Does Corruption Affect Health and Education Outcomes in the Philippines? Omar Azfar * Institutional Reform and Informal Sector University of Maryland at College Park College Park, MD 20742
Tugrul Gurgur Department of Economics University of Maryland at College Park College Park, MD 20742
We examine the effect of corruption on health and education outcomes in the Philippines. We find that corruption reduces the immunization rates, delays the vaccination of newborns, discourages the use of public health clinics, reduces satisfaction of households with public health services, and increases waiting time at health clinics. Corruption also has a negative effect on education outcomes: it reduces test scores, lowers national ranking of schools, raises variation of test scores across schools and reduces satisfaction ratings. We also find that corruption affects public services in rural areas in different ways than urban areas, and that corruption harms the poor more than the wealthy. JEL Classifications: H4, I1, I2. Key words: corruption, decentralization, health care, education, service delivery.
*
[email protected]. The research underlying this paper was supported by a grant from the World Bank, financed by the Netherlands Trust Fund. We are extremely grateful to Tony Lanyi for his management of this project, to Satu Kahkonen for survey design and implementation (assisted by the staff at Social Weather Stations), and to Diana Rutherford for excellent project support. We thank Shanta Devarajan, Malika Krishnamurty, Jennie Litvack and Ritva Reinika at the World Bank for their support. We are also grateful for comments on an earlier draft of this paper from those already mentioned, as well as from Roger Betancourt, and Peter Murrell and the attendees of the 2001 NEUDC conference. All errors are our own.
I. Introduction There are several mechanisms by which corruption might undermine service delivery. Corruption can increase the cost to consumers if a bribe is demanded in addition to the official payment, which reduces demand for services and therefore may worsen health and education outcomes. If however corruption takes the form of the official pocketing the payment intended for the government, this reduces government resources allocated to service delivery, which would also worsen outcomes. As noted by Pritchett (1996) the relationship between public spending and outcomes is usually ambiguous in many countries, and this ambiguity may be a reflection of differences in the efficacy of spending due to corruption. A number of past studies have looked at the effect of corruption on public sector performance in health, education, infrastructure, etc. (Gray-Molina et al. (1999), Gupta et al. (2002), Reinikka and Svensson (2001), Rajkumar and Swaroop (2002)). In this paper we supplement these results and examine the effects of corruption at the local level in the Philippines on health and education services. This approach has the advantage that it keeps fixed a large number of variables that vary across countries and cause omitted variable problems and other econometric problems in cross sectional regressions. Five years after the democratic revolution in the Philippines, the Local Governments Act of 1991 devolved both political authority and administrative control of many health services and other subjects to the provincial and municipal level. Much of the corruption in the Philippines does appear to be at the local level: of the 336 corruption cases current in mid-2000, 49% were against municipal mayors- the level of government, which is our focus in this study (Batalla 2000). Many observers have stated that corruption is the root cause of continued poverty in the
1
Philippines (World Bank (2000)). All this makes our study of local level corruption and service delivery in the Philippines highly relevant in a country specific context. In addition studies such as this one might have global relevance in terms of the increasingly important question of the effect of corruption on service delivery. Our results resonate with cross-country results (e.g., Gupta et al. (2002)), which find a negative correlation between corruption and health outcomes. However, we acknowledge the difficulty of generalizing from cross-country results and one possibly unrepresentative country, and would prefer to replicate this study in other countries with a large number of local governments before making global prescriptions. This paper is structured as follows. We begin by describing the data. In particular we examine in detail the quality of the data on corruption and are reassured by a number of correlations across samples. The corruption perceptions of households, municipal administrators and municipal health officers are all correlated with each other and the corruption perceptions of households are highly correlated with the corruption perceptions of other households in their municipality. Emboldened by these findings, we begin to examine the consequences of corruption. Here we use several different outcome measures from different sources. We use households’ reports of waiting time, their satisfaction with government health services, access to public health clinics, immunization rate of children, and delay between the birth of a child and his/her immunization. In each case we find the expected negative and significant effect of corruption on performance. Our results for education are similar. We find a significant negative effect on test scores, national ranking of schools, variation of test scores within schools, and household assessments of satisfaction with public education. Our empirical analysis also highlights the
2
disproportionate burden of corruption on the poor. The perverse effect of corruption on health and education outcomes is also more serious in rural areas as compared to urban areas. We provide information about the experience of Philippines with decentralization reforms in the next section. The data and the econometric models are described in sections 3 and 4, respectively. We analyze the consequences of corruption in section 5. The robustness of our results is discussed in Section 6. A conclusion follows.
II. Country Background The Philippines is a country of 70 million people who live upon thousands of islands that lie between the Pacific Ocean and the South China Sea1. The larger of these islands have vast expanses of mountains and jungles that physically separate large populations. The sheer geography of the Philippines necessitated some form of decentralized or at least deconcentrated governance for centuries, but this was not always combined with devolution of political authority. Decentralization in the Philippines was mandated by the new democratic constitution of 1987. The Local Government Code (LGC) enacted in 1991 significantly increased the responsibilities and resources of sub-national governments: 77 provinces, 72 cities, 1526 municipalities and over 40,000 barangays or neighborhoods2. In addition, it mandated regular elections for local executives and legislative bodies. The Code devolved “basic services” to local governments—these include most health services along with such infrastructure provision as school, clinic, and local road building. Local government units (LGUs) have authority to create their own revenue sources (within firm limits), as well as to enter international aid agreements. 1
The country background is adapted from Azfar, O. et al. (2004). It corresponds to the situation in 2000, when the data was collected.
3
Municipalities have responsibility for primary health care, disease control, purchase of supplies and equipment necessary for this, as well as municipal health facility and school buildings. The barangay, the lowest formal level of government, is described in the Code as the “primary planning and implementing unit of government policies…” In practice, the barangays have little policymaking or planning capacity, although they have significant fiscal resources in comparison to their responsibilities. The President exercises “general supervision” of the legality and appropriateness of LGU actions. Prior to 1992, Philippine public finance was highly centralized, with the central government accounting for almost 92 percent of all public expenditure and more than 95 percent of all revenue collection. Over the following three years, decentralization reduced these levels to 87.4 percent and 94.6 percent, respectively (Manasan (1997)). Expenditures devolved to subnational governments covered a wide range of government activities, the most prominent of which was health, accounting for more than 53 percent of the devolved expenditures. Primary health care is significantly devolved in the Philippines, with staff being hired, fired, and paid (according to a nationally-defined scale, and mostly with central grant funds) by the local governments. Many localities use their discretionary resources to supplement the health staff salaries defined by the central government, while others attempt to deal with fiscal shortfalls by hiring fewer or cheaper health staff. The Local Government Code provides that provinces, cities, and municipalities are all to have Health Officers as well as Health Boards (the barangays provide only minimal health services). Assessments of decentralization’s impact on public health service provision in the Philippines are mixed, with experts concerned about deterioration in the technical quality and administration of the programs, but most people expressing more positive views. Despite 2
These numbers change as new units are created or old ones combined (Miller (1997)).
4
scandals in centralized medicine procurement, the purchase of many medicines has in fact been decentralized, and as a result many observers now say that medications are more appropriate and there is less leakage of resources out of the system than previously. On the other side, studies suggest that the Philippines made its most notable public health system advances in the 1980s— bolstering programs on malaria, immunizations, TB, maternal and child health, and other areas to counter a stagnation in health indicators from the late 1970s into the 1980s—and that things have slid since then (World Bank (1994)). By contrast, governance of public education is centralized under the administration of the Department of Education, Culture and Sports (DECS), but with some (at times significant) local input. The LGC assigns school building construction and repair to the local governments, and the center is responsible for practically everything else, including policy, curriculum, personnel, and operations. Local institutions with a formal role in education governance include the School Boards at provincial and municipal levels (mainly for programming the Special Education Fund or SEF, see below), and the Parent-Teacher Community Associations (PTCAs, essentially the same as PTAs elsewhere) for each school. Local officials such as governors and mayors are represented on the boards and often use them to exercise influence. As with health, there are no fees charged (formally) in the school system, though parents have to buy uniforms and pay modest PTCA fees. As envisioned in the Local Government Code, DECS chooses local school teachers and administrators in consultation with local School Boards. In practice governors and mayors do try to influence teacher hiring and transfers (see Azfar et al. 2003 for more details). Thus we may expect to find some impact of local governance on education outcomes. Furthermore, while the budgeting of financial resources in particular is highly centralized in the
5
Philippines, the local share of education finance has grown. There is a major tax earmark for education, the Special Education Fund (SEF), whose uses are determined by local school boards.
III. Data Description Our data is based on eight surveys undertaken in the spring of 2000. The sample covered 19 provinces and 80 municipalities from 11 regions. We surveyed 1100 households; 80 municipal administrators, 80 health officials and 80 education officials; 19 provincial administrators, 19 provincial health officials and 19 provincial education officials, 160 government health facility workers and 160 school principals –some private (50) and some public (110). The sample of households represents 19 provinces, 80 municipalities within them, and 301 barangays3 within those 80 municipalities. Households can be matched to either schools or health facilities at the barangay level. We begin by discussing our central variable of interest –corruption. We define corruption as the abuse of office for personal gain (Klitgaard et al. (2000)). Corruption manifests itself in several ways: through shirking, the sale of jobs, bribery, and the theft of funds and supplies. We asked questions about all these improprieties in the surveys of government officials. Results are presented in Table 1. There are reports of all kinds of corruption in each kind of government office –with the sole exception of the theft of supplies in the municipal education (DECS) office. Most kinds of corruption are more prevalent in the municipal administrators office than in other offices, perhaps due to the administrator’s office exerting more authority and thus having more opportunity to extract rents (see Azfar and Gurgur 2000 for more on this subject). Nineteen
3
Barangay is the smallest political unit into which cities and municipalities in the Philippines are divided. It is the basic unit of the Philippine political system. It often consists of less than 1,000 inhabitants residing within the territorial limit of a city or municipality and administered by a set of elective officials, headed by a barangay chairman.
6
percent of municipal administrators stated that there were cases of bribery in their office in the last year and a full 32% that there were instances of the theft of funds. By contrast only 2.5% of municipal health officers and 1.3 % of municipal DECS officers reported incidents of bribery in their office with 16.5% and 1.3%, respectively, reporting the theft of funds. We next created an index of corruption from its various components. This index is the normalized sum of the first seven variables in Table 1. This index is correlated at 0.5 or above with most of its components for both the municipal administrator and the municipal DECS officers. These high correlations, which reflect some positive link among the components of the index, are the first sign that our index is measuring some coherent underlying variable. We had also asked a general question on “how common is corruption in the municipal government” with four possible answers: non-existent; rare; common; very common. If our index were a good measure of corruption we would expect it to be correlated with the answer to this question. In fact the indices are highly correlated with the answers to this general question with a significance level of 5% or less for all municipal officials and officials at public health centers. Moreover the corruption perceptions of households and public officials are positively correlated: We constructed a “public officials corruption index” combining the answers of public officials working at public schools, health clinics and municipalities. The resulting index is correlated at 0.28 (p-value=0.01) with households’ corruption perceptions. We also constructed measures of other aspects of public sector management and civic community, which are categorized in three groups:
7
1. School/Health Center and Community Resources: physical and human resources of schools or health centers (capacity index); size of school/ or health center, location (urban area); education level of households; financial wealth of households 2. Voice, Exit, Civic Participation (social capital) variables: newspaper readership among households; voter turn-out; effect of ethnicity in voting; existence of private schools/health centers in the area; distance to closest health facility 3. Institutions: Internal accountability of schools, health facilities, and municipalities (accountability index); frequency of audit by upper government; autonomy of schools/health centers and municipalities in decision making Most of these variables, which are described in Table 2, are composite indices between 0 and 1 and are constructed using several survey questions4. Our measures of performance in health and education services include both hard data obtained from the Ministry of Health and the Ministry of Education in Philippines and users’ perceptions and responses that we derive from the Household Survey. These variables are explained in Table 3.
