1 -welfare states in developing countries

8/2/2019 1 -Welfare States in Developing Countries

1/20

Welfare States in Developing Countries:Unique or Universal?

Nita Rudra University of Pittsburgh

Do varieties of welfare capitalism exist in the developing world? This analysis challenges scholars of comparativepolitical economy and international political economy who treat the political economies of less developed countries

(LDCs) as more or less identical to one another or, at the other extreme, as nations marked by tremendous diversity.This paper is one of the first attempts to highlight systematic differences among the political economies of thedeveloping world, particularly with respect to their distribution regimes. Using cluster analysis, the results illustratethat welfare efforts in LDCs are either directed towards promoting market development (a productive welfare state),protecting select individuals from the market (a protective welfare state), or both (a dual welfare state). The discovery

of distinct patterns of welfare regimes in LDCs presents hitherto unknown implications for the influence of domesticpolitics and policies in late twentieth-century globalization.

Existing scholarship overlooks the possibility ofvarieties of welfare capitalism in the developingworld. In contrast, discussions abound regard-

ing identifiable and systematic differences in domesticinstitutional arrangements within the advancedcapitalist countries, particularly with respect to theirdistribution regimes. Is it feasible that distributionregimes of relatively poor countries also fall into dis-tinct patterns? By implication or design, studies in

comparative political economy (CPE) and interna-tional political economy (IPE) treat less developedcountries (LDCs) as more or less identical to oneanother or, at the other extreme, as nations marked bytremendous politico-economic diversity. It is startlingthat neither scholars nor policymakers have a clearsense of peer groups among developing nations,outside of broad, amorphous categories such as regionor level of economic development. This analysis chal-lenges long-standing conceptions of LDC politicaleconomies by providing theoretical and empiricalsupport for systematic divergence in their choice setof social welfare policies. The discovery of distinctpatterns of welfare regimes in LDCs presents hithertounknown implications for the influence of domesticpolitics and policies in late twentieth-centuryglobalization.

The gross lack of efforts to investigate common-alities among developing countries has its roots inCPE convergence debates that for decades focusedonly on developed nations. Convergence is defined asthe tendency of societies to grow more alike, todevelop similarities in structures, processes and per-formances (Kerr 1983, 3). From the 1960s until theearly nineties, scholars believed that only postindus-trial societies could experience convergence, since

(successful) industrialization requires a particulararrangement of social and economic forces.1 Theinference was that LDCs were marked by extremedivergence. In other words, developing nationsshould be vastly different from one another becausethey are in the early stages of economic development.

An IPE strand of the convergence debate emergedin the early to mid-nineties that, in time, came toinclude less developed countries more directly. Con-cerns of institutional convergence began assumingcentral importance in the literature, given trendstowards market integration. Regardless of the level ofeconomic development, the issue was whether nationsharmonize their domestic practices and socialarrangements in response to the challenges of growthand prosperity in the global economy. As the debateevolved, the identification of set constellations of pro-

1These analyses follow from stagist or modernization theories of the 1960s.

The Journal of Politics, Vol. 69, No. 2, May 2007, pp. 378396

2007 Southern Political Science Association ISSN 0022-3816

378


2/20


3/20

Organization for Economic Cooperation and Devel-opment (OECD) advance two impressions of LDCpolitical economies: (1) the types are endless;4 and (2)welfare states are precluded because of low economicdevelopment. First, the logic implies that, if nationswith high standards of living exist in a homogenous

world, then countries with low standards of livingmust live in a vastly heterogeneous one. Levelingforces of industrialization are hypothesized to produceconvergence in OECD social structures and policies,such as pluralistic decision making, the ability of thestate to extract resources, and a preponderance ofcommitted industrial workers (for examples, see Formand Bae 1988; Kerr 1983). The existence of welfarestates is also one of the by-products of industrializa-tion. Only nations at high levels of economic develop-ment can form a welfare regime (Cutright 1965;Wilensky 1975). While these arguments are plausible,

CPE scholars have not tested their arguments as theyapply to the LDCs, and IPE takes opposing positions.

International political economy scholars implythat the challenges of growth in a globalizing economyensure significant similarities between the politicaleconomies of the many LDCs, and the existence of anLDC welfare state is implicitly assumed (Avelino,Brown, and Hunter 2005; Cerny 1995; Garrett 2001;Rudra 2002; Wibbels 2006). Since LDCs face similareconomic challenges (e.g., demand for capital, largepools of surplus labor), they are expected to convergeon neoliberal policies for the purposes of attracting

capital and promoting exports. The negative correla-tion between expanding markets and social spendingin LDCs provides confirmation of this hypothesis(Garrett 2001; Rudra 2002; Wibbels 2006).5 However,by focusing on social spending per se, IPE scholarspresuppose the existence of the LDC welfare state

without investigating its particulars. With little senseof the welfare states salient characteristics, it is unclearhow or why less social spending is necessarily associ-ated with an embrace of market-friendly neoliberalpolicies. Leaders may very well engage in low (ordecreasing) social spending while promoting illib-

eral welfare measures, such as public employment orlabor market protections.6 The LDC convergencequestion thus remains unresolved.

Ultimately, the problem in both camps is thatthe systematic divergence hypothesis has not beenexplored in LDCs. The discovery of distinct patterns ofproduction and distribution regimes has more or lessput to rest expectations of convergence in the devel-oped nations, but left open the question as applied toLDCs. Nevertheless, these advancements in the OECDliterature hold important lessons for developing coun-tries, as is often the case. First, studies indicating sys-

tematic divergence in the advanced economies suggestthat the level of development does not necessarilypredetermine the configuration of national politicaleconomies. Both Esping-Andersen (1990) and Halland Soskice (2001) reveal that national politicaleconomies are what determine economic performanceand social well-being, and not the reverse. Second, thedetection of either distinct patterns of distributionregimes or production regimes can confirm the sys-tematic divergence hypothesis. This analysis focuseson distribution regimes to continue engagement withconvergence debates in IPE literature. Finally, as

Esping-Andersen (1990) demonstrates, the theoreticalsubstance of welfare states is of import to politicaleconomy, along with the level of expenditures.

Contemplating SystematicDivergence in LDCs: Patterns of

Welfare Regimes

Resolving the convergence debate in developing coun-tries has significant implications for policy and poli-

tics, particularly given trends in market integration. Ifearly CPE is indeed correct and extreme divergenceprevails, policy decisions of LDC governments arewithout any (extraterritorial) bounds.7 This would

4The literature on developmental states serves as an exception tothe assumption of extreme divergence amongst developing coun-tries. However, the limitation is that this literature effectivelydepicts the political economies of a select few Northeastern Asiancountries; by default, the rest of the developing world falls into a

single residual category defined by the absence of some basic insti-tutional characteristics essential for growth. In other words, in thisliterature, the majority of LDCs are identified on the basis of whatpolitical institutions they do not have rather than what they dohave.

5Social security policies in particular are regarded as poor marketdisciplining devices on labor. The upward pressure on labor costsand dampening effects on work incentives discourage exports. Therevenue generation required for social security through taxation isexpected to discourage both productive and financial capital.Although tests of this convergence thesis in the developing worldare limited, IPE discussions of convergence as applied to nationsmore generally are vast (for examples, see Cerny 1995; Evans 1997;Gill 1995; Steinmo 1994).

6Note that reductions in public employment and labor marketderegulation are basic components of the structural adjustmentprograms (i.e., neoliberal reforms) advocated by the InternationalMonetary Fund and World Bank.

7Recall that proponents of the convergence hypothesis in CPEsuggest that high levels of economic development provide theprimary context to which policy makers respond. This is extra-

380


4/20

impose strong limitations on researchers and policymakers committed to encouraging development inlower-income countries. Exemplars of history, themissions of international financial institutions, andgeneralized policy prescriptions lose persuasion infavor of wait and see. The opposite occurs if IPE

scholars are correct and convergence exists, suggestingthat domestic structures and processes are meaning-less, and policy responds primarily to the laws of inter-national economics. However, if systematic divergencecharacterizes the developing world, policy makers areresponsive to local needs and politics (divergenceexists), as well as some transnational forces, such assurvival in a global economy (it is systematic).

Questioning CPE Convergence Scholars

This analysis challenges the contention of CPE schol-

ars that welfare states are necessarily a postindustrialphenomenon. The historical experience of the OECDnations coupled with specific challenges of twentieth-century globalization have made it impossible forLDC governments to ignore embedded liberalism orcalls to maintain social stability alongside marketexpansion (see Ruggie 1982, 1994).8 The repercussionsof nineteenth-century globalization, which focused onusing state intervention to maintain market-drivenequilibria instead of social protections, are wellknown: domestic unrest; economic breakdown; andinterstate rivalries ultimately leading to World War I

(Polanyi 1944).9

Largely in reaction to this experience,governments of OECD nations in the post-World WarII period formalized their welfare regimes for the pur-poses of social welfare and stability (i.e., twentieth-century globalization).10 It is thus plausible, as Collierand Messick (1975) show, that the successful workingsof welfare systems in advanced economies have pro-vided important precedents for todays LDCs.11 Incontrast, when todays advanced economies firstembarked on the journey to industrialization, no realprecedents for a welfare state existed.

