SinceFootnote 1 the Reagan Administration of the 1980s, policy debates in Washington have been broadly framed in terms of the appropriate share of economic activity devoted to the public sector. As Ezra Klein put it, “American politics is one long argument about what government should or shouldn't be doing, and how it should or shouldn't be doing it.”Footnote 2 Those on the left generally call for an expansion of the government's share of the economic pie, while those on the right search for ways to strangle what they view as an emerging Leviathan. Economists and political scientists have engaged in this contentious discussion even longer, ranging from Anthony Downs asking if the government is inevitably too small in democracies to James BuchananFootnote 3 asserting nearly the opposite.Footnote 4 The result is that a variety of explanations for the growth of a government—ranging from “Wagner's Law” to the role of interest organizations in promoting expansive governmentFootnote 5—have been developed and tested. Indeed, a comparison of two studies of government growth separated by a quarter-century—Lowery and Berry and Garrett and Rhine —suggests that our stock of theory about the growth of government has changed little over the last several decades.Footnote 6
Yet, most of these theoretical accounts—as well as the empirical analyses designed to test them—may be seriously misplaced due to their almost exclusive attention to real as opposed to nominal changes in the scale of public sector activity relative to the size of the economy as a whole. That is, most of the hypothesized explanations for government growth seek to explain why governments provides more goods and services.Footnote 7 But much of the nominal growth of U.S. government spending from the end of World War II to the mid-1970s—from two thirds to three quartersFootnote 8—was not in the form of the public sector doing more relative to the size of the economy as a whole (i.e., real growth in government), but was instead in the form of the things a government does becoming more expensive than activities in the private sector (i.e., government cost growth).Footnote 9 And this pattern did not change after the 1970s, despite many changes in the nature of economic production in the public and private sectors.Footnote 10
Baumol's work plays an important role in understanding government cost growth.Footnote 11 Baumol's theory accounting for government cost growth assumes that the economy is composed of two sectors—one relatively more capital intensive and the other relatively more labor intensive. The former tends to have rapid increases in productivity via capital improvements, while the latter has few productivity gains. The increases in productivity in the capital-intensive sector support wage increases. Then, to avoid a flow of labor from the labor-intensive sector to the capital-intensive sector as employees seek higher wages, the labor-intensive sector must also increase wages. The result is ever-increasing costs in the labor-intensive sector even if the quality and quantity of goods and services in this sector remain constant. If, as Beck and Baumol suggest, the public sector is more labor-intensive and the private sector is more capital-intensive, then the cost of goods and service in the public sector will increase over time relative to costs in the private sector.Footnote 12
A large and growing body of work supports Baumol's explanation for government cost growth.Footnote 13 Indeed, based on a survey of the literature, Nordhaus concludes that, “Baumol's hypothesis of a cost-price disease due to slow productivity growth is definitively confirmed by the data.”Footnote 14 As productivity increases in the more capital intensive private sector, wages increase. As private sector wages increase, labor competition between the public and private sectors leads to corresponding wage increases in the more labor-intensive public sector. The latter, however, are funded through higher taxes rather than via increased productivity. The result is that the costs of government for a constant market basket of goods and services tend to increase over time relative to costs in the private sector—“Baumol's disease.”
Our paper has two objectives. Our primary purpose is to assess what secondary impact, if any, government cost growth has had on the real size of a government. Baumol himself, at different times during his long career, has made contrary predictions in this regard. Baumol predicts that cost growth will lead to an inevitable decline in governments’ real share of economic activity.Footnote 15 Faced with rising costs for public sector goods and services, consumer-voters will shift spending from the former to the latter. In contrast, an older Baumol believes that government cost growth should not limit the real size of the public sector.Footnote 16 That is, because of productivity improvements in the private sector, we will be able to afford paying more for a constant level of public goods and services. We assess the empirical veracity of these differing assessments of the implications of government cost growth—as well as two alternative expectations that cost growth has different implications for the scope of government purchases than for the real amount of government transfers. Our empirical analysis modifies Berry and Lowery's econometric models of the real growth of public purchases and transfers to assess the veracity of three possible effects of the growing cost of government goods and services relative to costs in the overall economy—a concept that we abbreviate as “government's relative cost.”Footnote 17 We argue that Berry and Lowery's models are especially well suited for this analysis because, unlike all of the other extant empirical models of government growth cited earlier, Berry and Lowery's models disaggregate total government spending into government purchases and transfer payments, categories of spending the real level of which may be quite differently influenced by the government's relative cost. A secondary purpose of our paper, albeit largely in notes, is to reassess Berry and Lowery's original findings about the economic and political factors influencing the real size of a government by replicating the estimation of their models—taking advantage of both twenty-five years of additional longitudinal data and the many improvements in time-series analysis.
In the next section, we develop alternative hypotheses about the effect of the cost of government goods and services relative to costs in the overall economy on the real size of a government, deriving these propositions from predictions made by Baumol over the course of his long career. Then we review Berry and Lowery's econometric models, before modifying them to permit tests of our competing hypotheses. We present our empirical findings and conclude by discussing whether the younger or older Baumol had it more nearly right when considering the impact of cost disease on the size of government.
