Econometrics of Income Distribution: Toward More Comprehensive Specification of Institutional Correlates.

Durham, J. Benson

Journal of Economic Literature Classification Numbers: B40, C50, D33

1. Introduction

Many social scientists argue that certain economic and political "institutions" affect economic outcomes. Perhaps following North and Thomas (1973), the most developed literature in this genre focuses on the effect of institutions on growth and investment as an extension of or an alternative to the neoclassical model. For example, econometric and qualitative studies evaluate the relative economic efficiencies of dictatorships and democracies (Barro, 1996; Przeworski and Limongi, 1993; Durham, 1998 and 1999), labor market centralization (Garrett and Lange, 1985 and 1986; Alvarez, Garret and Lange, 1991; Hicks, 1994; and Headey, 1970) and fiscal federalism (Weingast, 1993).

Unfortunately, fewer theoretical and empirical studies address the relationship between these institutional variables and income distribution. While the literature and econometrics on regime-type and income equality is comparatively extensive (Sirowy and Inkeles, 1990), there is a dearth of evidence regarding other institutions that quite conceivably affect income distribution, particularly labor market organization and fiscal federalism. Put differently, previous studies suffer from specification bias with respect to institutions. Such error, as Bollen and Jackman suggest in a previous institutional inquiry, "not only misrepresents the relationship directly involved but also biases the remaining coefficients that are being estimated (1985, p. 441)." In short, given Deininger and Squire's (1996) recent advances in data collection, this paper endeavors toward a more comprehensive specification using both cross-sectional/time-series (panel) and time-series econometric designs.

Section 2 outlines the arguments regarding regime-type, and Section 3 discusses labor market organization and fiscal federalism. Section 4 reviews common economic determinants, and Section 5 comments on the data and empirical design. Section 6 presents the econometric results from the panel design as well as time-series equations for Taiwan, the United Kingdom and the United States. Section 7 concludes.

2. Regime-Type

Economists and political scientists alike debate the relative economic efficiencies of regime-types. The most noted inquiry addresses whether democracies or dictatorships better promote growth and investment. Very briefly, the primary causal concept in this literature concerns the relative utility of executive discretion or freedom of action. According to the authoritarian view, dictatorships better govern economies because they are not susceptible to the "political business cycle"--democracies must spend scarce resources to win elections. The democratic view contrarily argues that democracies solve the veritable "credible commitment problem." Democratic procedures check arbitrary (and kleptocratic) sovereigns and thereby secure property rights.(1)

2.1. Theory and Evidence on Regime-Type and Income Distribution

With respect to income distribution, while there are some dissenters, a majority of scholars argue that democracies are more sensitive to pressures among the polity to redress unequal distribution of resources. Similar to the debate on capital allocation and accumulation, the causal mechanism similarly addresses discretion or freedom of action, which is notably a continuous and not a dichotomous variable. Democracies are vulnerable because they must win votes to lengthen tenure, and therefore vote maximizers redistribute scarce resources to the majority. This rationale closely relates to the reformist objective to extend the democratic franchise beyond property owners--conservatives oppose extending political rights to protect privileged economic: groups. Due to their (comparative) inclination toward majoritarianism and solicitation of the mean voter, democracies more likely represent lower income groups than dictatorships.

Moreover, authoritarian regimes, presumably and particularly those buttressed by the wealthy, are more insulated from popular egalitarian demands and thus less apt to redistribute income. In fact, rather than focus on the majoritarian nature of democracy, which is quite objectionable given the routine occurrence of coalition governments (Bollen and Jackman, 1985, p. 440), another key facet of the democratic view is that particular dictatorships exacerbate material inequalities. Perhaps Marx's The Eighteenth Brumaire of Louis Bonaparte and classical Marxists interpretations of "fascism" are seminal exemplars. Autocrats possess the potential to pursue policies that suppress a majority in favor of a select minority, because no political mechanism holds them accountable. Scholars associated with "dependency theory," for example, suggest that dictators collaborate with multi-national corporations to develop industrial enclaves bereft of satisfactory backward linkages (Evans, 1979). Bollen and Jackman cite Brazil in the 1960s, Chile under Pinochet, Iran under the Shah, as well as more traditional monarchies that purportedly fit such a pattern, including Nepal and Saudi Arabia (1985, p. 439).

The dissenting view argues that authoritarian regimes more ably redistribute wealth because material inequalities worsen political inequalities in democracies. As Beitz (1982) argues, democratic institutions do not adequately represent the poor, and therefore the system does not effectively redistribute resources. Less affluent citizens cannot take advantage of electoral mechanisms, he reasons, as economic inequalities reproduce and exacerbate political inequalities. Poorer groups lack either class consciousness or effective organizational representation. Close to an agnostic perspective, while they do not necessarily endorse authoritarian regimes, some Marxist versions argue that the political system is irrelevant because it simply reflects the interests of the capitalist class. Thus, democracies per se scarcely address inequalities. As Jackman and Bollen (1985) note, a more benign view of capitalist development argues that technology shapes the distribution of material and political resources simultaneously, which suggests that any empirical regularity between regime-type and income distribution is spurious.

Turning to empirical evidence, Perotti (1996) finds some data to support the prevailing democratic view. Democracies in general have more equal distributions of income, which implies that polyarchies redress income inequality via the "political mechanism." But another finding besets this conclusion--the "income effect." That is, more developed countries also have more equal distributions of income, at least according to the imposed functional form of the level of development. The most advanced countries, as widely noted, are democracies, and therefore Perotti (1996), Sirowy and Inkeles (1990, pp. 134-36) and others cannot distinguish the democracy and income effects on income distribution. For example, Perotti finds that "if one splits the sample of countries into rich and poor, the results mimic those obtained when splitting the sample into democracies and non-democracies (p. 20)." This largely conforms to the "technological perspective" and the "skeptical model" that Sirowy and Inkeles characterize, as "any association between democracy and socioeconomic equality is spurious since both are determined by the level of a nation's economic development (1990, p. 136)."

But these entangled effects are perhaps due to economists' and other social scientists' near ubiquitous treatment of regime-type as a dummy variable. Simply, the dichotomous treatment of regimes is too narrow, as subtypes of democracy and dictatorship quite conceivably better redistribute resources, as the previous discussion of particular authoritarian forms suggests. In addition, the critical underlying concept of discretion--or insulation from pressures for more equitable income distribution in this case--is a continuous and not a dichotomous variable. Sovereigns do not simply lack or possess freedom of action. Rather, there is a continuum across space and time.

2.2 Regime-Type Measures

The following econometric analysis alternatively considers two types of regime-type proxies. The first refers to commonly used ex post outcome proxies that endeavor to capture myriad political and social aspects of "democracy," a concept with several meanings. The second regards ex ante institutional measures that isolate subtypes of dictatorships and democracies that facilitate or inhibit income equality. Rather than endorse any particular proxy, the primary aim of this paper is; to include different outcome and institutional measures to enhance the sensitivity analysis.

2.2.1. Ex Post Measures: The Freedom House and Gastil Indicators

Among outcome proxies, the Polity III measures employ nine indicators of "institutionalized authority characteristics." These indicators, in turn, endeavor to capture three concepts: "the influence dimensions of authority," "the recruitment of chief executives" and "government structure (Jaggers and Gurr, p. 470)." Another measure in this genre, perhaps most popular, is Gastil's dual indices of "political" and "civil" rights from Freedom House. Political rights refer to the "capacity of individuals to participate freely and effectively in the selection of policy-makers and binding decision rules affecting the national, regional and local community." Civil rights "are defined by the freedom of individuals to develop views, institutions and personal autonomy apart from the state (Jaggers and Gurr, p. 474)."

Again, while economists routinely construct dummy transformations to capture the effect of regime-types, commentary on the ordinal forms of these indices is instructive(2). Regressions that include ordinal expressions capture the degrees of executive discretion, but scholars should interpret unit changes in both variables carefully. As Clague et al. note, the indices capture outcomes of processes rather than "defining features or requisites ora democratic political system (1996, p. 251)." Few comparative economists or political scientists have an intuitive understanding of ordinal differences across cases according: to either scale. More importantly, policy entrepreneurs cannot replicate such outcomes, but they can purport to craft procedural institutions that distribute resources more effectively.

