Diaz-Cayeros, Estvez and Magaloni, Ch.

From Clientelism to Entitlements:

The politics of social transfers in Mexico, 1989-2006 *
Alberto Diaz-Cayeros
Stanford University
Federico Estvez
ITAM
Beatriz Magaloni
Stanford University

Chapter for discussion at the Comparative Politics Workshop UCLA

This is the third chapter of a book length manuscript. More chapters of the manuscript are
available at www.stanford.edu/~albertod/conference
Draft, Comments Welcome

Funding for this project was provided by the Center for Democracy, Development and
the Rule of Law, the Department of Political Science and the VPUE grant at Stanford
University and the Rockefeller Foundation Bellagio Center. Various chapters and
fragments of the project were presented over the years in numerous venues. We thank in
particular participants in seminars at the Social Science History Institute and the Political
Science Department at Stanford, UCLA, Berkeley, the World Bank, the Midwest
Political Science Association Meetings, the American Political Science Association
Meetings and the Latin American Studies Association Meetings. Superb research
assistance has been provided by Sandra Pineda, Marcela Gmez, Arianna Snchez, Jorge
Bravo, Katherine Kelman, Ana Gardea, Emmerich Davis, and Hamilton Ulmer.
79

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

This book analyses the political economy of social assistance programs in Mexico from
1989 until 2006. This period has witnessed impressive transformations. On the one hand,
the long-lasting rule of the Partido Revolucionario Institucional (PRI) came to an end in
2000 and Mexico went through a transition to democracy that entailed a fundamental
change in the workings of the basic institutional apparatus. On the other hand, the three
administrations during this period dramatically changed the existing social assistance
programs designed to improve the well-being of the poor and mobilize their electoral
support. There has been an important reduction of extreme poverty during the last
decade. However, social programs continue to be criticized for their presumed
manipulation by politicians seeking to obtain electoral support. This book analyses the
effectiveness of the various social assistance programs in Mexico and the political logic
driving each of them.

Chapter 3. The Welfare Effects of Social Transfers

1. Social assistance and public goods.

Social assistance programs are aimed at reducing poverty. Much has been learned
over the last decade regarding the evaluation of conditional cash transfer (CCT) programs
on individual welfare indicators. In particular, scholars have found creative ways to make
inferences based on the counterfactual of what would have happened with a
developmental outcome closely linked to well-being had a certain intervention not taken
place (Coady, 2000; Rawlings and Schady, 2002; Rawlings and Rubio, 2003; Duflo,
undated; Skoufias, 2001; de Janvry, 2004). Such work is based primarily on the
collection of carefully designed surveys that can allow a comparison between treated and
untreated groups in an experimental setup. Individual level welfare indicators in a panel
format or experimental settings such as those explicitly designed to evaluate
Progresa/Oportunidades (Wodon, 2001; IFPRI, 2000) are not available, however, to study

80

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

changes in local public good provision. As noted by Besley (2006), a general
shortcoming in development research is that we know far less about local public good
provision than about policy interventions aimed at income support, even though the
former have equally important effects on well-being.
This chapter analyzes the aggregate effects of policy interventions that have
externalities beyond their direct individual beneficiaries. We use municipal variance in
the provision of local public goods, in order to compare the social effectiveness of social
assistance programs over the medium term. Potable water, sewerage systems and
electricity grids are provided in Mexico as public goods (i.e. they are non-excludable and
non-rival in their consumption).1 These public services make up the bulk of municipal
public good spending. But these projects can have important leakages, in the sense that
non-poor households can benefit from the projects financed by social transfers. The issue
is whether one can measure the extent to which expenditure in these local public goods
was effective.2
The Mexican experience allows us to compare programs characterized not only
by radically different designs, in terms of targeting/universalism and the
discretion/formula-driven dichotomies, but also regarding the political environment
where the programs were implemented and the degree of decentralization in their
operation. Our goal is to compare the effectiveness of various interventions, paying
particular attention to the problems of selection bias that bedevil efforts at impact
evaluation. Although we do not measure directly improvements in individual well-being,

These are not pure public goods though: water systems can be privatized, user charges and fees charged,
and it is possible to have congestion.
2
And the degree to which there was leakage, meaning that the non-poor obtained benefits.

81

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

we are able to measure the impact of social spending in the supply of aggregate local
public goods.
More specifically, the chapter compares the aggregate effect of local public goods
expenditure within Pronasol with the FDSM/FISM and the decentralized municipal
public works expenditures on the improved provision of local public goods at the
municipal level. Pronasol was characterized by high centralization and discretion, while
FDSM/FISM is decentralized and formula based (see chapter 1). Our results suggest that
the effect of the discretional fund, Pronasol, is negligible once correcting for endogeneity.
By contrast, changes in the provision of potable water, electricity and sewerage in
aggregate municipal populations do respond to decentralized and formula based public
works appropriations. We estimate these effects, controlling for the impact of social
cleavages, democracy, participation, demography and economic development on public
service provision.

Scholars have addressed this question in the context of Mexico, providing

important insights. Hiskey (2003) analyzes the impact of Pronasol spending in the
provision of electricity, potable water, and sewerage at the municipal level in two
Mexican states, Jalisco and Michoacan. He finds that the effect of public spending
depended on the configuration of partisanship, in particular, the presence or absence of
multi-party competition. When there was multiparty competition, his estimations suggest
that Pronasol funds improved the access to public services, although the magnitude of the
effect seems to be extremely small. The analysis is suggestive, although it is not as well

82

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

specified econometrically or as comprehensive as some of the later work, which means
that it probably suffers from some omitted variable and selection biases.
Cleary (2002 and 2004) provides a much more nuanced and complete assessment
of the performance in public good provision for the full set of Mexican municipalities.
His econometric analysis uses log-odds ratios and a more comprehensive set of economic
controls. In contrast to Hiskey (2003), Clearys findings suggest that electoral
competition in Mexico does not mediate the impact of public spending on the provision
of public goods. In terms of public spending, he finds a strong effect of Pronasol funds in
the provision of electricity, a weak effect in sewerage, and no effect in potable water. His
study argues that forms of political participation, rather than partisan competition, are the
main determinants of differences in performance in public good provision.
Neither of these studies control or is aware of the problem of selection bias, which
has been central to the work on impact evaluation in development economics. In order to
address the endogeneity problem we use geographic instrumental variables that provide
predictions of the allocation of discretional funds that are not affected by the expectations
of changes in the provision of public goods. We build on previous research efforts,
providing what we believe is a more robust set of estimations of the effect of social
spending in local public goods provision in Mexico. But departing from the work by
Hiskey (2002) and Cleary (2002 and 2004), whose goal is primarily to address the effect
of electoral democracy and political participation in public good provision, we focus
primarily on the comparative assessment of various policy interventions.
The chapter is organized as follows. The next section provides an overview of the
main theoretical hypothesis advanced in the comparative literature to explain differences

83

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

in the provision of local public goods and services. The third section discusses the
empirical strategy of estimation used to assess the policy effectiveness of social spending
in local public goods provision. Following recent advances in development economics,
we estimate two stage least squares (TSLS) regressions, using geographic variables as
instruments.3 We follow this strategy in order to mitigate the endogeneity of public
spending that plagues most of the existing work. The fourth section discusses the results.
Our conclusions motivate the need for a political understanding of the allocation of
resources for poverty relief, which is the task we undertake in the following chapters.