Post-World War II globalization has been accom-panied by new challenges, particularly for late entrantsto the international market, rendering the economicand political costs of ignoring embedded liberalismvery high. First, the magnitude, complexity, andspeed of todays global financial, commodity, and

service market operations carry risks and uncertain-ties to citizens of all nations.12 Compared to theadvanced economies, LDCs are in a position ofmaximum uncertainty, since only a few developingnations can actually influence the markets in whichthey trade and invest (Waterbury 1999). Second, socialreactions to the market are a common thread in bothdeveloped and developing countries. This is evidencedin LDCs by the large number of labor and capitalstrikes in response to the adoption of neoliberal poli-cies. Third, although labor as a class is not strong andsuffers from collective action problems, there are

pockets of labor groups which can and have affectedsocial policies. Fourth, the relatively recent spread ofdemocracy and its link to embedded liberalism shouldnot be underestimated. The expansion of the right tovote puts all those negatively affected by globalizationin a better position to insist that international marketexpansion be moderated with the pursuit of otherobjectives.

In conclusion, nineteenth-century style stateinterventionism in the current era is just as unlikely inthe developing world as it is the OECD nations. Thecontention that embedded liberalism is common

practice among LDCs casts doubt on CPE predictionsregarding (the lack of ) welfare states and extremedivergence.

Questioning IPE Convergence Scholars

The first part of the IPE convergence argument seemsplausible; challenges to growth in a global economyare likely to affect domestic social policy decisions. Buthow convincing is their reasoning that internationalmarket pressures ultimately force universal acceptanceof market-friendly social policies? While some combi-

nation of markets and domestic interventionism forsocial welfare has been common to all countries post-WWII, the different historical, economic, and politicalrealities of LDCs suggest that their national socialsystems will differ not only from the OECD countries,but systematically vary from one another as well. Thisinvestigation rests on the premise that LDCs maintainsome form of capitalist market economy. In all the

territorial in the sense that decision making in the OECD nationsis driven by similar forces and thereby transcends national bound-aries. By inference, then, low levels of economic developmentprovide no specific context to which policy makers respond.

8The International Labor Organization has also been a leadingforce here since 1919.

9Examples of nineteenth-century state intervention are tariffs,access to capital, and encouragement of large-scale industries(Gerschenkron 1962).

10See Pitruzzello 2004 for a discussion of the differences betweennineteenth- and twentieth-century globalization.

11I am grateful to Benjamin Cohen for emphasizing this point.

12See Keohane and Nyes (2000) discussion of what is new (and notnew) about contemporary globalization.

welfare states in developing countries: unique or universal? 381


5/20

countries in the sample, private enterprises exist,and the market remains the principal means ofdistribution.

The capacity to commodify is likely to be the keyfactor differentiating LDC welfare states. This refers tothe degree to which government-backed social policies

ensure that the majority of people depend on wagelabor, with wage levels largely determined by marketforces (see Esping-Andersen 1990). Commodificationin this particular sense does not apply to the OECDssince the workforce is already proletarianized.13

Advanced welfare states in the postwar era haveinstead focused on counterbalancing proletarianiza-tion with decommodification, or permitting peopleto make their living independent of pure market forces(Esping-Andersen 1990).

Esping-Andersen argues that the first step in con-ceptualizing the welfare state involves locating the

primary source of tension that gave rise to its particu-lar political economy, or to the states larger role inmanaging and organizing the economy (1990, 2). Inthe early European experience, proletarianization wasthe major source of conflict (Esping-Andersen 1990;Koo 1990). However, concerns about the absence ofproletarianization, particularly in the postwar era,have been the focus of LDC political economies (Koo1990). This is chiefly because the progressive shift ofthe labor force from primary agricultural activities tosecondary manufacture and tertiary commerce andservices has not occurred as it did in Europe.14 At issue,

then, is not the elimination of internalclass, inequal-ity and privilege as it has been in the OECD nations(Esping-Andersen 1990), but rather minimizing exter-naldivisions between the rich and poor economies byexpanding wage labor and catching up with theindustrialized nations.

Significantly, business as well as labor is depen-dent upon LDC welfare states that focus on commodi-fication. Proletarianization in the current era arguablyrequires relatively greater state intervention. Thedemand for skilled labor has increased, and a minimallevel of education appears to be a prerequisite for

entering todays markets (Blunch and Verner 2000;Feenstra and Hanson 1996; Tendler 2003; Thompson1995).Wood and Ridao-Cano (1999),for example,findthat even in basic manufacturing sectors, workers inLDCs are generally low-skilled (not unskilled). This isin direct contrast to the experiences of early industri-

alizers, where private entrepreneurs needed much lessstate intervention to begin production (Gerschenkron1962). Several scholars have argued that ade-skillingof the workforce occurred in Europe during earlyindustrialization, and literacy rates actually declined(Nicholas and Nicholas 1992; Sanderson 1972; Stone1969).15 Households, churches, and Sunday schools,rather than the state, provided primary education(Nicholas and Nicholas 1992).16 According to the evi-dence presented by Goldin and Katz (1998), thecomplementarity between skill and technology did notbegin until as late as the twentieth century. 17

Point of Divergence

Despite the intense need for LDC governments tofocus on expanding wage labor, some countries in thepostwar era place substantially greater priority ondecommodification prior to full-scale commodifica-tion efforts. First, the latter is politically much moredifficult to achieve in some LDCs because of the mis-trust that emerged towards international markets inthe 1930s.Colonial interference and declining terms oftrade for agricultural exports in that decade hampered

the complementarity (real or perceived) betweeninternational market participation and the rapidexpansion of formal wage labor, at least in the earlystages. Second, precedents set by the experiences of theOECD nations matter (Collier and Messick 1975);pressures on all governments to provide some degreeof protection from market dependence intensified inthe postwar period. Finally, LDC labor is more depen-dent on a decommodifying welfare state than its

13Differing from commodification, proletarianization refers spe-cifically to the outcome, or the successfuldependence of the major-ity of the workforce on (formal) wage labor for survival; a stagelong past in OECD nations.

14See data presented by Erickson and Peppe (1976) that confirmthis trend in OECD countries. See Browning and Roberts (1990)for an alternative argument. In most LDCs, secondary sectoremployment remains limited, while the tertiary sector, distin-guished by large numbers of informal sector workers, has beenforced to absorb much of the rural surplus (Erickson and Peppe1976; Evans and Timberlake 1980; Koo 1990).

15This term refers to the replacement of skilled workers by a large

class of unskilled, subliterate factory operatives. See Nicholas andNicholas (1992).

16Some argue that, because of the laissez-faire tradition, stateshesitatedto intervene in education. Initiativesto do so began in thelate 1800s (Kiesling 1983). To give one important example, the firstreal nonprivate school in England was introduced as late as 1944with the 1944 Education Act. This permitted local authorities toestablish and maintain both primary and secondary schools(Morrish 1970, 83).

17Goldin and Katz describe technology-skill complementarity aswhen the skilled or more educated labor is more complementarywith new technology or physical capital than is unskilled or lesseducated labor (1998, 694).

382


6/20

early European counterparts. The former relies on thestate to represent its needs because workers sufferfrom both persistent collective action problems (seeBellin 2000), and the prolonged absence of a guaran-teed minimum income.18 Developing states areinclined to intervene to provide this minimum income(through public works projects, public employment,labor market protections, etc.), since the transforma-tion of surplus labor into formal wage labor has beenoccurring through the market process at an extremelyslow rate.19

In sum, it is feasible that some developing coun-

tries prioritize protection from markets even beforefull-scale proletarianization has been achieved (i.e., aprotective welfare state). If this is the case, not all LDCswill have productive welfare states which directwelfare efforts primarily towards encouraging wagelabor. The implication is that the relationship betweencommodification and decommodification in LDCsmay not be linear as it has been in the developedeconomies (see Figure 1).

Delineating Different Welfare Regimes inDeveloping Countries

A blueprint for an LDC welfare state that promoteseither commodification or decommodification per senever existed. In the postwar era, referring back to theprimary tension that drives political economies,LDC welfare states took qualitatively different formsdepending on how governments chose to address thelack of proletarianization and pursue their primaryobjective of creating a modern industrial order.20 Gov-ernment intervention in the economy was guided by

one of two goals: making firms internationally com-petitive or insulating firms from international compe-tition.21 Why political leaders pursued one strategyover another is based on a whole host of factors andexplored elsewhere (see Waterbury 1999).22 Central tothis investigation is that ruling elites pursued socialbenefits compatible with the chosen development

18For many of the early industrializers, agriculture played a strong

role in industrialization, while in the LDCs, as Bates (1981) argues,the popular strategy of rapid industrialization often came at a costto the efficiency of the agricultural sector. The end result is a largesurplus labor economy in which the absorption rate of labor ispersistently low. This is not to deny that much of Europe had alarge surplus (rural) population when welfare policies were firstadopted. However, as Pandit and Cassetti (1989) show, the leveland rate of absorption of labor into the manufacturing sector hasbeen considerably slower in the developing world compared to thenow developed countries. This has been further exacerbated bytrends in the twentieth century towards greater mechanization(Baer and Herve 1966).