2.2.2. Toward an Ex Ante Measure: The Effective Party/ Constitutional Framework Proxy

The Gastil and Polity III indicators are subjective outcome proxies that serve a variety of research objectives. As an alternative rather than a substitute, a continuous, objectively measurable index should substantively distinguish democratic and authoritarian regime subtypes. The focus of such an alternative is on institutions; rather than outcomes, which thereby addresses the implicit serviceable concern regarding whether constitutional framers can craft regimes more disposed to income equality. In other words, the quantitative inputs to the proxy more directly refer to constitutional choices.

Briefly(3), the institutional measure incorporates both the number of "effective" political(4) parties and the constitutional framework, which refers to the relative position of the executive vis-a-vis legislative branches of government(5). The fewer parties in government, the more discretion the regime has in policy-making. Importantly, the measure focuses on institutions and constitutional crafting because electoral laws, from the very stipulation of the existence of elections to combinations of "first-past-the-post" and proportional representation systems, have a direct effect on the number of parties(6). Moreover, framers directly stipulate the constitutional framework, a real choice among "parliamentary," "presidential," "authoritarian," and combinations thereof.

In addition to the effective party/constitutional frameworks measure, dummy variables that follow Linz and Stepan and Skach (1993) also explore that possibility of empirical regularities between income distribution and certain subtypes of democracy. While these dummy variables do not capture a continuum of discretion, this alternative nevertheless addresses the purported "perils" of presidential democracy (Mainwaring, 1993).

To briefly summarize the focus of these institutional proxies, the effective party/constitutional framework measure distinguishes both authoritarian and democratic subtypes and purports to capture the "political mechanism" through which governments process popular demands for income distribution. Among dictatorships, given that ideal parties "mobilize" or "channel" political sentiment from the polity(7), then the more embedded party would be more responsive to popular demands for redistribution(8). Thus, exclusively considering dictatorships, policy-makers in one-party regimes enjoy less discretion than their counterparts in party-less regimes. This measurement addresses Barro's (1996) recent observation that there is no econometric distinction among dictatorships. Moreover, the party distinction isolates dictatorships most disposed to perpetuating inequalities. Among democracies, the effective party/constitutional framework measure and dummy variables for parliamentary and non-parliamentary systems measure distinguish pluralistic regimes where the Polity III and Gastil measures do not. Therefore, given potential collinearity between the level of development and democracy, the measure enables tests of whether certain subtypes of pluralistic regimes more effectively promote equality.

A possible shortcoming of the ex ante proxy might concern internal party discipline with respect to dictatorships. Importantly, by definition parties do not constrain policy-makers from without in single-party systems. However, the single party might more likely contain heterogeneous sects or factions, thereby making within party discretion less acute compared to multiparty systems in which equally disparate factions might simply form an additional party. In short, this qualification might imply a "blip" in the interval between one and additional parties in the continuous proxy. On the other hand, democratic polities perhaps influence participating parties to adopt more pluralistic self-governance. Consider the opposition's common accusation in contemporary British politics, for example, that Blair runs (New) Labour like Lenin or even Stalin governed the Communist Party. After all, a literally, bloody "night of the long knives" seems far more likely to occur in a dictatorship compared to a democracy, which would ultimately suggest considerable within party discretion in authoritarian regimes. Unfortunately, either way, the continuous effective party/constitutional framework measure does not capture this possibly problematic distinction. But again, inclusion of the ex ante proxy in addition to the outcome measures does not necessarily imply an endorsement of the former but a more thorough sensitivity analysis of the (purported) effect of political regimes.

2.2.3. Alternative Functional Forms

Whatever the particular merits of each proxy, the continuous nature of policy-maker discretion suggests three hypotheses using each measure. First, the relation between regime-type and income distribution might be linear, which perhaps becomes less intuitive after certain values in the independent variable. One might question whether there is a constant slope effect after some critical number. Thus, failure to accommodate diminishing marginal returns recommends a linear-log form that tests whether the beneficial effect on income distribution becomes less pronounced, or "tails off" after some general threshold. Finally, the quadratic or U-shaped form tests whether there is an optimal ordinal value.

3. Additional "Institutions": Labor Markets and Fiscal Federalism

In addition to a more comprehensive assessment of regime-type proxies, another problem with previous studies is the omission of other related factors. "Regime" ultimately refers to the methods of access to the state, which in turn quite conceivably relates to other institutions that affect income equality. Unfortunately, the literature does not test for these institutional variables, which broadly describe the organization of the economy, and is therefore susceptible to specification bias.

3.1. Labor and Partisan Ideology: The Social Democratic Corporatist Model

For example, the organization of labor unions in a given country quite conceivably influences income distribution. After all, more equitable distribution of scarce resources is the raison d'etre of trade unions. Also, commentators commonly associate "left government partisanship," which describes the ideological composition of the state, with more equitable distribution of income among the working classes.

Quantitative literature on labor relations and the partisan characteristics of developed democracies, "the social democratic corporatist model (SDC) of national economic performance (Hicks, 1994, p. 190)," is abundant (Garrett and Lange, 1985 and 1986; Alvarez, Garret and Lange, 1991; Hicks, 1994; and Headey, 1970). Briefly, rather than income distribution, this perspective largely addresses growth and investment--strong labor organization and centralization leads to higher capital accumulation and growth, provided such labor structures attend "Left" partisan government tenure.

The SDC perspective focuses on two concepts. The first is how to characterize the labor market. Garrett and Lange argue that "centralized and dense unions ... appear conducive to the development of concertation between unions, employers' associations and government." Regulation of union behavior in the market both "promotes and is promoted by" a "virtuous circle" of economic policy (1985, p. 793). In this cycle, labor must realize the potential gains from wage restraint and not disrupt the flow of investment.

Partisan composition of government is the second consideration, as the interaction between union organization and government policy affects macroeconomic performance. In contrast to, say, a scenario with strong labor organization and Left control of government, countries with relatively weak labor movements combined with governments of the "Right" can still "pursue their partisan preferred macroeconomic strategies and achieve similarly beneficial macroeconomic outcomes (Alvarez et al., 1991, p. 539)." Crafts similarly suggests that "it must also be recognized that the effects of industrial relations structures are contingent on political circumstances (1992, p. 25)." In less compatible scenarios, trade unions less willingly show wage restrain and are more apt to behave militantly given a confrontational partisan government. An interaction term captures this dynamic, as the expression, momentarily ignoring other regressors, is

(1) Y = [[Beta].sub.0] + [[Beta].sub.c]C + [[Beta].sub.p]P + [[Beta].sub.c,p](C x P) + [Mu]

where Y is some macroeconomic performance variable, C is the union centralization index, P is the "Left" partisanship measure, and [b.sub.c,p] captures the interaction. Alternatively, Garrett (1995) adds C and P to capture "left labor power," with no interaction term, as in

(2) Y = [[Beta].sub.0] + [[Beta]c+p](C + P)+ [Mu].

Whether in (1) or (2), dense labor organization and Left partisanship should translate into more equitable income distribution, regardless of the regime-type mechanism, because unions help transfer wealth from owners to workers, and Left governments typically purport to represent lower income groups(9). The interaction term, however, more likely has a less pronounced impact, as conventional wisdom would suggest that governments of the Right and weak labor organizations should lead to greater inequality, regardless of the purported benevolent effect on growth and investment(10).

3.2. Fiscal Federalism

A third institutional consideration besides regime-type and labor organization regards "fiscal federalism," which ultimately refers to the abstract concept of multiple "sovereignty" among central and local governments. Qualitative research on this variable largely concerns economic expansion, but the organization of local and central governments quite conceivably affects income distribution. A very brief review of the economic performace argument is instructive.

Weingast claims that "market preserving" federalist organization of the state promotes economic growth. Departing from the "credible commitment problem," Weingast's "fundamental political dilemma of an economic system" is that "a government strong enough to protect property rights is also strong enough to confiscate the wealth of its citizens (1993, p. 287)." Democratic government is not enough to assuage sovereign temptation to renege on property rights. Federalism more thoroughly induces such constraint. The "decentralization of political power" alleviates the credible commitment problem because "no government or jurisdiction has a monopoly over regulation of the market, (which) induces political competition among jurisdictions, competition that limits the ability of any one state to impose debilitating regulation." In his empirical investigation, Weingast argues that federalism enhanced the economic development of(quite curiously) Britain(11) and the United States, and more recent work with Qian applies the concept to contemporary China vis-a-vis former communist Eastern Europe (1997, p. 86).