Baumol's disease and the real growth of government
As noted above, Baumol has been inconsistent in his expectations about whether and how the cost of government goods and services relative to costs in the overall economy should influence real spending. In some of his analyses of the cost disease, he suggests that as the real costs of public sector goods and services increase, consumer/voters should, at least in some cases, prefer to substitute private sector goods and services for those produced in the public sector, leading to a decline in the level of government services, and thus a decrease in the share of real economic activity devoted to the public sector.Footnote 18 Whether such substitution occurs depends, of course, on the price elasticity of public goods and services. Baumol, for example, cites higher education as a public service that is relatively price inelastic, noting that, “as productivity in the remainder of the economy continues to increase, costs of running the educational organizations will mount correspondingly.”Footnote 19 He also notes that “not all services in the relatively constant productivity sector of the economy face inelastic demand. Many of them are more readily dispensable … as far as individual consumers are concerned. As their costs increase, their utilization tends therefore to decrease into the category of luxury goods with limited markets or disappear almost completely.”Footnote 20 In such cases, reduced demand for price elastic goods and services could lead to sharp reductions in spending relative to the overall size of the economy. Thus, Baumol notes that, “spending on these services is apt to be cut back or, at best, increased by amounts that are barely sufficient to stay abreast of overall inflation.Footnote 21 As a result, the supply of these services may fall in both quality and quantity.” In assessing the severity of this consequence of the cost disease, however, it seems reasonable to assume that public goods and services are not uniformly price elastic or inelastic. Rather, they represent a range of elasticities. Thus, Baumol's disease should lead to downward pressure on the demand for public goods and services in general even if it is uneven across different types of goods and services.
Baumol also notes, however, that such downward pressure on demand could well be much more pervasive than warranted by variations in price elasticities, per se, if politicians and citizens misunderstand the full implications of Baumol's disease. This is, in a sense, a political rather than an exclusively economic interpretation of Baumol's analysis. On this critical point Baumol observed that:
The critical point here is that because politicians do not understand the mechanism and nature of the cost disease, and because they face political pressures from a similarly uninformed electorate, they do not realize that we can indeed afford these services without forcing society to undergo unnecessary cuts, restrictions, and other forms of deprivation.Footnote 22
We discuss the foundation of this political version of Baumol's disease more fully below. But, whether due to well founded economic choices concerning the proportion of public goods and services that are indeed price elastic and/or a political failure on the part of elected officials and their constituents to appreciate how the cost disease operates, real public spending as a proportion of economic activity might well decline over time. This reasoning prompts the following proposition:
Withering State Proposition: As the cost of government goods and services relative to costs in the overall economy (i.e., government's relative cost) increases, the real size of government declines.
In more recent work, however, Baumol contends that government cost growth should have little impact on real public sector spending.Footnote 23 This prediction is based on an expectation that productivity improvements in the capital intensive sector of the economy (the private sector in Baumol's stylized two-sector economy) make the country wealthier over time so that the public sector (the labor intensive sector in Baumol's model) can afford to spend more for a static level of real goods and services. Thus, Baumol notes that, “Despite ever increasing costs, [public] sector services still never become unaffordable to society. This is because the economy's constantly growing productivity simultaneously increases the community's overall purchasing power and makes for ever improving living standards.”Footnote 24 Indeed, Baumol anticipates that the government's share of gross domestic product (GDP) will rise to 60 percent over the next century with no change in real levels of public sector goods and services.Footnote 25 In making this prediction, Baumol is assuming that politicians and citizens can be fully informed about the true implications of the cost disease, an assumption that he acknowledges is fraught with difficulties.Footnote 26 Nevertheless, Baumol's more recent work is quite optimistic—at least on the economics if not the politics of the cost disease—and suggests the following expectation:
Surviving State Proposition: As the cost of government goods and services relative to costs in the overall economy (i.e., government's relative cost) increases, the real size of government should remain unchanged.
Yet, as noted above, there are reasons to believe that the Surviving State Hypothesis may be wrong. This returns us to the political implications of Baumol's disease. First, it views consumers as both rational and fully informed, so as to be aware of changing relative costs in the public and private sectors and the changing values of their income, and thus be able to avoid thinking exclusively about the size of government in nominal terms. However, there is considerable evidence that consumers, even when making routine consumption choices, both may not be fully aware of changing relative prices in many contextsFootnote 27 and are often susceptible to what Shafir, Diamond, and Tversky call a “money illusion” whereby they are attentive only to nominal costs.Footnote 28 This is not surprising; government goods and services—because of both the inherent separation of the decisions to spend and tax, and the collective nature of public sector fiscal choices—impose additional conceptual burdens on citizens’ capacities to think about marginal costs, burdens that may potentially result in a money illusion in the same manner in which they contribute to fiscal illusions.Footnote 29 Indeed, Baumol asserts that the quantities of many public services have already been unnecessarily cut in response to cost disease problems,Footnote 30 and recognizes that citizens and political elites may well be susceptible to a money illusion.Footnote 31
Second, the assumption that as the United States becomes wealthier it can afford to spend more on public goods and services may overlook a reluctance by all or even most voters to support devoting more of their wealth to the public sector. Inequality of incomes has grown in recent decades because most income gains have accrued to only the extremely wealthy.Footnote 32 Consequently, it is not at all clear that most voters will feel that they have the capacity to finance even a constant level of government goods and services at ever higher levels of taxes even if overall income is up. Thus, the Withering State Proposition—that government cost growth prompts a decrease in the share of real economic activity devoted to the public sector—remains plausible even if Baumol is correct in arguing recently that, at least at the societal level, “we can afford it all.”