19These government measures are decommodifying in the sensethat workers become less dependent on the market.

20Kurtz (2002), Haggard and Kaufman (n.d.), Kuruvilla (1996),and Huber (1996) make related arguments about the connectionbetween industrialization strategy and social policies. It is Batess(1981) proposition that an LDC governments primary social

objective is to create a modern industrial order.21While state intervention could be directed towards both goals,historically LDCs have tended to advance more in one area whileretreating in the other.

22One possible objection is that the causal arrows could be reversedand high levels of human capital (commodification) influencedLDCs to be more accepting of international market participation.While there might be some merit to this claim, a relatively highlyeducated workforce was no guarantee that LDCs would pursueoutward-oriented strategy. Argentina and Uruguay, for example,had the highest rates of literacy (91%) in the developing world inthe 1960s; however, they pursued the alternate strategy andrejected emphasis upon international market participation.

FIGURE 1 The Era of Embedded Liberalism: Welfare States in Developed and Developing Countries*

*Note that the OECD decommodification rankings (low, medium, high) are from Esping-Andersen (1990: 52).



7/20

strategy, and key to this compatibility was the coopta-tion of potentially powerful groups.23

Protective welfare states have roots in a politicaleconomy that historically eschewed emphasis oninternational markets and ultimately focused govern-ment efforts on decommodification. The focus upon

insulating domestic firms from international compe-tition allowed politicians to exercise maximum discre-tion and control over the economy, particularly in theearly stages. Absent the threat of international marketcompetition and pressures of cost containment, rulerscould provide allowances to both workers and firms inthe major industrializing sectors. Politicians had theflexibility to employ direct and immediate benefits toworkers contrary to employers economic interests,mostly because the latter were compensated throughother means (e.g., tariffs, subsidies).

Protective welfare states are consequently a

curious fusion of elements of socialism and conserva-tism. Like the OECD social democratic model, protec-tive welfare states have a strong distrust of markets.Both regime types claim to detest the dehumanizingeffects of unfettered capitalism. Commonalities withthe statist variant of the conservative model also exist,particularly with its emphasis on the preservation ofauthority (see Huber 1996). Conservative forces in theprotective welfare states fear that internationalmarkets can destroy their power and privilege.24

Leaders thus prefer social rights that simultaneouslypromote loyalty to the state and create divisions

among social groups such as labor and business. Full-scale commodification certainly would make it diffi-cult for the state to be the most dominant factor in theexpanding international economy.

Yet protective welfare states in developing coun-tries are distinct from both the social democratic andconservative welfare models in that emphasis ondecommodification occurred prior to proletarianiza-tion and, consequently, social rights have beendirected towards a small clientele.Welfare policies maynot be redistributive and beneficent, even though theyare often thought of in this way. Titmuss long ago

stated that when we use the term social policy wemust not [. . .] automatically react by investing it witha halo of altruism, concern for others, concern aboutequality and so on (1965, 27). Before proletarianiza-

tion occurs, making rights conditional upon labormarket attachment, some work performance andactuarialism results in welfare benefits for only a small,privileged stratum.

Productive welfare states, in contrast, prioritizecommodification, and initially evolved from systems

which actively encouraged participation in exportmarkets. The goal of encouraging international com-petitiveness of domestic firms creates an emphasis oncost containment and requires governments to surren-der some control over the economy.25 The range ofsocial policies is then much more limited as rulers areconstrained from pursuing worker benefits that areindependent of employer interests. In other words,those policies prevail that can successfully serve theinterests of workers and capital simultaneously.

Productive welfare states thereby share certain ele-ments with the liberal model. In contrast to its coun-

terpart, this regime type embraces some of thenineteenth-century liberal enthusiasm for the marketand self-reliance. The particular property of the liberalparadigm that ultimately comes to distinguish theproductive welfare state is the emphasis uponstrengthening the commodity status of labor in a glo-balizing economy. At the same time, the fundamentalpoint of departure from the liberal model is that thestate-market relationship is complementary ratherthan adversarial. Considerable public interventionaims to enhance international market participation.Social policies are circumscribed by this goal and

promote worker loyalty without hindering businessactivity. In contrast, the OECD nations were neverdriven to be productive welfare states per se becausethe proletarianization occurred gradually, spanningover two centuries, and required relatively less stateintervention.26

The tension in this model lies in reconciling thepush for dependence on wage labor with demands foremancipation from the market. As argued earlier, anexcessive focus on commodification puts system sur-vival at stake. Furthermore, although labor marketdualism will be less prominent given government

measures to provide capital with an abundance ofproductive workers, a clear class hierarchy still exists.

23This was feasible since the planned nature of industrializationmeant that governments more or less knew who the winnerswould be.

24See, for example, Esping-Andersens (1990, 41) discussion ofstatism and how it was feared that capitalism would destroy powerand privilege.

25For example, even if governments intervene in the setting ofprices and wages, their decisions will be constrained by consider-ations of international market performance.

26Recall that proletarianization was supported initially by privateand nongovernmental institutions (see note 14). Much of theOECD welfare states literature takes commodification as a given,and focuses instead on government attention to developing ahighly skilled workforce and training systems. See, for example,Hall and Soskice (2001).

384


8/20

Efforts to keep pace with an already industrializedinternational economy results in the rapid, simulta-neous expansion of white-collar and blue-collar work(see Koo 1990). The ongoing controversy then is theextent to which protective welfare policies can beselectively employed to ensure system longevity. Gov-

ernments of productive welfare states can attempt toaddress this problem through repression or by offeringsome minimum level of protective social benefits,usually for white-collar workers. While it is feasiblethat a protective welfare state could eventually evolveinto a productive welfare state, the reverse is unlikely tooccur.27

Cluster Analysis: TestingContrasting Hypotheses

Do developing countries display convergence orextreme divergence? Or, as this analysis posits, are theLDCs characterized by systematic divergence? Is itpossible to discern a distinct statistical pattern thatlends support to the idea that different welfare modelsin the developing world do exist and that they corre-spond to the protective and productive typology out-lined above?

Cluster analysis is a quantitative method that canhelp discriminate between the above hypotheses. Byallowing the classification of objects into relativelyhomogenous groups, this method can determine thenumber of LDC distribution regimes, if any. Eachgroup identified by cluster analysis is as internallyhomogenous as possible, but as distinct as possiblefrom all other groups. The technique is applied to findsimilarities between units under classification, ratherthan interrelationships among variables (factor analy-sis). The objective is to group n units into r clusterswhere ris much smaller than n (Lewis-Beck, Bryman,and Liao 2004). Cluster analysis is one of the mostpopular means of constructing a typology. Although itoriginated in psychology and anthropology, it has

since become a valuable tool in biology, geography,political science, sociology, economics, andmathematics.

To begin the search for natural groupings in thedata, a clustering method must be selected. Partition,or nonhierarchical, methods do not apply here, sincethe number of clusters is not known a priori. Instead,

I apply the hierarchical agglomerative linkage method,which considers each observation as a separate group.Next, the agglomerative algorithm considers N(N-1)/2 possible fusions of observations to find andcombine the closest two groups. This process repeatsitself until all observations belong to a single group,

and a hierarchy of clusters is created. To begin thisprocedure, however, computation of a similarity ordistance matrix between the entities is required. Iapply the most common representation of distance, orthe Euclidean distance (Aldenderfer and Blashfield1984; Everitt 1974), to calculate the distance betweenthe units. To give a simple example, if two cases areidentical, then the Euclidean distance between themwill be zero. The final product is a tree-like represen-tation of the data, or dendrogram, which illustrates thesuccessful fusion of countries. It is completed onlywhen all the countries are in one group.

Several agglomerative linkage methods exist incluster analysis. The most common are single linkage,complete linkage, average linkage, and Wardsmethod. These represent different mathematical pro-cedures to calculate the distance between clusters.However, following standard practice in the social sci-ences, and given the disadvantages of single and com-plete linkage (see Panel on Discriminant Analysis,Classification and Clustering 1989), Wards methodis used and the weighted average linkage method isapplied as a robustness check.28 Wards method isdesigned to optimize the minimum variance within

clusters and works by joining groups that result in thelowest increase in the error sum of squares (Aldender-fer and Blashfield 1984; Ward 1963). At each stage,after the union of every possible pair of clusters isconsidered, the method fuses the two clusters whoseincrease in the total within-cluster error sum ofsquares is minimal. Several studies have observed that,in comparison to the above-mentioned alternatives,Wards method ranks first in the recovery of true clus-ters (Blashfield 1976; Tidmore and Turner 1983).