Returning to this research question, which Weingast does not address, the effect of federalism on income distribution primarily addresses territorial inequalities. In many cases, regional income disparities exist within countries. For example, the southern regions of Italy and Spain are considerably less affluent than northern areas. With respect to theory, dependency theorists argue that "enclaves" of capitalist development naturally emerge in less developed nations. Also, the widely observed discrepancy between urban and rural areas in less developed areas, germane to Kuznet's trajectory hypothesis as discussed in the following section, necessarily implies regional considerations. Simply, a higher degree of fiscal federalism implies that these lesser developed regions by definition have less resources with which to redress, inequalities given the (poor) local tax base. Less centralization suggests that national governments transfer fewer resources across regions. Under these assumptions, federalism has a negative effect on income distribution.

The following equations employ two proxies that measure two different aspects of fiscal federalism. With respect to the first measure, as Lijphart (1984, p. 177-78), Oates (1972, p. 18), Casella and Frey (1992, p. 643), and Kee (1977, p. 80) suggest, the most valid approximations of fiscal federalism include the fractions of total revenues collected and total expenditures allocated by' the national government. Given these hypotheses, the revenue federalism measure should have a negative impact on income distribution, as decentralized national governments do not have the resources to address regional inequalities by transferring wealth from affluent to poorer areas.

The second measure of fiscal federalism regards the mismatch between spending and taxation levels. In some cases, spending is decentralized while taxation remains a responsibility' of central government. According to Garrett, such gaps between spending and revenue encourage fiscal irresponsibility and therefore poor macroeconomic performance. But this phenomenon should have a positive impact on income distribution. A large gap between local spending and revenue suggests that local governments spend fewer indigenous resources, which increases the possibilities for significant resources transfers across regions. The proxy is local spending minus; local revenues, divided by total government revenues(12), as in

(3) [G.sub.federal] = [E.sub.l] - [R.sub.l]/[R.sub.l] + [R.sub.c]

where [E.sub.l] is local expenditure, [R.sub.l] is local revenue, [R.sub.c] is central revenue, and [G.sub.federal] is thus the federal spending gap(13).

4. Economic Determinants

To summarize the discussion of institutional variables, the dual effects of income and democracy on income distribution bedevil economists who only employ dichotomous regime variables. Furthermore, inclusion of other variables that describe the organization of the economy explores how the broader institutional framework might affect income distribution. These tests more effectively specify the mechanisms through which institutions affect income distribution and advance a more comprehensive model relevant to normative economists.

But institutional variables are not the only determinants of income distribution. Similar to econometric studies of economic growth (Levine and Renelt, 1992), there is no universal consensus or "true model" of empirical correlates of income distribution. This specification includes regional dummies, a quadratic functional form of the initial per capita level of GDP the government spending ratio, population growth, and openness to international trade(14). Each factor should be described briefly in turn.

The regional variables refer to the observation that Latin American and African countries, which have also grown more slowly, have less equal distributions of income than other areas. Similar to Barro's (1996) inclusion of regional variables in growth regressions, such specification suggests that these dummy variables have some effect irrespective of the other determinants, but the precise causal process is ambiguous.

The quadratic form of the initial level of national income is a proxy for Kuznets' argument about the effect of urbanization. According to Kuznets, urbanization purportedly effects income distribution because urban areas are "characterized by more inequality relative to rural areas, but also by higher percapita income." Perotti explains that the combination of these effects implies that "urbanization increases inequality in lower levels of development, but inequality decreases as countries continue to grow (1996, p. 14)." While "urbanization" is a particular component of "modernization," to clearly distinguish the concept from the general level of economic development is difficult, and both Kuznets' and Perotti's diction in fact seems to confound the two concepts. Therefore, along with the poor temporal coverage of the World Bank figures for urbanization, which is suitable for Perotti's (1996) pure crosssectional design, the quadratic expression of initial level of GDP is a reasonable proxy given the unclear distinction. Again, the quadratic form, which Bollen and Jackman (1985, p. 441) employ, suggests a negative effect on equality as income increases from comparatively lower levels and a positive effect as income increases from higher levels.

Another control variable is the share of government spending to total national income. This factor purports to capture the extent to which government supply responds to subject population demand. The hypothesis is that a government more involved in the economy generally is more likely to specifically pursue income distribution through fiscal policy. An alternative approach is to include population demographics, such as the share of the population over 65 years of age. Inequality and income, the argument reasons, are lower among retirees, and the demand for social security increases as the population ages (Perotti, 1996). But to include such demographic figures assumes that governments in fact respond to such demands for redistribution. Insensitive to that assumption, the government ratio considers the actual supply of government involvement.

The specification in this paper also includes population growth, but not in the context of popular demand for government policy. Briefly, increased birth rates increase inequality because different income groups grow at different rates, as poorer segments of the population increase more quickly, thereby increasing the proportion of lower income groups. As Bollen and Jackman summarize, another facet of this consideration is that high population growth generally by definition accompanies higher fertility and lower mortality, which produces a younger population age structure among low-income groups (1985, p. 443). This second aspect complements Perotti's suggestion that income distribution is more even among older populations, but again, the emphasis is not on the demand side.

Finally, this specification of income distribution includes openness to international trade. Most commonly, the factor addresses the notion that free trade puts downward pressure on wages for lower skilled workers. But, this variable is also germane to the debate in comparative political economy about international influences and policy-making. As Gourevitch (1978) suggests, international politics affects domestic political structures, and the system has unit-level consequences--world politics affects internal policy choices. Briefly, in addition to the standard argument about wages, countries that are more open to trade and the international economy enjoy less policy discretion ceteris paribus. Such limits on policy choice might inhibit redistributive policies, as the French Socialists "right-turn" in the early 1980s under Mitterand illustrates. Moreover, the proxy is also germane to dependency theorists who argue that exposure to the world economy leads to the development of enclaves and indigenous under-classes. In short, given each of these varying rationales, the hypothesis suggests that openness worsens income inequality.

5. Data Disabilities

Before presentation of the results, detailed commentary on the data set is imperative. As scholars ubiquitously state in any empirical investigation of income distribution, to measure inequality is difficult because much of the evidence depends on the accuracy of public surveys, which are considerably less reliable in poorer countries. Furthermore, there are distinctions between income and wealth, pre- and post- tax income, wage and non-wage earnings and individuals and households that complicate both temporal and spatial measurement. The regressions in this paper use Deininger and Squire's (1096) data set.

Deininger and Squire cite three methodological issues that distinguish income distribution studies, for which this study controls. The first regards the difference between surveys based on household or individual data, and the authors find that "using the distribution of income across; households rather than persons as the basis for the Gini index results in a slightly lower value." But given the magnitude of the difference, approximately 1.69 without controls, Deininger and Squire conclude that "there is no reason to expect a large systematic bias in empirical work as a result of using both household-based an individual-based Gini coefficients (1996, p. 580)." The second distinction between surveys based on before- and after-tax data also potentially biases the estimates. Assuming that taxation redistributes resources from rich to poor, coefficients from after-tax surveys should be lower ceteris paribus(15). Deininger and Squire note, however, that the quantitative importance of the difference depends on both the progressive nature and effectiveness of the given tax system. They suggest that the effect is less pronounced in poorer countries because "the role of redistributive taxation is smaller in these countries (1996, p. 580)." The most important distinction according to Deininger and Squire is between income- and expenditure-based coefficients. Given that expenditure is based on net income after taxes and other expenses, coefficients should be lower than those based on gross figures, and their data render a 6.6 difference between expenditure and income-based coefficients.

Therefore, in addition to the economic variables, the controls also include dummy variables for the three distinctions based on household, tax and income data sources, which ensures that the remaining coefficients in the specification hold irrespective of the different survey procedures(16). This precaution roughly resembles Bollen and Jackman's design that includes a dummy variable for the household/individual distinction (1985, p. 442).