2. Heterogeneity, state capacity and democracy.

Well-being hinges not just on individual income, but on the access individuals
and households can have to public goods and services in the locality where they live.
Although public goods can sometimes be privately provided, in general, the provision of
facilities such as potable water, sewerage, electricity, schools and health clinics are made
by governments. What explains differences in the provision of those local public goods?
The most influential hypothesis in the last few years has been one associated with
Alesina, et. al. (1999), which proposes that greater social heterogeneity, as measured
through an index of ethnolinguistic fractionalization (ELF), makes it harder for
communities to provide public goods. Such failure is attributed to the idea that it is more

We also corrected the estimates for spatial autocorrelation using GeoDA. However, given that the results
did not change, we report the models without the spatial lag.

84

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

difficult to engage in collective action when co-ethnic groups distrust outsiders, and that
ethnic fragmentation might breed conflict.4
Studies linking ethnic diversity to public good provision generally take a
statistical approach. The association between polarization as measured by ELF indexes,
and failures to provide local public goods, as measured primarily by public expenditure
patterns, seems to be a robust finding. However, there is considerable disagreement in
this literature regarding the specific mechanisms through which heterogeneity influences
public good provision. It is possible that there are substitutions among public goods, in
the sense that some places might get less of some types of goods, but more of others
(Banerjee and Somanthan, 2001). It is also possible that ethnically heterogeneous groups
have a different profile of tastes for public goods than homogenous ones, so the result is
driven by preferences, rather than polarization and conflict (Banerjee and Somanthan,
2006; Banerjee, 2002). And the literature fails to specify whether ethnic differences in
fact become reflected into bureaucratic structures and the political system (Posner, 2004),
which are the arenas where the provision of public goods is decided.5 Nonetheless, this
literature suggests that a key determinant of success in local public goods provision is the
capacity of communities to work together with a common aim.

Studies finding evidence of the impact of social heterogeneity in public good provision across nations and
within countries include Alesina and La Ferrara, 2000; Kwaja, 2002; Miguel, 2004; Miguel and Gugerty,
2005; Dayton-Johnson, 2000; and Baqir, 2000.
5
Political scientists have engaged this literature quite extensively in their own work on ethnic conflict and
civil war (Weinstein, forthcoming; Fearon and Laitin, 2003). The most promising research agenda, seems
to be the move away from cross sectional variation to a focus in local experimental settings, in which
scholars have tried to understand the conditions under which communities are more able to create networks
of trust. Habyarimanna et. al., 2006, in particular, performed experiments in Kampala, Uganda, testing the
willingness of co-ethnics and non co-ethnics to cooperate. Also in an experimental set up, Wantchekon
(2002) tested the appeal of programmatic promises of public good delivery by presidential candidates
according to ethnic differences in Benin.

85

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

While in this literature social structures are the main explanation of failures in
public service provision, another alternative is to consider shortcomings in public
administration. There is a long tradition of research, especially among political scientists,
which has seen public good provision through the lens of state capacity (Kohli, 2000).
The general thrust of that literature has been to suggest that failures in the provision of
public goods reflect underlying problems arising from weak states that are incapable of
taxing, running a bureaucracy, or in general, fulfilling basic public functions. Although
there surely are variations in the administrative capacities of bureaucracies and service
providers, state capacity tends to be a difficult explanation to put to test. Very often states
are defined as incapable precisely because they do not provide public goods and services.
Moreover, Tendlers (1997) work on Cear in Brazil has shown that it is possible to
create an autonomous bureaucracy that can provide public goods effectively even in
conditions of what could be conducive to a weak state.6 Furthermore, it is usually very
hard to find a variable that measures state capacity which is not confounded with the
general level of wealth or development. Nevertheless, it is important to acknowledge that
public service provision might be better or worse due to differences in bureaucratic
performance. In the Mexican context, there is a wide variation in the administrative
apparatus of municipal governments. Around half of the municipalities can be thought of
as lacking state capacity, due to their small size, precarious public finances and the
deprivation of their inhabitants (Cabrero, 2004).

State capacity can be proxied through fiscal variables, in particular, the capacity of local governments to
collect revenues and spend in public goods. Zhurakvskaya (2000) has found, for example, that public good
provision in Russian cities did not respond to local tax collection efforts, because the center offset those
increases through withholding revenue sharing.

86

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

Bureaucracies respond to the incentives of the public officials that run them and
the politicians that oversee them (World Bank, 2004). Policy makers have increasingly
paid more attention to local power structures and corruption as explanations for the
difficulties governments face in providing public goods and services.7 In a particularly
poignant example, Reinikka and Svensson (2004) measured an astounding leakage of 87
percent in a program in Uganda meant to provide grants to schools for non-wage
expenditures. Such leakage was successfully reduced through greater citizen involvement
and information regarding the allocation of funds to the local schools (World Bank,
2006).8 Olken (2006) similarly found that the leakages in a poverty relief program
delivering rice in Indonesia were large enough that they offset the welfare gains from
having the program in place at all.
To a large extent these studies get leverage from analyzing the distribution of
funds across levels of government. The literature on decentralization is often premised on
the notion that local governments are better at providing public services than centralized
bureaucracies. In a study of Bolivian municipalities, for example, Faguet (2004) shows
that decentralization made public spending decisions more effective for the provision of
public goods. While decentralized public goods provision is often successful, the initial
optimism regarding the virtues of decentralization has been tempered by greater

Bardhan and Mookherjee (2005) have shown, in a formal model, that centralized systems of public
service delivery are more subject to corruption. However, they also note that local elites might capture
governments making them less efficient than a centralized arrangement. Besley and Coate (2003) have
provided a model in which the advantages of decentralization depend on legislative behavior and how
jurisdictional spillovers and conflicts arising from the variance in preferences over public good provision
across places are mediated by the political system. Despite these theoretical advances, we are only starting
to understand the links between democratic accountability, local public good provision and
decentralization.
8
In an empirical evaluation of Sens (1981) influential hypothesis that democracy prevents famines, Besley
and Burgess (2002) have shown that Indian states with greater freedom of the press are more likely to
deliver disaster relief. Besley and Prat (2001) have similar findings for a cross section of countries.

