surveys consistently find that how much one has to pay for a college education is a serious national issue. (1) Policymakers have responded to this concern. In 1997 Public Law 105-18 (Title IV, Cost of Higher Education Review, 1997) created an 11-member National Commission on the Cost of Higher Education. (2) More recently, in June 2005, Secretary of Education Margaret Spellings created a National Commission on the Future of Higher Education with a broad mandate to look into costs and accountability in higher education. When public angst is high and commissions are being created, good policy outcomes require a clear understanding of the forces behind the phenomena of concern. Unfortunately, there is little consensus and considerable controversy about the causes of the rapid increase in higher education costs.
In his July 1996 congressional testimony, David Breneman laid out the difficulty very neatly. He said that there are two competing theories explaining the rise of costs in higher education. The first relies on the insights of William Baumol and William Bowen about the cost difficulties faced by personal services industries (Baumol, 1967; Baumol & Bowen, 1966). As we will explain below, the ideas behind the "cost disease" explanation in higher education have a distinguished heritage in economics. The competing explanation is Howard Bowen's "revenue theory of costs" (1980). In Howard Bowen's view, the source of cost increases in higher education is the rising revenue stream made available to colleges and universities. Higher education institutions spend everything they can raise, so revenue is the only constraint on cost.
We have a number of goals in this article. The first is to explain the two competing approaches in some detail. To summarize our view, cost disease rests on a firmer behavioral foundation than Bowen's revenue theory. Despite that advantage, the choice between them ultimately is empirical. This is our second task. As Breneman noted in his testimony, "it is hard to test these two theories because for most of the post WWII era, higher education has experienced remarkable revenue growth" (1996, p. 60). The time series evidence on college costs is indeed compatible with both the cost disease and revenue theory explanations. We propose instead a cross-section test using disaggregated price data from a broad set of industries.
One important difference between these two theories is that the cost disease theory is based on similarities between higher education and other industries, while the revenue theory of costs is based on peculiarities of higher education as an industry. Howard Bowen is by no means alone in proposing higher education-specific explanations for cost increases. Malcolm Getz and John Siegfried (1991) list six competing explanations, one of which is cost disease. The other five are higher education-specific explanations: cost increases arising from a change in the product mix toward more expensive disciplines; cost increases arising from shortages of higher education inputs; cost increases arising from faculty and administrators in charge having inflated desires for quality; cost increases arising from poor management in higher education; and cost increases arising from government regulations creating expanded duties for higher education. (3) We will focus on Bowen's revenue theory of cost because, unlike the other higher education-specific causes in this list, it is overarching. It is not tied to a specific time frame. Like cost disease, the revenue theory is meant to explain the entire evolution of cost in this industry.
Before proceeding, we need to bring in the substantial theoretical and empirical literature on cost functions and production surfaces in higher education. This literature is a major contributor to our understanding of the forces affecting college costs. Some of its many strands include identifying average and marginal costs of university output as a whole and by department or cost source, measuring potential economies of scale and scope within a multiproduct setting, and estimating production surfaces to evaluate efficiency. (4)
Cost functions are estimated from data, and the approach presumes that firms attempt to minimize cost. If this assumption is true, we can give the cost function a technical interpretation--that is, it captures the production function that determines the choices available to universities. If it is not, then the cost function is only capturing behavior, not technology.
An extreme interpretation of the revenue theory might suggest that the operating principle in higher education is cost maximization. Clearly, this is not strictly true because universities do leave revenue on the table and some do not spend all they take in. As Brinkman notes, "higher education institutions neither minimize nor maximize costs; instead they operate within a range of accepted norms for production relationships, such as student-faculty ratios or lab space per student for instruction" (1990, p. 110). Thus, the evidence from estimates of cost functions may be compatible with either overarching theory. For this reason, despite the careful theoretical and empirical work in this literature, we need to step outside of its framework in order to subject the cost disease explanation of rising higher education costs and the revenue theory of cost to any form of critical test.
The difference between a higher education-specific explanation and an economy-wide explanation provides the basis for our test. If the revenue theory of costs or other higher education-specific explanations have great explanatory power, costs in higher education should follow an idiosyncratic time path. On the other hand, if the cost disease explanation dominates, the time path of costs in higher education should be very similar to the time paths of costs in industries that share the characteristics creating cost disease. Using cross-section industry data, we show that the evolution of cost in higher education is very similar to the evolution of prices in other service industries that use highly educated labor and strongly dissimilar to industries producing standardized manufactured goods. We can reject the hypothesis that higher education costs follow an idiosyncratic path.