6. Econometric Results

6.1. Panel Regressions

FGLS random-effects(17) pooled regressions, which include continental and time-specific dummy variables(18) to control for temporal effects, test spatially dominant panel designs, and the following empirical section on time series results addresses the unbalanced nature of the panels.

Turning to the panel results, Regression 1 is the base equation without institutional determinants, and some findings notably contradict persistent hypotheses. For example, the coefficients for the quadratic form of the level of income clearly contradict Kuznet's thesis, as the model produces a global minimum of approximately $15,433 in per capita income. This result suggests that the Gini coefficient decreases (income distribution improves) as income approaches the critical point, to the right of which the measure increases (income distribution worsens). This contradicts the view that inequality increases in the initial stages of development, but then improves during later phases. Also surprising, population growth correlates with more equitable income distribution, which is significant at the 10 percent level.

Some findings are consistent with enduring hypotheses. For example, openness to international trade seems to worsen income inequality, which might support the common view of wage pressures, dependency theory or simply the more general view that increased exposure to the world economy constrains (redistributive) social policy. Also consistent with theory, the coefficient for the government ratio enters the equation with the expected negative sign, as increased involvement in the economy correlates with decreases in the Gini coefficient(19). Also as expected, the regional dummies enter with characteristic signs, as Africa, the Americas and Asia have more uneven income distributions compared to Europe(20).

Finally, control dummies for the underlying data sources produce the expected signs and magnitudes, especially regarding expenditure data, which renders Gini coefficients approximately 6.4 lower on average, and personal income data, which renders figures about 1.5 lower on average. These parameters compare with the respective 6.6 and 1.7 raw means in the data set that Deininger and Squire report, and the dummy for pre- and post-tax is insignificant.

Turning to single-institutional equations, the Polity III dummy variable for "democracy" supports the prevailing regime-type perspective(21). According to Regression 2, democracies have Gini coefficients on average approximately 1.1 lower than dictatorships, ceterisparibus and notably including the income effect, which is significant at the 10 percent level. Ordinal forms of the Policy III scale do not produce more accurate estimates, as the linear, linear-log and quadratic expressions are all statistically insignificant according to Regressions 3 through 5. To address data source sensitivity, the Gastil dummy in Regression 6 also suggests that democracies have more egalitarian income distributions, but the figure is less robust and notably insignificant. The three functional forms of the ordinal Gastil measure also produce insignificant results according to Regressions 7 through 9.

The effective party/constitutional framework measure strongly suggests that there is no relationship between regime-type and income distribution. For example, no functional form of the number of effective parties is significant according to Regressions 10 through 12. Also, contrary to theory, single-party dictatorships do not affect more even distributions of income as the insignificant coefficient in Regression 13 is positive. Also, dummy variables for constitutional frameworks in Regressions 14 and 15 suggest that there is no difference among parliamentary and presidential democracies and the: residual authoritarian category. In general, these results imply that policy entrepreneurs and constitutional framers cannot craft methods of state access that compel more equitable income distribution.

But besides regime-type, again, other institutions are also important, and the regressions produce some unexpected results. For example, while the coefficients for SDC variables largely have the expected signs in Regressions 16 and 17, with the notable exception of the interaction term, none is statistically significant. This casts doubt on the intuitive perspective that strong labor movements and "Left" governments effectively achieve their stated egalitarian objectives.

Also, the revenue based federalism measure significantly correlates with income distribution but with the unexpected sign. As Regression 18 indicates, a 10 percent increase in the local revenue proportion corresponds with an approximate 0.6 lower Gini coefficient, which is safely significant. This contradicts the expectation that centralization leads to effective redistribution across disparate regions. The insignificant but unexpectedly positive coefficient for the federal spending gap in Regression 19 similarly vitiates the hypothesis.

Besides data quality, which unfortunately besets any econometric study of income distribution, the previous equations likely suffer from institutional specification bias. Therefore, the final panel equation exercise regresses Gini coefficients on every linear combination of institutional variables: each functional form of the number of effective parties, SDC expressions (1) and (2), and the two fiscal federalism considerations. Regression 20 is the model that explains the most variance in income distribution in the limited sample, and the results confirm the empirical relationship between the federal revenue ratio and income distribution. The more "complete" specification suggests that a 10 percent increase in the ratio corresponds with approximately 0.9 lower Gini coefficients. Also, the more comprehensive equation confirms the univariate single-institution models that regime-type, labor organization and government partisanship do not correlate with income inequality.

6.2. Individual Time-Series Equations: Taiwan, the United Kingdom and the United States

As previously noted, the panel regressions are unbalanced. The primary difficulty in the design, although Deininger and Squire hardly preclude such evaluation, regards the inconsistent temporal coverage. Nonetheless, a pooled design is still arguably the best strategy for general, sweeping inferences because, after all, variation in the exogenous variables through time across space should summarily affect income distribution. But to address these shortcomings with an alternative approach, this section presents individual time-series estimations for three countries with an adequate number of consecutive observations(22). In addition to a broadening the sensitivity, analysis, such a strategy might also address uneasiness associated with spatially defining and comparing certain "institutions," such as "political party," "federalism" or "centralized labor organization," across such a vast array of countries. Most important, however, these regressions might usefully suggest future qualitative research that thoroughly describes idiosyncrasies of economic and political institutions that affect equality in particular countries.

The equation for Taiwan that explains the most variance, Regression 21, does not clearly support common institutional or economic hypothesis regarding income distribution. Although the sign is positive, the KMT's fluctuating dominance, which ranged from approximately 1.16 to 1.24 effective parties during 1972 to 1992 did not have a demonstrative quantitative effect on income distribution. Curiously, the only statistically significant regressor is population growth, which, contrary to the hypothesis, correlated negatively with the Gini coefficient at the 10 percent level of significance.

Data for the United Kingdom generally support the SDC perspective Regression 22 suggests that decreased labor organization and Right-wing partisanship worsened income distribution from 1966 to 1990. This reflects common characterizations of the Thatcher years, 1979 to 1990, which witnessed steadily increasing inequality and partisanship. In fact, the lowest Gini coefficient is for 1977 under Labour, and the figure steadily increased to the greatest value in 1990 under 11 years of Tory rule. A problem with this summation is that the statistically significant interaction term suggests that cohesion between Right-wing government and weaker labor movements, perhaps pigeonholing "Thatcherism" for some commentators, also improved income inequality. In addition, the British case confirms the correlation from the panel regressions regarding the federal revenue ratio, which increased in the early 1970s and decreased in the late 1980s in Britain. Finally, the significant quadratic form of the number of effective parties curiously suggests that income distribution worsened until approximately 2.1 parties and improved with greater values, as the figure fluctuated from about 2.02 during 1966 through 1969 to 2.17 during 1987 through 1990. Also, openness to international trade, which peaked in 1974, correlated negatively with the Gini coefficient, in contrast to the base panel regression finding and the common notion regarding wage pressures.

Although inequality similarly increased steadily from 1966 to 1990, the American case does not corroborate the institutional findings with respect to Britain. Regression 23, which best fits the variance in income distribution in the United States, suggests that the federal spending gap, which peaked in 1982, correlated positively with the Gini coefficient, in contrast to the hypothesis. Also, two findings suggest that the United States does not conform to the base income distribution panel equation, Regression 1. First, increased government involvement in the economy curiously correlated with greater inequality, and, second, increased openness to international trade correlated with more equality, similar to the equation for the United Kingdom.

7. Conclusions

Again, the specifications do not posit a single "true" model of income distribution but more closely resemble an informal sensitively analysis of the effect of regime-type and other key institutions. The most robust conclusion among the institutional considerations regards fiscal federalism, as the results suggest that more decentralized countries correlate with greater equality, all things being equal. This finding, however, contradicts the view that greater centralization corresponds with more equitable income distribution due to resource transfers across disparate regions or to centralized redistributive policy aims. Also, among the more conspicuous ambiguous findings, regime-type does not seem to affect income distribution, and, surprisingly, most data produce no correlation between SDC variables and equality.

