Controlling coalitions: Social lending at the multilateral development banks

Abstract

Multilateral development banks (MDBs) dramatically increased social lending for health, education, and safety nets after 1985. Yet the great powers’ social policy preferences remained relatively static from 1980 to 2000. This contradicts the conventional view that powerful states control IOs. We argue that highly institutionalized IOs like MDBs require a complete model of possible member-state coalitions encompassing the preferences of all member states—not just major powers. We develop multiple measures of state preferences and include all member states in our coalitional model. We evaluate our model and alternatives with an analysis of more than 10,000 MDB loans from 1980 to 2000. We find that when we include all member states weighted by their voting shares, principal preferences are significantly related to lending outcomes.

Appendices

Appendix A: OECD Creditor Reporting System Categories Coded as “Social”

110:

EDUCATION

111:

Education, level unspecified

11110:

Education policy and administrative management

11120:

Education facilities and training

11130:

Teacher training Teacher education

11181:

Educational research

112:

Basic education

11220:

Primary education

11230:

Basic life skills for youth and adults

11240:

Early childhood education

113:

Secondary education

11320:

Secondary education

11330:

Vocational training

114:

Post-secondary education

11420:

Higher education

11430:

Advanced technical and managerial training

120:

HEALTH

121:

Health, general

12110:

Health policy and administrative management

12181:

Medical education/training

12182:

Medical research

12191:

Medical services

122:

Basic health

12220:

Basic health care

12230:

Basic health infrastructure

12240:

Basic nutrition

12250:

Infectious disease control

12281:

Health education

12282:

Health personnel development

130:

POPULATION POLICIES/PROGRAMMES AND REPRODUCTIVE HEALTH

13010:

Population policy and administrative management

13020:

Reproductive health care

13030:

Family planning

13040:

STD control including HIV/AIDS

13081:

Personnel development for population and reproductive health

160:

OTHER SOCIAL INFRASTRUCTURE AND SERVICES

161:

Employment

16110:

Employment policy and administrative management

162:

Housing

16210:

Housing policy and administrative management

16220:

Low-cost housing

163:

Other social services

16310:

Social/welfare services

16320:

General government services

16330:

Settlement

16340:

Reconstruction relief

16350:

Culture and recreation

16361:

Narcotics control

16362:

Statistical capacity building

16381:

Research/scientific institutions

420:

WOMEN IN DEVELOPMENT

42010:

Women in development

430:

OTHER MULTISECTOR

43010:

Multisector aid

43020:

Multisector aid for basic social services

43030:

Urban development and management

43040:

Rural development

43050:

Non-agricultural alternative development

43081:

Multisector education/training

520:

DEVELOPMENT FOOD AID/FOOD SECURITY ASSISTANCE

52010:

Food aid/Food security programmes

Appendix B: Constructing a Time-Series Index of Social Policy

To construct a social policy index (SPI), we used factor analysis to extract the common factor in our fourteen social variables. We first standardized these variables to the 1996 mean and standard deviations to preserve variation over time (i.e. we subtract the 1996 mean from all observations’ values for the social variables for all years and divide by the 1996 standard deviations). We then performed factor analysis to isolate the variable indicating social policy performance. The first factor has an eigenvalue above eight across all five datasets, while none of the remaining factors have eigenvalues above one. This strongly suggests a single key factor, which we have labeled “social policy” and use as our index values. The loadings are summarized below (see Table 3).

Table 3 Results from factor analysis (averaged across the five imputed datasets)

As noted, we are primarily concerned with policy preferences, not simply social outcomes. Since social outcomes are a result of both government policy and other factors exogenous to policy, we focused on those variables for which a compelling argument could be made that they were the product, at least to a significant degree, of government policies. Time series data on fourteen underlying variables exist in large sets for the 211 countries included in this study.

To deal with missing data, we employ multiple imputation as recommended by Gary King and collaborators (King et al. 2001). To impute the missing values, we used the program Amelia II, as it takes into account the unique nature of time-series cross-sectional data (King et al. 2001). For the imputation, we constrained all imputed values to fall within the 97.5th percentile and 2.5th percentile of the pre-imputed dataset. For the variables used to impute the missing data, see Table 4. We set the number of imputations at five.

Table 4 List of variables used for multiple imputation

The factor analysis was then performed on each of the five imputed datasets (as were the later logistic regressions). The social policy factor was generated in each case, and then the datasets were combined as per the usual multiple imputation rules.

Appendix C: Pivotal-Weighted Preferences

A more faithful modeling of coalition formation draws on coalition theory in comparative politics in order to specify which coalitions can be built given majority voting rules. We employ the pivotal player model in order to predict which coalitions among the possibilities are most likely to form. We assume a one-dimensional policy space, where actor preferences can be arrayed along a single policy continuum. This assumption is defensible given our consideration of the single policy dimension of social protection.²⁷

Second, we assume that our actors form voting coalitions, not governing coalitions. Because voting takes place on discrete policy questions that can be arrayed in unidimensional space, and the votes do not affect executive tenure (as in confidence votes), little incentive exists for the formation of voting coalitions that are non-contiguous.²⁸ For example, if a given measure is closer to the ideal point of actor A than actor B, and B votes in favor of the measure, then A should also vote in favor. Coalitions thus will not drop “dummies” or potential members that are not necessary to reach the majority threshold but that nevertheless are aligned in issue space.²⁹ While voting coalitions are arguably less stable than governing coalitions, they likely violate the assumptions of preference-aggregation models less than governing coalitions, where office holding is a key consideration.