This result has important consequences for how one might go about controlling costs in higher education. If cost disease is the primary long-term driver of real increases in cost per full-time equivalent student, then cost control cannot be achieved without productivity growth. The problem in higher education is that productivity growth often is synonymous with lower quality. Adding more students to each class can diminish the benefit for each student, leading to diminished outcomes and lower graduation rates. Increasing the number of courses a professor teaches would reduce research or community service, both of which are outputs of higher education. Productivity growth that is quality neutral or quality enhancing requires a change in the technology of service delivery.
The article follows in three additional sections. Section two provides a detailed discussion of the competing explanations for rapid cost increases in higher education. Section three contains our test. Section four discusses the policy consequences of our findings.
Competing Theories of College Costs
This section gives a more detailed account of the two competing theories explaining the rapid increase in costs in higher education. We begin with a simple expository relationship between unit cost, educational quality, and the technology of service delivery. This relationship serves as a framework for discussing both the cost disease and the revenue theories.
Figure 1 shows the constraint faced by a college or university. The unit educational costs-quality locus in the figure illustrates the simple idea that, within the existing technology for service delivery, a college or university can only achieve higher quality if it is willing and able to pay higher educational costs per unit. Two features of the figure deserve emphasis. First, it focuses on the costs of providing an education, so other costs such as those of housing or feeding students or fielding athletic teams are excluded. Second, the qualifier "within the existing technology" is crucial. Technology does not refer to new hardware or software alone. It refers to the entire currently understood process (or menu of ways) by which higher education services are delivered by universities. Improvements in technology that would shift the curve down are certainly possible. We focus here on the constant technology case to highlight differences in the two theories under discussion.
[FIGURE 1 OMITTED]
In brief, the revenue theory of costs can be summarized by saying that an institution chooses a point on this constraint based on what it can afford. In other words, given its revenue, the institution determines its costs. The presence of cost disease would lead this constraint to shift up over time. In this case, to maintain quality in the face of rising cost requires increased revenue. Without matching revenue increases from public appropriations, private giving, or tuition, quality must erode over time. The constraint also can be moved by productivity-increasing technological change. Cost-reducing technological progress in this sector would shift the constraint downward. This would permit higher quality at a constant cost per unit, lower cost at a constant quality, or some of both.
The cost disease explanation is traditionally traced to Baumol and William G. Bowen (1966) and Baumol (1967). Yet this work is strikingly similar to parallel research done in international economics by Bela Balassa (1964) and Paul Samuelson (1964). And Balassa's and Samuelson's arguments are a formalization of insights that trace back to the work of David Ricardo in the early nineteenth century (Ricardo 1821).
Cost disease is based on the idea that technological progress that increases labor productivity (and thus reduces unit cost) is not randomly distributed across industries and over time. The likelihood of productivity growth is related to how labor is used in the industry. "In some cases labor is primarily an instrument--an incidental requisite for the attainment of the final product, while in other fields of endeavor, for all practical purposes the labor is itself the end product" (Baumol, 1967, p. 416). Manufacturing is the prime example of the former, and higher education is an excellent example of the latter. (5) If you can cut the amount of labor that it takes to make most manufactured goods, competition in the long run transfers the higher productivity to workers in the form of higher wages and/or lower prices. On the other hand, for many services productivity gains are either hard to achieve or would be considered decreases in quality.
Despite their lagging productivity, personal service industries have to compete for workers with goods-producing industries. Because they are experiencing technological progress, the goods-producing industries will be giving substantial wage increases to their workers. The only way that service industries can compete for workers is by raising wages also, and this causes prices of services to rise much more rapidly than the prices of goods. This is the cost disease process.
Baumol provides an extreme example from the entertainment industry that is often repeated in discussions of cost disease. He notes that "a half hour horn quintet calls for the expenditure of 2.5 man hours, and any attempt to increase productivity here is likely to be viewed with concern by critics and audiences alike" (1967, p. 416). On the other hand, productivity gains are indeed possible in higher education. Technological innovations like closed-circuit television in the 1960s or Web-based distance learning today have the potential to increase productivity. Yet, at least to this point, the primary delivery vehicle remains the faculty member who interacts with students. An institution can increase class size to raise measured output (students taught per faculty year) or use less expensive adjunct teachers to deliver the service, but these examples of productivity gain are likely to be perceived as decreases in quality. An institution can also increase the number of courses each faculty member teaches per year, but not without having an impact on other attributes of output such as research or public service.