To speculate on further research, the general dearth of robust relationships in econometric models could reflect the assumption of homogenous expectations. In short, similar to studies of regime-type and growth, statistical designs do not sufficiently consider the ideology or beliefs of the very policymakers who operate vis-a-vis institutions(23). In short, assuming some "correct" measure of the complete ensemble of relevant institutions, policy--makers with varying ideas can pursue varying economic objectives that lead to different and unintended results. Two single-party dictators, for example, can purse equitable income distribution or contrarily protect economic elites. Likewise, democracies produce a wide variety of governing parties with myriad objectives. Thus, econometric models of income distribution assume homogenous expectations among economic policy-makers. While the SDC partisanship scale does cover the most advanced countries, most regressions, do not satisfactorily capture these differences.

Econometrically incorporating "ideas" is difficult, but some interaction term with institutions might better represent the "true" form. That is, income distribution depends on the joint coincidence of some optimal configuration of discretion and the sovereigns' objective to redistribute resources. As suggested elsewhere (Durham, 1998), the equation, momentarily ignoring other economic factors, might resemble

(4) Y = [[Beta].sub.0] + [[Beta].sub.ib](I x B) + [Mu]

where the effect of the given set of institutions, I, largely depends on prevailing beliefs, represented by B. Conversely, the impact of I depends on the magnitude of B. The coefficient, then, [b.sub.ib], captures the coincidence of institutional and ideology.

Besides the difficulty associated with classifying economic ideology, this approach might treacherously approach endogenous selection. Identifying those governments that "truly" intend to redistribute income but fail to do so would be difficult. However, the prevailing assumption of homogenous expectations in these models might help explain the dearth of compellingly robust findings for institutional variables. Also, as the discussion of time-series equations suggests, perhaps case studies that include such "thick description" of the interaction between ideas and economic as well as political institutions would be instructive.

[TABULAR DATA 1-2 NOT REPRODUCIBLE IN ASCII]

Table 3

The Effective/Party Constitutional Framework Measure FGLS Random-Effects Models 1960-1992

 Regression 13:
 [Beta] p value

Squared Effective Parties
Log Effective Political
 Parties
Effective Political Parties -0.112 0.617
(Critical Point)
Initial GDP Squared 2.99E-08(**) 0.011
Initial GDP -0.001(**) 0.000
(Income Critical Point) 17079
Government Ratio -0.116(**) 0.028
Population Growth -0.532(*) 0.068
Openness to Trade 0.013 0.212
Dummy for Africa 13.070(**) 0.000
Dummy for Central 8.008(**) 0.000
 and North America
Dummy for South 11.681(**) 0.000
 America
Dummy for Europe -6.803(**) 0.001
Dummy for Oceania 0.370 0.932
Dummy for Expenditure -8.238(**) 0.000
Dummy for Personal -1.798(**) 0.001
 Income
Dummy for Tax 0.601 0.370
Intercept 49.190(**) 0.000
Observations 548
Cases 80
Average Time Periods 6.850
[R.sup.2] Within 0.210
[R.sup.2] Between 0.571
[R.sup.2] Overall 0.519
[Chi square] 245.630
Prob. [Chi square] > 0 0.000
Hausman [Chi square] 43.440
Prob. Hausman [Chi square] >0 0.368

 Regression 14:
 [Beta] p value

Squared Effective Parties
Log Effective Political -0.075 0.320
 Parties
Effective Political Parties
(Critical Point)
Initial GDP Squared 2.95E-08(**) 0.012
Initial GDP -0.001(**) 0.000
(Income Critical Point) 17127
Government Ratio -0.115(**) 0.029
Population Growth -0.534(*) 0.066
Openness to Trade 0.013 0.208
Dummy for Africa 13.111(**) 0.000
Dummy for Central 8.145(**) 0.000
 and North America
Dummy for South 11.721(**) 0.000
 America
Dummy for Europe -6.838(**) 0.001
Dummy for Oceania -0.305 0.944
Dummy for Expenditure -8.313(**) 0.000
Dummy for Personal -1.770(**) 0.001
 Income
Dummy for Tax 0.688 0.307
Intercept 48.852(**) 0.000
Observations 548
Cases 080
Average Time Periods 6.850
[R.sup.2] Within 0.213
[R.sup.2] Between 0.568
[R.sup.2] Overall 0.519
[Chi square] 245.150
Prob. [Chi square] > 0 0.000
Hausman [Chi square] 32.180
Prob. Hausman [Chi square] >0 0.836

 Regression 15:
 [Beta] p value

Squared Effective Parties -0.021 0.801
Log Effective Political
 Parties
Effective Political Parties -0.007 0.988
(Critical Point) -0.170
Initial GDP Squared 2.99E-08(**) 0.011
Initial GDP -0.001(**) 0.000
(Income Critical Point) 17147
Government Ratio -0.117(**) 0.027
Population Growth -0.513(*) 0.080
Openness to Trade 0.013 0.205
Dummy for Africa 13.028(**) 0.000
Dummy for Central 7.992(**) 0.000
 and North America
Dummy for South 11.668(**) 0.000
 America
Dummy for Europe -6.742(**) 0.001
Dummy for Oceania -0.366 0.932
Dummy for Expenditure -8.181(**) 0.000
Dummy for Personal -1.818(**) 0.001
 Income
Dummy for Tax 0.580 0.391
Intercept 49.096(**) 0.000
Observations 548
Cases 080
Average Time Periods 6.850
[R.sup.2] Within 0.209
[R.sup.2] Between 0.573
[R.sup.2] Overall 0.521
[Chi square] 248.630
Prob. [Chi square] > 0 0.000
Hausman [Chi square] 22.420
Prob. Hausman [Chi square] >0 0.994

Notes: (*) 10 percent confidence interval

(**) 5 percent confidence interval

Table 4

Single-Party and Party-less Dictatorships and Constitutional Frameworks Dummy Variables FGLS Random-Effects Models

 Regression 13:
 1962-1992
 [Beta] p value

Dummy for Dictatorship
Dummy for Parliamentary
Dummy for Presidential
Dummy for Mixed
Dummy for Semi-Presidential
Single-Party 0.186 0.920
Dictatorship
Initial GDP Squared
Initial GDP 0.001(**) 0.009
(Income Critical Point)
Government Ratio -0.026 0.739
Population Growth 1.860(**) 0.041
Openness to Trade 0.000 0.996
Dummy for Africa 10.655(**) 0.000
Dummy for Central and 15.329(**) 0.000
 North America
Dummy for South America 22.539(**) 0.000
Dummy for Europe -8.202(**) 0.001
Dummy for Oceania
Dummy for Expenditure -1.354 0.508
Dummy for Personal Income -1.543 0.349
Dummy for Tax -0.829 0.565
Intercept 35.074(**) 0.000
Observations 123
Cases 40
Average Time Periods 3.075
[R.sup.2] Within 0.112
[R.sup.2] Between 0.856
[R.sup.2] Overall 0.860
[Chi square] 497.210
Prob. [Chi square] > 0 0.000
Hausman [Chi square] 11.530
Prob. Hausman [Chi square] >0 1.000

 Regression 14:
 1960-1992
 [Beta] p value

Dummy for Dictatorship 0.320
Dummy for Parliamentary -0.320 0.687
Dummy for Presidential -1.247 0.168
Dummy for Mixed -1.125 0.847
Dummy for Semi-Presidential -5.117(*) 0.006
Single-Party
Dictatorship
Initial GDP Squared 2.74E-08(**) 0.018
Initial GDP -0.001(**) 0.000
(Income Critical Point) 17770
Government Ratio -0.113(**) 0.032

Population Growth -0.608(**) 0.038
Openness to Trade 0.012 0.229
Dummy for Africa 13.401(**) 0.000
Dummy for Central and 8.179(**) 0.000
 North America
Dummy for South America 12.351(**) 0.000
Dummy for Europe -6.878(**) 0.001
Dummy for Oceania -0.596 0.893
Dummy for Expenditure -8.957(**) 0.000
Dummy for Personal Income -1.816(**) 0.001
Dummy for Tax 0.790 0.232
Intercept 49.351(**) 0.000
Observations 549
Cases 80
Average Time Periods 6.863
[R.sup.2] Within 0.237
[R.sup.2] Between 0.553
[R.sup.2] Overall 0.504
[Chi square] 251.910
Prob. [Chi square] > 0 0.000
Hausman [Chi square] 46.110
Prob. Hausman [Chi square] >0 0.345