More intriguingly, the possibilities for how cost disease bears on the size of the public sector may be more nuanced than the two competing hypotheses advanced by Baumol. Simply put, Baumol lumps almost all government activities together into the low-productivity sector of the economy.Footnote 33 This is certainly reasonable for domestic purchases by the public sector—e.g., services provided by teachers, prison guards, and police officers. But Baumol treats government transfers in the same manner. He writes, “Government welfare and related programs, which do not benefit from any significant source of productivity growth, are essentially handicraft activities whose technology remains fundamentally unchanged.”Footnote 34 This is, of course, correct with regard to spending to purchase the labor of public welfare officers who process transfer payments. But the bulk of transfer spending is not used to purchase such services. Rather, it is simply extracted from the private sector via taxes and returned to individuals to be spent in the private economy. Thus, actual transfers—public sector spending provided citizens for their own consumption—are not subject to the pitfalls of cost disease over and above how they bear on the economy on average since they are spent in that larger economy.
More to the point, government purchases and transfers may be, to some degree, substitutable. As discussed below, such substitution may be direct or indirect. But if this expectation is true, then the product substitutions that Baumol anticipates result from cost disease may not be from the public to the private sector, but instead entail government spending shifting from domestic purchases to transfers, thereby prompting an increase in the real level of transfers.Footnote 35 Importantly, this perspective is not inconsistent with Baumol's analysis. Indeed, Baumol discusses “hybrid industries” that rely on inputs from both productivity progressive and stagnant production processes.Footnote 36
Such substitution may come in two forms. The first and most obvious is a direct substitution of one provision mechanism for another. We noted above that transfers and purchases are in part substitutable. This is certainly true in many cases, such as the example of Veterans Hospital and Medicaid that we earlier cited. But in many other cases, such as the services provided by agencies like NASA or the Department of Justice—or simply building roads—this is doubtful. It is not at all clear what a substitutable transfer would entail in these cases. Thus, cost disease pressures are likely to continue to bear on many of the activities supported by domestic purchases, leading to either of our previous two propositions. But to the extent that some public goods and services are directly substitutable, we should see more reliance on transfers and less on direct production as Baumol's disease influences production choices.
Second, Baumol's disease may create incentives for a much more indirect type of substitution of transfers for direct production among entrepreneurial politicians. That is, elected officials need problems to solve for their constituents.Footnote 37 That is how they get reelected. In doing so, they are politically rewarded for spending and politically punished for taxing. Given the dilemma posed by this incentive system, Baumol's disease could alter the political calculus of deciding to address issues with transfer solutions or issues with solutions entailing direct government production of goods and services. It would do so by generating more “bang for the buck” over time from the former than the latter since the former is less influenced by Baumol's disease than the latter. Thus, political agendas may change due to Baumol's disease in such a way that transfers become a more prominent form of government spending. Given these two mechanisms, we can advance a third proposition involving both purchases and transfers, whereby cost pressures on the purchases or direct production side of the government budget ledger lead to its shrinking as spending is shifted to the transfers-side of the ledger.
Withering Purchases/Prospering Transfers Proposition: As the cost of government goods and services relative to costs in the overall economy (i.e., government's relative cost) increases, (i) the real level of government purchases declines, and (ii) the real level of government transfers increases.
The direct substitution of transfers for purchases may be only partial, however. It is also possible that there are real pressures on governments to grow in real terms even if the Surviving State Hypothesis of Baumol's later work is largely valid. Government's real share of economic activity need not remain constant. But even in the case of real government growth in such a surviving state condition in which Baumol's Cost Disease prevails, the cost of government purchases will increase relative to those of transfers. This could well create bias among public officials to rely on transfers to address pressures for new real growth rather than on real purchases spending. This leads to the following proposition.
Surviving Purchases/Prospering Transfers Proposition: As the cost of government goods and services relative to costs in the overall economy (i.e., government's relative cost) increases, (i) the real level of government purchases remains stagnant, and (ii) the real level of government transfers increases.