Cluster analysis will confirm the systematic diver-gence hypothesis if it reveals a distinct number of LDC

welfare regimes corresponding to the productive-protective dichotomy. However, if early CPE specula-tions of extreme divergence are correct, then clusteranalysis will demonstrate no identifiable pattern. Thenumber of clusters will be large, far outnumberingthe two patterns predicted in this analysis. Finally, thethird possibility, IPEs predictions of convergence, will

27In other words, if productive welfare states are successful, theycannot become protective, since the latter emphasizes decom-modification before commodification is complete.

28I use the weighted average linkage method so that if some of theclusters are small, the results will not be biased. This method givesequal weight to groups with small numbers of observations.



9/20

be confirmed if all developing nations fall into onecluster. At this point, however, the IPE literature lendsitself to significant ambiguity. Are LDCs likely to con-verge upon productive or protective welfare states?Recall that the central process underlying convergencetendencies is the challenge of growth in a global

economy. On the one hand, most IPE scholars implic-itly assume international market pressures will driveLDCs towards embracing social policies most similarto OECD liberal welfare states, which would result inproductive welfare states. But at the same time, as thisanalysis points out, since the 1930s many LDCs havefound that insulating themselves from internationalmarkets has been the best way to respond to the chal-lenges of growth in the postwar era. Consequently, it isfeasible that LDCs may instead have evolved into pro-tective welfare states.

Variables

The primary goal is to assess welfare priorities in LDCsand whether they follow the predicted pattern of privi-leging commodification or decommodification.Simply applying the most common methodexamin-ing government budget prioritiesis insufficient indeveloping nations for three reasons. First, to be con-sistent with Esping-Andersen, expenditures areepiphenomenal to the theoretical substance of welfarestates (1990, 19). Second, as the World Bank (1990)and the International Labor Organization (ILO; see

Figueiredo and Shaheed 1995) have pointed out, gov-ernments of developing nations often employ lessresource-intensive means to protect their workers,suchas labor market policies and public employment.29

Third, it is impossible to know whether spending isserving the desired goals, or serving clientelistic needs,as is frequently the case in LDCs (see Nelson 1999;World Development Report 2004). This issue is par-ticularly salient in evaluating goals for commodifica-tion. If government spending is high, but the allocatedresources are misused and have little effect on improv-ing the health and education of (potential) workers,

then the LDC cannot be a productive welfare state.The difficulty here is the dearth and reliability of

data that can capture such occurrences across LDCs.One solution, although imperfect, is to include a com-bination of policy, spending, and outcome variables.The other alternative is to wait for more effective insti-tutions to evolve and, as a consequence, more reliable

data. In such a case, the hazard is that policy makersand citizens of LDCs are likely to face the conse-quences of a vicious cycle involving insufficient data,neglect of important research and the persistence ofweak, ineffective institutions. Put simply, more effec-tive welfare institutions may be dependent upon

analyses such as this one,which attempt to make use ofavailable data.Exercising the first option, I build on the insights of

the most renowned experts of welfare in both devel-oped and developing countries, Esping-Andersen(1990) and Dreze and Sen (1989) respectively, to deter-mine the most appropriate indicators of LDC welfarestates.30 Spending and outcome variables are used tocapture extensive public efforts aimed directly atexpanding the basic capabilities of the population tosuit wage-labor markets. An emphasis on decommodi-fication is detected by pervasive policies and govern-

ment spending geared towards protecting individualsfrom the risks and uncertainties of the market. Protec-tive welfare policies are then more commonly associ-ated with (but not limited to) social-insurance typevariables.According to Esping-Andersen, decommodi-fication ultimately strengthens the worker andweakens the absolute authority of the employer (1990,22). While it is reasonable to expect some overlapbetween productive and protective variables inpractice, the division is driven by two very differentlogics and produces distinct sociopolitical outcomes(Dreze and Sen 1989; Esping-Andersen 1990).31 See

Appendix A (online at http://www.journalofpolitics.org) for more detailed explanations of the data sourcesand variables discussed below.

Variables Representing ProductiveWelfare States

Degrees of commodification are determined by thelevel of public investment in primary and secondaryeducation, and basic healthcare, as well as literacyrates, rates of infant mortality, and the percentage ofinfants vaccinated against diphtheria, pertussis, and

29Arguably, while these policies command relatively less govern-ment resources, they may ultimately be more expensive for thelarger society in LDCs.

30The productive-protective dichotomy builds on Dreze and Sens(1989, 16) distinction between promotion and protection.

31For example, some scholars argue that protective welfare benefits(e.g., pensions, labor protections) make workers more productivein the marketplace. The productive-protective dichotomy dealswith ideal types and, in reality, we should expect some overlap, ormodal tendencies of population distributions, between the polaralternatives.

386
http://www.journalofpolitics/http://www.journalofpolitics/


10/20

tetanus (DPT).32 I use the conventional method ofobserving government expenditures on education andhealth as a proportion of the total public budget sincethis is the most precise measure of government com-mitment(see Rudra and Haggard 2005). To put this inperspective, if education spending, for example, is

measured instead as a percent of GDP, countries likeKorea and Singapore fall in the same percentile rangeas developing countries such as Mali, Malawi, andLiberia. But measuring spending relative to the totalbudget more clearly reveals qualitative differencesbetween LDC welfare states, and it unveils a consider-able variance between countries. As a compromise, Iapply spending as a percentage of GDP and per capitaspending as robustness checks.

The outcome variables (literacy, mortality, andimmunization rates) help the analysis for two reasons.First, outcome variables reflect past policies. Nations

with a legacy of responsiveness to internationalmarkets are likely to have pursued market-promotingsocial policies at an early date. In other words, if LDCshave highoutcomes, it suggests previous leaders haveemphasized commodification. This is particularly rel-evant since, as mentioned earlier, once the process ofproletarianization is complete, it cannot thereafterbecome a protective welfare state. Second, outcomevariables can help see beyond the numbers. Publicofficials might be engaging in high levels of clien-telism, using resources for patronage purposes ratherthan affecting positive outcomes. From this perspec-

tive, high levels of spending alongside persistently lowoutcomes are telling, indicating weak governmentcommitment towards a productive welfare state.However, since other factors might determine out-comes in addition to government spending (efforts ofnongovernmental organizations, GDP, etc.), I drop allof the outcome variables and run the model again as acheck. The cluster groupings are almost identical,although, as predicted, the differences between clus-ters are less obvious but still statistically significant(seeAppendix B at http://www.journalofpolitics.org).

Variables Representing ProtectiveWelfare States

Five variables capture the extent to which LDC gov-ernments aim to protect workers from market risksand uncertainties: the extent of public employment;spending on social security and welfare (pensions,family allowances, unemployment, old age, sickness

and disability); housing subsidies; labor market pro-tections; and investment in tertiary education. As afinal point, while means-tested poor relief should alsobe included under protective welfare policies, cross-country comparable data are virtually nonexistent.33

Public employment is one of the most pervasive

methods of market protection in LDCs (see Rodrik2000). In some cases, it provides short-term security inearnings, such as hiring for public work projects, butin the larger number of instances, the public sectorprovidessecure jobs (Rodrik 2000; Gelb, Knight, andSabot 1991). As Robinson states:

The permanent status that many, in some cases themajority of, civil service employees enjoy means thatapart from dismissal for grave disciplinary reasons theyare assured of employment until retirement, providinga degree of protection and privilege not found in theprivate sector. (Robinson as quoted in Rodrik 2000,231)

Given that cross-country comparable data on publicemployment is extremely sparse, the percentage ofgovernment budget spent on employee wages andsalaries is used to estimate this variable.

Analyzing spending on social security andhousing as means to guard against income risks iscommon in the broader welfare literature. In LDCs,pensions tend to be the largest component of thisspending, ensuring a steady flow of income over alifetime, regardless of market shocks and uncertain-ties. Unemployment, family allowances, and sickness

protections, though less common, provide security inthe face of short-term absence from the market.Housing subsidies also help stabilize incomes(Chapman and Austin 2002; Renaud 1984). Higher-skilled workers, and especially civil servants, oftenreceive housing as part of their wage package.

Labor market protections are common LDCwelfare measures that help guarantee incomes byplacing institutionalized restrictions on firms hiringand firing decisions (Betcherman, Luinstra, andOgawa 2001). Data for such protections, however, arebeset with problems (Rama and Artecona 2002). One

reliable, albeit crude, indicator is the ratification ofILO conventions by nations. Enforcement standardsare relatively nil, and ratifications do not necessarilytranslate into policy innovations. However, recentresearch has shown that ratification has a significanteffect on labor costs (Rodrik 1996) and can reflect

32Note that immunization against measles is also included anddoes not affect the cluster groupings (not shown here).

33This should not affect the findings from this analysis, becausesuch policies are usually allotted a very small budget in LDCs(Subbarao et al. 1997). Targeting options in LDCs are expensivedue to the administration costs of identifying, reaching, and moni-toring the target population (Grosh 1994).

http://www.journalofpolitics.org/http://www.journalofpolitics.org/


11/20

internal political factors such as government prefer-ences or the power of left-wing parties (Brookmann2001). It is fair to assume then that labor market pro-tections will be relatively low in countries that haveratified a very low number of ILO conventions (e.g.,United States, Korea, Singapore).34

Lastly, the provision of free or heavily subsidizedtertiary education when primary or secondary leveleducation access is less than universal awards strongpromise of future income security to those who haveaccess to the former (see World Development Report2003).35 Particularly since high-skilled labor is in rela-tively high demand yet in scarce supply in the majorityof LDCs, such workers can secure great advantages inthe bargaining process.