IV. Econometric Model We estimate the effect of corruption (as well as other variables) on performance of health and education services using various methods, including random effect, tobit, ordinary-least squares and robust regressions, at the household, school, and municipality level as appropriate. The main justification for the random effects model is the individual units of analyses (households) in our sample nested within higher-level units of analysis, which we identify as
4
An earlier draft entitled “Decentralization and corruption in the Philippines” contains tables that describe these variables in considerable detail and a number of reliability tests. This paper is available upon request.
8
provinces. A province in Philippines is the largest unit in the political structure, comprising of a number of municipalities and (in some cases) cities with more or less homogenous characteristics, such as ethnic origin of inhabitants, dialect spoken, agricultural produce, etc. Therefore, it is possible that unobservable effects in each cross-section (province) may cause biased estimates. To address these concerns, we consider fixed-effects and random-effects models for econometric analysis. As long as it is theoretically and statistically justified, a random effects model is preferred to a fixed (within) effect model, since the latter ignores the information offered by the comparison between provinces and consequently it is less efficient. However, an important assumption behind random effects estimators is that random components of provincespecific effects are not correlated with the regressors. We test the appropriateness of this assumption using Hausman test, which compares fixed effects estimators and random effects estimators. Since fixed effects estimators are always consistent regardless of the orthogonality condition, a significant difference between the estimates indicates correlation between the regressors and the province-specific effects. If this is the case, we only report fixed effects regression results. Otherwise, we use maximum-likelihood estimation that fully maximizes the likelihood of the random-effects model. When our dependent variable is binary (e.g. use of public health facilities), we use conditional logit and random-effects probit for fixed-effects and random effects, respectively. In the latter, the likelihood is expressed as an integral computed using Gauss-Hermite quadrature approximation and its numerical fitness is checked by reestimating the model with different quadrature points and comparing the change in estimates. If the coefficients change by more than 1% the results are interpreted as unstable and not reported.
9
We should note that even when statistically justified, it is still theoretically debatable to justify orthogonality between unobserved province-specific effects and explanatory variables, since it is quite possible to imagine the confounding influence of these unobserved effects on at least some regressors in the model (for example, a drug problem in one province may affect the accountability or corruption variables). Hence, we always report estimates of fixed-effects model along with random-effects results. Almost in all cases we also try estimation via ordered probit method after transforming continuous dependent variables, such as NEAT scores, school rankings, and standard deviation of NEAT scores, into discrete ones. Although this approach lowers the information content of dependent variables, the coefficient estimates are less likely to be affected by potential measurement errors in dependent variables. When our dependent variable is a performance measure based on government statistics, the corresponding econometric model is either at school-level (e.g. test results) or at municipallevel (e.g. immunization rate). When survey questions are used as dependent variables, we estimate the model at the household level rather then aggregating the data in order to capture household-specific effects, such as child characteristics, household income or education. We also disaggregate the corruption variable based on location (urban vs. rural) and prosperity of municipalities (rich, middle-income, and poor) to understand whether corruption affects one area more or less than others depending on differences in regional characteristics.
10
V. Results V.1 Health Outcomes Our first performance measure is immunization coverage in the Philippines. The regression model for percent of children immunized is at the municipal level since the Ministry of Health provides data only on municipal averages. This limits our sample to 33 municipalities (we lose several municipalities because of missing data). The results are presented in Table 5. Local governments with high corruption level are less successful in providing immunization to their communities. One standard deviation increase in corruption reduces immunization rate by 11-19 %5. The robustness of results to outliers is confirmed by the use of robust regression and ordered probit models. Urbanization rates, unequal distribution of wealth, and distance to health centers (as reported by public health facilities) also negatively influence immunization coverage. Surprisingly, empirical results also indicate a negative partial correlation between immunization and local prosperity, for which we have no easy explanation. We do not observe any difference between rural vs. urban or rich vs. poor municipalities in terms of the effect of corruption. The coefficients are usually significant and similar in magnitude. Our next six dependent variables are derived from the Household survey and the regressions are run at the household level. Although it is possible to run the same regressions at more aggregate level (e.g. barangay or municipality), we prefer a household level analysis because it allows us to use of household-level variables as regressors, such as child or household characteristics. Additionally, to highlight the importance of community-specific variables, some household level variables are aggregated at the municipal level. We, first, run regressions using individual-specific variables alone (such as education, wealth, urbanization, social participation,
5
exp(1.6311 x 0.1060)-1= 0.1887 for fixed effects model; exp(1.0107 x 0.1060)-1 = 0.1131 for random effects model
11
and reading newspapers) and then we add municipal averages of these variables to capture community-specific effects. Regression results on vaccination of children (as reported by households) are presented in Table 6. The sample is restricted to households with children of at least one year old. This ensures that the households covered in our sample are the ones who should complete the vaccinations of their children. The simple correlation between this variable and the previous dependent variable is reasonably high (r = 0.46). The results show that the coefficient of corruption variable is significant at 5% and its magnitude is quite substantial. The odds of completing vaccination can decrease 1.8 to 4.2 times as a result of a one standard deviation increase in corruption6. The size of public health facilities in the area has a negative impact on immunization, suggesting that this variable is a proxy for excessive demand rather than adequate supply (If governments respond to excess demand by increasing capacity, but less than is needed, then larger facilities will be associated with more excess demand and hence lower immunization rates. Another explanation is diseconomies of scale in the provision of vaccines). Unlike the previous regression, this time wealth, rather than education of households, has a significant and positive impact on immunization. Both the existence of private health providers and distance to alternative public health facilities appear significant, albeit with wrong signs. It is possible that both variables, similar to size of health facilities, are correlated with demand for health services. Among public sector management variables, we find that the frequency of audit by central government and autonomy of local governments increases the immunization coverage. Next, we look at the causes of delay in vaccination of children. Our dependent variable is the log of time (in months) between birth of a child and his/her immunization provided that the child is at least one year old and that all immunizations have been completed. The variable is the
12
average of eight required immunizations (BCG, Polio 1-2-3, DPT 1-2-3, and Measles). The first two columns in Table 7 are fixed-effects, followed by random effects model. Since there is an inherent selection process in the dependent variable (i.e. dependent variable is observable only if all vaccinations are completed), we also use Heckman's selection model to check sample selection bias. However, we find that the correlation between the error terms of selection equation and outcome equation is negligible (rho=0.01.) and the results of the outcome equation are almost identical to ones obtained from fixed-effects model7. The corruption variable is significant at 5% in all three columns. One standard deviation in corruption increases the time between birth of a child and his/her immunization by 8-15%8. Existence of private health facilities marginally reduces the length of this time period; distance to alternative health clinics, on the other hand, has the opposite effect: infants are more likely to be vaccinated at later ages. We also find that corruption is more damaging in rural municipalities as compared to urban ones, but the difference is not statistically significant. The regression results on choosing public health facilities for immunization are shown in Table 8. As expected, wealth of a household is inversely proportional with his/her choosing public health service over private ones. Households are also more likely to go to public health clinics if wealth inequality in their municipality is higher or if there is no private health-service providers in the area. We find that corruption in the public sector is an important deterrent that discourages people to choose public health facilities. In areas where corruption is one standard deviation lower than the national average, households are 2.77 times more likely to choose
6
1 / exp(-5.6972 x 0.1060) = 1.8292 from column (1) and 1 / exp(-13.3963 x 0.1060) = 4.1372 from column (2). 177 observations are censored (households who have not yet completed vaccination of their children). The selection equation involves all explanatory variables in the outcome equation, except the province dummies. Wald test on independence of the equations (H0: rho=0): p=0.97. Average net marginal effect of corruption on dependent variable (i.e. selection plus outcome): 1.1564 with standard deviation 0.0005. 8 exp(0.7220 x 0.1060) -1= 0.0796 from column (3) and exp(1.3070 x 0.1060) -1 = 0.1486 from column (1). 7
13
public health facilities9. In poor municipalities, it is higher by as much as 7 folds10. We also use Heckman's selection model to check whether the results are influenced by sample selection bias, i.e. the decision to use public health facilities for immunization is influenced by households’ prior decision to immunize their children. We find that correlation between the error terms of selection equation and outcome equation is very small (rho=0.01) and the results of the outcome equation are almost identical to ones obtained from fixed-effects model11. In Table 9 we study the waiting time (as reported by households) in public health clinics. In addition to fixed-effects method, we also address sample selection issue using Heckman's selection model since households may select themselves out of using public health clinics. Although the correlation between selection and outcome processes is not significant, the magnitude of the correlation (rho=0.11) is reasonably large and warrants reporting the results in column (3). Due to concerns for perception bias on the behalf of households (that is, households facing frequent delays would be suspicious of corruption in health facilities), we only use the corruption perceptions of public officials our corruption measure. Corruption variable is significant at 5 % when community-specific effects are not considered (first column), but becomes only marginally significant when these effects are added to the model (column 2 and 3). When we disaggregate the corruption variable based on location (urban vs. rural) and prosperity of municipalities (rich, middle-income, and poor), we find that the influence of corruption on waiting time is negligible in rich and/or urban communities. However, in rural regions one
9
exp(9.621 x 0.1060)=2.7727 from column (2). exp(18.3355 x 0.1094)=7.4328 from column (2). 11 118 observations are censored (households who have not yet completed vaccination of their children). The selection equation involves all explanatory variables in the outcome equation, except the province dummies. Wald test on independence of the equations (H0: rho=0): p=0.95. Average net marginal effect of corruption on dependent variable (i.e. selection plus outcome): -9.6211 with standard deviation 0.0005. 10
14
standard deviation in corruption can increase waiting time as much as 28% and in poor municipalities 15%12. In Table 10 we use the incidence of being denied a vaccine from public health facilities (as reported by households). In addition to fixed- and random-effects model, we also use Heckman's selection method, because the outcome variable is observable only if a household chooses to use public health facilities. We again replace the composite corruption index with the corruption perceptions of public officials, because the denial of a vaccine could affect a household’s corruption perceptions. All estimation methods indicate that the corruption variable is significant at 5% or less. A one standard deviation worsening in corruption may raise the likelihood of being denied vaccines by 1.23-1.47 times13. In urban areas this increase is more significant and can be as much as 2.13 fold14. Interestingly, the probability of being denied a vaccine is positively related to the size of health facilities. As we discussed previously, the size of health facilities is likely to be proportional to the demand for health services. The results also show that patients are less likely to be denied a vaccine from public health clinics, if private centers exist in the area or alternative private clinics are close enough. Our last health-related performance variable is households' satisfaction with public health clinics. Once again, in addition to fixed- and random-effects models, we use Heckman's selection model to address sample selection problems (use of public health facilities is a prerequisite for reporting satisfaction rating). Our corruption index excludes households' corruption perception. The results, reported in Table 11, show that corruption has a statistically significant effect on households' satisfaction with public health clinics. A one standard deviation improvement in corruption can raise satisfaction ratings as much as 29% overall, 33% in rural areas and 48% in 12 13
exp(2.2428 x 0.1108)-1=0.2513 and exp(2.2428 x 0.1218)-1=0.1519 from column (1) exp(3.8730 x 0.1001)=1.4736 from column (2) and exp(2.0821 x 0.1001)=1.2317 from column (4).