The pivotal-players model emphasizes the role of strategically well-positioned actors in the coalition-formation process. Within the many possible connected majority coalitions that might form, some potential members might be “pivotal” in the sense that the combination of their centrist position and their size makes them very attractive coalition partners, and it will prove difficult to form connected coalitions without them. Thus, pivotal players can veto a large set of the possible coalitions that might form and can extract policy concessions from their coalition partners that their size alone would not necessarily predict. Large centrist players thus have a strong advantage over similarly large extremist players. We use a weighting formula to capture states’ centrality to coalition formation based on their pivotalness.

Specifically, for each bank year we arrayed all countries in each executive director’s constituency from highest to lowest based on the hybrid proxy measure of preferences for social loans. We then summed all possible values of the voting shares of countries adjoining one another, creating a matrix of all potential coalitions. For all of the coalitions where the sum of voting shares was greater than .50 of the constituency’s total vote, we computed the consequence to the potential coalition of the defection of the partner at each of the two extreme ends of the coalition. If the defection of a partner on one of the ends of the potential coalition would cause the coalition’s collapse (vote shares fell below a majority plus one), we counted this as an instance where the defecting country would prove pivotal.

We summed all such instances where the given country proved pivotal to a potential coalition and gave each country a “pivotalness” score within the constituency. In general, this design meant that countries proved more pivotal where they had large vote shares and/or where their scores were near the center of the constituency’s continua for the social indices. We then weighted all countries’ social preference by the pivotalness score. Finally, we summed the products of all of the countries’ social index values multiplied by their pivotalness scores to produce an aggregate preference for social lending for each constituency by bank year. Using these values for each constituency, we then repeated this process for each executive board for each bank year.³⁰

In Table 5 we provide a simple example of this technique for the 2000 constituency at the African Development Bank that included Belgium, Italy, France, Spain, and Switzerland. First, the members of the constituency are arrayed according to their preferred outcome along the SPI scale from 1.08 to 1.46.³¹ Next, we count all possible contiguous coalitions, where the combined voting shares are at least fifty percent plus one. Next, we counted all “critical coalitions” where, if one of the members on either end of the coalition defects, the coalition collapses. In this example there are four such potential critical coalitions: Switzerland-Belgium-Italy, Switzerland-Belgium-Italy-Spain, Belgium-Italy-Spain-France, and Italy-Spain-France. By dropping off of the end of the first two coalitions, Switzerland could prove pivotal in two instances. Switzerland’s vote share and position situates it strategically, providing advantages in coalition formation. Likewise, Italy and France could prove pivotal in two instances. Belgium and Spain are not pivotal in any instance. Thus, the total number of potential critical defections—instances of pivotal players—is six.

Table 5 Pivotal players for France-Germany-UK-USA constituency at the African Development Bank, 2000

We then weight each country’s ideal point by the “pivotalness” share. For example, Italy’s ideal point of 1.402 is multiplied by its pivotalness share of .33 (2 of 6 pivotal instances), producing a product of .36. Finally, we sum each of these products to produce an overall preference for the constituency, which in this example is 1.30.

Critically, coalition formation at the development banks occurs in two stages. First, most member governments have formed semi-permanent coalitions with a few to several other governments, usually in the same region, and they jointly delegate voting authority to a single ED charged to represent their collective interests on the executive board. At any given bank, all but a few countries share executive directorates with multiple other countries in this fashion. The member governments delegating to a common ED (henceforth, constituency), can meet and vote on lending policies prior to executive board meetings. In this way countries in a constituency can instruct their shared ED, who usually comes from the largest member country in the constituency. In the second stage, the EDs, with marching orders in hand from their constituencies, then all meet together on the executive board to make policy decisions for the bank as a whole.

This two-stage process requires that we model both stages of coalition formation. First, we model the coalition formation process that consists of the member states delegating to their ED. In this case, the potential coalition partners are the member-states of each constituency. We calculate the pivotal players for each constituency at each bank for each year. We then derive a unique preference for each constituency by constructing a weighted average based on the degree to which each member country could prove pivotal in constituency voting, as discussed above.

In step two, we follow this same procedure, this time using the EDs (each of whose preferences for percentage of social loans we derived in the previous step) as the potential coalition partners. We calculate the pivotalness score for each of these EDs with the same procedure used when analyzing individual member states in each constituency as discussed above. Finally, as above, we weight each ED’s derived preference by their pivotalness shares and sum the weighted values to produce a single predicted social policy preference for each bank for each year.