At roughly the same time Baumol was developing his cost disease theory, international economists were grappling with a related phenomenon. One of the oldest stylized facts in economics is that the cost of living is systematically higher, and the value of money is correspondingly lower, in countries with higher average standards of living. In other words, $1,000 buys more in Djakarta than in Detroit. The Penn World Tables calculates national price level information from disaggregated microeconomic data. (6) In Figure 2, the most recent base year data from 2000 show the clear relationship between level of development and national price level for a fixed basket of goods and services.
As long ago as 1817, David Ricardo noted this phenomenon and identified the probable cause. Ricardo claimed that "The prices of home commodities ... are higher in those countries where manufactures flourish" (Ricardo, 1821, chap. 7, para. 35). The term "home commodities" is Ricardo's language for nontradable goods and services. Goods may be nontradable because of large transport costs relative to value. Services often are nontradable because of their mode of delivery--you have to go to the provider. Ricardo asserted almost 200 years ago that the price level would be higher in countries that were further up the development ladder, and the reason would be that richer nations' nontradable goods and services would cost more locally than would the corresponding goods and services in poorer nations' domestic markets.
[FIGURE 2 OMITTED]
This claim is the international cross-section counterpart to cost disease. In 1964 Bela Balassa and Paul Samuelson simultaneously advanced the proposition that the positive correlation between the price level and real per capita income could be explained by productivity differentials between nations. The average level of labor productivity is higher in richer nations than in poorer nations. This is why the richer nations are richer. But Balassa and Samuelson argue that the productivity advantage is concentrated in tradable goods, not in nontradable services. In the absence of significant barriers to trade, international trade tends to equalize the prices of tradable goods across countries. This means that wage rates will be higher in countries that have higher labor productivity in these tradable goods. Higher wages also push up service prices in richer countries because there is no service sector productivity advantage in richer countries to match their advantage in tradable goods. In plain language, a half-hour hair cut or hourlong university lecture should cost less in a poorer country, but a Toyota or a barrel of oil would not. Since the overall price level is a weighted average of prices for tradable goods and nontradable services, the price level should be higher in richer nations. This became known as the Balassa-Samuelson hypothesis, which is one of the most well-established propositions in international economics. Clearly, if the cost disease phenomenon were present in each nation, then as countries become richer (through productivity growth in manufacturing), their service prices indeed would tend to rise.
Revenue Theory of Cost
Howard Bowen summarizes his theory this way:
On the whole, unit cost is determined neither by rigid technological requirements of delivering educational services nor by some abstract standard of need. It is determined rather by the revenue available for education that can be raised per student unit. Technology and need affect unit costs only as they influence those who control revenues and enrollments. (1980, p. 18)
It is easy to see why this argument was named the revenue theory of cost. Using Figure 1, universities see the quality/cost locus as a constraint. They work assiduously to loosen the revenue constraint since that is the path to higher quality. Since universities spend all they are given, the gain in quality from the last dollar of spending may be positive but low in comparison to the social value of the same public dollar spent somewhere else, like health care or K-12 education. In Bowen's view, public restraint is a guarantor that keeps universities from wasteful overspending.
There are two ways to interpret the revenue theory of cost. First, it might be trivial. In a nonprofit setting, costs equal revenue, so in each period the revenues available determine the costs that can be expended. This is not very illuminating. Bowen had something more in mind. By claiming that the determination of unit cost is separable from "rigid" technology or "abstract standard of need," he puts revenue in control and ignores or downplays other factors.
Because revenue is the constraint on costs, Bowen expects colleges and universities to do everything they can to loosen the constraint. His third "law" of higher education costs states "Each institution raises all the money it can" (1980, p. 20), yet a look at tuition-setting behavior shows that universities are not revenue maximizers. The fact that selective universities commonly draw students from their waiting lists is evidence that excess demand exists for places at those schools. (7) Universities with excess demand could increase charges without suffering any excess capacity. One could argue that raising prices might decrease the yield of high quality students and that this would harm the overall quality of the institution. Indeed, this is true. Many institutions practice need-blind admissions in order to attract the best possible student body. These institutions clearly leave revenue on the table. This behavior suggests that colleges and universities maximize some measure of excellence, prestige, or quality, but not revenue. This is Bowen's first "law": "The dominant goal of institutions are educational excellence, prestige, and influence" (p. 19). The difficulty is clear; there are conflicts among Bowen's "laws." The institution can maximize quality, or it can maximize revenue. It cannot do both. And at least in setting tuition, the maximization of quality trumps the maximization of revenue.