15
poor municipalities15. We also find that the influence of corruption is slightly lower in urban and/or rich municipalities, though the difference is not statistically significant. In summary, our econometric analysis reveals that corruption has a significant and negative effect on all health-related performance variables. In most cases the coefficient is significant at 5 percent or less. The robustness of these results is checked using various estimation techniques and different model specifications, which is discussed in the next section in more detail. The corruption variable remains the single most important factor that influences health outcomes in a consistent basis. We also find that demand for public health care is more “corruption-elastic” in urban areas, i.e. corruption reduces households’ use of public health facilities and the likelihood of getting vaccines from public clinics in urban areas, but it has less influence over these two variables in rural areas. However, the effect of corruption on waiting time in public health clinics is less significant in the urban areas. We attribute these results to the presence of alternative health facilities in urban areas - either in the form of private health care providers or other public health facilities. On the other hand, citizens in rural areas with rampant corruption suffer with more waiting at public health clinics, late immunization of infants and report less satisfaction with public health services. We also run regressions to understand the effect of corruption in rich, middle-income and poor municipalities. Regardless of the relative prosperity of a municipality, corruption hurts satisfaction with public clinics, immunization rate of children, and average age of infants to get vaccines. Poor and middle-income municipalities also report more waiting at public clinics and more frequency of being denied vaccines when corruption is epidemic. Corruption in public
14
exp(7.9596 x 0.1001)=2.1217 from column (2). exp(2.5575 x 0.1001)=1.2901 from column (2); exp(3.0092 x 0.0945)=1.3289 from column (1), and exp(3.6137 x 0.1094)=1.4849 from column (1).
15
16
clinics is also more likely to deter households living in poor municipalities from using public clinics and forces them to opt for self-medication. Our results also show that voting is not an effective mechanism to discipline local governments –one possible explanation is that people might be voting on the basis of factors other than improvements in service delivery. Exit mechanisms, in the form of existence of private health facilities and accessibility of alternative health clinics, seem to have a more significant impact, especially on the satisfaction with public health care, vaccinations of infants at earlier ages, and accessibility of public health clinics for immunization. When it comes to public sector management variables (accountability mechanisms, audit by central government, and autonomy of local governments), we observe mixed results. The index that we constructed for accountability of health facilities and municipalities remains insignificant in most regressions and has wrong sign when it becomes significant. Audits by the central government have negative effects on several health outcomes. Higher autonomy in local governments improves immunization rate (as reported by households), raises accessibility of public health facilities for immunization and reduces the time length between birth of an infant and his/her immunization, but it also leads to longer waiting times.
V.2 Education Next, we look at the education-related performance variables. All econometric models, except households’ satisfaction ratings, are at barangay levels. We start with households' satisfaction with public schools. The results are reported in Table 12. The estimation methods we use include fixed- and random-effects to capture regionspecific factors, as well as Heckman's selection method to address school selection process that
17
precedes satisfaction ratings. As we did before, we exclude households' opinion on corruption and use only public officials' perceptions to prevent a biased link between the dependent variable and the regressor. We find that the corruption variable appears significant at 5% in randomeffects model, but only marginally significant in fixed-effects and Heckman's selection models. The results do not show any statistically significant partial correlation between corruption and households' satisfaction with public schools in urban areas or rich municipalities. Existence of private schools in the area and frequency of reading newspapers are two other variables with significantly negative influence over schools' satisfaction ratings. It is possible that both variables enable households to compare the performance of their schools with alternatives and make them more critical in their assessments. Next, we move to more objective education outcomes in public schools. Dependent variables are at the school level and they include various measures of success in a nation-wide exam (NEAT). The most obvious measure of the quality of service delivery is the mean of the NEAT score and the closely related school ranking. Variation of students' test scores within a school is another outcome variable. It shows whether a school is able to provide equal opportunity to its students to achieve their potential. We measure variation in terms of standard deviation of test results. However, since the standard deviation of a variable is likely to be proportional to its mean, we also use the coefficient of variation (standard deviation divided by the sample average) as another measure of variance. The results on percentage of student passing NEAT are presented in Table 13. We find that corruption in the public sector, in particular in rural municipalities, significantly reduces the success rate of students. A one standard deterioration in corruption around its mean causes a 12% decrease in number of students passing NEAT. There is no evidence that the effect of corruption
18
differs between affluent municipalities and poor municipalities. Among other significant variables we find that schools with better financial and personnel capacity are able to raise the percentage of students passing NEAT. Barangays with better-educated households also witness better test scores –this may represent a parent-child human capital transfer. Higher voter turnout pushes test scores up, whereas influence of ethnicity in elections and wealth inequality within municipalities hurt education results. Exit mechanisms (existence of private schools in the area), accountability mechanisms, and audit by central government are mostly ineffective. As shown in Table 14 average NEAT scores of students are also adversely affected by corruption in public sector. A one standard increase in corruption reduces NEAT scores as much as 11%. Comparing the effect of corruption in rich and poor municipalities, we observe that corruption has slightly more negative effect in rich municipalities. Education of parents and autonomy of public schools in decision-making are found to improve test scores, whereas wealth inequality and voting based on ethnicity reduces school performance. Another outcome variable that we used in our regressions is the national ranking of schools based on NEAT scores (Table 15). The effect of corruption is even more striking: a one standard deviation increase in corruption increases school ranking as much as 93% (higher ranks correspond to worse performance). Corruption is especially damaging in rural areas, where the magnitude of corruption coefficient is 2-3 larger. In addition to performance variables measuring level of success in NEAT scores, we also look at the variation of scores within schools. As shown in Table 16, the standard deviation of NEAT scores rises where corruption is more persistent, especially in rural areas. The other two significant variables are school capacity index and ethnic divisions (proxied by ethnicity considerations in voting decisions). Schools that enjoy better financial and personnel resources
19
are produce less equal outcomes. One possible explanation is that better off parents are more motivated to capture resources when capacity is higher. Pervasiveness of ethnic divisions in communities also has a similar effect. Since it is possible that standard deviation of test scores may be inflated by higher test scores, we also look at the coefficient of variation, which divides sample standard deviation by sample average. The results, reported in Table 17, are similar. Corruption and voting based on ethnicity have still negative impact on education outcomes. Capacity also appears to be related to greater inequality in outcomes.
VI. Robustness VI.1 Selection We have shown that there is a significant partial correlation between 13 dependant variables that measure various aspects of health and education services and corruption perceptions, after controlling for capacity (based on measures of human and physical capital), adult education levels, urban residence, living standards (as proxied by assets), inequality, existence of private sector competition, voting and media exposure, accountability measures, and local autonomy. To check the robustness of the regression results we use a robust regression method that involves down weighing observations that resemble outliers. It begins by estimating the regression, calculating Cook’s D distance statistics and excluding any observation with D>1. After that a weight is assigned to each observation inversely proportional to its residual. The appropriateness of this robust regression method may be questionable as the number of observations with small weights increases (we report the portion of the sample with weights less than 0.75). As one can see from the regression tables, robust estimates are very close to the ones
20
we obtain from other estimation methods. Hence, we conclude that our results are not greatly influenced by any outlier in the sample. Another concern that we address is the sample selection problem. When we regress, for example, waiting time in public health clinics or satisfaction with public schools on a set of explanatory variables, the sample space is limited with households that choose public service providers over the private ones. When such a selection process exists, linear regression estimates have to be adjusted for the fact that the dependent variable is the outcome of a nonrandom selection process. To tackle the issue of sample selection bias in regression results, we use Heckman’s selection method (Heckman (1979)). The basic idea of a sample selection model is that the outcome variable, y, is only observed if some criterion, defined with respect to a variable z, is met. The original Heckman’s method involves a two-stage approach: In the first stage, a dichotomous variable z determines whether or not y is observed; in the second state, the outcome variable y is estimated conditional on its being observed. The two error terms corresponding to selection equation and outcome equation are assumed to be correlated and having a joint bivariate normal distribution. We estimate the parameters by fully maximizing the joint likelihood function, which takes into account the heteroscedasticity of the error terms16. These estimates are consistent and asymptotically efficient under the assumption of normality17. The similarity in results, in particular for the coefficient of corruption variable, reinforces our conclusion that the negative and significant effect of corruption on health and education outcomes does not arise from a selection process.
16
Note that if a variable appears in both the selection and outcome equations the coefficient in the outcome equation is affected by its presence in the selection equation as well. Therefore, after reporting the estimates for outcome equation, as a footnote we also report the net effect of corruption variable that takes into account its effect in selection process 17 However, if the normality assumption is violated or there is a model specification error in one of the equations, then the estimated will be biased. For that reason, we choose to use other estimation methods instead of solely reporting the results obtained from Heckman’s selection method.
21
VI.2 Causality Another concern in econometric studies is about whether the partial correlation reflects a genuine causal relationship. For example, poor service delivery may be a cause of corruption in the public sector, as well as being a consequence. It is also possible that some common source of respondent bias, like pessimism about the performance public sector, has led to worse perceptions of corruption. To tackle these problems, we use four approaches: First, to correct potential “cynicism” of respondents towards government and the corruption level, we used a survey question that is supposed to be answered similarly (in the absence of individual bias) by all respondents: “the extent of corruption in central government”. We assumed that the average score represents the true corruption level in central government and thus, the difference between a person’s responds and the average score reflects the “pessimism” or “optimism” of that person. Using that difference as a discount factor, we updated the corruption index . We found that perceptions on national corruption is highly correlated with perceptions on local corruption (r = 0.3010, significant at 1%). Second, we applied the standard statistical approach to resolve the reverse causality problem by using instrumental variables. Following Mauro (1997) and Friedman et al. (2000), we used the ethnic fragmentation at each municipality as an instrument for corruption at that municipality. The ethnolinguistic fractionalization variable was computed as one minus the Herfindahl index of ethnolinguistic group shares, and reflected the probability that two randomly selected individuals from a population belonged to different groups18. The data for both variables were obtained from the 2000 Census of Population. As an additional instrument we also used a survey question that measures the extent to which ethnicity affects voting in local elections.
18
The exact formula is 1 − ∑ i si2 where si is the share of group i in the jurisdiction.
22
Simple correlations between these two variables and the corruption index are 0.41 and 0.47, respectively. The R square at the first stage was 0.37. We corroborated the validity of the instruments with an OIR-Hausman test19. In our third approach, we transformed the dependent variables into zero to one interval, which enabled logit specification. Then, instead of trying to find an instrument that is conceptually linked to corruption, we used the polynomials of exogenous variables as instruments thanks to non-linear estimation required by logit specification20. The R square at the first stage was 0.65. We verified the validity of overidentifying restrictions using Hausman test21. The results are summarized and compared with the results from the base models in Table 18. In general, when a dependent variable is based on objective sources (such as test scores) rather than subjective sources (such as satisfaction with public schools) the results are quite similar in magnitude and significance level. For example, according to our base results a 10 percent increase in corruption reduces immunization rates (as reported by the Ministry of Health) 3.6-3.8 percent. When IV approach is used, that figure is in the range of 2.2-3.0 percent. When a dependent variable reflects the observations of households, the results that are based on “cleaned” corruption perceptions somewhat differ from the base results: the magnitudes of the coefficients are lower and their statistical significance is a little weaker. Overall 9 out of 13 coefficients remain significant at 5% if the “cleaned” corruption variable is used. IV results, on 19
The test is based on regressing the residuals from the main structural equation on the entire set of exogenous variables. Under the null hypothesis of overidentifying restrictions, the test statistic, NR2 (N is the sample size and R2 is the uncentered the goodness of fit from the regression of residuals on all the instruments) has a χ2 distribution with K-T degrees of freedom, where K is the number of exogenous variables and T is the number of endogenous variables. If the instruments are excluded from the structural equation correctly, the set of instruments should have no explaining power over the residuals and consequently R2 should be low. The p-value of the test statistic is 0.76 for ethnic fragmentation and 0.52 for the ethnic voting. Hence, we failed to reject the null hypothesis that the overidenfying restrictions are valid. 20 Some dependent variables (such as test scores or immunization rates) can easily be transformed to 0-1 interval by dividing by 100. For some other variables (such as ranking in NEAT), we divided the observations by their maximum.