87

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

awareness that a crucial aspect that determines whether a local government is capable of
providing public goods is the accountability of local politicians to citizens (World Bank,
2004; Bardhan, 2005).
As a general statement, democratic forms of government seem to be more
effective at providing public goods. The general question of whether political regimes
determine the provision of public goods has been tested across countries by Baum and
Lake (2001). In particular, they found that access to health care, drinking water, and
school enrollment improves with democracy. They also report findings on outcome
variables such as infant mortality rates and life expectancy. The outcome variables could
be attributed to many other causes, not just related to public policies, but to the general
effect of democracy on economic growth and population dynamics (see Przeworski et.
al., 2000; Navia and Treufel, 2003; Pritchard, 2000). However, the overall effects of
democracy on public service provision such as health clinics, immunization and access to
safe drinking water seem to be relatively robust.
At the local level, Chhibber and Noorudin (2004) found that Indian states where
patterns of electoral competition are stiffer, in the sense that the incumbent faces strong
contestation from one single challenger, are more likely to provide public goods.
However, they also find that public good provision in multiparty competition decreases,
because in fragmented races politicians do not need to provide as many public goods in
order to build all encompassing coalitions. In the case of Mexico, Hiskey (2003) argues
that in more competitive electoral environments, measured through the effective number
of parties, public service provision is more sensitive to public spending. Cleary (2004)
does not find this effect of electoral configurations on public service provision, although

88

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

he shows that variables related to political participation, such as literacy and turnout,
improve public service delivery.9 Closely related to the argument of participation, in the
case of India, Chattopadhyay and Duflo (2001) have argued that female leadership in
Village Councils led to greater investments in infrastructure related public goods (water,
fuel and roads), while men tend to favor investments in education.10 In a similar vein,
Pande (undated) has shown that the representation of scheduled castes in the Indian states
improved the provision of education and land reform, which are policies that the poor
would presumably favor.11 The general thrust of these empirical contributions has been to
suggest the need to explore the effect of accountability processes (democracy arguably
being the most important) and citizen participation in the provision of public goods.
The literature thus offers primarily four explanations for the differential provision
of public goods. First, it is possible that more homogenous communities are more capable
of providing public goods. Second, governments that in some measurable way are more
capable might be better able to provide public goods. Third, local power structures and
the relationship between local and national political and administrative actors might
influence the opportunities for rent seeking and capture in the provision of public goods.
Finally, democratic accountability in various guises (electoral or participatory) might also
determine the success or failure in the provision of public goods. Keeping in mind the
interaction between spending patterns and factors that measure these prevailing

Platteau (2004) criticizes, however, participatory models of local level development.

Kearny and Lott (1999) find that across countries greater female representation leads to larger
government.
11
Foster and Rosensweig (2001) have shown that in India democracy means the empowerment of landless
workers, which has led to land reform and higher public good provision, although not necessarily higher
productivity. Bardhan and Mukherjee (2006), however, find no trade-off between land reform and increases
in income.
10

89

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

explanations, in this chapter we compare the impact of different social spending
strategies in local public good provision.

3. The determinants of public goods provision

In the Mexican context, access to safe drinking water, electricity, and sewerage,
among other local public goods depends upon provision by municipal governments.
While education and health facilities are also crucial inputs impacting well-being, their
provision in Mexico has been primarily decided in a relatively centralized fashion. Even
after the decentralization of education in 1993 and the decentralization of federal health
systems a few years later, most of the financing of those two public goods remains
federal (Diaz-Cayeros and Courchene, 2003) and the decisions regarding the location and
support of schools and health clinics were primarily made by state and federal, not
municipal governments. In the empirical test that follows, we limit ourselves to municipal
services, not including education and health provision. This is due mostly to the difficulty
in finding appropriate quantitative indicators of school and health clinic coverage, and
that, as argued by Cleary (2004), municipal services are where one might expect the most
intense scrutiny on the part of citizens.12
The general econometric framework used in most of the studies of public good
provision estimate regressions of the following form:
Public Good = 0 + 1S + 2H + jP + kE +
Where S is some measure of public spending, H is a measure of heterogeneity or
diversity, such as an ELF index, P are political variables, such as mandated female
12

While a more comprehensive analysis would ideally compare poverty rates, changes in malnutrition,
infant mortality or morbidity, or the income effects generated by the provision of public goods on
households.

90

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

participation or the effective number of parties; and E are economic controls that proxy
for development levels and state capacity. The public good provision itself is usually
measured as the relative coverage, in percentage terms, of a certain public service.
Following recent advances in development economics, we argue that the main
problem with estimating the effects of a social assistance program spending in the
provision of local public goods in this form is one of endogeneity: namely, while we seek
to understand how much a given service, say potable water, improved as a consequence
of public spending, the decisions regarding the allocation of discretional funds across
municipalities are most certainly influenced by the conditions of the water infrastructure
around the country and the likely impact policy makers and politicians believe spending
in a particular place will have in the delivery of those social services. This endogeneity
problem can be corrected, but if it is not, the estimations of the effects of policies may be
seriously biased.
The same is true for other public services, such as electricity and sewerage. For
example, if a politician wants to claim credit for a large improvement in the coverage of
electricity in the few years of his short term in office, he might prefer to allocate funds to
urban places that already have a relatively well developed grid, so that a large number of
dwellings can improve very quickly. Sewerage could be more urgently needed in some
regions where unsanitary conditions would produce a cholera outbreak or some other
public health crisis. Policy makers in this case might actually be able to allocate resources
with the highest priority to the places where no improvement has been observed in the
past. The selection of where the locate electricity and sewerage projects might be
determined precisely by the dependent variable we want to explain.

91

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

These examples suggest that the best estimation strategy is not to run an OLS
regression, even with appropriate control variables. An additional problem in some of the
literature is that there are omitted variable biases if one tries to estimate the level of
public good provision. Attributing the level of provision, which is a stock accumulated
through time, to the flow of public spending made in a specific time interval, would make
a wrong inference in several ways.13 The level of public service provision across
municipalities might be determined by many decisions taken by households and
individuals privately. A home might have access to safe drinking water because a rich
household invests on a well, or because there is a public intervention that connects homes
to the water system. Although electrification requires public initiative, the percentage of
homes with access to electricity often depends on private provision through theft of
electricity from the distribution grids. This is particularly noticeable in poor semi-urban
areas. Moreover, water delivery has been successfully privatized in many municipalities,
so that access to drinking water can be determined by a combination of private and public
investment. Teasing out the public vs. private efforts in provisions of public goods
measured as levels is extremely hard.
Public spending in the past surely accounts for most of the provision of local
public goods, while current spending should only impact the level marginally. If we do
not have data available measuring past spending and past private provision, there is a
serious omitted variable bias in using levels as the dependent variable. The strategy
suggested by Banerjee (2002) is to measure public service provision as a first difference,

13

For a creative use of the history of public spending, matching it to the provision of public goods in the
Italian regions, through a perpetual inventories approach, see Golden and Picci (2006).