We can attempt to resolve the conflict in Bowen's "laws" without losing the spirit of his argument. He is saying that institutions maximize "educational excellence, prestige and influence" facing a revenue constraint, and they do what they can to loosen that constraint without doing damage to their main objective. The instances in which institutions fail to maximize revenue are simply times in which doing so would do damage to the quality of the education they could offer. Also, it is worth noting that tuition revenue is probably the only type of revenue that institutions are not interested in maximizing. Larger donations and larger state appropriations are always preferred to smaller ones.
Given Bowen's argument, the difficulty that policymakers really have with colleges and universities concerns aspirations of quality. Colleges and universities want ever-increasing quality, but policymakers are not convinced that these quality gains are worth the associated expense. The only way that Bowen sees to control the institutions is to control their revenue. On the other hand, if cost disease is real, then public institutions are condemned to perpetual decline relative to private colleges and universities so long as private donors think differently about quality than do state legislators (see Kane, Orszag, & Gunter, 2003).
The Two Theories and the Time Series Data
Bowen and other authors who put forward higher education-specific explanations for increases in higher education costs were quite familiar with the cost disease explanation. On occasion, these analysts went to some lengths to explain why they did not endorse it. Two of these discussions deserve some scrutiny.
Massy (2003) gives two reasons for why he discounts cost disease both as an explanation of the past and as a forecast of the future. First, he claims that Baumol-style cost push can account for only a small fraction of the current increases in higher education cost. His argument is problematic for a number of reasons. He assumes that nonfaculty labor costs, which comprise 10-20% of educational and general expenses, are not rising at rates similar to faculty salaries. This is very unlikely, since a large number of administrators, laboratory technicians, librarians, health and counseling staff, and information technology support personnel are very highly educated. He assumes also that the rest of an institution's costs are subject to normal productivity gains, which is a heroic assumption since many nonfaculty activities provided by colleges and universities are themselves services. Massy also emphasizes the possibility of future productivity growth within higher education. This is indeed the only way to break the grip of cost disease, and there is some scope for productivity change in all economic activities, but the possibility of productivity growth in the future is no reason to dismiss the importance of the lack of productivity growth in the past.
Bowen's rejection of cost disease is rooted in the time series behavior of real cost per full-time equivalent student. His basic claim is that the cost disease explanation is inconsistent with the broad pattern of data on higher education costs. Figure 3 presents the time series data for Real Educational and General Expenditures (E & G) per student for 1929-1995. (8) These data cover all higher education institutions, public and private, including two-year and four-year institutions. Bowen relies on these observations as the basis for his claim that the cost disease explanation is unsatisfactory.
[FIGURE 3 OMITTED]
There is one obvious anomaly in the data. It occurs in 1943, which is out of line with the surrounding data points (the early data are biannual). This anomaly is caused by a precipitous drop in enrollment, no doubt caused by World War II, accompanied by much less severe drops in expenditures. Otherwise, the year-to-year changes in the data are fairly consistent. They show level real E & G spending per student in the 1930s and 1940s. This period was followed by a sustained rise from roughly 1950 to 1970. Thereafter, real expenditures per student stopped increasing and fell slightly until the early 1980s. Real expenditures resumed their upward march in the early 1980s at a rate that is as rapid as the rate observed in the 1950s and 1960s.
The potential for difficulties with the cost disease explanation are concentrated in the 1929-1982 period. During that period, the entire cost rise was concentrated in a burst of activity between 1950 and 1970. In constant dollars, educational expenditures per full time equivalent student remained roughly constant between 1931-32 and 1949-50. Using Bowen's adjustments for changes to the composition of the student population (the increasing proportion of more expensive graduate students), real expenditures per FTE student in higher education actually decreased. Real expenditures then doubled over the period 1949-50 to 1969-70. (9) Between 1970 and 1982, real cost per FTE student again declined. (10) These periods of the declining real costs for higher education are what caused Bowen to dismiss and Thomas Kane (1999) to question the importance of the cost disease explanation.