Designing an empirical test
We seek to test the propositions presented in the previous section about the effect of a government's relative cost on the real size of a government. An appropriate test is complicated by the existence of many alternative explanations for government growth. Put simply, a suitable empirical test must control for many variables that have been predicted to influence the real size of government. Our approach is to modify econometric models developed by Berry and Lowery—which include independent variables reflecting a variety of extant explanations for government growth—by incorporating a government's relative cost as an additional independent variable.Footnote 38
Berry and Lowery's models have several advantages over the other econometric models discussed and cited in the introduction.Footnote 39 The first concerns the dependent variable. Most empirical models of government growth have as dependent variables a measure of the total nominal size of a government. In contrast, Berry and Lowery's models use deflated indicators of the real size of a government as dependent variables, which is consistent with testing our propositions that pertain to the effect of government's relative cost on the real size of a government (or the real level of transfers or purchases).Footnote 40 Further, almost all prior empirical analyses of government growth employ government's total share of the economic pie as dependent variables. In contrast, Berry and Lowery disaggregate government spending into purchases and transfers, and develop separate models for each category. Their dependent variables are the real level of government spending on domestic purchases or on transfers, respectively, as a proportion of GDP. To be able to test our propositions, then, we too must employ disaggregated dependent variables. Thus, Berry and Lowery's models suggest that the dependent variables in our propositions should be the domestic purchases share of GDP in year t to (PURCHASES)Footnote 41 and the transfers share of GDP in year t (TRANSFERS t).Footnote 42
Turning to the independent variables, Berry and Lowery address pervasive multicollinearity by estimating separate models reflecting each of two, broad, accounts of the forces shaping government growth in the United States.Footnote 43 Their excessive government models (of both government transfers and purchases) suggest that forces within a government drive public sector growth, while their responsive government models (of both government transfers and purchases) assume that government growth arises from external pressures. The fact that Berry and Lowery specify and test “responsive” and “excessive” versions of each of their models (for both transfers and purchases) is important for us, since modifying their models by adding a government's relative cost as an independent variable allows us to determine whether our empirical results about the effect of a government's relative cost are robust to two quite different sets of control variables.
Berry and Lowery test several extant theories of government growth with four models.Footnote 44 Since detailed rationales for the selection of variables and their operational definitions can be found in Berry and Lowery's published paper, we offer only capsulated descriptions here.Footnote 45 The first, the responsive government model of domestic purchases contains variables reflecting two explanations of government growth. The first is Wagner's Law.Footnote 46 The core of Wagner's Law is the supposition that government spending is income elastic; as income increases, the demand for government spending increases even faster given that many of the amenities sought as a population becomes wealthier tend to be public goods such as clean air and good roads.Footnote 47 Another part of Wagner's interpretation concerns societal interdependencies that may demand public goods solutions in the form of domestic purchases.Footnote 48 In particular, the formation of new households in the population requires public spending on infrastructure,Footnote 49 and an increase in the population that is young requires spending on schools and then prisons,Footnote 50 all of which lead to increases in public purchases relative to the size of the economy as a whole. The second explanation captured in this model concerns the political party controlling the government in question.Footnote 51 When electorates choose Democrats, this can be viewed as demand for greater levels of domestic purchases.Footnote 52 In contrast, the election of Republicans indicates a demand for lower spending.Footnote 53 Thus, Berry and Lowery's responsive government model of domestic purchases is specified in the following equation.Footnote 54
$$\eqalign{& PURCHASES_t = b_0 + {\rm} b_1 INCOME_{t - 1} + {\rm} b_2 HOUSEHOLD_{t - 1}, {\rm} \cr & + B_3 YOUNG_{t - 1} \quad + {\rm} b_4 PARTY_{t - 1}} $$The second model, the excessive government model of domestic purchases, highlights forces internal to governments, especially the roles of self-interested bureaucrats and elected politicians. This model assumes that government employees tend to vote as a block for candidates supporting a larger public sector. Therefore, as public employees become a larger share of the electorate, more such candidates win office and—when elected—reward their supporters with an expansion of government.Footnote 55 This model also reflects the efforts of elected officials to expand purchases while avoiding the resulting blame by adopting fiscally illusionary tax systems. Systems that are highly illusionary rely heavily on withholding,Footnote 56 debt financing,Footnote 57 and a complex array of taxes,Footnote 58 thereby leading voters to think that they are receiving such good value for their money that they will demand ever more government purchases.Footnote 59 Finally, the model reflects competing hypotheses developed by Cameron and Marlow;Footnote 60 Cameron suggests that more centralized fiscal systems, as manipulated by self-serving politicians, are better able to restrain purchases, while Marlow postulates nearly the opposite.Footnote 61 The excessive government model of domestic purchases, then, is as follows.