Analysis Results

The results of the cluster analysis are shown in Tables 1and 2. Because cluster analysis has a low tolerance formissing data, the final sample size is 32. This sample isstill marked by regional and economic diversity, and

thus remains fairly representative of LDCs. Each vari-able represents data averages for 1990 through 1997(the latest date to which cross-national data is avail-able for a large number of LDCs). Results are analyzedin the following three steps: (1) assess how manycluster groups exist; (2) determine which countries fall

into each cluster; and (3) evaluate the characteristicsof each cluster and its member countries to assesswhether or not they confirm systematic divergence.

The first critical step is to determine the numberof clusters present in the data. The number is a ques-tion of particular interest, since it could providesupport for CPE, IPE, or systematic divergence. If, asimplied by CPE, no cluster structure is shown, thenefforts to identify a few broad categories of welfarestates among LDCs are meaningless. The distinctionsbetween countries are greater than the similaritiesbetween them. At the other extreme, a single cluster

would imply that developing countries as a whole area relatively homogenous group. The IPE view thenprevails, and state intervention to create a modernindustrial order has had more or less the same welfareconsequences in all developing countries.

Over 30 stopping rules (procedures to determinethe number of clusters in a data set) are applicable incluster analysis. Fortunately, Milligan and Cooper(1985) conducted a well-known study to distinguishamong them and assess which criteria provide themost valid test for the existence of a cluster. Theirexperiment suggests that the Duda and Hart (1973)

procedure was one of the best. Duda and Hart Je (2)/Je(1) estimates are presented in Appendix C (http://www.journalofpolitics.org).36 Results from thismethod surprisingly indicate that a three-group solu-tion is most distinct in this hierarchical cluster analy-sis, contrary to the expected two ideal regime types.The next step is to determine which countries are ineach cluster. Table 1 presents the country members ofthe three clusters. This pattern reveals that, althoughregion plays a role, it is not a predominant factor in thegroupings. While only Latin American countries com-prise cluster 2, the members of clusters 1 and 3 repre-

34As a robustness check, I also run labor regulation data con-structed by Botero et al. (2004). The drawbacks for using this dataset are that it focuses on one year, and data is missing for threeLDCs from the sample. Excluding the major outliers, the correla-tion between the ILO data in the sample and Botero et al. is 0.65.

The final results differ in that Panama, Paraguay, Greece, Colom-bia, and Thailand fall from cluster 1 (productive) to cluster 2(dual).

35Demographics play an important role in determining levels ofpublic spending on education and social security. Note that, inaddition to assessing the different levels of education spending(primary, tertiary) as a percentage of total government expendi-tures, they are also measured as a percentage of GDP, and spendingper student relative to GDP per capita. Although the final clusterresults are minimally affected by the alternative specifications, thelatter is emphasized since it is the only measure that takes demo-graphics into account. Several countries in Africa and South Asia,for instance, show average levels of spending on primary educa-tion. Yet because these LDCs have the highest growth rates ofschool-age population, the number of children actually benefitingfrom state assistance is quite small, and the lack of funds is creatingan education crisis. Evaluating LDCs on the basis of per studentspending provides a more accurate assessment of commitment toprimary education. Zambia, Bangladesh, and Malawi are excellentexamples. This measure also effectively captures disproportionatespending on small populations of students enrolled in tertiaryeducation. For social security and welfare, however, controlling fornumber of beneficiaries wasmore complex, since thedata does nottell us the number of aged persons receiving these benefits. Inaddition, this category is not limited to pensions. Nonetheless, toget a general sense of the impact of elderly demographics, a vari-able is created by dividing the social security and welfare data bythe proportion of aged persons over 65. The results are verysimilar, with only Panama dropping from cluster 1 to cluster 2.

36Duda and Harts ratio criterion is Je (2), which is the sum ofsquared errors within a cluster when the data are broken into twoclusters. Je (1) provides the squared errors when one cluster exists.The three-group solution is most distinct here since the sum ofsquared errors (Je (1)) increases substantially in the four-groupsolution. The conventional rule for deciding the number of groupsis to determine the largest Je (2)/Je (1) value (.6622) that corre-sponds to a low pseudo T-squared value (5.1) and has a higherT-squared value above and below it.

388
http://www.journalofpolitics.org/http://www.journalofpolitics.org/http://www.journalofpolitics.org/http://www.journalofpolitics.org/http://www.journalofpolitics.org/


12/20

sent Africa, the Middle East, East Asia,37 LatinAmerica, and South Asia. Income effects appear toplay a relatively larger role, although, again, not a deci-sive one. Cluster 3 contains only low-income andlower-middle income countries. However, cluster 1reveals a more economically diverse set of countries,ranging from lower-middle income to high incomeLDCs. This finding shows that poorer nations can alsosuccessfully promote commodification. Cluster 2similarly contains both low-middle and high-middle

income countries.The existence of three clusters fundamentally

challenges the extreme divergence and convergencehypotheses. The next logical question is whether thestatistical analysis supports systematic divergence, orpredictions of the two ideal typesproductive andprotective welfare states. To assess this, the clusters areranked according to their levels of welfare effortstowards protection and production. Decile data arecomputed for each welfare variable, and then eachcluster (and country) is ranked from 1 to 10. Forexample, the first decile is the point with 10% of the

data below it and 90% above it. It is given the lowestscore of 1. The ninth decile is the point with 90% ofthe data below it, while the score given to values withinthe top 10% is 10. Table 2 displays these values. Thegreatest weight is placed on the cluster averages, sincethe statistical procedure uses algorithms to differenti-ate the most homogenous groups. It is noteworthy that

differences between deciles tend be quite significant.38

The average for each country within the cluster isimportant, but each welfare category contains infor-mation that should not be overlooked. See AppendixD (http://www.journalofpolitics.org) for a graphicalrepresentation of the results (dendrogram).

Focusing on the cluster averages, several patternsemerge. Clusters 1 and 3 appear to favor the produc-tive and protective components of welfare, respec-tively. Cluster 2, in contrast, favors neither welfare statecategory. This discovery reveals that some LDC welfarestates take dual roles in the postwar economy, raisingquestions about whether dual welfare state status istransitory. A detailed breakdown of the clusters isgiven below.

Cluster 1 clearly privileges commodification overdecommodification. As would be expected, several ofthe East Asian economies, as well as some Latin

American countries noted for their emphasis on edu-cation (e.g., Costa Rica), fall into this category. Theaverage scores for commodification are higher thanthe average scores for decommodification in most ofthe member countries. Panama and Paraguay appearto be anomalies, since their scores do not appear toreflect the prioritizing of productive welfare activities.However, this turns out not to be too surprising for, aswe shall see, further robustness checks reveal that thesetwo countries (along with Greece) appear to be sensi-tive to model specification and fluctuate between clus-ters one and two.

In cluster 3, empirical evidence for the LDCwelfare paradox is highly suggestive: poorer countries,which arguably need productive welfare states themost, appear to be expending the least effort towardsthis goal. Attention to housing and tertiary educationseems to factor in most prominently in the protectivewelfare states. The outcome variables are telling. Forseveral LDCs (e.g., Egypt, Lesotho, Morocco), despitetheir high spending on primary education, literacyrates remain low. This suggests that funds are eitherbeing used for clientelistic purposes or are simplyincommensurate with the level of need. Health spend-

ing also appears to be regressive in relation to outcomevariables. On the other hand, several LDCs in thiscategory rank in the top percentiles for protectivecategories such as wages and salaries and tertiaryeducation.

Finally, cluster 2 appears to place emphasis onproductive and protective activities, yet average scores

37Indonesia is originally included in the cluster 3 but has to bedropped from the analysis because data necessary for robustnesschecks are missing.

38For example, LDCs falling in the sixth decile for primary educa-tion spend almost 30% more per student per capita than LDCs inthe fifth decile.

TABLE 1 Cluster Groupings

Cluster 1 Cluster 2 Cluster 3

Chile Argentina Bolivia

Colombia Brazil Dominican RepublicCosta Rica Mexico Egypt

Cyprus Uruguay El SalvadorGreece India

Israel Iran, Islamic RepKorea, Rep. LesothoKuwait MoroccoMalaysia Tunisia

Mauritius Turkey Panama ZambiaParaguay ZimbabweSingapore

Sri LankaThailandTrinidad and Tobago

http://www.journalofpolitics.org/http://www.journalofpolitics.org/


13/20

TABLE2

ClusterAnalysis*

Country

Commodification

Country

Average

Decommodification

Country

Average

Imm

unization

Infant

Mortality

Literacy

Primary+

Secondary

Ed

ucation

Sp

ending

Health

Spending

Health

Spending

ILO

Conventions

Wagesand

Salaries

Spending

SS&

Welffare

Spending

Tertiary

Education

Spending

Cluster1(Productive)

Chile

9

8

8

2

10

7

7

7

3

10

2

6

Colombia

2

6

6

7

9

6

5

8

2

4

3

4

CostaRica

7

8

8

5

10

8

1

7

10

8

5

6

Cyprus

10

9

9

9

5

8

6

7

6

9

1

6

Greece

4

9

9

3

7

6

3

9

4

8

2

5

Israel

10

10

7

9

6

8

8

6

1

9

4

6

KoreaRep.