23
the other hand, are still closer to the base results, suggesting that reverse causality is not serious problem.
VII. Conclusion In this paper we used data from 80 municipalities in the Philippines to assess the impact of corruption in local governments on health and education outcomes. Our results showed clearly that corruption undermines the delivery of health services in the Philippines. We used seven different measures for the quality of health services: six of them from Household Survey (immunization of children, delay in vaccination of children, waiting time of patients, accessibility of health clinics for treatment, choosing public health clinics for immunization, and satisfaction with public health clinics) and one from the Ministry of Health (municipal average of immunization rate of children). In each case regression results indicated a significant and negative effect of corruption on the quality of health services. We also found that corruption affects health outcomes in rural areas in a different way than the urban areas. Demand for public health care is more “corruption-elastic” in the urban areas, whereas rural areas with rampant corruption suffer with more waiting at public health clinics, late immunization of infants, and less satisfaction with public health services. There are also important poverty relevant effects.
Regardless of the prosperity of a
municipality, corruption hurts satisfaction with public clinics, immunization rate of children, and leads to delays in vaccinations (the magnitude of this effect is more significant in poor municipalities). However, unlike rich municipalities, poor and middle-income municipalities also report more waiting at public clinics and a higher frequency of being denied of vaccines when
21
The p-value of the test statistic was between 0.59 and 0.91 depending on the instrument tested.
24
corruption is widespread. Corruption in public clinics is also more likely to deter households living in poor municipalities and forces them to opt for self-medication. Our results on education were similar. We used seven measures for the quality of education: various measures of test scores and household assessments of the quality of education. Corruption hurts all education outcomes. However, its negative effect is more prevalent in rural areas than the urban. We observed, on the other hand, minor differences between rich and poor municipalities in terms of the effects of corruption. The coefficients of corruption variables are significant in both cases and similar in magnitude Taken together our results do suggest that corruption undermines the delivery of health and education. This complements cross-country findings on the subject, and adds to the expanding list of ways corruption undermines welfare.
25
References Acemoglu, D., Johnson S., Robinson, J. A. (2001) Colonial origins of comparative development: An empirical investigation. American Economic Review 91: 1369-1401 Azfar, O. (2003) Review of corruption and the delivery of health and education services. Mimeo., IRIS, University of Maryland, College Park Azfar, O., Gurgur, T (2000) Decentralization and corruption in the Philippines. Mimeo., IRIS, University of Maryland, College Park Azfar, O., Gurgur, T., Meagher, P. (2004) Political disciplines on local government: Evidence from the Philippines. In M. S. Kimenyi and P. Meagher (ed) Devolution and Development, Governance Prospects in Decentralizing States. London, Ashagate Batalla, E. (2000) The social cancer: Corruption as a way of life. Philippine Daily Inquirer, August 27 Department of Interior, Philippines, Local Government Code of 1991, Republic Act No. 7160, Department of Interior and Local Government, Manila Friedman, E., Johnson, S., Kaufmann, D., Zoido-Lobaton, P. (2000) Dodging the grabbing hand: the determinants of unofficial activity in 69 countries. Journal Of Public Economics 76 (3): 459-493 Gray-Molina G., De Rada E. P., Yánez E. (1999) Transparency and accountability in Bolivia: Does voice matter?. Working Paper No. R-381, Inter-American Development Bank, Washington, D.C. Guerrero, L.L., Rood, S. A. (1999) An explanatory study of graft and corruption in the Philippines. Social Science Information 27(1): 110-141 Gupta, S., Verhoeven, M., Tiongson, E. (2002) The effectiveness of government spending on education and healthcare in developing and transition countries. European Journal of Political Economy 18: 717-737 Klitgaard, R., MacLean-Abaroa, R., Parris, H.L. (2000) Corrupt Cities: A Practical Guide to Cure and Prevention, Boston, ICS Press Heckman, J (1979) Sample selection bias as a specification error. Econometrica 47: 153-161 Manasan, R. (1997) Local government financing of social service sectors in a decentralized regime: Special focus on provincial governments in 1993 and 1994. Discussion Paper 97-04, Philippine Institute for Development Studies, Manila Mauro, Paolo (1997) Corruption and growth. Quarterly Journal of Economics 110(3): 681-712
26
Miller, T. (1997) Fiscal federalism in theory and practice: The Philippines Case. USAID Economists Working Papers Series No. 4, Washington, D.C. Pritchett, L. (1996) Mind your P’s and Q’s: The cost of public investment is not the value of public capital. Policy Research Working Paper 1660, World Bank, Washington, D.C. Rajkumar, A.S., Swaroop, V. (2002) Public spending and outcomes: Does governance matter? Mimeo., World Bank, Washington, D.C. Reinikka, R., Svensson, J. (2001) Explaining leakage of public funds. World Bank Policy Research Paper 2709, Washington, D.C. World Bank (1994) The Philippines Devolution and Health Services: Managing Risks and Opportunities, Washington, D.C. World Bank (2000) Combating Corruption in the Philippines, Washington D.C.
27
Table 1: Corruption Index and Its Components Mun. Health
Mun. Adm..
Mun. DECS
Public School
Private School
Pub. Health C
Mean Statistics Proportion of People Who Get Paid but Don’t Show Up
2.56
6.33
0.00
-
-
-
Paid to Obtain Jobs
2.95
3.80
5.00
8.87
3.40
-
Bribery Happened in the last year
2.53
18.99
1.25
0.92
4.08
-
Theft of Funds Happened in the last year
16.45
31.65
1.25
1.83
4.08
-
Theft of Supplies Happened in the last year
16.23
15.38
0.00
1.83
6.12
-
Frequency of Theft of Funds
3.80
9.09
14.67
1.53
1.36
20.37
Frequency of Seeking Informal Payments
4.49
10.68
12.00
0.92
2.04
15.51
Corruption in the National Government
74.00
69.23
62.77
66.35
80.85
62.72
Corruption in the Provincial Government
59.43
43.86
37.96
50.65
69.57
46.71
Corruption in the Municipal Government
43.42
29.32
24.79
36.86
62.32
36.78
Corruption in the Barangay Government
38.96
28.85
22.22
24.76
48.89
24.12
0.24*
0.41*
-
-
-
-
0.11
0.02
0.15
0.45*
0.22*
-
0.27*
0.72*
0.44*
0.53*
0.68*
-
0.22
0.68*
.
0.32*
0.64*
-
Bribery Happened in the last year
0.26*
0.68*
0.52*
0.44*
0.65*
-
Frequency of Theft of Funds
0.49*
0.52*
0.44*
0.37*
0.22
0.29*
Frequency of Seeking Informal Payments
0.72*
0.50*
0.51*
0.32*
0.10
0.33*
Corruption in the National Government
0.22
0.11
0.14
0.00
0.09
0.26*
Corruption in the Provincial Government
0.21
0.24*
0.18
0.20*
0.09
0.24*
Corruption in the Municipal Government
0.42*
0.32*
0.36*
0.07
0.11
0.74*
Corruption in the Barangay Government
0.40*
0.26*
0.41*
0.13
0.10
0.36*
Correlation between Corruption Index and Other Corruption Measures1 Proportion of People Who Get Paid but Don’t Show Up Paid to Obtain Jobs Theft of Funds Happened in the last year Theft of Supplies Happened in the last year
Correlation between Corruption Indices Municipal Health Officials
1.00
Municipal Administrators
0.17
1.00
Municipal DECS Officials
0.17
0.17
1.00
Public Schools
0.18
0.04
0.24*
1.00
Private Schools
0.16
0.24
0.42*
0.13
1.00
0.30*
0.03
0.16
0.25*
0.35*
1.00
0.19
0.15
0.18
0.18
0.04
0.20
Public Health Clinics Corruption Perception of Households 1
When the correlation of corruption index with one of its components is reported, the corruption index does not include that component.
28
Table 2: Description of Explanatory Variables Variable School capacity index
Health center capacity index
Description Based on 15 questions measuring the training, education, turnover rate, and motivation of teachers, class size, adequacy of school supplies, equipment, textbooks, and number of teachers (Source: Public Schools Survey) Based on 11 questions measuring the training, education, turnover rate, and motivation of personnel, computerization of records, adequacy of medicines, vaccines, equipment and number of personnel (Source: Health Clinics Survey)
School size
Number of students (Source: Public Schools Survey)
Health center size
Number of personnel (doctor, nurse, midwife, other) (Source: Health Clinics Survey)
Urban
In the Philippines, “urban” areas fall under the following categories: (1) have a population density of at least 1,000 persons per square kilometer, (2) at least six establishments (commercial, manufacturing, recreational and/or personal services), (3) at least three of the following: town hall, church, public plaza, market place, or public building
Education of households
Mother can write (Source: Households Survey)
Ownership of durable goods and services Wealth inequality within municipality Involvement in social organizations
Average ownership of the following 12 items: electricity, water, toilet, radio, television, refrigerator, oven, microwave oven, video player, computer, credit card, and motor vehicle (Source: Households Survey) Standard deviation of “ownership of goods/services” within municipality (Source: Households Survey) Average involvement/association with the following organizations/groups/activities: PTA, church, mothers club, women club, youth groups, farmers cooperatives, business organizations, labor unions, homeowner clubs, community clubs, etc. (Source: Households Survey)
Reading newspapers
Frequency of reading local and national newspapers (Source: Households Survey)
Vote in elections
Frequency of voting local and national elections (Source: Households Survey)
Voting affected by ethnicity Private service providers in the area exists Distance to closest health facility Frequency of audit by central government Accountability index
Whether the ethnicity of candidates affects voting of household (Source: Households Survey) Whether private service provider (health or education) exist in the municipality (Source: Public Schools Survey and Municipal Health Officials) In kilometers. Source: Public Schools Survey Frequency of visits by provincial or municipal officials. Public Schools Survey and Health Clinics Survey Based on 10 questions measuring the existence and enforcement of written targets, frequency of evaluations, inventory control, and record-keeping (Source: Public Schools Survey and Health Clinics Survey)
29
Table 2: Description of Explanatory Variables (cont.)