$$\eqalign{&PURCHASES_t =b_0 + {\rm} b_1 CIVGOVEM_{t - 1} + {\rm} b_2 DEBT_{t - 1}, {\rm} + B_3 WHELD_{t - 1} \cr & + {\rm} b_4 COMPLEXITY_{t - 1} + {\rm} b_5 REVCEN_{t - 1} + {\rm} b_6 AID_{t - 1}} $$Model 3 is the responsive government model of transfers, which includes the INCOME variable from the responsive government model of domestic purchases, based on the expectation that if greater income equity is itself a public good,Footnote 62 citizens should demand higher relative levels of transfers as they become wealthier.Footnote 63 Governments controlled by Democrats are expected to increase relative transfers, while Republican governments are expected to hold the line on transfers.Footnote 64 The model also assumes that the share of the population receiving transfer benefits influences transfer spending.Footnote 65 Furthermore, it is assumed that the effect of share of the population receiving transfer benefits is greater when Democrats control the government.Footnote 66 Finally, the model reflects the role unemployment benefits plays as an automatic stabilizer. Thus, the responsive government model of transfers can be specified as the following.
$$\eqalign{& TRANSFERS_t = {\rm} b_0 + {\rm} b_1 INCOME_{t - 1} + {\rm} b_2 VETPOP_{t - 1}, {\rm} + B_3 OLDPOP_{t - 1} \cr&+ {\rm} b_4 POORPOP_{t - 1} + {\rm} b_5 PARTYNAT_{t - 1} + {\rm} b_6 DEMELEC_{t - 1} + {\rm} b_7 REPELEC_{t - 1} \cr& + {\rm} b_8 VETPOP_{t - 1} \, {^\ast} \, PARTYNAT_{t - 1} + {\rm} b_9 OLDPOP_{t - 1} \, {^\ast} \, PARTYNAT_{t - 1} \cr & + {\rm} b_{10} POORPOP_{t - 1} \, {^\ast} \, PARTYNAT_{t - 1} + {\rm} b_{11} UNEMPL_t} $$Last, and in contrast, the excessive government model of transfers emphasizes narrow self-interest on the part of government officials. The model assumes that elected officials, acting as political entrepreneurs, increase transfer benefits to groups with substantial voting power to try to attract their votes.Footnote 67 Moreover, politicians time the benefit increases for maximum electoral advantage by granting them during election years.Footnote 68 Also, the model includes the same set of variables in the excessive government model of domestic purchases designed to reflect political manipulation of fiscal illusions and the impact of fiscal centralization.Footnote 69 Therefore, the excessive government model of transfers is as follows.
$$\eqalign{& TRANSFERS_t = {\rm} b_0 + {\rm} b_1 VET:VOTE_{t - 1} + {\rm} b_2 POOR:VOTE_{t - 1}, {\rm} \cr &+ B_3 OLD:VOTE_{t - 1} + {\rm} b_4 FARM:VOTE_{t - 1} + {\rm} b_5 REVCEN_{t - 1} + {\rm} b_6 AID_{t - 1} \cr &+ {\rm} b_7 DEBT_{t - 1}, {\rm} + B_8 WHELD_{t - 1} + {\rm} b_9 COMPLEX_{t - 1} + {\rm} b_{10} NUMELEC_1 \cr & + {\rm} b_{11} VET:VOTE_{t - 1} \, {^\ast} \, ELECTION_{t - 1} + {\rm} b_{10} POOR:VOTE_{t - 1} \, {^\ast} \, ELECTION_t \cr & + {\rm} b_{13} OLD:VOT_{t - 1} \, {^\ast} \, ELECTION_t \; + {\rm} b_{14} FARM:VOT_{t - 1} \, {^\ast} \, ELECTION_t} $$Berry and Lowery's tests of these models for the period 1948–82 data found strong support for several aspects of the responsive government models and much less support for the excessive government models.Footnote 70 But their empirical results are now quite dated. As a byproduct of using Berry and Lowery's models as a base for testing our own propositions about the impact of government's relative cost on the real size of a government (or on the real levels of purchases and transfer), we will be able to replicate the estimation of Berry and Lowery's models over the longer period (1947–2012) for which we have data. In this replication, as reported largely in notes, we will also be able to take advantage of improvements in time series methods since their results were published.
To test our Baumol's Disease propositions, we add a measure of the cost of government goods and services relative to costs in the overall economy (government's relative cost) to each of the four models presented above. For this purpose, we use a variable constructed by Semat et al.Footnote 71 Their study reported strong support for the primary effect of Baumol's Disease on government costs—that the relative price of government goods and services increase over time largely due to productivity differentials between the public and private sectors. Their measure of government's relative cost is the ratio of the implicit price deflator (IPD) for government purchases to the implicit price deflator for GDP. This variable—to be denoted Government's Relative Price—measures the cost of a fixed “market basket” of government goods and service relative to the cost for a fixed basket of goods and service in the U.S. economy as a whole. As seen in figure 1, the value of Government's Relative Price has increased rather steadily from 1948 to 2010.
Figure 1: Government Cost Growth, 1948–2010
Now that we have operational definitions of the real level of government transfers (measured by TRANSFERS), the real level of government purchases (measured by PURCHASES), and the cost of government goods and services relative to costs in the overall economy (measured by Government's Relative Price), we can derive testable predictions from the four propositions introduced earlier:
Withering State HypothesisFootnote 72: As Government's Relative Price increases, both PURCHASES and TRANSFERS decline.