In the aftermath of the Second World War, public funds began flowing into higher education. This period saw the expansion of the role of government that persists to this day. Public higher education expanded dramatically, and cost per FTE student rose. The federal government also played a supportive role during the postwar period, first with the GI Bill, then with the financial aid introduced in the National Defense Education Act in 1958 and the Higher Education Act of 1965, both of which aided students in private institutions as well as public institutions, and finally through increases in federal research funding. Rising real appropriations came to an end with what we now call the tax revolt. (11) The rate of cost increase in this period is thus seemingly a function of the revenues made available through the political process. This is Bowen's interpretation (see Bowen, 1980, pp. 37-47). Our review of the time series evidence is quite different. The existence of these periods during which real spending per student declined does not necessarily mean that the cost disease explanation fails for higher education. (12)
First, the decade starting in 1972 was a period of slow productivity growth. Bureau of Labor Statistics data for output per hour in manufacturing show that productivity grew 3.04% from 1960 to 1972, 1.81% from 1973 to 1981, and 3.16% from 1982 to 1995 (Bureau of Labor Statistics, 2007). The cost disease explanation relies on rising productivity as the engine for rising real wages, which generates rising costs in service industries. Absent rapid growth in real wages, there will not be rapid growth in costs in an industry like higher education. On the basis of the productivity data alone, we would expect less rapid increases in higher education costs during the 1973-1982 period.
Second, there were important changes in the relationship between wages and education. Between 1970 and the early 1980s, the average earnings of male workers with 5 or more years of college education fell approximately 20% in real terms. Faculty salaries tracked downward with them. (13) Between 1970 and 1982, faculty salaries at public four-year institutions had fallen almost 25% in real terms. The fall at private universities was slightly greater at 30%. These reductions in the returns to higher education and the associated decreases in faculty salaries have roots in the changing structure of the overall economy, and they clearly affected costs in higher education. Combined with the decline in productivity growth, the reductions in the real salaries of important workers in higher education are consistent with a considerable slowing of the growth in real prices in higher education. The returns to education rebounded in the 1980s, and this is an important part of the reason why college costs started to increase in real terms.
Third, the data for higher education in Figure 3 cover both two-year and four-year institutions, and the second period of decline in the real costs in higher education was a period of very rapid expansion in two-year institutions. Costs per student at two-year institutions are significantly lower than costs per student at four-year institutions. If the data on real cost per FTE student had been calculated using a constant mix of institution types, cost would have grown 1% more rapidly and the measured cost per FTE from the early 1970s to the early 1980s would have shown a much less significant decline. (14) In summary, there are several arguments that a proponent of the cost disease explanation for higher education costs can use to explain the declining real costs in higher education in the 1970s in Figure 3.
In addition, a focus on the periods of decline of Figure 3 is not the only way one can approach the data. It is possible to view the entire sweep of the historical evidence by breaking the data since 1929 into two distinct periods, using 1981 as a break. From 1929 to 1981, real cost per FTE equivalent student grew at an annual rate of 1.66%. From 1981 through 1995, the rate of cost increase accelerated to 2.74%. This acceleration in real cost after the early 1980s has occurred despite the restraint in state appropriations to public colleges and universities.
Perhaps the single most salient structural explanation for this break in the 1980s is the evolving economic return to higher education. The period of slower cost increase was dominated by what Claudia Goldin and Robert Margo (1992) have called "The Great Compression." In 1940, an American male at the 90th percentile of the income distribution earned five times as much as a man at the 10th percentile. By 1950, the gap had shrunk to a factor of three. In terms of years of schooling, between 1940 and 1950 there was a 13% decrease in the wage premium for college graduates. Goldin and Margo estimate that almost half of the compression was due to falling returns to schooling. The Great Compression also had staying power. Male wage differentials in 1975 were very similar to their 1945 levels. This extraordinary smoothing of the income distribution went into reverse starting in the late 1970s. By 1999, the 90-10 gap for male workers had risen to 5.4. Again, much of this increased income dispersion results from a rising earnings gap between college graduates and those with a high school degree or less. Thomas Lemieux (2006) presents evidence indicating that the rise in the 90-10 gap is accentuated for more highly educated workers.
Highly trained labor is an integral component of producing higher education. Wages and benefits comprise 70 to 80% of a university's operating budget. Most of that labor expense results from the industry's intensive use of highly educated labor. Faculty and administrators are the most obvious source of cost, but much of the support staff at a university also has a university degree or more. This includes everything from librarians and IT personnel to departmental executive secretaries.
Our objective in the foregoing discussion was not to make an argument for one or the other of the explanations for the rise in higher education costs, but rather to indicate that it is very difficult to use the time series evidence to sort out which of the two theories provides the more satisfactory explanation. The periods during which real higher education costs declined clearly cast some doubt on an explanation that seems to point to continually increasing real costs, yet there are other factors that make these periods of decline plausible in the context of the cost disease explanation. The time series evidence in Figure 3 is not sufficient to allow one to distinguish between the two explanations.
