Alternative Options for Deflating Education Expenditures Over Time
Economic Policy Institute
There is widespread interest in the problem of how to compare nominal education spending figures from different points in time or place. There are two distinct policy concerns involved:
a) Policy makers want to know if, at a single point in time, federal aid to education is being distributed fairly between localities. If the cost of education in different states or regions differs, then a given number of dollars in aid to one location will purchase a different quantity of real resources than that number of dollars in aid will purchase to another location. A similar question arises in large or diverse states, where the cost of living (and thus the cost of education) may vary considerably by urbanicity or geographic location. If these states seek to equalize spending or state aid between districts, an equalization of nominal dollars may not provide an equalization of real resources.
b) Policy makers want to know if, for a particular district, state or nation, the productivity of education spending is growing or declining over time. An industry's productivity grows if its outputs grow faster than its inputs. Education analysts have no satisfactory way to measure the industry's output, although test scores are used as a proxy. But even if this problem were addressed satisfactorily, we would still not know whether the productivity of education was growing or declining unless we can properly measure inputs. This is because, in any geographic location, the value of dollars spent will change over time because of inflation. Assume, for example, that measured school outputs have been unchanged from Year 1 to Year 2, but per-pupil spending has doubled. If inflation from Year 1 to Year 2 has been 100 percent, then school productivity will have been unchanged because output did not grow and neither did input (grow or shrink). But if inflation from Year 1 to Year 2 has been 50 percent, then school productivity would have been cut in half. Thus, the proper measure of inflation is necessary to make accurate assessments of historical changes in education productivity. Because the willingness of the public and legislators to increase education spending is dependent, in part, on judgments about whether past increases have been well spent or wasted, a proper analysis of inflation has great practical importance.
This problem of making proper inflation adjustments as a basis for making judgments about productivity exists in all economic sectors, not only elementary and secondary education. In the public sector generally, there is widespread policy concern about the extent to which expenditures have apparently increased in recent decades, without an apparent corresponding improvement in the quality or efficiency of the services provided. Americans pay higher taxes and receive public services whose quality, when not in decline, does not seem to improve commensurate with our higher payments. It is not only school officials, but all government, whose credibility is low, in part because Americans believe their tax revenues simply disappear into a bloated, bureaucratic hole: In the last quarter century, government spending jumped from 26 to 31 percent of our gross national product, while schools are not noticeably better, police protection has apparently declined, mail is delivered less often, streets are dirtier, and roads have deteriorated. This apparent conflict between rising public expenditures and declining quality of public service may be one of the causes of the resistance to taxation which increasingly affects public decision-making. If inflation in public services has been greater than experts usually estimate or than the public perceives, then real expenditures in public services may have increased less than public debate assumes. A proper understanding of recent inflation in public services is critical to decision-making about future appropriations because legislators generally must decide how many future dollars would be required to provide real increases in services, over and above the funds required to offset inflation. In general, this estimate must largely be based on patterns of inflation from the recent past.
There is also widespread policy concern about the extent to which nonpublic human services expenditures have also apparently increased in recent decades, and there is great confusion in our public debate about the extent to which these expenditures represent real increases or simply compensate for inflation. The clearest example of this is in medical care: considerable political energy was expended in the last year over whether various proposals to budget more funds for Medicare represented "cuts" from previous funding levels or simply "restrained growth" in funding. Much of the debate over President Clinton's failed proposal to provide universal health care coverage concerned the extent to which various elements of his (and others') proposals would provide real new health care services to Americans, or would, instead, stimulate greater inflation in health care resulting in more money being spent for the same services.
In sum, there are two clearly distinguishable problems in education cost adjustment theory. The first is a cross-sectional problem: adjusting nominal dollars so that the real purchasing power of expenditures can be compared between different geographic locations at a given point in time. This is related to the widely appreciated differences in the "cost of living" in different areas. The second is a longitudinal problem: adjusting nominal dollars so that the real purchasing power of expenditures can be compared between different points of time for the same geographic location. This is related to the widespread appreciation of the effects of "inflation." For overwhelming practical reasons, solving these two problems may require different conceptual approaches. We will return to this point later in this paper.
