Published by EH.NET (June 2004)
Peter H. Lindert, Growing Public: Social Spending and Economic Growth since the Eighteenth Century, Volume 2: Further Evidence. New York: Cambridge University Press, 2004. v + 230 pp. $70 (cloth), ISBN: 0-521-82175-4.
Reviewed for EH.NET by Robert A. Margo, Department of Economics, Vanderbilt University.
Earlier this year I reviewed Volume 1 of Peter Lindert’s new book, Growing Public: Social Spending and Economic Growth since the Eighteenth Century, for EH.Net. This is a book about why countries spend more today on public schools, welfare, public pensions, and redistribution in general than in the past. Volume 1’s intended readership is the general public and thus technical apparatus is kept at a bare minimum. For social scientists eager to see how the sausages (data) are made or who otherwise prefer first order conditions and regression equations to prose, there is Volume 2. This is a (self-contained) review of Volume 2 but to fully understand what is going on, the reader should refer to my earlier review of Volume 1. Lindert is Distinguished Professor of Economics at the University of California, Davis, and a Research Associate of the National Bureau of Economic Research.
Volume 2 is divided into a preface, seven substantive chapters, and seven appendices. The chapter numbers begin with thirteen. (Volume 1 ends with chapter twelve.) Chapter 13 presents the theoretical framework that underlies the analysis in Volume 1. Chapters 14-18 present the regressions that correspond with the analogous chapters in Volume 1. Chapter 19, “Reconciling Unemployment and Growth in the OECD” (with Gayle J. Allard) appears to be stand-alone (although clearly related).
Just as it is impossible to figure out fully what is going in Fogel and Engerman from reading just Volume 2 of Time on the Cross, one cannot fully appreciate Lindert from reading just Volume 2 of Growing Public. Busy social scientists can get the gist, however, by reading Chapters 13, 14, and one of 15-17 (probably 15 is best), and maybe 18 (the effects of social spending on growth).
In the long run, the biggest contribution that economic history makes to scholarship is the production of fresh data. Whether or not one agrees with Lindert’s analysis, one can thank him for (and use) his data. These data consist of three country-level panels. The first pertains, approximately, to the 1880-1930 period; the second two, approximately, to the post-1960 period. The principal data are available on-line (at Lindert’s website at the University of California at Davis). Some of the historical schooling data are printed in the appendices, along with detailed source notes for them — not enough (in my opinion) to literally reproduce each figure with ease, but more than enough for those who wish to probe deeper into the sausage making.
Chapter 13, as just noted, presents the theoretical framework, a version of Gary Becker’s well-known “pressure group” model. According to this model, individuals have a utility function defined over their own consumption, and possibly that of two other groups, one of which is to receive a positive transfer from the government and the other is to be taxed. These “caring” coefficients are positive; that is, the individual values positively the transfer to the subsidized group but also has her utility reduced from the fact that the taxed group has to finance the transfer. Utility is also a negative function of the deadweight losses occurring because of the transfer. Individuals can undertake actions (spend money) to favor or oppose the transfer, or possibly do nothing. The government in this framework is a passive actor; that is, there is an assumed function that maps the value of the transfer into the aggregate amounts spent in favor or in opposition. Under certain conditions, a “pure strategies” Nash equilibrium will exist.
This simple framework yields predictions that, broadly speaking, conform with the historical evidence on social spending (for example, if individuals are very poor they will likely place a high value on their own consumption, and choose to not participate rather than lobby in favor or oppose a policy). On the other hand, the model is not well suited, in my opinion, for understanding changes over time in voting rights. Preferences are also taken as given, although there is ample evidence from the historical record that whether or not a polity views “outsiders” as worthy of support (or derision) can be manipulated by politicians.
The chapters on social spending (15-17) all share more or less the same basic regression specification (18 is also similar in this regard). Rather than review each separately, I think it is more helpful to elucidate the specification, focusing on the implementation in Chapter 15. Although the criticisms made below are specific to Chapter 15 some of them are also relevant, albeit in different form, to the other chapters.
The basic specification consists of two structural equations:
Social spending = a + b*voting rights + c*per capita income + d*growth in per capita income over the previous ten years + other X’s + error
Growth = e + f*social spending + other Z’s + error
Growth and per capita income are endogenous in the social spending equation, while social spending is endogenous in the growth regression.
