Diaz-Cayeros Estevez Magaloni
Diaz-Cayeros Estevez Magaloni
Diaz-Cayeros Estevez Magaloni
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
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.
80
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
82
83
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
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
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
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
88
89
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
91
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
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
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
95
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 +
+
96
97
-.5
.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
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
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
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
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
22
102
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
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
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
106
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
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
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
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
33
111
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
113
.5
.5
1
1
sinagua90
.5
sinagua95
.5
sinagua00
.5
SinAgua05
.5
0
0
.5
.5
114
.5
.5
1
1
sindren90
.5
sindren95
.5
sindren00
.5
SinDren05
.5
0
0
.5
.5
115
.5
.5
1
1
sinelec90
.5
sinelec95
.5
sinelec00
.5
SinElec05
.5
0
0
.5
.5
116
117
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
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
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