Autonomy index
Average of schools/health centers autonomy in service management and municipal’s autonomy in personnel management (Source: Public Schools Survey, Health Clinics Survey, Municipal DECS Survey, and Municipal Health Officials Survey)
Corruption perception of officials within their organization
Average of 7 questions on bribery, theft of supplies, theft of funds, and purchase of jobs. (Source: Public Officials Surveys)
Corruption index of households
Average of corruption perception of officials (municipal health, municipal DECS, municipal administrations, public schools, and health clinics) within their organization and corruption perception within other public offices within their municipality. (Source: Public Officials Surveys) Average of two questions: frequency of corruption within municipality and being aware of any corruption incident within municipality. (Source: Households Survey)
Age of the child
In months. (Source: Households Survey)
Gender of the child
Girl. (Source: Households Survey)
Corruption index of officials
30
Table 3: Description of Performance Measures Health Outcomes Immunization rate of children Immunization of children Delay in vaccination of children Choosing Public Health Clinics for Immunization Waiting time of patients Accessibility of Health Clinics for Treatment Satisfaction with public health clinics
Description Percent of children immunized against BCG, polio, DPT, measles, and hepatitis B. Municipal average, 1995. (Source: Ministry of Health, Philippines) Ratio of vaccines received by the infant to number of immunizations required (BCG, Polio 1-2-3, DPT 1-2-3, and Measles. (Source: Household Survey) Time, in months, between birth of a child and his/her immunization, average of eight required immunizations: BCG, polio 1-2-3, DPT 1-2-3, and measles (Source: Household Survey) Source: Household Survey Time, in minutes, to get treatment from public health clinics (Source: Household Survey) Households being denied vaccine from public health clinics (Source: Household Survey) Source: Household Survey
Education Outcomes
Description
Percentage of students passing National Elementary Assessment Test, NEAT
NEAT is the national examination, which aims to measure learning outcomes in the elementary level in response to the need of enhancing quality education as recommended by the Congressional Commission on Education. It is designed to assess abilities and skills of Grade VI pupils in all public and private elementary schools) test scores and household’s subjective ratings of primary education. School average, 1997-98 (Source: Ministry of Education)
Average score in NEAT,
School average, 1997-98 (Source: Ministry of Education)
National ranking of public schools in NEAT Standard deviation of NEAT scores within schools Coefficient of variation of NEAT scores within schools Satisfaction with public schools
1997. Source: Ministry of Education 1997. Source: Ministry of Education 1997. Source: Ministry of Education Source: Household Survey
31
Table 4: Descriptive Statistics Variables from Household Survey Education of households Ownership of durable goods Wealth inequality within municipality Involvement in social organizations Reading newspapers Vote in elections Voting affected by ethnicity
N
School size, school average Health center size Facility at urban area Private service providers in the area exists, municipal average Distance (in kilometers) to the closest health center
Frequency of audit by central government
Accountability index
Autonomy index
St.Dev.
Min.
Max.
Household average Municipal average Household average Municipal average
1118 81 1118 81
0.8462 0.8425 0.3666 0.3629
0.3610 0.1943 0.1923 0.1118
0 0.2143 0 0.0833
1 1 0.8333 0.5536
Municipal average
81
0.4702
0.2016
0.2057
1.2066
Household average Municipal average Household average Municipal average Household average Municipal average Household average Municipal average
1118 81 1114 81 1118 81 1118 81
0.0875 0.0886 0.2749 0.2719 0.8265 0.8303 0.3032 0.3031
0.1261 0.0622 0.2824 0.1617 0.3449 0.1226 0.4244 0.2680
0 0 0 0.0038 0 0.4286 0 0
1 0.2551 1 0.725 1 1 1 0.8571
Min.
Max.
Variable from Public Officials Survey Capacity index
Mean
N
Mean
Public schools Private schools Health Clinics Public Private
110 50 128 108 49
0.5356 0.6926 0.4842 613.31 298.20
0.1183 0.0926 0.1266 887.75 269.47
0.0962 0.4134 0.1667 61 46
0.8013 0.8215 0.8333 7156 1194
HC average
122
7.3033
7.9978
1
64
Public schools Private schools Health Clinics Schools Health Facilities
110 50 128 80 79
0.3964 0.7800 0.3395 0.2542 0.4494
0.4914 0.4185 0.4742 0.3204 0.4974
0 0 0 0 0
1 1 1 1 1
Municipal average
80
2.4686
1.9758
0.1833
10
Public schools Private schools Health Clinics Municipal average Public schools Private schools Health Clinics Municipal average Public schools Private schools Health Clinics Municipal average
110 50 128 80 110 50 128 80 107 48 123 78
0.5742 0.2889 0.4915 0.5955 0.7405 0.7278 0.7700 0.7458 0.5280 0.7951 0.6860 0.5288
0.2398 0.2145 0.2910 0.2183 0.0753 0.0852 0.0818 0.0764 0.1829 0.1827 0.1582 0.1822
0 0 0 0.1333 0.4296 0.4667 0.5619 0.4290 0.1667 0.5 0.35 0.1667
1 0.8577 1 1 0.8914 0.9078 0.9357 0.9032 0.9167 1 1 0.9331
32
St.Dev.
Table 4: Descriptive Statistics (cont.) Corruption Indicators
N
Mean
Household average
1059
0.2995
0.3106
0
1
Urban average Rural average Rich municipalities Middle-income m. Poor municipalities Public schools Private schools
646 419 23 32 15 110 50
0.3305 0.2511 0.3432 0.3119 0.2789 0.0661 0.0444
0.3199 0.2893 0.1611 0.1724 0.1294 0.0781 0.0968
0 0 0.0496 0 0.0952 0 0
1 1 0.75 0.5833 0.5953 0.3056 0.5
Corruption perception of health center officials
HC average
113
0.1770
0.2360
0
1
Corruption perception of households
Municipal average
81
0.2968
0.1633
0
0.75
Corruption perception of all officials
Municipal average
81
0.2493
0.1001
0.0194
0.4770
Corruption perception of officials and households
Municipal average
80
0.2736
0.1060
0.0277
0.5530
Corruption perception, Households Corruption perception, Households
Corruption perception, Households
Corruption perception of school officials
Performance Variables
St.Dev.
Min.
Max.
N
Mean
St.Dev.
Min.
Max.
Public Schools Private Schools Public Schools Private Schools Public Schools Private Schools Public Schools Private Schools Public Schools Private Schools Household average Municipal average
58 17 41 16 40 12 40 12 40 12 705 80
83.9684 95.2412 75.6227 88.9222 5026.45 2696.17 14.8995 18.6683 0.1981 0.2001 0.8415 0.8429
18.7758 6.1650 18.8423 5.6236 3165.75 1406.26 5.0878 3.8924 0.0851 0.0510 0.2160 0.1192
36 80.5 18.58 79 61 935 7.25 11.47 0.0610 0.1027 0 0.5238
100 100 100 100 10158 6128 25.11 23.88 0.3509 0.2750 1 1
Municipal average
37
75.8219
18.1507
25.8652
97.9652
Ratio of vaccines received for 8 required immunizations, Household Survey
Household average Municipal average Household average Municipal average
954 80 1067 80
18.6803 18.9283 0.8942 0.8952
27.0840 13.1081 0.2386 0.0965
0 2.6254 0 0.5095
180 70.4545 1 1
Ratio of vaccines received from public health facilities for 8 required immunizations, Household Survey
Household average Municipal average
1069 81
0.8916 0.8925
0.2749 0.1155
0 0.5644
1 1
Time (in months) between birth of a child and his/her immunization, Household Survey
Household average Municipal average Household average Municipal average Household average Municipal average
801 79 908 80 962 80
4.8404 5.4851 0.3073 0.3002 0.7914 0.7874
6.0327 2.4672 0.4616 0.2786 0.2556 0.1169
0 2.9460 0 0 0 0.3889
95.75 18.6462 1 1 1 0.9762
% of students passing NEAT 97-98, school average Average NEAT score 97-98, school average National ranking in NEAT 97, school average Standard deviation of NEAT scores 97, school average Coefficient of variation of NEAT scores 97, school average Satisfaction with schools, Household Survey % of children immunized against BCG, Polio, DPT, Measles, and Hepatitis B Waiting time (in minutes) in public health centers, Household Survey
Denied vaccine from public health facilities for any reason, Household Survey Satisfaction with public health centers, Household Survey
33
Table 5: Immunization of children, 1995 Dependent variable is the log of percent of children immunized against BCG, Polio, DPT, Measles, and Hepatitis B – municipal average (Source: Ministry of Health, Philippines); *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; VARIABLES Health center capacity index, municipal average Health center size, municipal average Health Facilities / Community Resources
Household at urban area, municipal average Education of households, municipal average Ownership of durable goods, municipal average Wealth inequality within municipality Involvement in social organizations, municipal average Reading newspapers, municipal average
Voice, Exit, Participation
Institutions, Corruption, Decentralizati on
Vote in elections, municipal average Voting affected by ethnicity, municipal average Private health facilities in the municipality exists Log of distance to closest alternative health center, municipal average Frequency of audit by central government, municipal average Accountability index, municipal health officials, municipal average Autonomy index, municipal health officials, municipal average Corruption perception of officials and households, municipal average N, Adj. R2
RANDOM EFFECTS2 0.2471 (0.71) -0.0773 (-1.57)† -0.2118 (-2.30)** 0.8112 (3.81)*** -1.9691 (-3.39)*** -0.5957 (-2.05)** -1.5870 (-2.41)** -0.1317 (-0.54) 0.2533 (0.72) -0.0066 (-0.05) 0.1119 (1.66)* -0.1661 (-1.86)* 0.0565 (0.30) -0.4044 (-0.82) -0.1080 (-0.46) -1.0107 (-2.66)*** 33
ROBUST REGRESSION3 0.5057 (1.62)† -0.0341 (-0.77) -0.2062 (-2.47)** 0.8856 (4.61)*** -1.1808 (-2.25)** -0.3984 (-1.52)† -1.5059 (-2.53)** -0.3426 (-1.54)† 0.7032 (2.21)** 0.0703 (0.52) 0.0882 (1.45) -0.3171 (-3.92)*** -0.0814 (-0.48) -0.4076 (-0.91) -0.3184 (-1.49) -1.0647 (-3.10)*** 33
ORDERED PROBIT4 2.4876 (0.78) -0.8363 (-2.98)*** -1.6804 (-1.96)** 6.9780 (2.82)*** -13.7100 (-5.91)*** -4.8980 (-3.05)*** -10.6867 (-1.46)† -1.8277 (-0.91) 4.1686 (0.98) -1.0334 (-1.39) 1.0002 (3.32)*** -1.8912 (-2.27)** 1.3232 (1.92)* -4.1082 (-1.11) 0.6723 (-0.59) -8.8675 (-3.10)*** 33, 0.3343
-
15.0045
-
-31.6113
0.3733
0.0215
0.0032
0.0000
Corruption perception of officials and households, urban municipalities Corruption perception of officials and households, rural municipalities
-2.9060 (-2.39)** -2.5996 (-2.33)**
-1.1486 (-2.71)** -1.1078 (-2.95)***
-1.1817 (-1.71)* -0.3270 (-0.38)
-9.0088 (-4.79)*** -8.9292 (-6.48)***
Corruption perception of officials and households, rich municipalities Corruption perception of officials and households, middle-income municipalities Corruption perception of officials and households, poor municipalities
-2.7924 (-2.88)** -2.2979 (-3.04)** -2.2604 (-2.96)**
-0.5178 (-1.95)* -0.2902 (-1.20) -0.5318 (-2.03)*
-1.9784 (-3.19)** -1.8055 (-3.16)*** -1.9328 (-3.31)***
-9.3967 (-4.18)*** -10.6547 (-3.01)*** -12.3669 (-4.00)***
Log likelihood p-value (significance of model)
Corruption5
FIXED EFFECTS1 0.0678 (0.11) 0.0057 (0.04) -0.8856 (-2.65)** 2.0097 (2.85)** -1.3582 (-1.19) -0.5639 (-0.90) -4.9362 (-2.19)* 1.4965 (1.68)† 1.1203 (1.30) 0.1678 (0.43) 0.4824 (2.20)* -0.4200 (-1.91)* 0.9000 (1.39) -1.6311 (-1.20) 0.4009 (0.84) -1.6311 (-1.86)* 33, 0.0818
1
Grouped by regions. F test that all fixed effects are zero: p=0.45 Grouped by regions, Maximum likelihood results. Hausman test on orthogonality condition: p=0.86. Breusch and Pagan Lagrangian multiplier test for random effects (H0: no random effect): p=0.99 with rho=0.00 (ratio of variance of random effect component to total variance). Random effects tobit gives the same as random-effects (MLE) . 3 Number of observations with weight less than 0.75: 3 4 Observations are independent across groups (regions) but not necessarily independent within groups. Dependent variable is divided into 5 categories.. 5 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
34
Table 6: Has Immunization, Household Survey Dependent variable is binary. It is equal to one if household has received vaccines for their children at least 6 of the required 8 (BCG, Polio 1-2-3, DPT 1-2-3, and Measles); and zero otherwise. The sample is restricted to households with children of age 1-12. *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; VARIABLES Health center capacity index, municipal average Health center size, municipal average Health Facilities / Community Resources
Household level Municipal average Household level Municipal average Household level Municipal average
Household at urban area Education of households Ownership of durable goods Wealth inequality within municipality
1.3642 (2.21)**
2.6038 (2.74)*** -2.3180 (-2.96)*** -0.0876 (-0.16) -5.7299 (-1.77)* 4.5879 (3.35)*** 34.3275 (2.85)***
-0.1907 (-0.40) 4.1989 (3.20)***
Voting affected by ethnicity, municipal average
-0.5748 (-0.37)
1.9996 (0.93)
Private health facilities in the municipality exists
-0.8506 (-1.43)
-2.8127 (-2.85)***
Log of distance to closest alternative health center, municipal average
0.7739 (1.30)
2.1887 (2.82)***
Frequency of audit by central government, municipal average
2.2830 (1.84)*
5.8406 (3.04)***
4.9985 (1.39)
2.0530 (0.41)
7.6240 (3.11)***
15.9203 (4.10)***
-5.6972 (-2.31)**
-13.3963 (-3.28)***
0.2103 (1.99)**
0.7567 (4.10)***
0.0855 (1.28)
0.2105 (1.29)
511, 0.2045
511, 0.2817
-113.1025
-102.1166
0.0000
0.0000
Corruption perception of officials and households, urban municipalities
-5.9325 (-2.25)**
-13.2233 (-3.07)***
Corruption perception of officials and households, rural municipalities
-5.6310 (-2.27)**
-12.9222 (-2.37)**
M i i lA -7.0356 (-1.89)*
M i i lA -35.1423 (-3.01)***
Corruption perception of officials and households, middle-income mun.