A Test of the Competing Theories
To separate these two explanations, we have to turn to cross-section data. The data come from the price indexes for Personal Consumption Expenditure by Type of Product generated by the Bureau of Economic Analysis of the Department of Commerce. (15) These data come from the Gross Domestic Product accounts, which record expenditures and prices for the final purchaser of the good or service. As a result, the classification of some product categories may seem strange. For example, the product category gas is classified as a service because the final purchaser is paying for the service of having natural gas delivered to his or her home. There are several service categories that have the characteristic that a large portion of the price of the service is bound up in the price of the product being delivered. Also, the name of several of the product categories starts out with "other," making the composition of the product category ambiguous. We have included an Appendix that lists the subcategories under these "other" product categories.
Using the lowest level of aggregation with continuous data from 1929, there are price indexes for 69 individual product categories, 13 of which are durable goods, 17 of which are nondurable goods, and 39 of which are services. (16) We can compute the rate of increase of prices for all 69 of these product categories. We will be comparing the behavior of these prices with the behavior of cost per full-time equivalent student because there is no time series evidence for costs in these industries. In higher education, subsidies allow colleges and universities to set prices below costs. (17) Most other firms are not provided subsidies, so prices exceed costs because the unsubsidized industry has to return a profit to its owners. Our maintained assumption will therefore be that there are not systematic changes in the profitability in the industries producing the goods and services that mask the underlying time series behavior of costs. This allows us to compare costs in higher education with prices in other industries.
We are not able to directly test the revenue theory of costs. As we noted earlier, Bowen did not properly specify an objective function that guides university behavior, which renders his theory difficult to frame as a testable hypothesis. Yet one characteristic that it shares with several other explanations of costs in higher education is that it is a higher education-specific theory. It relies on factors affecting the revenues in higher education to explain the behavior of cost in higher education. If higher education-specific factors are the primary driver of college and university costs, it would be merely a coincidence if the prices of any of the other product categories in the data had a time pattern similar to the time pattern of costs in higher education. The revenue theory is silent about the products whose price behavior should be similar to higher education costs. On the other hand, the explanation of higher education cost increases based on cost disease makes a prediction. It predicts that costs in higher education will have a time path that is very similar to the time path of the prices of product categories for personal services, particularly personal services that depend upon highly educated labor. This difference in the prediction gives us a chance to sort out which of the theories is more consistent with the data.
We recognize that other industries are no different from higher education; the explanation of the time path of costs in any industry likely incorporates a combination of industry-specific factors and more general economy-wide factors. In some instances, the industry-specific factors will be very important. For example, the deregulation of the airline industry affected costs and prices, and changes in Medicare and Medicaid regulations may have influenced the cost and prices of physicians. In every industry the industry-specific factors could well be dramatically more important than any economy-wide factors. If that is the case, we will not find any patterns in the data. The industries whose time paths of cost are similar to higher education will be a random group. For us to find important economy-wide influences, these factors have to be important for cost changes both in higher education and in the other product categories.
There are two ways to think about similar time paths. First, one could simply look at the rate of change over a representative time period. The goods with similar time paths to higher education would be the goods whose increase in real prices was similar to the increase in real cost per student in higher education. Second, one could compare the shape of the time path of prices for goods and see how close it is to the time path of costs of higher education. The major difference in the two approaches is that the first only uses data from the end points of the time series being compared, while the second uses the information in the intervening years.
To construct the measure of how "close" two time series are, we divided the time period from 1949-50 to 1995-96 into 11 four-year long time segments(e.g., 1949-50 to 1953-54, 1953-54 to 1957-58, etc.). For each of our time segments, we computed a measure of real price in the second year relative to the real price in first year; for example, we divided the real price of a product category in 1953-54 (to more closely match academic years we averaged of the two years price indexes) by the real price of that product category in 1949-50. We computed cost indexes for higher education in the same manner. We then computed the absolute difference between the price index of each product category and the cost index for higher education over the 4-year period. If the rate of change of prices for a particular product over a four-year period was identical to the rate of change of higher education costs per student, the two measures would be identical and the absolute difference would be zero. The absolute differences would grow as the rates of change in the two series differed. To compute our final measure of the closeness of the two series, we averaged the absolute differences over the 11 four-year time segments covering 1949-50 to 1993-94.
We recognize that the choices of four-year time segments and starting in 1949 are somewhat arbitrary. We did robustness checks using ten-year, six-year, and two-year time segments and series that started in 1929. The results from these exercises were not qualitatively different from the results we report below. Starting the series from 1949 does allow us a relatively long series while avoiding the potential anomalies that come from including the Depression and the years strongly influenced by the Second World War and subsequent demobilization.