Surviving State HypothesisFootnote 73: As Government's Relative Price increases, both PURCHASES and TRANSFERS are unchanged.
Withering Purchases/Prospering Transfers Hypothesis: As Government's Relative Price increases, (i) PURCHASES declines and (ii) TRANSFERS increases.
Surviving Purchases/Prospering Transfers Hypothesis: As Government's Relative Price increases, (i) PURCHASES remain stagnant and (ii) TRANSFERS increases.
Testing the enhanced government growth models: Data and estimation issues
Given that the CITIBASE data set used by Berry and Lowery is no longer publicly available, we recreated each of their measures from the United States’ Bureau of Economic Analysis’ (BEA) National Income Product Account (NIPA) tables, with some supplemental data from the Statistical Abstract of the United States.Footnote 74 Annual observations were collected for the time period of 1947 to 2010, with the exception of models including Tax Complexity, for which disaggregated data was unavailable prior to 1959.Footnote 75 Inclusion of dummy variables did not indicate any significant difference in how the independent variables bear on the dependent variables between the period examined by Berry and Lowery and the expanded time period examined here.Footnote 76 Thus, we are confident that the replication portion of this analysis matches well with Berry and Lowery's.Footnote 77 The core dependent variables in the analysis are the same as Berry and Lowery's: the real share of GDP devoted to public sector domestic purchases (PURCHASES) and public sector transfers (TRANSFERS), the appropriate deflation of which has already been discussed.Footnote 78 As evident in figure 2, the use of different deflators for the numerators of the two measures has a significant impact on their temporal paths. There is a steady rise in the real share of GDP devoted to transfers, which is the same as its nominal share since both are deflated by the GDP IPD. And while the nominal GDP share of purchases is nearly flat across the whole time period, its real values indicate a sharp decline from approximately a third of GDP in the 1950s to roughly a fifth of GDP at the end. This difference in the real and nominal values of purchases GDP share is a result of employing different deflators for the two elements of the ratio measure of purchase share and clear evidence of the primary impact of Baumol's Cost Disease on government purchases. And the key independent variable for our extension of the original analysis is RELATIVE PRICE, the ratio of the IPD for government purchases to the IPD for GDP, as seen in figure 1.
Figure 2: Nondefense Public Sector Shares of GDP, 1948–2010
Two estimation problems were given considerable attention. First, time series methods have progressed significantly since Berry and Lowery's analysis. More specifically, we address the possibility of our results being excessively influenced by the persistence of our variables across time. Augmented Dickey-Fuller tests indicated that we could not reject the assumption of the presence of a unit root process within any of the raw dependent variables; differencing each series allowed us to reject this assumption. Independent variables have also been differenced in the models reported below; Engle-Granger tests for co-integration (using Schaffer (Reference Schaffer2010)) indicated that we could not assume our variables were co-integrated. Given these limitations in the data, our analysis necessarily focuses on short-term, year-to-year changes in our independent variables driving short-term, year-to-year changes in the growth of each variety of spending, assuming, in other words that the impact of Baumol's disease on real spending is almost immediate and does not accumulate over time. This, of course, is a very rigorous test of our hypotheses, a test that militates against our finding supportive results. However, our use of differenced variables also means that we rely far less upon autoregressive (AR) corrections than Berry and Lowery: Bausch-Godfrey tests indicated that the models show no evidence of serial correlation. Taken together, this analysis is more rigorous than the original study.
Second, as with the original study by Berry and Lowery, collinearity posed a significant concern for interpreting the results on our explanatory variables.Footnote 79 In order to account for this, we estimated alternative specifications of the models in which variables exhibiting a strong possibility of collinearity (as indicated by exceptional Variance Inflation Factor values), where one or another collinear variable was excluded in turn. Only models in which such exclusion led to a variable exhibiting statistical significance not evident in the fully specified model are reported here.
Results
The results for the responsive government domestic purchases model are presented in models 1 and 2 of table 1, with the second adding Relative Price to the initial specification. The replication results in the models are similar, but somewhat weaker for the responsive government model of domestic purchases than those reported by Berry and Lowery.Footnote 80 More important for our present purposes, there is no evidence that changes in government's relative costs are strongly related to year-to-year changes in GDP share of real domestic purchases. In model 2, the coefficient for RELATIVE PRICE—our measure of this concept—is not significant, and it is wrongly signed in any case in respect to the Withering State Hypothesis.Footnote 81 Thus, the steady decline in the real share of GDP devoted to public sector purchases over time observed in figure 2 cannot be attributed to the secondary impact of Baumol's cost disease. The evidence of a null effect is consistent with either the Surviving State Hypothesis or the Surviving Purchases/Prospering Transfers Hypothesis, each of which predicts that the expansion of societal wealth derived from more technologically progressive sectors of the economy provides sufficient revenue to maintain public goods purchases provision over time, despite the cost pressures arising from Baumol's Disease.Footnote 82
Table 1: Tests of the Responsive Purchases Model
Two Tailed Test: *p < 0.1, **p < 0.05, ***p < 0.001.