-7.2202 (-2.15)**
-34.3271 (-3.07)***
Corruption perception of officials and households, poor municipalities
-6.8302 (-2.17)**
-34.5367 (-2.98)***
Accountability index, average of HCs and municipal health officials, municipal average Autonomy index, average of HCs and municipal health officials, municipal average
Log of age of the child (months), household level Child: Girl, household level
p-value (significance of model)
2
-3.9131 (-4.31)***
4.5756 (1.66)*
Vote in elections
Log likelihood
1
-1.7328 (-3.41)***
6.3598 (2.08)**
N, Adj. R2
Corruption2
3.3257 (1.00)
-3.1618 (-2.11)** -16.4441 (-2.00)** 0.9562 (1.19) -0.3857 (-0.07) 0.2161 (0.41) 3.4025 (1.14)
Corruption perception of officials and households, municipal average Child Characteristics
4.1292 (1.49)†
1.5009 (1.22)
Reading newspapers
Institutions, Corruption, Decentralization
FIXED EFFECTS LOGIT1
-2.1090 (-1.60)†
Household level Municipal average Household level Municipal average Household level Municipal average
Involvement in social organizations
Voice, Exit, Participation
FIXED EFFECTS LOGIT1
Corruption perception of officials and households, rich municipalities
0.6579 (0.86)
Grouped by provinces. Results obtained from random effect probit are not stable and not reported. Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported.
35
Table 7: Delay in Receiving Immunization, Household Survey Dependent variable is the log of time (in months) between birth of a child and his/her immunization, average of eight required immunizations (BCG, Polio 1-2-3, DPT 1-2-3, and Measles) . The sample is restricted to households with children of age 1-12. *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%
FIXED EFFECTS1
FIXED EFFECTS2
RANDOM EFFECTS3
Health center capacity index, municipal average
-0.3992 (-1.27)
-0.4183 (-1.31)
-0.1048 (-0.40)
Health center size, municipal average
0.1012 (1.53)†
0.0705 (1.00)
0.0621 (1.24)
0.0410 (0.54)
-0.0258 (-0.30) 0.0337 (0.42) 0.0443 (0.51) 0.1825 (0.64) -0.0964 (-0.48) 0.2011 (0.29)
0.0390 (0.53) -0.0482 (-0.64) 0.0635 (0.76) -0.0031 (-0.02) -0.0733 (-0.38) 0.1283 (0.22)
VARIABLES
Health Facilities / Community Resources
Household at urban area Education of households Ownership of durable goods
Institutions, Corruption, Decentralization
Child Characteristics
0.0461 (0.58) -0.0823 (-0.42)
Wealth inequality within municipality
0.1812 (0.82)
0.1087 (0.37)
0.0510 (0.26)
Involvement in social organizations
0.1760 (0.77)
0.4278 (1.84)* -1.2490 (-2.03)** -0.0037 (-0.03) 0.1222 (0.35) -0.0792 (-0.89) 0.0205 (0.07)
0.0698 (0.21)
0.3842 (1.60)† -2.5145 (-2.86)*** 0.0512 (0.39) 0.4030 (0.88) -0.0794 (-0.87) 0.1802 (0.53)
Voting affected by ethnicity, municipal average
0.2825 (1.49)†
0.3765 (1.89)*
0.3062 (2.32)**
Private health facilities in the municipality exists
-0.1444 (-1.83)*
-0.1548 (-1.70)*
-0.1159 (-1.69)*
0.0935 (1.12)
0.1504 (1.65)*
0.1242 (1.72)*
-0.0152 (-0.10)
-0.0119 (-0.07)
-0.1102 (-0.86)
-0.2358 (-0.56)
-0.0611 (-0.14)
0.2579 (0.71)
-1.2179 (-3.69)***
-1.4809 (-4.28)***
-0.7358 (-2.69)***
1.3070 (3.16)**
1.1567 (2.60)***
0.7220 (2.24)**
0.2103 (1.99)**
0.2014 (1.90)*
0.2299 (2.28)**
0.0855 (1.28)
0.0946 (1.42)
0.0933 (1.45)†
No 400, 0.0601
No 400, 0.0608
0.0061
0.0050
No 400 -295.5047 0.0171
1.2343 (2.92)***
1.0946 (2.42)**
0.6619 (2.29)**
1.3477 (3.23)***
1.2471 (2.73)***
0.7278 (1.91)*
Reading newspapers Voice, Exit, Participation
Household level Municipal average Household level Municipal average Household level Municipal average
Vote in elections
Household level Municipal average Household level Municipal average Household level Municipal average
Log of distance to closest alternative health center, municipal average Frequency of audit by central government, municipal average Accountability index, average of HCs and municipal health officials, municipal average Autonomy index, average of HCs and municipal health officials, municipal average Corruption perception of officials and households, municipal average Log of age of the child (months), household level Child: Girl, household level Dummy variables N, Adj. R2 Log likelihood p-value (significance of model) Corruption perception of officials and households, urban municipalities Corruption perception of officials and households, rural municipalities
0.0355 (0.28)
Corruption perception of officials and households, rich 1.7017 (3.14)*** 1.6361 (2.59)*** 0.7960 (2.29)** municipalities Corruption perception of officials and households, 1.6095 (3.36)*** 1.5542 (2.72)*** 0.6942 (2.12)** middle-income municipalities Corruption perception of officials and households, poor 1.6684 (3.42)*** 1.6121 (2.77)*** 0.7207 (2.17)** municipalities 1 Grouped by provinces. F test that all fixed effects are zero: p=0.01. 2 Grouped by provinces. F test that all fixed effects are zero: p=0.00. 3 Grouped by provinces, Maximum likelihood results. Hausman test on orthogonality condition (H0: orthogonality condition valid): p=0.09. Breusch and Pagan Lagrangian multiplier test for random effects (H0: no random effect): p=0.51 with rho=0.03 (ratio of variance of random effect component to total variance). 4 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. Corruption4
36
Table 8: Choosing Public Health Facilities for Immunization, Household Survey Dependent variable is the ratio of vaccines received from public health facilities for 8 required immunizations (BCG, Polio 1-2-3, DPT 1-2-3, and Measles). The sample is restricted to households with children of age 0-12. *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; VARIABLES Health center capacity index, municipal average Health center size, municipal average Health Facilities / Community Resources
Household at urban area Education of households Ownership of durable goods
Household level Municipal average Household level Municipal average Household level Municipal average
Wealth inequality within municipality
-0.1256 (-0.16)
-0.4430 (-0.77)
-0.6137 (-1.04)
0.0061 (0.04)
-0.6367 (-1.12)
-0.9674 (-1.30) -0.1665 (-0.31) -0.0294 (-0.04) -2.4603 (-0.79) -3.0396 (-2.47)** 3.0128 (0.45)
-0.0334(-0.14) -0.0908 (-0.44) -0.1409 (-0.41) -0.3494 (-0.42) -1.3362 (-2.15)** 1.7438 (0.84)
3.8177 (1.50)†
4.5383 (1.52)†
1.9295 (2.44)**
2.2395 (1.43)
1.2948 (1.60)† -3.1694 (-1.67)* -0.1919 (-0.54) -0.8655 (-0.88) -0.1712 (-0.54) -0.2751 (-0.32)
-0.2229 (-0.35) -3.0696 (-2.67)***
-1.0514 (-0.55)
0.4858 (1.13)
Private health facilities in the municipality exists
-1.8540 (-2.83)***
-2.1657 (-2.52)**
-0.9006 (-4.27)***
-1.2550 (-1.77)*
-0.7372 (-0.87)
-0.2708 (-1.06)
0.2818 (0.21)
0.4556 (0.30)
0.4556 (0.91)
2.7805 (0.79)
2.8597 (0.74)
2.8597 (0.33)
0.6828 (0.37)
1.6089 (0.71)
0.3834 (0.55)
-7.0758 (-2.49)**
-9.6210 (-2.68)***
-1.5018 (-1.91)*
Log of age of the child (months), household level
-0.2818 (-0.43)
-0.3553 (-0.54)
-0.0612 (-0.18)
Child: Girl, household level
0.6189 (1.53)†
0.6789 (1.67)*
0.3896 (1.86)*
437, 0.2060
437, 0.2238
556
-106.0927
-103.7105
-139.2398
0.0000
0.0001
0.0000
Corruption perception of officials and households, urban mun.
-7.1009 (-2.46)**
-9.7165 (-2.69)***
-1.6439 (-1.78)*
Corruption perception of officials and households, rural mun.
-6.7644 (-2.43)**
-9.0763 (-2.52)**
-1.4435 (-1.53)†
Corruption perception of officials and households, rich mun.
7 0758 ( 2 49)** -5.8557 (-1.40)
9 6210 ( 2 68)*** -12.2945 (-1.89)*
M i i lA -2.4526 (-2.02)**
Corruption perception of officials and households, middle-inc. mun.