92

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

in order to ensure that we do not wrongly attribute the initial levels of provision to current
spending.14

a) The dependent variable.

The dependent variable we use in the analysis is an index of the change in the
provision of drinking water, electricity and sewerage in the Mexican municipalities.15 We
estimate changes in the logistic transformation of the index between 1990 and 2000
according to census data, so the dependent variable takes the form:
Index = log(I00/(1-I00)) - log(I90/(1-I90))
Where: Ixx=(%water coverage + % sewerage coverage + % electricity coverage)/3
for year xx. The appendix presents an explanation for why we did not use the mid census
estimations (conteos) and it also provides a set of scatterplots that give a good idea of the
distribution and patterns of this dependent variable from 1990 to 2005.
Instead of making individual estimations for each service, we averaged the
provision of the three services in each municipality (although individual estimations for
each service are available upon request).16 Table 3.2 presents some of the descriptive
feature of the index, comparing the data for 1990 and 2000, and the relative change in
14

We employed this strategy in Diaz-Cayeros and Magaloni (2003), and so does Hiskey (2003). The
alternative is to do as Cleary (2004), who estimates levels of public service provision, but keeps the initial
level in the right hand side. Such strategy produces a higher R2, without changing the substantive findings.
15
We do not use the marginality or welfare indexes that have been calculated in Mexico by INEGI and
CONAPO, because those factor analyses are not strictly comparable across years. More importantly, these
indexes include too many census indicators, many of which are related to private welfare, rather than
public good provision (for example, the population earning less than one minimum wage or the
construction materials of their home). We do not use a Human Development Index (CONAPO, 2000;
UNDP, 2005) because it measures individual welfare, rather than public services, and we do not have
reliable estimations for the HDI at the municipal level for 1990. Our index is highly correlated with any of
those conventional measures of development.
16
We do not perform some factor analysis or other data reduction method because we believe it is far more
transparent to simply average the three services weighting them equally. It is important to note, however,
that it is much more expensive to provide sewerage than electricity; and that the demands among citizens
are most intense for the case of potable water.

93

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

those census years. The mean value of local public good provision increased in the
decade, and the standard deviation was reduced. Perhaps more importantly, there is a
greater improvement among the lower end of the distribution than in the higher end.

Table 3.2 Descriptive statistics of Dependent variable

mean
standard
deviation
d1 (10%)
q1 (25%)
median
q3 (75%)
d10 (90%)

index
1990
58.9%

index
2000
69.7%

change
10.8%

20.5%

17.7%

-2.8%

30.0%
44.5%
60.4%
74.3%
85.1%

44.5%
57.4%
72.5%
84.4%
91.1%

14.5%
12.9%
12.1%
10.1%
5.9%

There are very large differences in the provision of public services between poor
and rich municipalities. For example, the first decile of the distribution (d1) had an
average coverage of 30 percent in local public goods, while the top decile (d10) had
almost three times greater coverage at 85 percent. However, the gap has somewhat
narrowed, since the improvements in local public goods have progressed more rapidly
among the lower half of the distribution. However, the median municipality, even in
2000, would fail to provide these essential public goods to around one fourth of its
inhabitants.
We make a transformation of these percentages into log-odds ratios. One of the
main problems with an OLS estimation using percentages is that we can predict
implausible values outside of the [0,1] interval. The transformation to log-odds ratio is
preferable to the extent that it is more sensitive to differences in the low and the high
ends of the variable (Cleary, 2004). However, a general problem with a logit formulation
94

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

is that it suffers from a well known aggregation problem. If there is heterogeneity in
public service delivery in the localities that comprise the municipality, the model can be
correct for the municipal level, but not for the lower levels of aggregation (Mukherjee et.
al. 1998). This might be particularly problematic if the peripheral areas of a municipality
have a much lower provision of public service than the urban centers (cabeceras). This
aggregation problem is not an issue in the linear probability model, so there is a tradeoff
in the choice of estimation form.
There is a great deal of variation in the dependent variable (see appendix graphs).
Municipalities that show large improvements are not just limited to the poor areas of the
country, but comprise also dynamically growing regions in the West and North of the
country. Improvements in electricity and drinking water provision are dispersed
throughout the country, with no specific area accounting for them in particular.
Municipalities that exhibit little or no improvement are usually places where migration
flows have outpaced the supply of public service delivery (a negative change would mean
that the rate of creation of new households has outpaced the provision of the services).
Thus, improvements in social conditions may be clustered in some areas. Moreover, there
are geographic reasons why it is more difficult to provide public services in some
territories than in others.
In order to ensure that our estimations do not suffer from spatially autocorrelated
errors, we tested for spatial correlation. We calculated a (queen) proximity matrix of
order 2, using GeoDA, in which both the contiguous and the next contiguous municipal
values in the dependent variable were taken into account in order to perform a Moran I

95

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

test. We did not find evidence of diffusion effects or other neighbor influences, so we
report models without a spatial lag.

b) Independent variables.
In terms of the independent variables, we include the initial level of local public
goods provision, an endogenous and two exogenous components of social spending, two
measures of fractionalization, two measures of political accountability, and a set of socioeconomic controls.
Our estimations take the following form:
Public Good =

0 +
1 Initial Level of Provision +
2 Discretional Social Spending (Instrumented) +
3 Formula-Based Social Spending +
4 Local Public Works Budget +
1 Religious Fractionalization +
2 Indigenous Population +
3 Population +
4 Population growth +
1 Alternation Before 1994 +
2 Alternation +
3 Illiteracy +
+

We describe these independent variables below. Even though we are measuring

improvements in local public goods as first differences, one must control for the initial
level of provision (Diaz-Cayeros and Magaloni, 2003; Banerjee and Somanathan, 2006;
Hiskey, 2003; Cleary, 2004). The reason for doing this is that it is often easier to improve
public services in places where there is virtually no provision of public services than to
expand to 100 percent coverage. We have estimated changes in access to public services
as convergence equations: the level of public service provision in 1990 should determine

96

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

the pace at which those services are improved. If convergence exists, well endowed areas
can increase access more slowly than less privileged ones. The coefficient for the initial
level of public good provision, which we expect to be negative, would indicate that the
municipalities lagging behind in public services are the ones that are showing the largest
improvements.
This is the pattern suggested by figure 3.1, which shows the unconditional
convergence of the index of public service provision across municipalities from 1990 to
2000. The horizontal axis depicts the average percentage of homes with access to
drinking water, sewerage and electricity, while the vertical axis shows how much that
percentage changed over the decade. The downward sloping trend of the data indicates
that poor places are catching up. It is not easy to say whether this convergence process is
sufficiently fast. One should note that there are municipalities with negative changes in
the level of provision, which are worse off in 2000 than in 1990. Most of these are very
small municipalities. The convergence seems to be clearer for the municipalities with
more than 50 percent coverage in 1990 than those below. But there is clearly a greater
density of observations aligned in a downward sloping pattern, which is not just an
artifact of the fact that provision cannot go beyond 100 percent.