Values under coefficients are, first, standard errors and then t-values.
The models in table 2 test the excessive government domestic purchases model. Overall, as seen in the first column of the table, we find quite weak support for the excessive government model of domestic purchases, a pattern that is consistent with the findings reported in the original study.Footnote 83 More importantly for our present purposes, the inclusion of our measure of government's relative cost in model 2 of table 2 again yields evidence consistent with either the Surviving State Proposition or the Surviving Purchases/Prospering Transfers Proposition, echoing the results for the responsive government purchases model in table 1. That is, changes in the relative costs of producing public sector goods and services do not seem to be associated with changes in the share of GDP devoted to real domestic purchases.
Table 2: Tests of the Excessive Purchases Model
Two Tailed Test: *p < 0.1, **p < 0.05, ***p < 0.001.
Values under the coefficients are, first, standard errors and then t-values.
Some variables divided by constants for readability.
Next, we turn attention to our models of government transfers, starting with the results for the responsive government models, which are presented in table 3. Model 1 includes a neutral measure of presidential election accumulation, while model 2 contains separate measures for presidential elections won by Democrats and Republicans. In terms of the replication models, the results were somewhat weaker than those reported by Berry and Lowery, especially so in terms of the Wagner's Law income variable and party control of government.Footnote 84 But the results also generated consistent results for several other variables. Overall, then, these results provide some support for the responsive government interpretation of changes in the share of GDP devoted to real transfer spending.Footnote 85 When Relative Price is added (i.e., models 3 and 4), the results are similar to those originally reported by Berry and Lowery. But the inclusion of RELATIVE PRICE, our measure of the cost pressures associated with Baumol's disease, matters. Both RELATIVE PRICE coefficients are significant and positively signed. This suggests that changes in the costs of goods and services produced by the public sector are associated with increases in transfer spending as a share of GDP. This, of course, is in accordance with our Prospering Transfers Proposition, but inconsistent with both of Baumol's expectations: the Withering State Proposition and the Surviving State Proposition.
Table 3: Tests of the Responsive Transfers Model
Two Tailed Test: *p < 0.1, **p < 0.05, ***p < 0.001.
Values under the coefficients are, first, standard errors and then t-values.
Some variables divided by constants for readability.
Our results for the final model, the excessive government model of transfers, are presented in table 4. Model 1 presents the full replication,Footnote 86 for which none of the coefficients is statistically significant at even the 0.10 level, in part at least due to collinearity.Footnote 87 These results are consistent with Berry and Lowery's lack of empirical support for the excessive government transfers model.Footnote 88 But when we add Relative Price (see model 4), its coefficient is positive and statistically significant, indicating that cost pressures on purchases are associated with increased spending on transfers relative to GDP. The inclusion of RELATIVE PRICE, combined again with the exclusion of Debt in model 5,Footnote 89 also generated significant estimates for both Poor Vote Share and Taxes Withheld. The coefficient for Taxes Withheld becomes significant and negatively signed, in contrast with both the findings of the original study and the expectations of the excessive government transfers model. But Poor Vote-share, by contrast, does generate a significant and positively signed coefficient as would be expected by the excessive government model. In sum, then, our results provide modestly more support than in the original study for the excessive government model of transfers when RELATIVE PRICE is added to the model. More importantly, it provides strong evidence that cost pressures on purchases are associated with increased transfer shares of GDP, lending support to the Prospering Transfers Proposition, but inconsistent with the Withering State and the Surviving State Propositions. Indeed, taking into account the results of all of our tests—across tables 1, 2, 3, and 4—the only one of the four propositions considered in this paper consistent with all empirical evidence about the effects of government's relative cost is the Surviving Purchases/ Prospering Transfers Proposition. As the cost pressures associated with Baumol's disease rise (i.e., as the cost of government goods and services increases relative to costs in the overall economy), (i) the real level of government purchases is stagnant, while (ii) the real level of government transfers increases.
Table 4: Testing the Excessive Model for Transfers
Two Tailed Test: *p < 0.1, **p < 0.05, ***p < 0.001.
Values under the coefficients are, first, standard errors and then t-values.
Some variables divided by constants for readability.
Conclusion
With regard to the replication function of our analysis, the doubling of the time period and the use of more rigorous time series methods produced results largely consistent with those of Berry and Lowery for both the responsive government and excessive government transfers models—somewhat supportive of the former, but largely unsupportive of the latter.Footnote 90 There was a modest weakening of support in our results for the responsive government model in comparison to the earlier study. But this was certainly not counterbalanced by any discernible improvement in the excessive government models in our results compared to those of Berry and Lowery's original results.