5

10

10

6

1

6

4

1

1

5

1

2

Kuwait

7

9

3

8

3

6

6

3

4

5

9

5

Malaysia

8

8

5

8

4

7

9

3

5

3

8

6

Mauritius

6

7

5

7

8

7

7

4

7

7

9

7

Panama

5

6

7

4

10

6

6

9

9

9

4

7

Paraguay

1

5

7

2

4

4

5

5

9

6

7

6

Singapore

8

10

6

5

5

7

9

3

6

1

5

5

SriLanka

7

7

6

7

3

6

3

4

3

7

8

5

Thailand

8

5

8

6

8

7

6

2

7

1

3

4

TrinidadandTobago

6

7

10

3

8

7

10

2

9

7

7

7

ClusterAverage

6

8

8

6

6

7

6

5

5

6

5

5

Cluster2

(Dual)Argentina

3

6

9

6

1

5

3

9

4

10

2

6

Brazil

2

4

5

2

5

4

1

10

1

9

5

5

Mexico

5

5

6

1

2

4

4

10

4

8

3

6

Uruguay

9

7

10

4

4

7

1

10

2

10

3

5

ClusterAverage

5

6

8

3

3

5

2

10

3

9

3

5

Cluster3(Protective)

Bolivia

1

1

4

1

3

2

2

6

7

8

6

6

DominicanRepublic

1

3

4

1

9

4

10

4

5

2

1

4

EgyptArabRep.

4

2

1

8

2

3

8

8

4

4

6

6

El.Salvador

3

4

2

1

9

4

5

1

10

3

1

4

India

1

2

1

4

1

2

8

5

1

2

8

5

IranIslamicRep.

10

4

2

3

7

5

9

1

10

5

7

6

Lesotho

3

1

3

10

9

5

5

1

7

1

10

5

Morocco

6

2

1

10

2

4

2

6

8

4

8

6

Tunisia

9

5

1

9

6

6

7

8

7

6

9

7

Turkey

2

3

4

5

3

3

3

5

9

2

5

5

Zambia

4

1

2

2

6

3

2

6

2

1

10

4

Zimbabwe

6

3

5

10

6

3

10

1

8

5

10

7

ClusterAverage

4

3

3

5

5

4

6

4

7

4

7

6

*Clusteranalysisresultsusingwardslin

kagecommandinIntercooledStata8.

StoppingruleisDudaandHart.

Averagesareroundedtothenearestintegertofacilitatecomparability.Housing,

health,socialsecurityandwelfarearerankedaccordingtopercentageoftotalgovernmentspending.

Educationvariablesar

erankedasspendingperstudent(referto

endnote35).Notethat,

asexplainedlaterinthetext,Panama,

Paraguay,andGreecesmembershipinclu

ster1isnotrobust.

390


14/20

for both welfare categories are moderate (i.e., five).This cluster is more appropriately labeled a weak dualwelfare state, since thesecountries placemore emphasison the proletarianization process than the protectivewelfare states, but significantly less than the productivewelfare states. In terms of commodification, the differ-

ence between cluster 2 and cluster 1 is that both healthand education are stressed in the latter.39 Uruguay is anexception. However, its level of health spending is lowrelative to cluster 1. Brazils profile is also distinct inthat, although education spending is low, and outcomevariables are not as high as other members of cluster 2,its literacy rates outrank similar middle-income LDCsin cluster 3. Most striking is that, on the protective side,cluster 2 ranks in the highest percentile for labor pro-tections (ilocnv) and social security and welfare.On theother hand, average scores for housing, wages and sala-ries, and tertiary spending are considerably lower than

in clusters 1 and 3.The existence of cluster 2 is an important revela-

tion. Based on the theoretical discussion, we canexpect that governments of weak dual welfare states inthe early postwar period were not completely hostileto international markets. It is certainly possible to beprimarily inward-oriented but, at the same time,encourage some export competitiveness. Cluster 2,then, represents a combination of the two ideal regimetypes: social policies that respond to the demands ofcapital and the needs of labor groups. Consequently,relative to the other two clusters, we might expect

heightened political competition for scarce publicresources. Partisan politics, for example, may bevibrant in these countries. One optimistic scenario is ifpartisan politics can successfully steer greater produc-tive welfare efforts, they can offset the tendencytowards elitism engendered by early decommodifica-tion policies. This raises questions about the transitorynature of this cluster; weak dual welfare states couldget mired in partisan politics that ultimately perpetu-ate the status quo (if capital and/or protected laborwins) or they could move incrementally towards pro-ductive welfare status (if structurally unemployed

labor groups win).

Robustness Checks

Do the cluster results hold up to changes in the condi-tioning information? Results for the cluster groupingsand member countries are highly robust to threeimportant changes. First, I run the analysis using an

alternative to Wards method. One common problemassociated with Ward is that it tends to be heavilyinfluenced by outliers (Ketchen and Shook 1996). Tocheck this, I use instead the weighted average method,which gives groups equal weighting in determining thecombined group regardless of the number of observa-

tions in each group. Given that differing clusteringmethods most often produce different results, Lorr(1983) suggests that similar results from two distinctmethods provide great confidence that the underlyingstructure is being recovered. As a second and thirdrobustness check, I substitute the welfare variablesmeasured relative to GDP and GDP per capita in placeof those measured as a proportion of total publicexpenditures. With the exceptions of Greece, Paraguayand Panama, which fall into cluster 2 (instead of cluster1), the results are identical in both models.

Initial Interpretation of the Results

The statistical estimates suggest that developing coun-tries tend to favor productive or protective welfarestates. Scholars from a variety of disciplines have longrecognized the intrinsic and instrumental values ofboth productive and protective types of social legisla-tion (see, in particular, Dreze and Sen 1989). So why,then, the ultimate trade-off between commodificationor decommodification efforts in LDCs? Close atten-tion to how historical legacies of managing state-

international market tensions have affected welfarestates sheds some insights on these results. Building onthe neoclassicalpolitical economy(NPE)and historicalinstitutionalist literatures, it can be understood that theinitial choice of developmentstrategy and complemen-tary welfare policies create distributional coalitions,which thereafter have a vested interest in maintainingexisting institutions and reinforcing them.40 Thisanalysis thus presents the possibility that institutionalcontinuity is linked to the role of positive feedbackeffects from the original distribution regimes.

The NPE approach maintains that state interven-tion encourages the formation of narrow interestgroups that engage in rent-seeking behavior.41 Gov-

39Notice that, in cluster 1, the spending or outcome variables (orboth) in health and education are high.

40This proposition derives from arguments that state interventioncreates distributional coalitions (see Colander 1984), and alsobuilds on the institutionalist theories of path dependence (seeThelen 1999, 2004).

41Rent-seeking refers to lobbying activities triggered by differentlicensing practices of governments. The increased income gains ofthe beneficiary occur at a loss to the greater society. See, forexample, Colander (1984).



15/20

ernment intervention in protective welfare states ini-tially creates social policies that cater to the groupsempowered (directly and indirectly) by minimal inter-national market exposure (i.e., workers in the civilservices, military, urban formal sector and salariedworkers). These distributional coalitions make itincreasingly difficult for the government to engage inthe significant amount of redistribution required topromote commodification. Productive welfare states,on the other hand, introduce benefits acceptable to

employers struggling to compete in the internationaleconomy. Demands for greater labor benefits are sub-sequently met with stiff political resistance. Leadersare ultimately loath to pursue policies that alienatetheir traditional support groups and increase socialinstability. As a consequence, this self-reinforcingprocess suggests that once welfare regimes are institu-tionalized, actors and interests may undergird theirexistence.42

Testing the precise causal relationship linkingindustrialization strategies, welfare regime types, dis-tributional coalitions, and path dependence is beyond

the reach of this paper. However, one way to assess ifthere is some link between LDCs initial decisionsregarding extent of participation in internationalmarkets and the welfare regimes that evolve (andpersist) is to compare early development strategieswith the recent (1990s) cluster groupings. Signs ofsuch a connection can be taken as a preliminary indi-

cation that social actors who benefited from the origi-nal welfare arrangements made reversals increasinglyunlikely.