In a report we issued in November (Rothstein and Miles 1995), we began to deal with the problem of making longitudinal adjustments for inflation in education. Following a path suggested by William Baumol, we noted that inflation in school spending would normally be higher than the consumer price inflation with which most of us are familiar; so, to understand what portion of the nominal spending increases for education we should attribute to inflation, we sought to use a more appropriate index than the "consumer price index" used to measure inflation in the economy as a whole. For purposes of that report, we utilized a modified version of the "services" index calculated by the Bureau of Labor Statistics (BLS).
We will not review the details of that argument here, but have attached the relevant sections of that report as Appendix 1 and Appendix 2 to this paper. We plan to continue to work on these issues, and we know that others, more expert than ourselves, have done and continue to do important work here. In this paper, we state some of further questions we are now exploring and describe our current thinking about how to answer these questions.
Question 1: Does a specific inflation index for education mask the public choices we make?The report by Hanushek et al. (1994) states that productivity of public elementary and secondary schools is declining. Hanushek's analysis is based on his claim that real expenditures have tripled since 1960. This claim, in turn, assumes that it is appropriate to compare current expenditures to those in 1960 (and other years), after adjusting earlier expenditures by the "Gross National Product deflator." For practical purposes, this adjustment is similar to the more common adjustment made by other analysts (see, for example, Odden 1992, 10) who use the "consumer price index" to convert nominal to real dollar expenditures.
As noted, we have argued that because education is an inherently low productivity industry in the sense that cost efficiencies are hard to achieve, analysts should not assume education faces an average inflation rate. A consumer price index measures the average inflation of all goods and services, weighted by their importance in the consumption of urban families. A GDP deflator measures the average inflation of consumption, investment, government purchases and net exports in the economy. We suggest that a "net services" index corresponds more closely to the inflation facing industries such as education where cost efficiencies are hard to achieve.
The inflation rate chosen makes a large difference in one's measurement of school spending. Switching from the average consumption index, the CPI-U, to the net services index lowers the estimate of the real growth of per pupil spending over the 1967-91 period from 99.2 percent to 61.1 percent, a growth roughly 40 percent less. Using a GDP price index would suggest 121 percent growth, or double that shown if inflation were measured by net services.
In response, Hanushek and Rivkin (1996, 4) note that "if school expenditure is deflated by an output deflatorsuch as the GNP deflatorchanges in the series of real expenditures indicate changes in society's resources that are devoted to education." This, they add, "yield[s] an indication of society's overall resource investment in schooling." Tracking society's investment is useful, but this is not the issue addressed in our previous report where we examined how much the inputs into the education process grew: did schools have more teachers, books, facilities, etc. with which to educate students and from which we can expect better education outcomes? Hanushek and Rivkin's method can't answer this question.
Second, if education and GDP are both adjusted for inflation by the same index (the GDP deflator), then the computation of education spending as a share of GDP is equivalent to a simpler calculation where no adjustment for inflation is made (i.e., just use nominal dollars). That education's share of GDP in nominal terms is essentially what one expects given Baumol's disease, as would be true in many industries (depending on demand elasticities) which have low productivity. In these situations, more spending (proportionately) is needed each year in order to keep the same real resources (staff, facilities, etc.) available to students. It is possible that education's share of nominal GDP will grow while its share of real (inflation-adjusted) GDP will not, a manifestation of higher inflation in education.
Does this mean that the growth of education's share of GDP, or total spending, squeezes out other spending or consumption? It certainly means we spend more nominal dollars on education, but the pattern of productivity and inflation across sectors described by Baumol means that spending can decline in sectors with above average productivity growth.
Consider two extreme examples, education and personal computers. Because of different rates of technological change (see Appendix 1), inflation has been much higher in education than in manufactured products like personal computers. The cost of delivering education services has increased relatively rapidly, while the cost of comparable quality computers has actually declined. Does the fact that we now spend more of "society's resources" on education mean that we must sacrifice spending on personal computers? Not at all. We can spend more on education precisely because we do not need to spend more on computers, as computers become less expensive.