 Regression 15:
 1960-1992
 [Beta] p value

Dummy for Dictatorship 0.687
Dummy for Parliamentary
Dummy for Presidential -0.927 0.427
Dummy for Mixed -0.804 0.890
Dummy for Semi-Presidential -4.796(**) 0.015
Single-Party
Dictatorship
Initial GDP Squared 2.74E-08(**) 0.018
Initial GDP -0.001(**) 0.000
(Income Critical Point) 17770
Government Ratio -0.1113(**) 0.032

Population Growth -0.608(**) 0.038
Openness to Trade 0.012 0.229
Dummy for Africa 13.401(**) 0.000
Dummy for Central and 8.179(**) 0.000
 North America
Dummy for South America 12.351(**) 0.000
Dummy for Europe -6.878(**) 0.001
Dummy for Oceania -0.596 0.893
Dummy for Expenditure -8.957(**) 0.000
Dummy for Personal Income -1.816(**) 0.001
Dummy for Tax 0.790 0.232
Intercept 49.030(**) 0.000
Observations 549
Cases 80
Average Time Periods 6.863
[R.sup.2] Within 0.237
[R.sup.2] Between 0.553
[R.sup.2] Overall 0.504
[Chi square] 251.910
Prob. [Chi square] > 0 0.000
Hausman [Chi square] 42.130
Prob. Hausman [Chi square] >0 0.509

Notes:

(*) 10 percent confidence interval

(**) 5 percent confidence interval

Table 5
SDC Variables and Fiscal Federalism FGLS Random-Effects Models

 Regression 16: Regression 17
 1966-1990 1966-1990
 [Beta] p value [Beta]
Federal Spending Gap
Federal Revenue Ratio
SDC Labor Organization -0.156 0.750
SDC Partisan Index -0.409
SDC Interaction Term 0.380
SDC Left-Labor Power -0.422
Initial GDP Squared 4.21E-08 0.135 3.30E-08
Initial GDP -0.001(*) 0.094 -0.001
(Income Critical Point) 14078 14471
Government Ratio -1.313 0.115 -1.283
Population Growth -0.023 0.832 -0.082
Openness to Trade -0.088(**) 0.000 -0.084(**)
Dummy for Africa
Dummy for Central and
North America
Dummy for South
America
Dummy for Europe
Dummy for Oceania
Dummy for Expenditure
Dummy for Personal -8.557(**) 0.000 -7.917(**)
Income
Dummy f or Tax 1.722(*) 0.056 1.588(*)
Intercept 46.523(**) 0.000 45.457(**)
Observations 159 159
Cases 13 13
Average Time Periods 12.231 12.231
[R.sup.2] Within 0.088 0.115
[R.sup.2] Between 0.796 0.779
[R.sup.2] Overall 0.692 0.695
[Chi square] 283.260 282.970
Prob. [Chi square] > 0 0.000 0.000
Hausman [Chi square] 160.910 154.700
Prob. Hausman [Chi
 square]> 0 0.000 0.000

 Regression 18:
 1960-1992
 p value [Beta] p value
Federal Spending Gap
Federal Revenue Ratio -5.608(**)
SDC Labor Organization
SDC Partisan Index 0.343
SDC Interaction Term 0.323
SDC Left-Labor Power 0.146
Initial GDP Squared 0.265 4.64E-08(**) 0.014
Initial GDP 0.196 -0.002(**) 0.000
(Income Critical Point) 19922
Government Ratio 0.129 1.360(*) 0.069
Population Growth 0.490 0.063 0.316
Openness to Trade 0.000 -0.036(**) 0.005
Dummy for Africa -7.333(**) 0.000
Dummy for Central and -5.385(**) 0.000
North America
Dummy for South -3.651(**) 0.000
America
Dummy for Europe 14.627(**) 0.000
Dummy for Oceania 4.633(**) 0.000
Dummy for Expenditure 9.786(**) 0.000
Dummy for Personal 0.000 3952(**) 0.(i)02
Income
Dummy f or Tax 0.086 3.60(*) 0.051
Intercept 0.000 48.792(**) 0.000
Observations 264
Cases 39
Average Time Periods 6.769
[R.sup.2] Within 0.084
[R.sup.2] Between 0.824
[R.sup.2] Overall 0.747
[Chi square] 641.400
Prob. [Chi square] > 0 0.000
Hausman [Chi square] 117.000
Prob. Hausman [Chi
 square]> 0 0.000

 Regression 19:
 1960-1992
 [Beta] p value
Federal Spending Gap
Federal Revenue Ratio
SDC Labor Organization
SDC Partisan Index
SDC Interaction Term
SDC Left-Labor Power
Initial GDP Squared 5.01E-08(**) 0.008
Initial GDP -0.002(**) 0.000
(Income Critical Point) 20298
Government Ratio 1.014 0.181
Population Growth 0.112(*) 0.084
Openness to Trade -0.032(**) 0.011
Dummy for Africa -9.184(**) 0.000
Dummy for Central and -5.535(**) 0.000
North America
Dummy for South -3.278(**) 0.001
America
Dummy for Europe 17.150(**) 0.000
Dummy for Oceania 4.040(**) 0.001
Dummy for Expenditure 10.003(**) 0.000
Dummy for Personal 3.171(**) 0.035
Income
Dummy f or Tax 4.595(**) 0.010
Intercept 47.323(**) 0.000
Observations 262
Cases 38
Average Time Periods 6.895
[R.sup.2] Within 0.066
[R.sup.2] Between 0.841
[R.sup.2] Overall 0.740
[Chi square] 610.240
Prob. [Chi square] > 0 0.000
Hausman [Chi square] 100.920
Prob. Hausman [Chi
 square]> 0 0.000

Notes:

(*) 10 percent confidence interval

(**) 5 percent confidence interval

Table 6

Complete "Institutional" Specification and Individual Time-Series Equations: Taiwan, the United Kingdom, and the United States

 Regression 20: Regression 21:
 Taiwan
 1966-1990 1972-1992
 [Beta] p value [Beta]
Squared Effective Political
 Parties
Log Effective Political - -1.983 0.246
 Parties
Effective Political Parties 16.732
(Critical Point)
Federal Spending Gap
Federal Revenue Ratio -8.769(**) 0.004
SDC Labor Organization 0.173 0.735
SDC Partisan Index -0.465 0.290
SDC Interaction Term 0.431 0.257
SDC Left-Labor Power -0.353
Initial GDP Squared 2.99E-08 0.307
Initial GDP -0.001 0.185 0.000
(Income Critical Point) 15983
Government Ratio -0.080 0.503 0.749
Population Growth -1.268 0.125 -3.925(*)
Openness to Trade -0.071(**) 0.000 0.002
Dummy for Expenditure
Dummy for Personal -9.811(**) 0.000
Income
Dummy for Tax 0.631 0.545
Intercept 52.326(**) 0.000 5.619
Observations 157 19
Cases 13
Average 12.077
Time Periods
[R.sup.2] Within 0.128
[R.sup.2] Between 0.823
[R.sup.2] Overall 0.719
[R.sup.2] 0.439
Adjusted [R.sup.2] 0.223
[Chi square] 307.720
Prob. [Chi square] > 0.000
Hausman [Chi square] 300.600
Prob. 0.000
Hausman [Chi square]> 0
F Statistic 2.040
Prob. F Statistic > 0 0.140
Durbin Watson 2.546
Transformed Durbin 1.928
 Watson
PraisWinston rho 0.055

 Regression 22:
 United Kingdom
 1996-1990
 p value [Beta] p value
Squared Effective Political 516.215(**) 0.010
 Parties
Log Effective Political -
 Parties
Effective Political Parties 0.142 -2172.507(**) 0.010
(Critical Point) 2.104
Federal Spending Gap
Federal Revenue Ratio -27.978(**) 0.018
SDC Labor Organization -6.230(**) 0.012
SDC Partisan Index -7.693(**) 0.003
SDC Interaction Term -6.406(**) 0.003
SDC Left-Labor Power
Initial GDP Squared
Initial GDP 0.439 0.000 0.162
(Income Critical Point)
Government Ratio 0.148 -0.291 0.281
Population Growth 0.066 -0.198 0.901
Openness to Trade 0.966 -0.135(**) 0.003
Dummy for Expenditure
Dummy for Personal
Income
Dummy for Tax
Intercept 0.753 -23.296(*) 0.070
Observations 25
Cases
Average
Time Periods
[R.sup.2] Within
[R.sup.2] Between
[R.sup.2] Overall
[R.sup.2] 0.979 0.988
Adjusted [R.sup.2] 0.960 0.982
[Chi square]
Prob. [Chi square] > 0.000
Hausman [Chi square]
Prob.
Hausman [Chi square]> 0
F Statistic 53.760 163.100
Prob. F Statistic > 0 0.000 0.000
Durbin Watson 1.665 0.960
Transformed Durbin 2.271
 Watson
PraisWinston rho 0.374