To get some sense of initial postwar developmentstrategies, I examine the level of manufactured exports(as a percentage of GDP) in each country at the earli-est dates available and compare this to their 1990scommodification-decommodification scores.43 Aftercrises erupted in many LDCs following initial experi-mentation with import substitution, most settled

upon their distinct industrialization strategies by thelate 60s and early 70s. Unfortunately, export data fromthe 1970s is the earliest available for most developingcountries. Economies that focused on orienting firmstowards international markets are expected to reflecthigh levels of manufactured exports. Figure 2 lendssupport to the assertion that more inward-orientedLDCs (low manufactured exports) in the earlierdecades tend towards protective welfare regimes (lowcommodification scores) in the present. The chartthus provides first indications of a connectionbetween early development strategies, the implemen-

tation of (initially) compatible social policies, and thedistributional coalitions that evolve to defend it. Ofcourse, correlations do not account for control vari-

42This path dependence can be disrupted by significant events suchas repressive dictatorships or economic crises. See, for example,Collier and Collier (1991).

43I focus on manufactured export ratios instead of trade ratios toobtain a more precise indicator of industrialization strategy. Forinstance, LDCs that export primary products but adopt inward-oriented industrialization strategies have high trade ratiosthat would make them appear outward-oriented. Thecommodification-decommodification scores were calculated bysubtracting each countrys decommodification score from itscommodification score in Table 2.

FIGURE 2 Welfare Regimes and Early Development Strategies

ZimbabweZambia

Uruguay

TurkeyTunisia

Trinidad and

Tobago

Thailand

Sri Lanka

Singapore

Paraguay

PanamaMorocco

Mexico

Mauritius

Malaysia

Lesotho Kuwait

Korea

Israel

Iran, I.R. of

India

Greece

El Salvador

Egypt

Dominican Republic

Cyprus

Costa Rica

Colombia

Chile

Brazil

Bolivia

Argentina

-5

-4

-3

-2

-1

0

1

2

3

4

5

1970 Manufactured Exports (logged)

Commodification-DecommodificationScore

-2 -1 0 1 2 3 4 5

Productive welfare states are represented in bold, dual welfare states are underlined and protective welfare states are italicized.

392


16/20

ables that might also affect these outcomes (e.g.,inequality, country size, political freedoms, partisan-ship) and possibly explain outliers.44

This exposition does not allow any analysis ofinstitutional change. Clearly some LDCs have experi-enced changes in their welfare regimes such that they

no longer correspond to their early development strat-egies. For example, countries such as Colombia andCosta Rica pursued mostly inward-oriented develop-ment strategies in the early postwar era, and yet theyare productive welfare states.45 Already steps ahead,research on OECD distribution regimes has convinc-ingly shown how endogenous political dynamics canalter supporting coalitions and their functional rolesto produce very different institutional arrangements(see Pierson 2004; Thelen 2004). The role of distribu-tional coalitions in creating lock in has only beenimplied here; more rigorous theoretical analysis and

testing of both institutional reproduction and trans-formation are required. It would also be worthwhile toanalyze how democracy alters or reinforces tendenciestowards path dependency.

Implications and Next Steps

This study challenges prevailing CPE and IPE concep-tions of developing political economies by illustratingsystematic divergence in their welfare states. Contraryto CPE expectations, welfare states are not necessarilyby-products of postindustrial development, and theycluster into three distinct welfare regime types. Find-ings from this analysis also question IPE convergencepredictions by demonstrating that LDCs maintainqualitatively different kinds of distribution regimes inthe current era of globalization. As suggested by IPE,pressures of international market competition imposeimportant constraints on policy makers in terms ofthe choices made on how best to strengthen their posi-tion in the global economy. But, at the same time, thevariation among the types of welfare states implies

that local needs and politics continue to serve asimportant sources of diversity. The LDC welfare statethus remains a key institution to manage the tensionsand dilemmas that emerge from exposure to the

international economy. The existence of three clustersof welfare regimes well into the twentieth century inti-mates that LDCs, similar to the developed nations,demonstrate a sustained capacity to formulate system-atically different social policies to align economy andsociety.

Before stressing the broader implications,however, it is important to be clear about the theoreti-cal and empirical limitations of the analysis. The pro-posed causal explanation linking LDC policy makers,development strategies, distributional coalitions, andwelfare-regime type is tentative and begs furtherexploration. Empirically, findings from this studycan only be suggestive. The unavailability of dataimpresses strong limitations on the number of obser-vations included and, consequently, the kinds of sta-tistical tests employed. Future research would greatlybenefit from more extensive and reliable time-series

data.Nonetheless, this first attempt to uncover variet-

ies of welfare capitalism in LDCs based on existingdata is provocative. To the extent that the analysis hashit upon key differences in welfare regimes, importantpolicy and research implications emerge. First, it illus-trates how and why popular social policy reformstouted by international financial institutions (e.g.,increased investment in education, labor market lib-eralization) may be more successful in some countriesthan in others. Second, the possibility of distinct LDCwelfare regimes signals that developing countries

should also be included in some of the central,ongoing debates in political economy (e.g., varieties ofcapitalism, political economy of welfare state develop-ment). For consideration in the varieties of capitalismliterature, this analysis suggests that key strategic inter-actions or relationships, may be just as important andidentifiable in LDCs as in the OECD nations. Forexample, strategic interactions may occur in protectivewelfare states between domestic capital, the protectedlabor strata, and government. Productive welfarestates, on the other hand, may be influenced by stra-tegic interactions between governing elites and the

owners of domestic and foreign capital. The welfaremodels not only set the broad parameters of socialpolicy debates, they can also help identify which actorswill be at a strategic advantage. In this way, a newapproach to exploring how politics matters in LDCscan be pursued. Such analyses can provide greaternuance to understandings of LDC domestic politics byhighlighting axes of social conflict beyond the tradi-tional capital-labor dichotomy.

The stage is set to ask even more specific policy-oriented questions. From a policy perspective,we want

44Note that, as the previous section details, categorization of Para-guay and Panama in cluster 1 is not robust.

45Significantly, these countries contrast with LDCs that remainedprotective welfare states in the nineties, even after switching tooutward-orientation strategies as early as the 1980s (e.g., Turkey,Morocco, India).



17/20

to know the trade-offs, if any, between equity, growth,and poverty in one political economy versus another.Do the productive welfare states have less poverty butmore social stratification problems than the protectiveones? Conversely, do higher poverty levels drivenations towards protective or productive welfare

states, or vice versa?Finally, for IPE, an important future researchimplication is that we now have added tools to addressthe globalization question. This analysis suggests thatsimply analyzing the level of government expendituresin the globalizing LDCs, as previous studies have done,can be meaningless. Rather, recognizing the differentdomestic arrangements related to welfare in LDCsprovides a more precise way to assess whether thehistoric choices of these nations (productive, protec-tive, dual) will continue to remain durable in the faceof rising international market competition. As stated

above, the cross-sectional nature of this analysis pro-vides important insights into LDC political econo-mies, but it prohibits a more complete test of IPEconvergence theories. The fundamental challengeahead is to determine whether or not developingnations can maintain their welfare institutions into thetwenty-first century, particularly as internationalmarket pressures intensify with the entrance into themarket of such potential powerhouse countries asChina and India.

Acknowledgments

The author is grateful to Barry Ames, David Bearce,Clifford Bob, Silvia Borzutzky, Sarah Brooks, RossBurkhart, Charli Carpenter, Siddharth Chandra, Ben-

jamin Cohen, Kate Floros, Michael Goodhart, IanGough, Mark Hallerberg, Evelyne Huber, WilliamKeech, Scott Morgenstern, Layna Mosley, IrfanNooruddin, John Odell, Anibal Perez, Simon Reich,James Robinson, Sebastian Saiegh, Lyle Scruggs,Martin Staniland, John Stephens, Ravi Sundaram,

Cassandra Thomas, Daniel Thomas, Michael Waller-stein, Kurt Weyland, the participants of the DukeSummer Institute June 2004, and the Consortium inLatin American & Caribbean Studies at the Universityof North Carolina at Chapel Hill and Duke UniversityNovember 2005 for their helpful comments on earlierversions of this paper. Garth Olcese and Chris Belascoprovided research assistance.

Manuscript submitted 21 June 2005Manuscript accepted for publication 26 May 2006

References

Aldenderfer, Mark S., and Roger K. Blashfield. 1984.Cluster Analy-sis. Beverly Hills: Sage Publications.

Avelino, George, David Brown, and Wendy Hunter. 2005. TheEffects of Capital Mobility, Trade Openness, and Democracyon Social Spending in Latin America, 19801999. American

Journal of Political Science 49 (3): 62541.Baer, Werner, and Michael E. A. Herve. 1966. Employment and

Industrialization in Developing Countries. The QuarterlyJournal of Economics 80 (1): 88107.

Bates, Robert. 1981.Markets and States in Tropical Africa. Berkeley:University of California Press.

Bellin, Eva. 2000. Contingent Democrats: Industrialists, Laborand Democratization in Late-Developing countries. WorldPolitics 52 (2): 175205.

Betcherman, Gordon, Amy Luinstra, and Makoto Ogawa. 2001.Labor Market Regulation: International Experiment in Pro-moting Employment and Social Protection. World Bank SocialProtection Discussion Paper Series No. 0128. Washington, DC:World Bank.

Blashfield, R. K. 1976. Mixture Model Tests of Cluster Analysis:Accuracy of Agglomerative Hierarchical Methods. Psychologi-cal Bulletin 83: 33788.