-8.2597 (-2.03)**
-16.7260 (-2.29)**
-3.6351 (-2.82)***
Corruption perception of officials and households, poor mun.
-8.1685 (-2.05)**
-18.3355 (-2.34)**
-2.6540 (-2.30)**
Accountability index, average of HCs and municipal health officials, municipal average Autonomy index, average of HCs and municipal health officials, municipal average Corruption perception of officials and households, municipal avg.
N, Adj. R2 Log likelihood p-value (significance of model)
Corruption3
-0.4558 (-0.19)
0.7164 (0.33)
Frequency of audit by central government, municipal average
Child Characteristics
0.6018 (0.29)
-0.5451 (-0.36)
Vote in elections
Log of distance to closest alternative health center, municipal avg.
Institutions, Corruption, Decentralizatio n
RANDOM EFFECTS PROBIT2
Voting affected by ethnicity, municipal average
Reading newspapers Voice, Exit, Participation
FIXED EFFECTS LOGIT1
2.4628 (1.51)† -8.1298 (-1.20) -0.5057 (-0.76) 6.0114 (1.29) -0.1995 (-0.34) 0.8049 (0.33)
Involvement in social organizations
Household level Municipal average Household level Municipal average Household level Municipal average
FIXED EFFECTS LOGIT1
1
-0.3444 (-0.54)
Grouped by provinces. Grouped by provinces. Likelihood ratio test of random effects (Ho: rho=0): p=1.00 with rho=0.00, where rho is the ratio of variance of random effect component to total variance. Results obtained from random effect probit are not stable and not reported 3 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
37
Table 9: Log of Waiting Time in Public Health Clinics (HC), Household Survey *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; FIXED EFFECTS1
FIXED EFFECTS2
HECKMAN SELECTION MODEL3
Health center capacity index, municipal average
-0.4908 (-1.25)
-0.4764 (-1.21)
-0.4327 (-1.10)
Health center size, municipal average
-0.1146 (-1.30)
-0.0859 (-0.93)
-0.0874 (-0.98)
0.0491 (0.53)
-0.0475 (-0.45) 0.2809 (2.92)*** -0.0433 (-0.39) 0.1696 (0.48) -0.3012 (-1.24) -2.1794 (-2.63)***
-0.0488 (-0.48) 0.2754 (3.05)*** -0.0388 (-0.41) 0.1228 (0.38) -0.3125 (-1.27) -2.2524 (-2.94)***
0.1758 (0.59)
-0.5006 (-1.29)
-0.4578 (-1.38)
-0.3703 (-1.32)
0.9194 (2.39)**
-0.2621 (-0.89) 1.2979 (-1.26) 0.1590 (0.99) 1.4947 (3.00)*** 0.0563 (0.54) 0.6201 (1.55)†
-0.2427 (-0.78) 1.3722 (-1.27) 0.1559 (0.97) 1.5020 (2.85)*** 0.0585 (0.57) 0.5867 (1.60)†
-0.0850 (-0.37)
0.0079 (1..03)
-0.0014 (-0.01)
-0.0282 (-0.32)
0.0.912 (0.90)
0.0.642 (0.70)
0.2006 (1.93)*
0.1488 (1.30)
0.1152 (1.07)
-0.1249 (-0.62)
-0.2366 (-1.16)
-0.1877 (-0.89)
0.6854 (1.28)
0.4931 (0.91)
0.3184 (0.62)
0.7143 (1.85)*
0.6374 (1.54)†
0.6473 (1.64)*
1.4276 (2.82)***
0.9779 (1.87)*
0.9490 (1.81)*
No
No
Province
836, 0.0300
836, 0.0394
972
VARIABLES
Health Facilities / Community Resources
Household at urban area Education of households Ownership of durable goods
Household level Municipal average Household level Municipal average Household level Municipal average
Wealth inequality within municipality Involvement in social organizations Reading newspapers Voice, Exit, Participation
Institutions, Corruption, Decentralizati on
Vote in elections
Household level Municipal average Household level Municipal average Household level Municipal average
Voting affected by ethnicity, municipal average Private health facilities in the municipality exists Log of distance to closest alternative health center, municipal average Frequency of audit by central government, municipal average Accountability index, average of HCs and municipal health officials, municipal average Autonomy index, average of HCs and municipal health officials, municipal average Corruption perception of officials, municipal average Dummy variables N, Adj. R2 Log likelihood p-value (significance of model) Corruption perception of officials, urban municipalities Corruption perception of officials and households, rural municipalities
Corruption4
-0.0444 (-0.42) -0.3463 (-1.47)†
0.3231 (2.09)**
-
-
-1465.5970
0.0013
0.0000
0.0000
0.5733 (0.86)
0.3441 (0.51)
0.3139 (0.48)
2.2428 (3.42)***
Corruption perception of officials, rich municipalities Corruption perception of officials, middleincome municipalities Corruption perception of officials, poor municipalities
1
1.6155 (2.38)**
1.5855 (2.16)**
M i i lA 0.7972 (1.20)
M i i lA 0.1575 (0.23)
M i i lA 0.1630 (0.21)
0.9691 (1.64)*
0.3900 (0.64)
0.3960 (0.60)
1.1751 (1.91)*
0.9888 (1.76)*
0.8950 (1.70)*
Grouped by provinces. F test that all fixed effects are zero: p=0.00. Grouped by provinces. F test that all fixed effects are zero: p=0.00. Hausman test on orthogonality condition (H0: orthogonality condition valid): p=0.00. Since orthogonality condition is not satisfied, random effects model is not used. 3 Outcome equation. Maximum likelihood results. Standard errors are corrected for heteroscedasticity. 136 observations are censored (households who do not use public health clinics). The selection equation involves all explanatory variables in the outcome equation, except the province dummies. Correlation (rho) between the error terms of selection equation and outcome equation is 0.11. Wald test on independence of the equations (H0: rho=0): p=0.25. Net marginal effect of corruption on dependent variable (i.e. selection plus outcome, average of observations): is 0.9102 with standard deviation 0.0184. 4 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
38
Table 10: Denied Vaccine from Public Health Clinics for Any Reason, Household Survey *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; VARIABLES Health center capacity index, municipal average Health center size, municipal average Health Facilities / Community Resources
Household at urban area Education of households Ownership of durable goods
Institutions, Corruption, Decentraliza tion
RANDOM EFFECTS PROBIT2
HECKMAN SELECTION PROBIT3
3.4269 (2.82)***
3.7453 (2.91)***
2.5618 (3.77)***
2.1468 (3.06)***
0.2686 (0.91)
0.1207 (0.35)
0.0540 (0.36)
0.0337 (0.18)
-0.9027 (-2.95)***
-1.4596 (-3.66)*** 0.1661 (0.52) -0.8132 (-2.21)** -1.4079 (-1.20) -0.8132 (-1.40) 8.9441 (3.30)***
-0.8555 (-4.56)*** 0.1350 (0.80) -0.4167 (-2.05)** -0.6511 (-1.20) -0.5917 (-1.30) 5.9200 (4.33)***
-0.7781 (3.54)*** 0.1137 (0.62) -0.4604 (-2.25)** -0.7918 (-1.05) -0.6592 (-1.59)† 5.7879 (4.27)***
0.7646 (-2.27)** -0.5778 (-0.75) -1.6516 (1.68)*
4.6247 (3.80)***
2.6807 (4.58)***
2.6787 (3.33)***
Involvement in social organizations
-0.9912 (-1.22)
2.9469 (2.23)**
-0.5624 (-1.13) -0.6373 (-0.43) -0.1090 (-0.33) -0.0703 (-0.09) -0.1776 (-0.93) 2.1406 (2.81)***
-0.5845 (-1.19) -0.8842 (-0.56) -0.3165 (-0.29) -0.8844 (-0.35) -0.1825 (-0.98) 2.0467 (2.59)***
-0.5693 (-1.13) -0.3916 (-0.43) -0.0823 (-0.33) -0.0169 (-0.09) -0.2092 (-0.93) 2.0094 (2.81)***
Voting affected by ethnicity, municipal average
3.0119 (4.10)***
3.6173 (4.27)***
2.0893 (5.47)***
2.0267 (4.66)***
Private health facilities in the municipality exists
-0.2387 (-0.79)
-0.9029 (-2.47)**
-0.6749 (-3.68)***
-0.5637 (-2.54)**
0.4071 (1.14)
1.0225 (2.51)**
0.8190 (4.01)***
0.6733 (3.00)***
0.2412 (0.37)
0.5625 (0.76)
0.7871 (2.22)**
0.4538 (1.37)
1.7525 (1.06)
2.8980 (1.57)†
1.7173 (1.92)*
1.5046 (1.42)
-2.1384 (-1.55)†
-2.3572 (-1.51)†
-1.3566 (-1.86)*
-1.4138 (-1.70)*
3.5513 (2.15)**
3.8730 (2.17)**
2.1423 (2.58)***
2.0821 (2.02)** Province
Vote in elections
Household level Municipal average Household level Municipal average Household level Municipal average
Log of distance to closest alternative health center, municipal average Frequency of audit by central government, municipal average Accountability index, average of HCs and municipal health officials, municipal average Autonomy index, average of HCs and municipal health officials, municipal average Corruption perception of officials, municipal average
-0.1058 (-0.20)
No
No
No
N, Adj. R2
674, 0.1267
674, 0.1636
674
972
Log likelihood
-252.13683
-242.2574
-303.9477
-673.5797
0.0000
0.0000
0.0000
0.0000
Corruption perception of officials, urban municipalities
6.3411 (4.67)***
7.9596 (4.26)***
5.8309 (4.76)***
5.9758 (4.14)***
Corruption perception of officials, rural municipalities
1.4970 (1.67)*
1.1204 (1.02)
1.6764 (1.37)
1.1719 (1.55)
Corruption perception of officials, rich municipalities
2.6849 (1.16)
1.8064 (0.81)
-
0.5286 (0.86)
Corruption perception of officials, middle-income municipalities
3.5628 (1.84)*
2.0468 (1.28)
-
1.3328 (1.24)
Corruption perception of officials, poor municipalities
3.4072 (1.74)*
2.8182 (1.75)*
-
1.7834 (1.82)*
Dummy variables in the model
p-value (significance of model)
Corruption4
FIXED EFFECTS LOGIT1
Wealth inequality within municipality
Reading newspapers Voice, Exit, Participatio n
Household level Municipal average Household level Municipal average Household level Municipal average
FIXED EFFECTS LOGIT1
1
Grouped by provinces. Grouped by provinces. Likelihood ratio test of random effects (Ho: rho=0): p=0.00 with rho=0.46, where rho is the ratio of variance of random effect component to total variance. When corruption variable is disaggregated based on the prosperity of municipalities (rich, middle income, and poor), the results become unstable and are not reported. 3 Outcome equation. Maximum likelihood results. Standard errors are corrected for heteroscedasticity. 298 observations are censored (households who do not use public health clinics). The selection equation (results are not reported) involves all explanatory variables in the outcome equation, except the province dummies. Correlation (rho) between the error terms of selection equation and outcome equation is 0.10. Wald test on independence of the equations (H0: rho=0): p=0.25. The coefficient of corruption variable in the selection equation is 2.2024 with standard error 0.5457 and significant at 1%. 4 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