Table 1 presents the results of these calculations with the product categories listed in increasing order of the mean absolute deviation. In this way, the product categories whose time series price behavior was most similar to the time series behavior of costs in higher education are at the top of the table. To make them stand out, we have listed the service industries in boldface type and the aggregate measure in ALL CAPS. The third column of the table gives the other comparison, a measure of the real price change over the entire time period. For example, the 1.9185 in the first row of the third column tells us that the product category "Expense of handling life insurance and pension plans" rose 91.85% in real terms over this time period.
An inspection of the table indicates that the product categories at the top of the table, those whose pricing behavior is most similar to the behavior of costs per student in higher education, are not a random selection from the product categories. There are 69 product categories, 13 of which are durable goods, 17 of which are nondurable goods, and 39 of which are services. The top 20 product categories in the table contain 18 services and two goods (magazines, newspapers, and sheet music and Tobacco products). The probability of 20 random draws yielding 18, 19, or 20 services from a population that contains 39 services and 30 goods is .0003. (18) This result is sufficient for us to reject the hypothesis that the product categories whose pricing behavior most resembles costs in higher education are randomly drawn from services and goods.
The statistical test simply used the distinction between goods and services. The cost disease explanation is based on characteristics of personal services, not simply services. For higher education, the hypothesis should be that costs rise in a similar fashion as the prices of personal service industries that utilize highly educated labor. Because there is no clear-cut way to define the exact set of product categories with the desired characteristics, we cannot offer a statistical test.
We think that an inspection of the table, however, is sufficient to demonstrate the plausibility of the prediction. The top of the table includes several product categories that should be dominated by the types of service providers in question: expenses of handling life insurance and pension plans (statisticians and actuaries); dentists, physicians, other professional services (chiropractors, medical laboratories, and optometrists, etc.); hospitals; and legal services. Interestingly, admission to specified spectator amusements, which includes legitimate theaters and opera and entertainments of nonprofit institutions (except athletics), which is the subject of Baumol and Bowen's 1996 book, is quite high on the list. Also, one of the two goods among the top 20 on the list--magazines, newspapers, and sheet music--is a product, not a service, but nevertheless its production relies on a considerable amount of highly educated labor. Clearly, there are product categories at the top of the table that are not from personal service industries that utilize highly educated labor. Tobacco products, water and other sanitary services, and mass transit systems are obvious examples. The presence of these product categories, however, is not evidence against our hypothesis. We are not suggesting that personal service industries that rely on highly educated labor are the only industries that experience cost changes similar to higher education. The important point is that product categories from industries that rely on highly educated labor do cluster at the top of the table.
In general, the services that are further down the list either are personal services that utilize less well-educated labor (e.g., barbershops, beauty parlors, and health clubs; other personal care services; and domestic services) or are not personal services (e.g., railway, telephone and telegraph; bus; airline; electricity; and gas). It is perhaps surprising that the product category nursery, elementary, and secondary schools, and the product category other education and research, which includes commercial and vocational schools and foundations and nonprofit research, are not further up the list. The fact that some of the subgroups within this product category employ relatively unskilled labor (nursery schools) or college graduates whose salary lags the mean for that group (vocational and commercial schools, elementary and secondary schools) may account for the whole category's place on the list.
The major exception to these generalizations is the product category brokerage charges and investment counseling, which is a personal service that is typically provided by highly educated professionals. Pricing behavior in this product category is clearly out of line with the pricing behavior of the other personal services provided by highly educated labor and with costs in higher education. The reason why brokerage services are different is easy to see. This is an industry that has experienced dramatic productivity gains associated with online trading of stock and bonds--to make a trade one no longer has to meet a broker face to face, or even over the phone--and the process of completing the trade involves many more computers and fewer face-to-face exchanges. (19) The cost disease story relies on lagging productivity growth; at least in the computer age, brokerage charges and investment counseling does not satisfy this condition.
This analysis is not without its flaws. At times, the product categories confound things by lumping together the services of highly educated workers with the services of workers with much less education. For example, as the Appendix table shows, the product category other recreation includes veterinarians and videocassette rental. Still, the picture that this analysis paints should be clear. On the whole, costs in higher education behave much more similarly to costs in industries providing services rather than goods and, among service providers, higher education costs are much more like costs of services provided by highly educated workers than they are like costs of services provided by workers with lower levels of education. Clearly, there are industry-specific effects both in higher education and in the other industries, but the cost disease phenomenon is strong enough to show through in our analysis.