Blunch, Niels-Hugo, and Dorte Verner. 2000. Is Functional Lit-eracy a Prerequisite for Entering the Labor Market? An Analysisof the Determinants of Adult Literacy and Earnings in Ghana.World Bank Working Paper No. 2410. Washington, DC: WorldBank.

Botero, Juan, Simeon Djankov, Rafael La Porta, Florencio C.Lopez-de-Silanes, and Andrei Shleifer. 2004. The Regulationof Labor. Quarterly Journal of Economics 119 (4): 133982.

Brickman, Ronald, Sheila Jasanoff, and Thomas Ilgen. 1985. Con-trolling Chemicals: The Politics of Regulation in Europe and theUnited States. Ithaca, NY: Cornell University Press.

Brookmann, Bernhard. 2001. The Ratification of ILO Conven-tions: A Hazard Rate Analysis. Economics and Politics 13 (3):281309.

Browning, Harley, and B. Roberts. 1990. Urbanization, SectoralTransformation, and the Utilization of Labor in LatinAmerica. Comparative Urban Research 8 (1): 86102.

Cerny, Philip. 1995. Globalization and the Changing Logic ofCollective Action. International Organization 49 (4): 595625.

Chapman, David, and Ann E. Austin. 2002. Higher Education inthe Developing World: Changing Contexts and InstitutionalResponses. Westport, CT: Greenwood Publishing.

Colander, David C. 1984. Neoclassical Political Economy: TheAnalysis of Rent-Seeking and DUP Activities. Cambridge, MA:Ballinger Publishing Company.

Collier, David, and Richard Messick. 1975. Prerequisites versusDiffusion: Testing Alternative Explanations of Social SecurityAdoption.American Political Science Review 69 (4): 1299315.

Collier, Ruth Berins, and David Collier. 1991. Shaping the Political Arena: Critical Junctures, the Labor Movement, and RegimeDynamics in Latin America. Princeton: Princeton UniversityPress.

Cutright, Phillips. 1965. Political Structure, Economic Develop-ment, and National Social Security Programs. American

Journal of Sociology70 (5): 53750.

Dreze, Jean, and Amartya Sen. 1989. Hunger and Public Action.Oxford: Clarendon Press.

Duda, Richard O., and Peter E. Hart. 1973. Pattern Classificationand Scene Analysis. New York: John Wiley & Sons.

394


18/20

Erickson, Kenneth Paul, and Patrick Peppe. 1976. DependentCapitalist Development, U.S. Foreign Policy, and Repression ofthe Working Class in Chile and Brazil. Latin American Perspec-tives 3 (1): 1944.

Esping-Andersen, Gosta. 1990. The Three Worlds of Welfare Capi-talism. Princeton: Princeton University Press.

Evans, Peter. 1997. The Eclipse of the State? Reflections on State-ness in an Era of Globalization World Politics 50 (1): 6287.

Evans, Peter, and Michael Timberlake. 1980. Dependence,Inequality, and the Growth of the Tertiary: A ComparativeAnalysis of Less Developed Countries. American SociologicalReview 45: 53151.

Everitt,Brian. 1974.Cluster Analysis. NewYork: John Wiley& Sons.

Feenstra, Robert C., and Gordon H. Hanson. 1996.Globalization,Outsourcing, and Wage Inequality.American Economic ReviewPapers and Proceedings 86: 24045.

Figueiredo, J. B., and Z. Shaheed.1995. Reducing Poverty throughLabour Market Policies. Symposium On Poverty: NewApproaches to Analysis And Policy. Geneva: ILO.

Form, William, and Kyu Han Bae. 1988.Convergence Theory andthe Korean Connection. Social Forces 66 (3): 61844.

Garrett, Geoffrey. 2001. Globalization and Government SpendingAround the World. Studies in Comparative International Devel-opment35 (4): 329.

Gelb, A., J. B. Knight, and R. H. Sabot. 1991. Public SectorEmployment, Rent Seeking and Economic Growth. Economic

Journal101 (408): 118699.

Gerschenkron, Alexander. 1962. Economic Backwardness in His-torical Perspective. Cambridge, MA: Belknap Press.

Gill, Stephen. 1995. Globalization, Market Civilization, and Dis-ciplinary Neoliberalism. Millenium: Journal of InternationalStudies 24 (3): 399423.

Goldin, Claudia, and Lawrence F. Katz. 1998. The Origins ofTechnology-Skill Complementarity. The Quarterly Journal ofEconomics 113 (3): 693732.

Grosh, Margaret E. 1994. Administering Targeted Social Programsin Latin America: From Platitudes to Practice. World BankRegional and Sectoral Studies Series. Washington, DC: WorldBank.

Gough, Ian, and Geof Wood with Armando Barrientos, PhilippaBevan, Peter Davis, and Graham Room. 2004. Insecurity andWelfare Regimes in Asia, Africa and Latin America: Social Policyin Development Contexts. Cambridge: Cambridge UniversityPress.

Haggard, Stephen, and Robert Kaufman. n.d. Revising Social Con-tracts: Welfare Reform in Latin America, East Asia and EasternEurope. Forthcoming.

Hall, Peter, and David Soskice. 2001.Introduction. In Varieties ofCapitalism, eds. Peter Hall and David Soskice. Oxford: Oxford

University Press, pp. 170.Huber, Evelyne. 1996. Administering Targeted Social Programs in

Latin America: Neoliberal vs. Social Democratic Models. InWelfare States in Transition, ed. Gosta Esping-Andersen.London: Sage Publications, pp. 14191.

Huber, Evelyne, and John Stephens. 2001. Development and Crisisof the Welfare State. Chicago: University of Chicago Press.

Kaufman, Robert, and Alex Segura-Ubiergo. 2001.Globalization,Domestic Politics, and Social Spending in Latin America: ATime-Series Cross-Section Analysis, 19731997. World Politics53 (4): 55387.

Keohane, Robert, and Joseph S. Nye. 2000. Globalization: WhatsNew? Whats Not? (And So What?)Foreign Policy118: 10419.

Kerr, Clark. 1983. The Future of Industrial Societies: Convergence orContinuing Diversity?Cambridge: Harvard University Press.

Ketchen, David J., Jr., and Christopher Shook. 1996. The Appli-cation of Cluster Analysis in Strategic Management Research:An Analysis and Critique. Strategic Management Journal 17(6): 44158.

Kiesling, Henry J. 1983. Nineteenth-Century Education Accord-ing to West: A Comment. The Economic History Review 36 (3):

41625.

Kitschelt,Herbert, Peter Lange,Gary Marks, and John D. Stephens.1999. Continuity and Change in Contemporary Capitalism.Cambridge: Cambridge University Press.

Koo, Hagan. 1990. From Farm to Factory: Proletarianization inKorea. American Sociological Review 55 (5): 66981.

Kurtz, Marcus. 2002. Understanding the Third World WelfareState after Neoliberalism: The Politics of Social Provision inChile and Mexico. Comparative Politics 34 (3): 293313.

Kuruvilla, Sarosh. 1996.Linkages between Industrialization Strat-egies and Industrial Relations/ Human Resource Policies:Singapore, Malaysia, the Philippines, and India. Industrialand Labor Relations Review 49 (4): 63557.

Lewis-Beck, Michael S., Alan Bryman, and Tim Futting Liao, eds.2004. The Sage Encyclopedia of Social Science Research Methods.Vol. 3. Thousand Oaks, CA: Sage Publications.

Lorr, Maurice. 1983. Cluster Analysis for Social Scientists. SanFrancisco: Jossey-Bass Publishers.

Milligan, G., and M. C. Cooper. 1985. An Examination of Proce-dures for Determining the Number of Clusters in a Dataset.Psychometrika 50: 15979.

Morrish, Ivor. 1970. Education Since 1800. London: Allen andUnwin.

Nelson, Joan M. 1999. Reforming Health and Education: The WorldBank, The IDB, and Complex Institutional Change. OverseasDevelopment Council. Baltimore: Johns Hopkins UniversityPress.

Nicholas, Stephen J., and Jacqueline M. Nicholas. 1992.Male Lit-eracy, Deskilling, and the Industrial Revolution. Journal ofInterdisciplinary History23 (1): 118.

Pandit, Kavita, and Emilio Casetti. 1989. The Shifting Patterns ofSectoral Labor Allocation During Development: Developedversus Developing Countries. Annals of the Association of

American Geographers 79 (3): 32944.

Panel on Discriminant Analysis, Classification and Clustering.1989. Discriminant Analysis and Clustering. StatisticalScience 4 (1): 3469.

Pierson, Paul. 2004. Politics in Time: History, Institutions and SocialAnalysis. Princeton: Princeton University Press.

Pitruzzello, Salvatore. 2004. Trade Globalization, Economic Per-formance, and Social Protection: Nineteenth-Century British

Laissez-Faire and Post-World War II U.S.-Embedded Liberal-ism. International Organization 58 (4): 70544.

Polanyi, Karl. 1944. The Great Transformation. New York: Farrara

1 -welfare states in developing countries

Documents