In summary, we do not accept the Hanushek-Rivkin attempt to defend their adjustment of education spending by the GNP deflator, rather than a services deflator more appropriate to education, by arguing that this method best illustrates social choices. If one wants to analyze the growth of inputs available to schools then it is necessary to take into account the inherent difficulties of achieving cost reductions in education, a factor which leads to higher inflation facing schools. The fact that education's share of spending has grown is just another manifestation of Baumol's disease. The fact that education's share of nominal spending has grown tells us nothing about whether its share of real resources has grown.
Question 2: Is the inflation in education best measured by examining changes in the prices of education inputs, like teachers and textbooks?In short, the answer to this question, we think, is "no," despite the fact that we ourselves use, in our own work, the term "inflation" to describe input price changes in education.
The reason for attempting to measure inflation in education is to measure the growth of inputs (i.e., translate increased spending on inputs into a "real" growth of inputs). There is no developed theoretical consensus about how to measure productivity (and thus inflation) in public or private services. In the manufacturing sector, the task is relatively straightforward. Economists calculate the value of enterprise shipments and subtract the cost of purchased inputs, yielding a resulting "value added" which includes the productivity of the enterprise's labor and capital. In public sector services like schools or welfare services, however, there are no shipments generating revenues from which purchased inputs can be subtracted. Thus, we are faced with the challenge of directly deflating nominal value-added, a challenge not faced in the manufacturing sector where real value added is a residual after real purchased inputs are subtracted from real shipments.
Note that the private sector methodology depends on the valuation of both purchased inputs and purchased outputs. But there is no way to price the outcomes of education. Thus, were it even possible to accurately count the changing nominal prices of real resources purchased by schools (inputs other than employment related costs), and to separate these prices into a "real" component (increased resources) and a component which represents price increases for the same resource, we would still not have an estimate of real value-added because such an estimate requires a valuation of shipments or output which is unavailable in education.
Question 3: Do price increases necessarily reflect "inflation" if the price increases do not result from either new resources or higher quality?As we hope to show, this is another way of posing the question which has recently been emphasized by Chambers and Fowler: "What is the difference between `expenditure' and `cost'?"
We begin to answer this question by asking why Policy makers and the public want to know the education inflation rate. The reason, it seems to us, is "accountability." We want to know how much of the price increase of education (rising per pupil spending) is the "fault" of elementary and secondary institutions, and how much is beyond their control. If the price of education has gone up because school administrators have "had to" pay more for education inputs, our first inclination is to increase the amount of money we give schools, to compensate educational institutions for their higher expenses. But if the price of education has gone up because school administrators have chosen to spend more money, then we may want schools to demonstrate improved outcomes to justify this increased spending.
A complication arises, however, when we try to define what it means to "choose" to spend more money. Clearly, if administrators add more resources (for example, lowering class size by adding more teachers), this is a choice for which we hold administrators accountableoutcomes should improve as a result. Or, if administrators add more money by upgrading the quality of resources (for example, hiring teachers with more advanced degrees, or from more prestigious universities, for whom higher salaries must be paid), this too is a choice for which we should hold administrators accountable.
But what if per pupil spending goes up because school administrators decide to pay school teachers at above market rates? The higher salary level might be more than is necessary to attract the desired quality of college graduates into the teaching profession, or it might be more than is necessary to attract better quality teachers from neighboring school districts (because salaries in the district are already higher than those in neighboring districts). In these cases, economists would say teachers receive "rents" in addition to their market wages.
The question we pose is this: Should "rents" paid to teachers or to other education inputs be considered a cost over which education institutions have no control? When we apportion the increases in prices of school inputs into the expenditures attributable to more (or higher quality) resources or to higher prices paid for the same resources, into which category should "rents" be assigned?
In our view, "rents," because they are within the control of education institutions and are not externally imposed higher costs, should be counted as real expenditures, not attributable to inflation. In other words, if school districts choose to spend more than is necessary for a given collection of education inputs, districts should be accountable to the public for improved results from such decisions, in a way that districts should not be held accountable for price increases of inputs which are beyond the districts' control. And, we emphasize again, when we say that districts should be accountable for unnecessary expenditures we do not suggest that these expenditures are wrong or that the public should prohibit them. Necessity is not the only basis for public decision-making. We would also add that the change in the size of "rents" in education over time may not be quantitatively large enough to materially affect inflation measures.