97

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

Convergence in Local Public Goods

-.5

Change in level of provision (2000-1990)

0
.5

Average Potable Water Electricity and Sewerage

.2
.4
.6
.8
Level of Provision in 1990 (percentage)

Figure 3.1

A. SOCIAL TRANSFERS
Discretional social spending is the endogenous variable of social spending, which
we instrument through geographic variables. The discretional social spending is
measured as the log of average per capita Pronasol local public goods expenditure,
discussed in the previous chapter, measured in real terms (pesos of 1994). Notice that we

98

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

are not using the overall expenditure of the program, since we are well aware that private
good transfers should in principle have little if any effect on local public good
provision.17 Figure 3.2 shows the distribution of this variable across the country. The map
suggests that virtually every municipality in the country received some funds, but that
there was significant variance in the amount of funds distributed to each place.
Figure 3.2

An important feature to note in the map, in contrast with the maps presented in the
previous chapter, is that there is a greater concentration of funds in relatively rich
municipalities, and that there is no clear clustering of funding in the South, where poverty
is widespread.
17

For the sake of comparability with a study like Clearys (2004), the appendix provides a set of
estimations for total Pronasol expenditure.

99

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

The literature on impact evaluation has become increasingly more demanding in
the use of instruments as a perquisite to obtain relatively solid inferences concerning the
impact of public spending in public goods. Zhuravskaya (2000), for example, uses the
Soviet legacy of industrial output over agricultural production as an instrument of
spending, in order to assess its impact on public good provision. Banerjee and Ayer
(2005) take advantage of the different political organizations and land tenure
arrangements among Indian princely states during the colonial era in order to asses the
effect of land distribution on economic performance. And Paxton and Schady (2002) use
Fujimori vote as an instrument to predict Foncodes resource allocations in Peru.18
In the case of Mexico, several geographic variables turn out to be good predictors
of expenditure allocations, but they are not correlated with improvements in public
service delivery. Hence they provide us with excellent instruments. In the estimation
reported in the next section, we kept the following four instruments:
1) Rainfall (average in the centroid, in millimeters)
2) Access to railroads (Euclidean distance of border of municipality to the
closest railroad track, in kilometers)
3) Access to cities (Euclidean distance of border of municipality to the closest
city of more than 100,000 inhabitants, in kilometers)
4) Rugged terrain (variance of altitude for the municipality as a whole in meters)
These instruments turn out to be good predictors of the spending variables
because policy makers probably took into account the profile of poverty when deciding
how to allocate funds. In some ways they all reflect the isolation of some communities

18

Instrumenting with vote results would be somewhat odd in out study, since our goal is to understand the
interaction between social transfers and politics: we hence cannot claim electoral patterns are exogenous.

100

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

which are trapped in poverty. While the overall level of public service provision is highly
correlated with poverty, and the geographic instruments are also correlated with poverty,
this is not the case for changes in the provision of public services.
We use the log of the per capita average funds allocated to the FDSM/FISM
between 1996 and 2000 (in real terms) as a measure of Formula Based Social Spending.19
There is no endogeneity problem in the allocation of these funds. The formula for
allocation of FDSM created in 1997 was based on calculating gaps in public service
provision and poverty, similar to a Foster-Greer-Thorbecke poverty index (Mogolln,
2000; Levy, 2001). These gaps provided estimates of so called Deprivation Densities
(Masas Carenciales Estatales), which determined the amount of funds states would
receive. For the distribution from the states to the municipalities, governors could use the
same federal formula, or a simpler allocation formula based on very similar variables to
those in our index. 19 states used the simpler formula, while 12 states used the more
targeted federal formula. Nonetheless, there was no discretion in the allocation of these
funds to the municipalities. During the first years of the FDSM/FISM a third of the funds
were distributed to states in equal shares, regardless of population, but state governments
were bound to distribute the resources to their municipalities by formula.20
Public expenditure through federal programs is complemented with municipal
funds. In fact, local governments in Mexico are responsible for the provision of drinking

19

We have no breakdown of social spending data for 1995 at the municipal level, although there is
anecdotal evidence suggesting a very large drop in federal allocations, as a consequence of the peso crisis.
That year Pronasol was abandoned by the incoming Zedillo administration, to be substituted by a new fund
for social infrastructure. A third of the formerly Pronasol money included in budgetary item 26 was
decentralized in 1996; and the decentralized share became 2/3 in 1997 The FDSM, renamed FISM in 1999,
became the most important federal transfer to municipalities.

20

The formulas have not changed since, only the census indicators used to calculate them have been
updated. The lump sum for each state was reduced to .5% of the funds, and eliminated after 2001.

101

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

water, sewerage and public lighting, among other local services (See Rodriguez, 1995).
Public works spending (gasto en obras pblicas y fomento) by municipalities constitutes
the part of their budget they invest, as distinct from expenditure in debt service and
general administration (mostly comprised by payrolls). On average, municipalities during
this period spent about half of their budgets on public works. Their budgets were much
smaller than the overall Pronasol funds, and relatively similar to FDSM/FISM funds.21
We include the municipally allocated public works spending (Local Public Works
Budget), measured as the logged average per capita allocation from 1989 to 2000.22 We
expect all the expenditure variables to have a positive sign, suggesting that when more
money is spent, the percentage of households without basic public services should be
reduced. We have no a priori expectations about whether discretional Pronasol funds,
formula based social infrastructure federal transfers, or municipal public works spending
should have a larger effect.

B. SOCIAL CLEAVAGES
In what regards the measures of social polarization, religious competition is the
most salient cleavage that drives conflict in Southern Mexico (Trejo, 2005). The
connection between religion and conflict is related to the church active construction of
social networks for pastoral purposes, spurred by the pressures of religious competition.
21

Public works expenditure by municipalities amounted to around 27 percent of Pronasol funds.

Since Pronasol often involved matching funds from the municipality for the federally financed projects,
in a previous version (Diaz-Cayeros and Magaloni, 2003) we included an interaction term of Pronasol
funds multiplied by public works spending. That interactive term showed a statistically significant effect,
but extremely small, so we decided not to include it in the current specification.