More importantly for our purposes, our extension of the Berry and Lowery specifications was essential for more than purposes of replication.Footnote 91 That is, to validly assess the several competing hypotheses on the secondary impact of Baumol's Cost Disease on the real GDP share of public sector spending, we need to account for the other reasons the real levels of transfer and purchases spending might change. We have argued that Berry and Lowery's model provides the best available vehicles for doing this.Footnote 92 Thus, we added to the Berry and Lowery specifications a measure of cost disease pressures—RELATIVE PRICE—to assess how they influence the proportion of GDP devoted to domestic purchases and to transfers. Our results indicate, first, that the older, presumably wiser, Baumol was right is suggesting that changes in government purchases share of GDP should not be greatly influenced by cost disease pressures.Footnote 93 Neither of the RELATIVE PRICE coefficients in the responsive and excessive government models is significant.
Yet, our changes in real transfers findings suggest that neither the optimistic nor the pessimistic Baumol were correct in assessing how cost pressures on government purchases might influence relative transfer shares. Instead of a negative estimate, as hypothesized in Baumol's earlier work, or a nonsignificant estimate, as hypothesized in Baumol's later interpretation, RELATIVE PRICE generated a positive and significant estimate in both the excessive and responsive government models of the GDP share of transfers.Footnote 94 This is consistent with our—if not Baumol's—view of public spending as a hybrid production process in which cost disease pressures on the purchases side of the ledger create incentives to rely more on transfer spending, which is less subject to cost disease pressures, when public spending solutions of pressing policy problems are called for.
Taken together with the temporal trends on domestic purchases and transfers GDP shares in figure 2, our results suggest that long-term budget developments in the United States are dominated by two characteristics. The first is a holding of the line on the nominal purchases share of GDP, which, of course, entails a real reduction in the purchase share over time. Thus, while cost disease pressures may have played a role in capping the nominal government purchase as a share of GDP, thus leading to a decline in the real purchases share of GDP, they have not played a direct role in the year-to-year pace of declining real purchase shares. The causes of that decline lie elsewhere than Baumol's cost disease per se, although, as we have discussed, a money illusion may well play a role. But the secondary consequences of Baumol's cost disease do matter more directly for the transfer shares of GDP. Our results suggest that over time, as cost disease pressures intensified on the purchases side of the ledger, government has increasingly relied on transfer spending. In short, cost disease pressures on purchases have created a growing bias toward greater reliance on transfers, a bias that has grown with the increasing impact of Baumol's disease on purchases over time.
This is important in that Paul Krugman, Ezra Klein, and Mark Thoma have noted that one of the key budgetary trends in the United States is the government's steady progression toward becoming what amounts to an insurance company with an army attached.Footnote 95 Our results in the transfers share models perhaps account for one reason why this is so. Compared to domestic purchases, transfers are less directly plagued by cost disease pressures. Given the accumulation of cost pressures on purchases, this should lead over time, to increased reliance on transfers and, thus, a shift in the balance of purchases and transfers in the budget. The good news from the perspective of the more optimistic Baumol is that this substitution, along with greater societal wealth generated by more technologically progressive sectors of the economy, provides another reason why Baumol's disease need not be viewed as leading to an inevitable withering of the state. Indeed, our results almost certainly underestimate the power of such substitution in preserving the scale and scope of the public sector since we considered only spending.Footnote 96 Tax expenditures too should be viewed as transfers in lieu of using direct purchases as policy solutions.Footnote 97
This does not mean, of course, that the substitution of transfers for purchases offers an ideal solution to Baumol's disease in the public sector. First, we have noted earlier that in some and perhaps many cases, direct production of goods and services may not be readily substitutable. Second, while we found that cost disease pressures on purchases do seem to stimulate greater relative transfer spending, they do not, consistent with the older Baumol, seem to diminish domestic purchases per se.Footnote 98 That is, across the purchases and transfers models, we found strong support for the Surviving Purchases/Prospering Transfers Proposition. Thus, the substitution of transfers for purchases that has occurred represents an addition to the public budget. To the extent that the real size of the overall budget is a problem, and it certainly is to those on the right, the substitution pattern we have observed increases the overall size of government. And third, it is not clear all substitutable transfers actually avoid Baumol's disease. In many cases, they are likely to do so. In programs such as general welfare, Social Security, and nutrition support, transfers entail spending across the array of goods and services provided by the private economy. Thus, they should bear cost disease pressures in the same manner as the general economy, which is why we deflated the numerator of transfers share by the IPD for GDP. But for other types of transfers, including Medicaid and school vouchers, such transfers merely shift spending from a technologically stagnant area of the public sector to technologically stagnant private production. In such cases, the cost pressures associated with Baumol's disease are not likely obviated. They are merely relocated. In sum, the pattern of substitution of transfers for domestic purchases may help alleviate the impact of Baumol's disease, but they will almost certainly not eliminate its full impact.
Appendix A: Supplemental replications of Berry and Lowery 1987
Table A1 Results of replication of Berry and Lowery 1987 models for capital and government employment
Appendix B: Description of variables
Table A2 Summary of variables used in models