If we want, therefore, to define inflation in education as only those price increases over which education institutions have no control, we cannot calculate it simply by compiling a weighted average of actual prices paid by educational institutions for their various inputs. We must find a way to estimate what those institutions "would have" paid if markets for the provision of each of those inputs were fully competitive.
The distinction we make here is similar to that made by Chambers and Fowler (1995) and by Fowler and Monk (forthcoming) between "expenditure" and "cost." As they see it, "cost" is the minimum school districts must pay to obtain needed inputs. "Expenditure" is what school districts actually do pay, including what they term "discretionary" factors in payment. They have assumed, then, the challenge of constructing an education "cost" index which consists only of those prices schools must pay.
Thus, Chambers and Fowler describe districts' competition for teachers in terms (among other factors) of the concentration of teachers in a county who work for a single district. As theory predicts, they find that teacher salaries are lower where large percentages of teachers in a county are employed by a few large districts. Teacher salaries are lower where districts have monopsonistic power over their employment. This is shown in table 3.1 of Chambers and Fowler (page 37): in counties where the largest district has no more than 5 percent of total county enrollment (and thus, class sizes and other factors being equal, employs no more than 5 percent of the county's teachers), teacher salaries are 7.9 percent higher than in monopsonistic counties where all teachers are employed by a single county-wide district.
We differ with Chambers and Fowler, however, in that they consider that the single district county has a teacher "cost" which is 7.9 percent lower than that of a district in the 5 percent enrollment category. In effect, they claim that the large district experiences a lower inflation rate than the small district. We, on the other hand, consider this negative rent imposed on teachers by the single district county by dint of its monopsonistic power to be a "discretionary" factor. If we assume that prior year expenditures for teachers in each district of type = 100 and we were to decompose per-pupil spending increases for the single district county, we would still assign 7.9 percent of the teacher cost to inflation, for this represents an increased cost the district would have had to pay were it behaving in a competitive fashion. Because of its market power, this district is able to hold its per-pupil spending increases below the rate of inflation, without any reduction in real resources provided to pupils.
We take the argument a step further. We can imagine a table similar to Chambers' and Fowlers' table 3.1 in which teacher costs were indexed, not by the concentration of enrollment (i.e., teachers) in a county's districts, but by the concentration of all college graduates in a county employed as teachers by school districts in that county. We suspect there would be similar results: counties in which a large proportion of college graduates were employed as school teachers would have lower average teacher salaries than would counties in which a small proportion of college graduates were employed as school teachers.
This suggests that to construct a specific education price index, it would be more appropriate to utilize, as the component representing teacher salaries, an index representing the prices (salaries) of all college graduates in a region who are substitutable for teachers. In other words, a teacher cost index, to reflect inflation in teacher salaries, should be based on the salaries of "comparable" workers, not on teachers alone. Only in this way can the effects of market imperfections in education be reduced.
We have used the example of concentration of teacher employment to illustrate these problems of calculating inflation because Chambers and Fowler have provided such useful data in table 3.1. However, we conclude this section of our discussion by observing that the concentration of teacher employment by a single district, or by all districts relative to other college graduates, is probably not the most significant "discretionary" factor which causes the actual increase in teacher salaries to deviate from the true inflation rate for teachers. The most significant market imperfection undoubtedly remains the cultural, historic, and current discriminatory practices that foreclose other traditionally "male" occupations to many female college graduates. This is probably the largest single factor causing salaries of college graduates generally to exceed salaries of comparable teachers. We cannot say whether, at the present time, this gender stereotyping causes a difference in rates of change in teachers' vs. comparable college graduates' salaries. But, to the extent that it does create different rates of change, an employment cost index that reflects comparable college graduates will contain a smaller proportion of women, and thus describe a truer measure of inflation, than an index of teachers alone.
While, as discussed in another section of this paper, we believe that a sectorally-specific inflation index may be too difficult to construct and may not be the most useful for policy purposes, we have no theoretical disagreement in principle with a sectorally- specific index, an education price index. Our point here is only that, if an education specific index is desired, its component parts should not be the prices of the actual inputs used by schools, but should be the prices of "comparables" or "substitutables" (weighted by the relative importance of these inputs in education), because only by using such surrogates can the impacts of wage setting in education and its quality effect be judged. Only in this way can an inflation index tell the public how much more schools have "had to" pay for similar resources.