“Voting rights” are measured by dummy variables for autocracies and whether or not women voted in the recent past; a cubic polynomial in percent enfranchised; and turnover of the chief executive. The autocrat dummy and the percent enfranchised are predicted values from a first stage regression. In some cases the actual equations estimated depart from this basic specification by leaving out one or more right hand side variables. For example, some equations include a full set of country and time period dummies, while others do not.
With regard to the rise of social spending Lindert’s key point is that voting rights matter (see my review of Volume 1). As voting rights are extended into the population, social spending rises, but it does not do so in a nice, linear fashion. Lindert attempts to capture this non-linearity with the cubic polynomial and the dummy variables. Using the regressions coefficients, Lindert presents (many) computations of “effects” — these are simulations as to what happens to a particular type of social spending if voting rights are extended. These computations clearly show that voting matters — for example, the public school enrollment clearly increases as the franchise is extended from 30 to 80 percent of the adult population (p. 36).
As with all econometrics a key issue is identification. What is Lindert’s identification strategy? In a phrase: arbitrary exclusion restrictions. The general strategy is diagrammed in Figure 14.1 (the specific implementation differs across time periods because of data availability). Per capita income depends on social spending and vice versa. Per capita income also depends on “accumulated prior capital,” a lagged Gershenkron-like “catch-up” variable in per capita income, various “global” variables, plus fixed country and time effects. Social spending also depends on demographics, religion, openness, “social affinity” (ethnic fractionalization), and the voting regime. “Prior accumulated human capital” does not appear in the spending equation, nor does the voting rights regime or religion appear in the growth regression. However, models of the intergenerational transmission of human capital would seem to suggest prior accumulations matter in social spending. Excluding voting rights from the growth equation (see Appendix Table D2) implies that autocracy does not affect growth. Recent work by Robert Barro, among others, suggests that religion should not be excluded from the growth regression.
The first stage regressions of the voting rights variables, as reported for the 1880-1930 period, are problematic. These equations, shown in Table 15.3, are regressions of the autocrat dummy and the percent enfranchised on lags of these variables plus lagged per capita income, lagged school enrollments, lagged percent urban, a variable measuring the growth of the global economy plus two dummies for whether the country recently lost a war. There are no country fixed effects in these regressions.
The use of lagged values of voting rights as instruments is not very convincing if there is any serial correlation in social spending. The issue is not whether past values of voting rights predict current values — this is not surprising, particularly when there are no country fixed effects in the regression — but whether transitions from one regime to another can be viewed as credibly exogenous to the process that drives social spending. It is difficult to imagine why the percent urban in the recent past affects voting rights but not social spending or growth. Wars sometimes occur for unpredictable reasons. Autocrats (and democracies) need soldiers (and others) to fight wars. When a war is won the victor might extend political rights to the previously disenfranchised out of gratitude. The losing side might be forced to extend the franchise as a condition of surrender. So war is a candidate instrument for voting rights — and it “works,” in the sense that the two dummies are significant (positive) predictors of percent enfranchised (but not the autocrat dummy — nothing matters here, except for lagged values of the voting regime). However, it is easy to imagine that war might affect social spending or growth in a direct manner, or indirectly through subtle changes in demographics that are not adequately captured by the controls Lindert includes.
I am not saying we should be dismissive of Lindert’s conclusions merely because of quibbles with the econometrics. Personally I believe that voting rights matter for the rise of social spending, and I also believe that any deleterious effects of social spending on growth have been greatly exaggerated. However, some (technically obsessed) readers, unaware of the pain and suffering necessary to produce data like those analyzed in this book, might be less forgiving out of hand and that is unfortunate. I wish the presentation in Volume 2 were clearer and the justification of the identification strategy more convincing. I also wish the implementation of the econometrics was more straightforward. (For example, I would have preferred starting with a “reduced form” social spending regression on just the voting rights regime, to which is then added county and time dummies, and maybe country-specific linear time trends. If voting rights survive this stringent an econometric test, we can take notice. If it does survive we could add more variables or try some instruments for voting rights, consider adding growth as an endogenous variable, and so on).
My review of Volume 1 concluded that it was important and deserved a wide readership. Although I am troubled by some of the econometrics in Volume 2, that does not change my assessment of Volume 1. The data are a fundamental and lasting contribution. Regardless of what one thinks of the specifics of the empirical analyses, there is surely much to the story. In any case, because the data are available everyone can run their own regressions.
Robert A. Margo is Professor of Economics and of History at Vanderbilt University, and Research Associate, National Bureau of Economic Research. He is the editor of Explorations in Economic History.