22

102

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

Trejo (2005) shows that as a reaction to Evangelical inroads in their dioceses, the
Catholic Church in Chiapas became more attentive to the needs and demands of poor
indigenous communities. This does not mean that religious affiliation becomes a source
of violence or contention, but rather that conflicts between citizens and state structures
might be more intense in places with more active religious competition. An indigenous
cleavage is less prevalent, since localities tend to be more homogeneous in this ethnic
dimension than in the religious one. We have hence calculated a measure of religious
fractionalization, which is an index of how divided religious belief in a municipality is.
We calculate the index defined as RF=1- pi2, where p is the share of each of the
religious affiliations reported in the 1990 census (Protestantism, Catholicism, Judaism,
other religion and agnostic).23
The second indicator of social cleavages we use is the share of Indigenous
Population, defined as monolingual individuals according to the 1990 census. While
there is a very large correlation between poverty and being indian, the indigenous
character of a community might not reflect divisiveness, but in fact improve the provision
of public goods. This might be particularly so in places where social organization makes
collective endeavors less subject to shirking and opportunism. For example, the most
tightly knit communities in Oaxaca, use the Tequio as a mechanism for the provision of
public goods and services.24 Those municipalities often use a system of rotation in public
offices, instead of having municipal presidents selected through elections (elecciones por
23

Perhaps reflecting the salience of this issue, the 2000 census includes a new category, breaking down
Protestantism by Evangelical and non-Evangelical. The fractionalization index of 2000 increases compared
to 1990 even without the finer categorization. However, it is likely that the largest increases in religious
fractionalization occurred in the 1970s and 1980s.
24
The Tequio is a form of communitarian cooperation for the provision of local public goods. It involves
compulsion, in the sense that members of the community must devote some of their labor for a collective
enterprise, but it is voluntary to the extent that it accords with the traditional values of most members in the
community.

103

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

usos y costumbres). Such system has turned out to be a very effective governance
device.25
Births and migration might exert pressure on the existing physical infrastructure,
so that it should be more difficult for places where population growth is high to keep up
with the provision of public services. We hence include a control for demographic growth
(Growth population), measured as the rate of population change between 1990 and 2000,
and the initial size of each municipality, as measured by its logged Population in 1990.
We expect the demographic growth variable to have a negative sign, suggesting that it is
more difficult to incorporate households without public services where demographic
pressures are high. We have no expectation in the sign of population, given that it might
be easier to provide public services in large cities; but it is also possible that there is far
more room for improvement in small places. However, demographic features should be
related to social complexity and a larger number of cross cutting cleavages (Cox, 1997).

C. DEMOCRACY
Accountability in public service delivery might be influenced by political
participation and electoral competition (Hiskey, 2003; Cleary, 2002 and 2004). The main
purpose of this chapter is not to evaluate these hypotheses, but to assess the overall effect
of policy interventions. We control, however, for variables that reflect democracy and
25

It would be possible to use census data to construct an index of ethnic fractionalization going beyond the
division between indian and non-indian. However, this would involve a rather laborious process, since the
data is not in electronic form (the 2000 census does provide a breakdown of the two main indigenous ethnic
groups in electronic form). While the lack of electronic data has not stopped us at other stages of this
project, we judged that the ethnic hypothesis is somewhat peripheral to the project, and probably not true
for ethnic affiliations, therefore not worth this extra effort. We leave it for other researchers to verify
whether this is the case.

104

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

political change. The simplest measure of whether a municipality in Mexico can be
accountable to citizens is alternation in power. If the PRI had lost office at the municipal
level during the decade we analyze, following Przeworski et. al. (2000), we consider that
a transition to democracy has actually occurred. We provide two measures of political
alternation, one for municipalities that voted for a non-PRI mayor during the Salinas
period, during the first half of the decade (Alternation Before 1994), and those that voted
for some non-PRI mayor at any point in the decade (Alternation).26
The final independent variable we use is illiteracy rates. This variable can be
thought of as a control for the political awareness and citizen demands for public services
(Banerjee, 2004). But it might simply reflect overall levels of development. Cleary
(2004) argues that although literacy is correlated with development, it can measure some
of the participatory elements of democracy. We thus include Illiteracy, measured as the
percent of population over 15 that could not read and write in the 1990 census, as an
indicator of local empowerment.

4. Results and discussion

We now proceed to discuss the main findings. The estimations were all run in
STATA with robust standard errors. The first regression reported in table 3.2 is a nave
OLS estimation, in which each of the spending variables is included without taking into
account potential endogenity problems. The strong and significant sign of Discretional

26

Hiskey (2003) and Cleary (2004) test their hypotheses in an interactive way, showing that the effect of
spending is mediated by the type of electoral competition. Our main concern is not this interactive effect, so
we prefer to keep the simpler formulation of a direct control for the possible effect of democratic
accountability on public goods provision.

105

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

Social Spending would suggest that Pronasol had positive welfare effects. The size of the
effect cannot be interpreted directly, given that the dependent variable is measured as
changes in the log-odds ratios, but it suggests that for the same amount of money,
Pronasol spending was more effective than the funds channeled by local governments for
public works. The Municipal Public Works Budget is also positive and significant, but the
magnitude of its coefficient is half of that for Pronasol.
Hence one could tentatively conclude that a centralized discretional resource
allocation was more effective than the decentralized choices made by municipal
governments. The coefficient for the Formula-Based Social Spending comes out as
significant, but negative! This coefficient would imply that places with larger spending
end up with a worse provision of public services. As we discuss below, the coefficient is
the result of the specific formulas that underlie the distribution of FISM.

106

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

Table 3.2 Determinants of the Change in Local Public Goods Coverage

Initial Level of
Provision
Discretional
Social Spending
Formula-based
Social Spending

(1)
OLS

(2)
OLS

(3)
TSLS

-0.474
(0.014)**

-0.402
(0.025)**

-0.403
(0.025)**

0.108
(0.014)**

0.063
(0.015)**

0.028
(0.073)

0.269
(0.116)*

0.276
(0.122)*

0.058
(0.011)**

0.042
(0.013)**

0.047
(0.017)**

-0.141
(0.071)*

-0.143
(0.078)

-0.12
(0.09)

0.739
(0.154)**

0.824
(0.198)**

0.846
(0.206)**

0.128
(0.033)**

0.12
(0.030)**

0.119
(0.030)**

-0.038
(0.021)

0.003
(0.021)

-0.002
(0.023)

-1.543
(0.123)**

-1.822
(0.169)**

-1.839
(0.176)**

0.018
(0.015)

0.067
(0.015)**

0.052
(0.034)

-0.161
(0.047)**

-0.039
(0.051)

-0.04
(0.053)

1.396
(0.242)**

-2.332
(0.820)**

-2.16
(0.943)*

-0.292
(0.031)**

Residual
Formula
Local Public
Works Budget
Religious
Fractionalization
Indigenous
Population
Alternation
Before 1994
Alternation

Illiteracy

Population

Population
Growth

Constant

Observations
2381
2365
2364
R-squared
0.39
0.38
0.37
Robust standard errors in parentheses
* significant at 5% level; ** significant at 1% level
TSLS run instrumenting Pronasol expenditure with
Rainfall, Distance to City, Distance to Railroad Track
and Rugged Terrain