Question 4: Can an education price index be properly used to interpret changes in spending for components of education spending?Hanushek and Rivkin not only adjust total per pupil spending by the GNP deflator, based on the argument on "opportunity costs" described above, they then go on to adjust specific components of education spending by this deflator as well, an operation which we can't understand, even in their own terms of social choices. Thus, they note, the real inflation adjusted (based on the GNP deflator) "daily wage" of teachers has risen from $34.20 in 1890 to $182.80 in 1990. We regard the deflation of one specific input, like teacher salaries, by an economy-wide deflator as being even less meaningful than the deflation of a single sector like education by an economy-wide deflator. Indeed, we think that it is not even meaningful to deflate the input by an education specific deflator.
Assume that we have an education price index (or as we suggest below, a broader services index) by which we can track changes in real education spending over time. What if we want to know how much teacher salaries have risen over time or how many teachers can be hired based on a certain salary poolwhat deflator should we use?
Our answer to this question depends on why we want to know. Here are the possible answers:
- If we want to know whether teachers generally are overpaid or underpaid in market terms, we would calculate their real salary patterns using an employment cost index for comparable workers (college graduates). As explained in the previous section, use of such an index would effectively explain whether schools were using monopsonistic power to "underpay" teachers, or whether teacher unions were using monopolistic power to win "rents" for teachers.
- If we want to know whether teachers pay has kept up with (or exceeded) the "cost of living," we would deflate their salaries by the consumer price index, for this would tell us whether their salaries in different periods enabled them to purchase more or less of the typical collection of goods and services purchased by urban consumers.
- If we want to know whether teacher salaries are a greater or smaller share of total school expenditures than they were in an earlier period, we would not deflate the salaries at all. We would simply calculate the share in nominal terms. Note here that, as we described above, if teachers represent a greater share of all school expenses, this does not represent districts' greater opportunity cost for hiring teachers. If the employment cost index for comparable college graduates rose faster than the overall education cost index, and if the book publishing index rose more slowly than the overall index, districts could spend a relatively larger share of their total expenditures on teachers, and a relatively smaller share of their total expenditures on textbooks, without having to give up real textbook resources in order to meet their teacher payrolls.
Question 5: Should the "Net Services Index" be extended and made more generally available?In Where's the Money Gone? we calculated the real growth of per pupil elementary and secondary education spending from 1967 to 1991 by subtracting the cost increases attributable to inflation. As table 1 shows, we concluded that the inflation rate for services like education was an average of 6.7 percent a year, compared to 5.8 percent for consumer purchases and 5.4 percent for the GNP.
Since the publication of this result, we have received inquiries from many scholars and practitioners who wanted to know if we could either provide a "net services index" for other locations and/or time periods, or whether we could provide a relatively simple guide for how these scholars or practitioners could make the calculations themselves.
We made these calculations for the nation as a whole, as well as for each of 9 sample districts. We calculated inflation by taking the "Services" index published by the BLS, and then removing from this index the items attributable either to medical care or to shelter rent. In practice, because the BLS already publishes a "Services, Less Medical Care Services" index, it was necessary for us to remove the shelter rent components, using raw data provided to us by the BLS. The specific methods used are described in Appendix 2. The process was cumbersome and time consuming, largely because the weights of rent and medical care in the overall services index changed at various times during the 24 year period we studied.
After all this was done, however, we found that the "net services index" rose at approximately the same rate as the "services" index before medical care and shelter rent were extracted; over the entire 24 year period, the services index rose less than 1 percent more than the net services index. (With 1967=100, the 1991 index number for net services (national) was 503; for all services (national) it was 508). While medical care services had more rapid inflation than services generally, shelter rent had less rapid inflation than services generally, and these mostly cancelled each other out.
This was also the case for the local indices we constructed, but to a lesser extent. Some local net services indices varied by as much as 8 percent from the corresponding local services indices. Still, these were not large differences over a 24 year period. Thus, we concluded, given the parallel trends, that it might be easier for future research simply to rely on the service index.