 Regression 23:
 United States
 1996-1990
 [Beta] p value
Squared Effective Political 0.207
 Parties
Log Effective Political -
 Parties -7.979
Effective Political Parties
(Critical Point) 3.116(**) 0.001
Federal Spending Gap
Federal Revenue Ratio
SDC Labor Organization
SDC Partisan Index
SDC Interaction Term
SDC Left-Labor Power -0.353 0.174
Initial GDP Squared -9.40E-08(**) 0.000
Initial GDP 0.004(**) 0.000
(Income Critical Point) 22127
Government Ratio 0.476(**) 0.001
Population Growth 1.003 0.333
Openness to Trade -0.152(**) 0.038
Dummy for Expenditure
Dummy for Personal
Income
Dummy for Tax
Intercept -7.913(**) 0.000
Observations 25
Cases
Average
Time Periods
[R.sup.2] Within
[R.sup.2] Between
[R.sup.2] Overall
[R.sup.2]
Adjusted [R.sup.2]
[Chi square]
Prob. [Chi square] > 0.000
Hausman [Chi square]
Prob.
Hausman [Chi square]> 0
F Statistic
Prob. F Statistic > 0
Durbin Watson
Transformed Durbin
 Watson
PraisWinston rho

Notes:

(*) 10 percent confidence interval

(**) 5 percent confidence interval

Notes

(1.) Recent data suggest that the level rather than the rate of economic, development affects regime-type (Durham, 1998 and 1999). Therefore, as the econometric section more thoroughly addresses, collinearity between national income levels and democracy is a potential problem in broadly specified income distribution models. The empirical data presented in this paper directly addresses this concern.

(2.) Both measures are inherently subjective, and temporal reliability is questionable. Brunetti persuasively argues that the "dominating" Gastil scale is "rather subjective (1997, p. 166)," and the Freedom House judges readily admit that "changes in information and judgment since 1973 make many ratings not strictly comparable from year to year (Gastil, 1987, p. 53)."

(3.) The construction of the proxy is described in more detail in Durham (1998 and 1999).

(4.) The number of parties follows Laakso and Taagepera (1979). as the "effective number of parties is the number of hypothetical equal-size parties that would have the same total effect on fractionalization of the system as have the actual parties of unequal size (1979, p. 4)." The effective number N is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [P.sub.l] represents the share of the total number of parliamentary seats that the lth party holds. Data come from the Chronicle of Parliamentary Elections and Developments Inter-Parliamentary Union, vols. 2-27 (1967-1993); The Political Handbook of the World (various issues); and Machie and Rose (1991).

(5.) In Stepan and Skach's terms, a "pure parliamentary regime" is a system of "mutual dependence," where parliamentary majorities must support executive power, and the executive office "has the capacity to dissolve the legislature and call for elections." A "pure presidential regime" is a system of "mutual independence," where both the executive and legislative powers have "fixed electoral mandates" that are the sole sources of "legitimacy (Stepan and Skach, 1993, pp. 3-4, emphasis added)." Similarly, "dictators," according to common indicators, who permit the existence of a multi-party parliament, occupy the "executive" branch.

While the raw number of effective parties captures discretion for parliamentary regimes, the figure adjusts for these non-parliamentary categories because the executive branch does not derive its legitimacy from the legislature. Therefore, the overall measure for these cases is the average number of effective parties across the legislative and executive branches.

(6.) The effective party component of the measure is also serviceably relevant to dictatorships, who may or may not choose to surround themselves with a party.

(7.) Some scholars object to the term "party" in a non-democratic setting. There are two rejoinders. First, to apply the term exclusively to democracies necessarily excludes similar analysis of less developed areas, as if poorer countries are "underdeveloped" politically (Ranney, 1968). Second, and more persuasive, as Durverger (1954) notes, the internal organization of parties in single or multiple party regimes do not differ widely.

(8.) The effective party/constitutional framework measure does not purport to capture interest group pressures or the impact of (marginal) political parties that the electoral system excludes from participation. Perhaps the Gastil and Polity III data sets better capture such influences. Again, the alternative is not necessarily a superior proxy for all purposes but perhaps more effectively addresses implicit serviceable concerns. The focus of this institutional measure is largely on the "party-in-the-government," as defined in Ranney (1968).

(9.) Data come from Garrett (1995).

(10.) In some cases conventionally perceived "right-wing" parties, particularly in Europe, entertain strong corporatist convictions. In other words, a consensus regarding social welfare policy can characterize a polity across partisan lines, such as in, say, Austria or Denmark. However, the indices that Garrett employs take social policy into account, as, but one example, (pre-Thatcher) Conservative governments in the United Kingdom are considered to the left of Republican administrations in the United States. In other' words, the partisan measure is not a relative index within countries but an absolute scale: across national boundaries. Nonetheless, the left-labor index in (2) is perhaps the most effective test.

(11.) Bernholtz (1993) criticizes Weingast's characterization of the British state, which scholars more conventionally consider as comparatively centralized.

(12.) Federalism data come from various issues of the UN's Yearbook of National Account Statistics.

(13.) The concept of fiscal federalism recommends brief discussion of other multiple sovereignties. In addition to local governments, supranational institutions potentially have a very significant impact through (primarily) international trade and (at least in the case of the European Union) social policy. Moreover, transfers between local governments can have considerable impact on economic outcomes. Some regressions attempt to capture these effects with dummy variables for the relevant international organization (EC membership) and consideration of the relative importance of international trade. Importantly, however, the notion of fiscal federalism focuses on the constraints on national government, the primary locus of policy-making.

(14.) These data come from the Penn World Table 5.6. See Summers and Heston (1991) for a description.

(15.) This view ignores the prevalence of tax evasion of wealthy individuals, a fundamental problem (especially in developing countries).

(16.) Deininger and Squire suggest three strategies for empirical investigation of their data. They write that "it would be prudent to examine whether such results hold for (a) the raw data, (b) data that have been adjusted for differences between expenditure-based and income-based coefficient, and (c) data consistently based on a common definition (1996, p. 582)." The following regressions follow (b). For an empirical examination that follows each of these alternatives see Durham (1998). Notably, these alterations do not significant affect the results.

(17.) To test the appropriateness of a random-effects model, the regression tables list the Hausman (1978) specification test, which evaluates the equivalence of parameters from random- and fixed-effects models. But whatever the formal equivalence of the coefficients, the strong emphasis on cross-sectional institutional variance in these hypothesis suggest that inclusion of country-specific dummy variables is largely counterproductive.

(18.) An alternative temporal consideration is also instructive and perhaps more theoretically appealing--the effect of election years. But with respect to all relevant institutional specifications (Regressions 1 -19, except Regression 13, which exclusively considers dictatorships), the election year dummy is statistically insignificant. Data are available upon request.

(19.) The government ratio less military spending is perhaps a more accurate proxy for public social welfare expenditure. Unfortunately, the distinction limits the data somewhat (from 568 to 384 observations in Regression 1), but nonetheless, the results are generally consistent using the two proxies. In fact, of the 19 (univariate) regressions, only two results differ in terms of significance, albeit in ambiguous directions. Curiously, the government ratio less military spending is significant in Regression 17 but insignificant in Regression 18, unlike the gross proxy.

(20.) Other dummy variables are perhaps instructive in lieu of controlling for regional effects. For example, a dummy variable for membership in the European Community complements in analysis of fiscal federalism by controlling for a influence supranational organization that facilitates significant international resource transfers, as discussed previously. Surprisingly, the variable produces rather ambiguous results, as the coefficient is expectedly negative in regressions that include the Gastil index but positive in equations that include the SDC variables. All other regressions produce statistically insignificant results. Also, dummy variables for Communist regimes, particularly the Soviet bloc and China, consistently suggest in every equation that such "Marxist-Leninist" systems have more equitable income distributions.