107

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

The estimation shows that the improvement in public good provision in Mexican
municipalities is powerfully influenced by the Initial Levels of Provision, suggesting that
in the overall process of modernization, there has been convergence of public good
provision across municipalities.27 In what regards the other independent variables, the
results are consistent with our expectations.
There is evidence suggesting that indian communities as proxied by
monolinguism are better at providing public goods to themselves.28 Religious
fractionalization, instead, shows a negative effect on public good provision. It is likely
that religious fractionalization plays an even larger role in polarizing communities if a
subsample of the Southern Mexican municipalities is analyzed. As discussed above, this
variable is positively correlated with violence, implying that more conflictive places are
less effective at providing public goods. Demographic pressures seem to make it harder
for municipalities to provide public goods, as reflected in the negative coefficient of the
population growth variable.
Democracy has a peculiar effect: a positive impact in municipalities in the around
300 municipalities which made a transition to democracy before 1994, and no impact in
the rest. This might be due to two different explanatory variables. The first is the age of
democracy, leading to governments which are more effective at public good provision
because they have more experience with democratic accountability. The other possibility
is that types of municipalities that exhibited alternation before 1994 were different from
27

The speed of convergence for public service delivery in Mexico is relatively fast: we have estimated that
the half-life of unconditional convergence (i.e. the time it takes for half of the initial gap to be eliminated)
is of around 11 years for water and 9 years for electricity.
28
This might be due to the community mechanisms that allow those municipalities to overcome collective
action problems and generate a greater trust in their government. Of course those municipalities also
happen to be relatively poor and deprived, which might be a factor that simply indicates their convergence
speed is faster than that of the rest of the country.

108

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

the rest. For example, they might be the places where civil society was more organized to
fight authoritarianism. By contrast, municipalities that experienced alternation after 1994
did so in this interpretation due to a tipping phenomenon triggered by the peso crisis and
the splits within the PRI (Magaloni, 2006). In this interpretation, pre-1994 alternation
municipalities did not only have a longer experience with democracy, but also stronger
civil society and citizens embedded in social movements and other forms of civic
engagement. The illiteracy coefficient is highly significant and negative, giving support
to Cleary (2004) who suggests that places where citizens are more empowered through
education they can exercise their voice more effectively, and hence improve their
provision of public goods.
Before continuing with the estimation that controls for endogeneity, we must
discuss an amendment to the variable measuring formula based allocations. The negative
coefficient of the convergence effect gives a hint as to why we obtain the unexpected
negative coefficient for this variable. The problem is that the formula of FISM is
extremely correlated with the initial levels of public service provision in fact, it is close
to measuring the same variable, and hence its negative sign. Following Mogolln (2002),
we can reconstruct the allocation of FDSM/FISM, by running a regression of the
distribution of funds on the census indicators that are included in the so called second,
simplified formula.29 This involves making a linear approximation which is not accurate
for the states that distribute funds to their municipalities through the more targeted

29

The fact that the match is so large should provide more confidence in the notion that we can treat this
spending as exogenous, although of course the formula itself might reflect considerations of benefiting
some areas of places more than others.

109

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

federal formula, but still reflects the bulk of the allocations. 30 Such estimation is
presented in the appendix. The residual of that estimation measures how much the
allocations depart from a simplified formula, and hence it provides a proxy for the
fraction of the FISM allocation that is generated by a highly targeted formula, using
deprivation densities (i.e. a formula that takes into account the distribution of poverty,
and gives greater weight to the poorest households), rather than just a linear combination
of average census variables.
Going back to table 3.2, the second column provides the same nave OLS
regression, but now using the residual of FDSM/FISM obtained from the appendix
regression, instead of the total funds, as the independent variable measuring formula
based spending.31 The effect of Formula-Based Social Spending is now positive,
significant, and much larger than the one for Pronasol. The Municipal Public Works
Budget retains its magnitude and significance, and so do most of the other coefficients.
Now the general conclusion seems to be that centralized, but highly targeted programs
might be the most effective to improve local public goods provision; while both a
discretional program and the decentralized decisions of mayors might have similar
effects.
But even the modest effect of Pronasol turns out to be an artifact of endogeneity.
This is the issue that is addressed in the third column, which instruments Pronasol
spending through the use of the four geographic variables, namely distance to the closest
city of more than 100,000 inhabitants, distance to a railroad track, rugged terrain and

30

The states that use the more targeted formula are Aguascalientes, Coahuila, Chiapas, Guanajuato,
Hidalgo, Mxico, Michoacn, Nayarit, Puebla, San Luis Potos, Sonora and Tamaulipas.
31
In order to keep all the expenditure independent variables in the same logarithmic metric, we made a log
transformation of the form: lresidual=log(1582+residual).

110

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

average rainfall. In the instrumental variables estimation the effect of Pronasol is now
very small and not statistically significant. That estimation retains most results for the
other independent variables. The coefficient for religious fragmentation, however, fails to
reach statistical significance at the 95 percent level.32 There seems to be no effect of
demographic pressures in the failure to provide local public goods once discretional
spending was instrumented.
The analysis hence suggests that the impact of Pronasol was fairly limited. The
programs investments were not geographically targeted to the places with the greatest
needs, as we will see in the coming chapters of the book, but to places where the PRI saw
its electoral future most threatened, predominantly in cities. FISM performs much better
than the decentralized decisions made by mayors in their own budgets. This could be a
consequence of a crowding out phenomenon, in which mayors see little reason to
devote sizeable parts of their budget to public works when they know they have the
earmarked FISM funds available.33
Improvements in public goods and social conditions are also attributable to
economic and social convergence processes that, while perhaps connected with
government action, are not directly attributable to the specific distribution of social
spending spent across municipalities. The notion of convergence implies that the
municipalities with the worst provision of public goods are the ones that can improve it
more quickly. This natural process of convergence in development indicators is fastest
in Mexico for the provision of electricity and drinking water, and practically non-existent
for sewerage. Patterns of investment, which are spread across the country, can contribute
32

But it is quite close, in the 85 percent range.

While FIS greatly improved targeting, its decentralized control by mayors might still have an urban bias.
Such bias would limit its effectiveness in the improvement of public good provision.

33

111

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

to such convergence. A relatively small level of public investment might have a greater
impact in poorer municipalities than in richer ones.34 To sum up, we find that the highly
decentralized and discretional program, Pronasol, was disappointing in its improvement
of welfare through public good provision. The transition towards a formula based
program, FISM, was welfare enhancing. A full fledged political economy explanation of
the limited impact of Pronasol and its transformation into FISM is carried out in the next
chapters of the book. It will come as no surprise that clientelism and pork barrel politics
played a key role in the design and implementation of Pronasol.

34

The opposite effect is also theoretically possible: where there is a relatively good provision of public
goods it might not be so expensive to extend the coverage; while in places with almost no public services
the fixed costs might be very high.