We emphasize, however, that the rough correspondence between the services and net services index, both nationally and in subnational areas, is purely coincidental. There is no economic phenomenon that we can think of that would explain why shelter rent and medical care inflation would move in opposite directions of roughly the same magnitude. If indices were desired for other locations, or other time periods, the coincidence might be duplicated or it might not.
We are currently in the process of updating the net services index for the 1995-96 school year, and will be interested in seeing whether the coincidental correspondence of the services and net services index continues to hold in the more recent period. If we could be confident that the unamended services index presented an accurate reflection of inflation in elementary and secondary education, this would greatly simplify our work and that of other analysts.
We hope to test the correspondence of the services and net services index for as many years prior to 1967 as it is possible to do. We also hope to test this correspondence for intermediate periods, such as periods dating from 1970, 1975, 1980, and 1985. If the rough parallelism holds for these earlier and intermediate periods, we would recommend use of the easily accessible services index for adjustment of "real" education expenditures.
This will not enable us to understand the growth of education spending as far back as 1890, as Hanushek and Rivkin wish to do, but it will cover most of the years with which current debates about education productivity are concerned. An index going back to 1890 would necessarily be speculative, based not so much on data, as on investigations of economic historians whose interpretation of economic trends might be used to establish relationships between a surrogate services index and the growth of GNP.
Question 6: Is the "Net Services Index" (or all services index) preferable to a specific education price index for understanding inflation in education?We think yes, for two practical reasons. First, because government statistical agencies, like the BLS, have not published or even computed price indices which use the relative importance of specific education inputs, we believe it to be practically impossible for education researchers to reconstruct the prices of comparable inputs sufficiently far back in time to be useful. The only effort to do so, that of Kent Halstead (1983), resulted in an education price index going back only to 1975. Since, for example, considerable public debate now takes place about education's purported productivity decline since the 1960s, Halstead's index is not adequate to inform participation in that debate.
We do not, however, disagree with current efforts to create a cross sectional cost of education index, without historical data, such as that partially proposed (for teachers) by Chambers and Fowler (1995). Indeed, we are great admirers of these efforts. As we indicated earlier, these remarkable efforts will prove enormously useful to equalization and other fund distribution tasks. But examination of the enormity of the task attempted in Chambers and Fowler must lead to the conclusion that such a task would not be possible for historical data, with its need for very specific data on things like crime rates, amenities, etc. Therefore, even if desirable, construction of such an index for understanding inflation is not practical.
As a more practical alternative, we urge the use of a broader services type index which reflects inflation in services like education. While such an index may differ in important respects from a more specific education index, both in the types of inputs counted and their relative importances, this index is likely to be a more accurate surrogate for a sectorally- specific index than anything else now available or likely to be so. It is certainly likely to be more accurate than either the consumer price index or the GNP deflator, which most education analysts have inappropriately been satisfied with.
Second, problems of inflation affect not only education but other similar human services: child welfare services, law enforcement services, etc. Given the difficulty of constructing a sectorally- specific education index going back very far in time, it is practically inconceivable that analysts could develop similar indices for each of these sectors. It should be relatively easy to test whether the types of inputs and their relative importances are similar in each of these human services. We suspect that they are and if this suspicion is correct, public policy debates would benefit considerably from having a single human services index that could be used to understand how the real costs of human services in education and other similar sectors have changed.
Chambers, J. and W. J. Fowler, Jr. 1995. Public School Teacher Cost Differences Across the United States. U.S. Department of Education, National Center for Education Statistics, NCES 95-758.
Halstead, D. K. 1983. Inflation Measures for Schools and Colleges. Washington, DC: U.S. Department of Education, Office of Educational Research and Improvement, National Institute of Education.
Hanushek, E. A. et al. 1994. Making Schools Work: Improving Performance and Controlling Costs. Washington, DC: The Brookings Institution.
Hanushek, E. A. and S. G. Rivkin. 1996. Understanding the 20th Century Explosion in U.S. School Costs. Rochester Center for Economic Research. Working Paper No. 388.
Odden, A. R., ed. 1992. Rethinking School Finance: An Agenda for the 1990s. San Francisco: JosseyBass Publishers.
Rothstein, R., and K. H. Miles. 1995. Where's the Money Gone? Changes in the Level and Composition of Education Spending. Washington DC: Economic Policy Institute.