(21.) As previously suggested, widely specified income distribution models may suffer from collinearity among initial GDP per capita, political regime-type measures and openness to international trade. Regressions that alternatively exclude trade and the functional form of GDP should ameliorate the difficulty in interpreting the coefficients for the regime variables. For example, Regressions 1 through 19 without openness to trade produce no changes in the significance tests of the coefficients. Furthermore, omission of the quadratic form of GDP per capita produces only three changes in the (14) relevant regressions, as the linear-log forms of the Gastil and Polity III indices (10 percent confidence) as well as the quadratic form of the Gastil measure (10 percent confidence) are significant.

(22.) While there are a few additional countries with more than 20 observations according to the Data Appendix (Bulgaria, Canada, India, and Japan), there are missing consecutive observations in each series. The regressions do not use interpolated data.

(23.) This focus is different from Romer's more general discussion of the role of "ideas" that encompass an economy. Previous models of growth include some 'variables with acknowledged respect to the broader notion of ideas. For example, as Romer discusses, openness to international trade can affect the benevolent "flow of ideas" across national boundaries. The following considers "ideas" with respect to the relevant actors that institutions restrain. Alesina et al. note the potential importance of governing ideologies in passing but do not consider any interaction with institutions (1996, pp. 205-06).

References

Alesina, Alberto et al. (1996). "Political Instability and Economic Growth," Journal of Economic Growth 1, 2:189-211, June.

Alvarez, R. Michael, Garrett, Geoffrey and Lange, Peter. (1991). "Government Partisanship, Labor Organization, and Macroeconomic Performance, American Political Science Review 85, 2: 539-56, June.

Barro, Robert J. (1996). "Democracy and Growth," Journal of Economic Growth. 1, 1: 1-27, March.

Beitz, Charles R. (1982). "Democracy in Developing Societies," In Gastil, Ed., Freedom in the World: Political Rights and Civil Liberties. New York: The Free Press.

Bernholtz, Peter, "Comment," Journal of Institutional and Theoretical Economics, 149, 1.

Bollen, Kenneth A. and Jackman, Robert W. (1985). "Political Democracy and the Size Distribution of Income," American Sociological Review 50, 438-57.

Casella, Alessandra and Frey, Bruno. (1992). "Federalism and Clubs: Toward an Economic Theory Of Overlapping Political Jurisdictions," European Economic Review 36, 313-31.

Chronicle of Parliamentary Elections and Developments, Inter-Parliamentary Union 2-7, 1967-1993.

Clague, Christopher et al. (1996) "Property and Contract Rights in Autocracies and Democracies," Journal of Economic Growth. 1, 2: 243-45, June.

Crafts, Nicholas F. R. (1992). "Institutions and Economic Growth: Recent British Experience in an International Context," West European Politics 15, 4:16-38, October

Deininger, Klaus and Lyn Squire. (1996). "A New Data Set Measuring Income Inequality," The World Bank Economic Review 10, 3: 565-91.

Durham, J. Benson. (1998). Economic and Political Institutions and Growth: Some Econometrics, Ph.D. diss., Columbia University.

--. (1999). "Economic Growth and Political Regimes," Journal of Economic Growth, forthcoming.

Duverger, Maurice. (1954). Political Parties. New York: Wiley.

Evans, Peter. (1979). Dependent Development. The Alliance of Multinational, State, and Local Capital in Brazil. Princeton: Princeton University Press.

Garrett, Geoffrey. (1995). "Capital Mobility, Trade, and the Domestic Politics of Economic Policy," International Organization 49, 4: 657-87, Autumn.

Garrett, Geoffrey and Lange, Peter. (1986). "Performance in a Hostile World: Economic Growth in Capitalist Democracies," World Politics 38, 4:517-35.

--.(1985). "The Politics Of Growth: Strategic Interaction And Economic Performance In The Advanced Industrial Democracies, 1974-80," Journal of Politics 47, 792-827, August.

Gourevitch, Peter. (1978). "The Second-Image Reversed: The International Sources of Domestic Politics," International Organization 32, 881-912.

Hausman, J. A. (1978). "Specification Tests in Econometrics," Econometrica 46, 1251-71.

Headey, Bruce. (1970). "Trade Unions and National Wage Policies," Journal of Politics 32, 407-32.

Hicks, Alexander. (1994). "The Social Democratic Corporatist Model of Economic Performance in The Short- And Medium-Run Perspective." In Thomas Janoski and Alexander Hicks, Eds., The Comparative Political Economy of the Welfare State, pp. 189-217, Cambridge: Cambridge University Press.

Jaggers, Keith and Gurr, Ted Robert. (1995). "Tracking Democracy's Third Wave with the Polity III Data," Journal of Peace Research 32, 4: 469-482.

Kee, Woo Kik. (1977). "Fiscal Decentralization and Economic Development," Public Finance Quarterly 5, 1: 79-97, January.

Kuznets, Simon. (1955). "Economic Growth and Income Inequality," American Economic Review 45, 1-28.

Laakso, Markku and Taageper, Rein. (1979). "Effective Number of Parties: A Measure With Application to West Europe," Comparative Political Studies 12, 3-27, April.

Levine, Ross and Renelt, David. (1992). "A Sensitivity Analysis Of Cross-Country Growth Regressions," American Economic Review 82, 4: 942-63, September.

Lijphart, Arend. (1984). Democracies: Patterns of Majoritarian and Consensus Government in Twenty-one Countries. New Haven: Yale University Press.

Machie, Thomas and Rose, Richard. (1991). The International Almanac of Electoral History.' Fully Revised Third Edition (Congressional Quarterly Inc.).

Mainwaring, Scott. (1993). "Presidentialism, Multipartism, and Democracy," Comparative Political Studies 26, 198-228.

North, Douglass C. and Weingast, Barry R. (1989). "Constitutions and Commitment: The Evolution of Institutions Governing Public Choice in Seventeenth-Century England," Journal of Economic History 49, 4: 803-32, December.

Oates, Wallace. (1972). Fiscal Federalism, Harcout Brace Jovanovich Inc.

Perotti, Roberto. (1996). "Growth, Income Distribution, and Democracy: What the Data Say," Journal of Economic Growth 1, 2:149-87, June.

Przeworski, Adam and Limongi, Fernando. (1993). "Political Regimes and Economic Growth," Journal of Economic Perspectives 7, 3: 51-69, Summer.

Qian, Yingyi and Weingast, Barry R. (1997). "Federalism as a Commitment to Preserving Market Incentives," Journal of Economic Perspectives 11, 83-92.

Ranney, Austin. (1968). "The Concept of `Party'." In O. Garceau, Ed., Political Research and Political Theory, pp. 143-46. Boston: Harvard University Press.

Romer, Paul. (1993). "Idea Gaps and Object Gaps in Economic Development," Journal of Monetary Economics 32, 543-73.

Sirowy, Larry and Inkeles, Alex. (1990). "The Effects of Democracy on Economic Growth and Inequality: A Review," Studies in Comparative International Development 25, 126-57.

Stepan, Alfred and Skach, Cindy. (1993). "Constitutional Frameworks and Democratic Consolidation: Parliamentarism Versus Presidentialism, World Politics 45, 1-22.

Summers, Robert and Heston, Alan. (1991). "The Penn World Table (Mark 5): An Expanded Set of Intenational Comparisons, 1950-88," Quarterly Journal of Economics 61,327-68.

Weingast, Barry R. (1993). "Constitutions as Governance Structures: The Political Foundations of Secure Markets. Journal of Institutional and Theoretical Economics 149, 1395-1424.

Data Appendix

[TABULAR DATA NOT REPRODUCIBLE IN ASCII]

J. Benson Durham, Without implication, the author is grateful to Richard R. Nelson, Douglas A. Chalmers, Robert Y. Shapiro, Brendan O'Flaherty, Alessandra Casella, Lauren Paige Kennedy, and an anonymous referee for helpful comments on this study.

Columbia University3