112

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

Appendix to Chapter 3
A. Census vs. Conteo Indicators
We refrain from making comparisons between the census and the population count of
1995 (which we did in Diaz-Cayeros and Magaloni (2003)), because a careful
examination of the data suggests that the Conteos overestimate the actual improvement in
public service provision. The Conteo de Poblacin is in fact based on two surveys, the
Ennumeracin that counts all the households in the country, and the Encuesta
(n=80,000), which has detailed information on families and social services based on
2,500 questionnaires per state. The Conteo in this sense is not a full count of public
service delivery at the municipal level. For example, the average improvement in potable
water provision between 1990 and 1995 is calculated as 9.4 percent; and the increased
coverage in electricity is 10 percent. If these figures were correct, there was virtually no
change in the provision of those services between 1995 and 2000. Although there was an
economic crisis in 1995, such conclusion is not plausible (an inspection of the graphs in
the appendix suggests that a rather large number of municipalities would have a worse
service provision than five years earlier). A more plausible explanation is that the
techniques used to estimate public service provision in the Conteos at the municipal level
are somewhat biased. We can compare the provisions between 1995 and 2005, assuming
that the bias is systematic among Conteos. Preliminary analysis suggests that our main
results remain unchanged, although in that estimation we are unable to incorporate an
assessment of Pronasol spending, given that the program was ended by 1995.

113

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

Figure A1. Households Without Access to Potable Water 1990-2005
0

.5

.5

1
1

sinagua90

.5

sinagua95

.5

sinagua00

.5

SinAgua05

.5

0
0

.5

.5

114

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

Figure A2. Households Without Access to Sewerage 1990-2005
0

.5

.5

1
1

sindren90

.5

sindren95

.5

sindren00

.5

SinDren05

.5

0
0

.5

.5

115

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

Figure A3. Households Without Access to Electricity 1990-2005
0

.5

.5

1
1

sinelec90

.5

sinelec95

.5

sinelec00

.5

SinElec05

.5

0
0

.5

.5

116

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

B. Breakdown by programs
Figure A4 shows the per capita allocation of Pronasol funds accumulated in real terms
from 1989 to 1994 within the programs to improve sewage, drinking water and
electricity. The graph highlights that these per capita allocations differed depending on
the municipal level of development (in this case, ranked according to CONAPOs
poverty level in 1990). These allocations are compared to the funds for education
infrastructure and the municipal funds program, appropriations within Pronasol fully
controlled by the municipal governments. Fondos municipales was initially small, but as
pressures for decentralization and a desire for municipal government to have a greater say
in the selection of projects mounted, these resources became increasingly more
prominent. On average, fondos municipales were 20 percent of total Pronasol
expenditure. The education figures include only the funds geared toward repairing and
constructing schools, not the scholarships going to individuals (i.e. they include only the
local public good component of this expenditures). While the specific allocations
differed, according to the specific characteristics of the municipalities, overall around 17
percent of the total program (including private goods transfers) went for drinking water
and sewage and 13 percent to electricity. Unfortunately, a breakup of sewage and
drinking water expenditure separately was not available due to the way Pronasol reported
its programs. Elsewhere (Diaz-Cayeros and Magaloni, 2003) we estimated the effect of
each of the Pronasol expenditures separately: water and sewerage; electricity; municipal
funds; and the remainder of the funds. The instrumental variable approach does not
perform well with individual programs. We have no good instruments to explain the
breakdown of allocations by program, while the instruments work quite well for the
overall spending. Since it is quite likely that the developmental impact of the program did
not just depend on the expenditure of specific projects in isolation, but on the overall
package of investment in a given municipality, we prefer to make the analysis with total
public goods funds.

117

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

Figure A4
Average Allocations by Programs (According to Poverty
Level)
200
180
160

per capita 1993 pesos

140
120
Drinking Water and Sewage
Electricity
Education
Municipal Funds

100
80
60
40
20
0
very low

low

average
CONAPO poverty level

118

high

very high

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

C. Reverse Engineering the FDSM/FISM formula.
Table A4 shows the results of regressing INEGI Conteo indicators of the percentage
households without electricity and sewerage, the illiteracy rate, a poverty head count and
the inverse of the population, on the per capita FDSM/FISM funds (not logged). The
inverse of the population is included due to the transitional rule that gave states for
several years a fixed lump sum share of FISM, regardless of their size. That coefficient
measures the value of being a municipality, since it is the amount of funds each
municipality gets in FISM regardless of its size or deprivation indicators. The results
show higher coefficients for illiteracy and electricity than for poverty and sewerage.35
Table 3.3 Determinants of FDSM/FISM allocations (1996-2000)
Population
Weighted
No electricity 1995
445.33
(3351.08)**
No sewerage 1995
224.731
(3226.19)**
Illiteracy 1995
455.803
(2348.50)**
Poverty 1995
185.062
(1680.67)**
1/Population
373579.8
(2687.78)**
Constant
18.263
(941.51)**
Observations
8.06E+07
R-squared
0.81
Absolute value of t-statistics in parentheses
* significant at 5% level; ** significant at 1%
level

The parenthesis in table 3.3 reports the t statistics, which are huge, given that we are
reverse engineering how the formulas end up producing the allocations.

35

This is perhaps surprising, given that the federal formula gives almost half of the weight to poverty
(.4616), followed by the characteristics in the dwellings (0.2386, which are not included in the simplified
formula), next illiteracy (0.125), electricity (0.114), and sewerage last (0.0608). The simplified formula
gives equal weight to all four deprivation factors included in the estimation.

119

Diaz-Cayeros, Estvez and Magaloni, Ch. 3

C. Total Funds including private good provision
Table A1. Effects With Total Pronasol Expenditure

Logindex90
Pronasol
FISM
Residual FIS
Pub Works

(1)

(2)

(3)

OLS
-0.469
(0.014)**
0.09
(0.017)**
-0.282
(0.031)**

OLS
-0.402
(0.025)**
0.024
(0.016)

TSLS
-0.395
(0.013)**
0.115
(0.08)

0.278
(0.123)*
0.049
(0.013)**
0.848
(0.199)**
-0.116
(0.077)
0.05
(0.015)**
-0.037
(0.052)
0.119
(0.030)**
-0.003
(0.021)
-1.84
(0.170)**
-2.175
(0.865)*
2365
0.37

0.255
(0.045)**
0.041
(0.014)**
0.815
(0.165)**
-0.171
(0.087)*
0.094
(0.038)*
0.000
(0.000)*
0.117
(0.033)**
0.014
(0.025)
-1.802
(0.128)**
-2.786
(0.616)**
2364
0.36

0.069
(0.011)**
Monolingual
0.774
(0.155)**
Religious.Fractionalization
-0.129
(0.071)
Population
0.018
(0.015)
Pop. Growth
0.000
(0.000)**
Alternation 1994
0.124
(0.033)**
Alternation 2000
-0.032
(0.021)
Illiteracy
-1.594
(0.124)**
Constant
1.236
(0.242)**
Observations
2381
R-squared
0.39
Standard errors in parentheses
* significant at 5% level; ** significant at 1% level

120