The third column of Table 1 gives the real price change from 1949-50 to 1995-96 for each product category. The comparable number for real cost per full-time equivalent student in higher education is 2.7149, which is higher than all but three of the product categories in the table (legal services, net purchase of used autos, and water and other sanitary services). (20) The prices of the product categories in the top of the table, those more similar to higher education, clearly rose much more rapidly than the prices of the product categories further down the table. The average for the top 20 product categories is 1.9054 while the average for the bottom 20 product categories is .9082. Clearly, there are anomalies. Again, brokerage charges and investment counseling is unusual. It is near the bottom of the table, but its prices increased considerably. But the broad result should not be surprising to anyone familiar with the literature on cost disease or the Balassa-Samuelson hypothesis. The product categories at the top of our table are dominated by services, and these are just the product categories that should experience more rapid price increases.
Figure 4 helps us understand the information in the table. In this figure, we have graphed the real price information for selected product categories and for higher education costs for 1949-50, 1969-70, 1979-80, 1989-90, and 1995-96. We used 1949-50 as the base so each series starts at 1.00 in 1949-50. The 1995-96 entries, at the right-hand edge of the figure, are the values in the third column in the table.
There are several results illustrated by this figure. First, the prices of the three goods (food for off-premises consumption, shoes, and new autos) all decrease in real terms, so their time series behavior was not very similar to costs in higher education or to the other services represented. Second, although the total price increase of brokerage charges and investment counseling is quite close to the total change in higher education costs, the time series behavior of the two series is very dissimilar. The reason for this is clear. Brokerage charges rose dramatically and then fell after 1979-80, which is very different from the time path of higher education costs, which started to rise more rapidly after 1979-80. Third, the time paths of prices of the services utilizing highly educated labor, which includes legal services, physicians, and dentists, and the time path of the costs of higher education all increase in slope after 1979-80. In contrast, the time paths of the prices of services utilizing less highly educated labor, such as domestic service and barbershops, beauty shops, and health clubs, leveled off after 1979-80. This evidence is consistent with the importance of the increase in the returns to education starting in the 1980s.
[FIGURE 4 OMITTED]
This evidence persuades us that higher education-specific explanations are not the best way to think about higher education costs. In fact, there is good evidence that country-specific explanations are similarly deficient. As Baumol and Blackman (1995) note, the long-term growth rate of the real cost of higher education in the United States is quite average compared to other nations. The most striking comparison is between the United States and Japan. Baumol and Blackman use UNESCO data to calculate the growth rate of the real price of higher education between 1965 and 1988, which is the period when Japanese labor productivity in manufacturing was soaring relative to the United States. During these years, the average annual growth rate of labor productivity in manufacturing was 2.8% in the United States and 6.2% percent in Japan. The cost of higher education rose at an annual rate of 5.56% in Japan but only 2.91% per year in the United States.
One can approach the study of costs in higher education, or in any other industry, by focusing on the things that make the industry different from other industries or by focusing on the things that make it similar to other industries. Clearly, the best explanation should account for both the differences and the similarities. The empirical question is which is most important. Without clear evidence, one should be suspicious of arguments like the revenue theory of costs that focus solely on specific features of an industry. Our analysis should turn this suspicion into disbelief. Cost per student in higher education follows a time path very similar to the time path of other personal service industries that rely on highly educated labor. This is entirely consistent with the cost disease explanation of the rise in cost in higher education. This explanation is based on strong economy-wide influences that affect industries that tend to experience lagging productivity growth and that rely on highly educated labor, not on characteristics of higher education itself.
While this evidence should not lead one completely to dismiss higher education-specific factors as part of the explanation for the rise in college costs, it makes it exceedingly difficult to sustain the position that these explanations are the whole story. In our view, the correct way to view past experience is to recognize that higher education behaves much the same way as other personal service industries that use highly educated labor. This does not mean that there is no role for higher education-specific factors, but it limits their role. Higher education-specific factors represent reasons why the cost behavior of higher education might be slightly different from the norm, but only slightly different. The data clearly are telling us that the cost disease phenomenon is the dominant reason that higher education costs have risen in such a sustained manner over the past 80 years.
Our conclusions suggest fruitful avenues for future work that could integrate an economy-wide focus with the ongoing research on cost functions and production efficiency in higher education. For instance, the same forces that have been identified as cost drivers in higher education--such as growing administrative staffing (the administrative lattice) or shifting product mixes--may also be present in other similar service industries (see Zemsky, Wegner, & Massy, 2005). A careful disaggregated examination of the set of industries whose characteristics match the profile of cost disease would help