39
Table 11: Satisfaction with Public Health Clinics (HC), Household Survey *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; FIXED EFFECTS LOGIT1
FIXED EFFECTS LOGIT1
RANDOM EFFECTS PROBIT2
HECKMAN SELECTION PROBIT4
Health center capacity index, municipal average
1.2490 (1.47)†
1.0839 (1.25)
1.4103 (3.24)***
0.8189 (1.41)
Health center size, municipal average
0.3153 (1.65)*
0.3964 (1.93)*
-0.1088 (-1.41)
0.2425 (1.91)*
0.1560 (0.79)
0.1854 (0.76) -0.5862 (-2.74)*** 0.3217 (1.35) 2.9808 (3.62)*** 0.1326 (0.24) 2.1308 (1.10)
-0.0558 (-0.47) -0.1687 (-1.54)† 0.1777 (1.22) 0.1892 (0.61) -0.0166 (-0.05) 1.0044 (1.06)
0.1146 (0.78) -0.2882 (-2.96)*** 0.2146 (1.47) † 1.9088 (3.95)*** 0.0267 (0.07) 1.1415 (0.95)
-0.2960 (-0.45)
-0.1815 (-0.21)
-0.1388 (-0.43)
-0.1542 (-0.29)
0.2127 (0.34)
-1.0297 (-1.22)
0.4198 (0.64) -4.7302 (-2.01)** 0.1052 (0.30) -2.2968 (-1.85)* 0.2207 (0.97) -0.8234 (-0.92)
0.2664 (0.67) -1.4402 (-1.54)† 0.0640 (0.30) -0.9290 (-1.93)* 0.1026 (0.75) -0.0624 (-0.14)
0.3250 (0.78) -3.1022 (-2.22)** 0.0497 (0.22) -1.3075 (-1.74)* 0.1482 (1.04) -0.4458 (-0.79)
Voting affected by ethnicity, municipal average
1.4920 (2.84)***
1.6507 (3.13)***
0.3827 (2.04)**
1.0505 (3.26)***
Private health facilities in the municipality exists
0.78041 (3.94)***
0.8309 (3.51)***
0.1957 (1.87)*
0.5039 (3.51)***
-0.4339 (-1.89)*
-0.6265 (-2.43)**
0.0190 (0.16)
-0.3840 (-2.46)**
-0.2686 (-0.57)
-0.1736 (-0.35)
0.3086 (1.38)
-0.0462 (-0.15)
-2.0900 (-1.77)*
-2.2966 (-1.81)*
-0.9914 (-1.66)*
-1.4142 (-1.90)*
-0.0215 (-0.03)
-1.0253 (-1.09)
0.2507 (0.69)
-0.5921 (-1.00)
-2.0900 (-2.56)***
-2.5575 (-2.17)**
-1.4285 (-2.95)***
-1.7126 (-2.33)**
No
No
No
Province
843, 0.0448
843, 0.0714
843
972
-499.7276
-485.7732
-555.4976
-849.5293
0.0001
0.0000
0.0004
0.0000
-2.6112 (-1.82)*
-2.0433 (-1.36)
-1.0670 (-1.73)*
-1.4141 (-1.49)†
-3.0092 (-2.11)**
-3.1031 (-2.02)**
-1.9284 (-2.67)***
-2.0189 (-2.18)**
-3.0881 (-2.14)**
-2.1973 (-1.44)
-0.3206 (-0.55)
-1.4268 (-1.53)†
-3.5193 (-2.72)***
-2.7149 (-1.98)**
-0.9489 (-1.82)*
-1.7398 (-2.06)**
-3.6137 (-2.68)***
-2.4069 (-1.67)*
-0.8510 (-1.72)*
-1.5731 (-1.78)*
VARIABLES
Health Facilitie s/ Commu nity Resourc es
Household at urban area Education of households Ownership of durable goods
Household level Municipal average Household level Municipal average Household level Municipal average
Wealth inequality within municipality Involvement in social organizations Reading newspapers Voice, Exit, Particip ation
Instituti ons, Corrupt ion, Decentr alization
Vote in elections
Household level Municipal average Household level Municipal average Household level Municipal average
Log of distance to closest alternative health center, municipal avg. Frequency of audit by central government, municipal average Accountability index, average of HCs and municipal health officials, municipal average Autonomy index, average of HCs and municipal health officials, municipal average Corruption perception of officials, municipal average Dummy variables N, Adj. R2 Log likelihood p-value (significance of model) Corruption perception of officials, urban municipalities Corruption perception of officials, rural municipalities
Corrupt ion4
Corruption perception of officials, rich municipalities Corruption perception of officials, middle-income municipalities Corruption perception of officials, poor municipalities
0.5403 (2.40)** 0.1893 (0.37)
-0.2350 (-0.70)
1
Grouped by provinces. Grouped by provinces. Likelihood ratio test of random effects (Ho: rho=0): p=0.98 with rho=0.00, where rho is the ratio of variance of random effect component to total variance. 3 Outcome equation. Maximum likelihood results. Standard errors are corrected for heteroscedasticity. 129 observations are censored (households who do not use public health clinics). The selection equation (results are not reported) involves all explanatory variables in the outcome equation, except the province dummies. Correlation (rho) between the error terms of selection equation and outcome equation is 0.22. Wald test on independence of the equations (H0: rho=0): p=0.24. The coefficient of corruption variable in the selection equation is -1.4538 with standard error 0..4161 and significant at 5%. 4 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
40
Table 12: Satisfaction with Public Schools, Household Survey Dependent variable is divided into 2 groups. It is equal to one if household is very satisfied with the public schools, and zero otherwise *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; VARIABLES School capacity index, municipal average School size, municipal average School / Community Resources
Household at urban area Education of households Ownership of durable goods
RANDOM EFFECTS PROBIT2
HECKMAN SELECTION MODEL3
2.7692 (2.22)**
1.6209 (1.19)
0.6918 (1.10)
1.0339 (1.24)
0.4297 (2.07)**
0.2854 (1.28)
0.1621 (1.49)†
0.1554 (1.24)
-0.7724 (-2.99)***
0.0783 (0.26) -0.7203 (-2.56)*** 0.1843 (1.01) 0.7777 (0.77) 1.1385 (1.68)* 5.2166 (2.19)**
-0.0196 (-0.12) -0.3484 (-2.27)** 0.1230 (0.72) 0.0994 (0.21) 0.6927 (1.73)* 2.2871 (1.89)*
0.0828 (0.45) -0.4456 (-2.56)** 0.1160 (0.71) 0.5253 (0.88) 0.6858 (1.65)* 2.9695 (1.97)**
0.2628 (1.01) 1.5450 (2.42)** 0.6901 (0.79)
1.5028 (1.43)
1.1518 (2.25)**
0.9498 (1.41)
Involvement in social organizations
0.4549 (0.58)
0.1405 (0.14)
0.1258 (0.15) 1.2909 (0.41) -0.6366 (-1.53)† -2.2766 (-1.55)† 0.4428 (1.49)† 0.0997 (0.09)
0.0313 (0.06) 0.9141 (0.67) -0.4040 (-1.62)† -1.2322 (-1.77)* 0.2518 (1.42) 0.6027 (0.96)
0.0924 (0.16) 0.9689 (0.55) -0.4132 (-1.62)† -1.3947 (-1.54)† 0.2745 (1.42) 0.0287 (0.04)
1.7266 (2.72)***
2.0455 (2.85)***
0.8005 (2.45)**
1.1755 (2.60)***
-1.1943 (-2.59)***
-1.1594 (-2.52)**
-0.5536 (-2.09)**
-0.7117 (-2.53)**
0.4392 (0.67)
0.1204 (0.17)
0.2987 (0.77)
0.0608 (0.14)
0.6473 (0.34)
0.4510 (0.22)
-0.4280 (-0.43)
0.2424 (-0.20)
0.1190 (0.19)
0.2208 (0.33)
-0.0734 (-0.21)
0.1306 (0.33)
-1.4587 (-1.17)
-2.4418 (-1.81)*
-1.6045 (-2.22)**
-1.5545 (-1.85)*
644, 0.0604
644, 0.0717
644, 0.0717
1033, 0.0717
-339.5133
-335.4428
-397.3319
-1011.1560
0.0001
0.0019
0.0006
0.0000
-0.9123 (-0.59)
-1.6003 (-0.94)
-1.3814 (-1.50)†
-1.3814 (-1.50)†
-3.4814 (-1.90)*
-3.6512 (-1.82)*
-1.9250 (-1.77)*
-1.9250 (-1.77)*
7 0758 ( 2 49)** -1.9525 (-1.23)
9 6210 ( 2 68)*** -1.8847 (-1.10)
M i i lA -2.4319 (-3.19)***
M i i lA -1.2759 (-1.20)
-1.9839 (-1.37)
-1.8467 (-1.18)
2.4029 (-3.58)***
-1.2715 (-1.31)
-2.4602 (-1.70)*
-2.2350 (-1.41)
-2.6579 (-3.83)***
-1.4970 (-1.53)†
Vote in elections
Household level Municipal average Household level Municipal average Household level Municipal average
Voting affected by ethnicity, municipal average Private schools in the municipality exists
Institutions, Corruption, Decentralizati on
FIXED EFFECTS LOGIT1
Wealth inequality within the municipality
Reading newspapers Voice, Exit, Participation
Household level Municipal average Household level Municipal average Household level Municipal average
FIXED EFFECTS LOGIT1
Frequency of audit by central government, municipal average Accountability index, average of schools and municipal DECS officials, municipal average Autonomy index, average of schools and municipal DECS officials, municipal average Corruption perception of officials and households, municipal average N, Adj. R2 Log likelihood
-0.7786 (-1.95)*
p-value (significance of model) Corruption perception of officials and households, urban municipalities Corruption perception of officials and households, rural municipalities Corruption4
Corruption perception of officials and households, rich municipalities Corruption perception of officials and households, middle-income municipalities Corruption perception of officials and households, poor municipalities
1
Grouped by provinces. Grouped by provinces. Likelihood ratio test of random effects (Ho: rho=0): p=0.00 with rho=0.12, where rho is the ratio of variance of random effect component to total variance. 3 Outcome equation. Maximum likelihood results. Standard errors are corrected for heteroscedasticity. 389 observations are censored (households who do not use public schools or not report their satisfaction rating). The selection equation (results are not reported) involves all explanatory variables in the outcome equation, except the province dummies. Correlation (rho) between the error terms of selection equation and outcome equation is 0.01. Wald test on independence of the equations (H0: rho=0): p=0.98. The coefficient of corruption variable in the selection equation is –0.2029 with standard error 0.4887. 4 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
41
Table 13: Log of percentage of students passed NEAT, average of 1997 and 1998 *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; VARIABLES
0.6025 (2.43)** -0.0936 (-1.39) 0.1596 (2.30)** 0.2727 (1.96)** -1.3977 (-3.75)*** -0.9878 (-3.93)*** 0.6080 (1.11) 0.1673 (1.11) 0.3954 (2.00)** -0.2539 (-1.88)* 0.2660 (2.28)** -0.1090 (-0.64) -0.2240 (-0.51) 0.4310 (2.85)*** -0.9821 (-2.38)**
0.6349 (2.91)*** -0.1109 (-1.73)* 0.1711 (2.69)** 0.3761 (2.87)*** -1.4757 (-4.56)*** -0.9433 (-4.18)*** 0.6146 (1.38) 0.2803 (1.98)** 0.3542 (2.01)** -0.1958 (-1.56)† 0.2892 (2.71)** -0.1772 (-1.12) -0.1922 (-0.50) 0.5722 (3.92)*** -1.0915 (-2.68)**
Dummy variables in the model
NO
Region
NO
NO
Region
N, Adj. R2
50
50
49, 0.2436
49, 0.57
49
-
-
-62.3355
-
15.0553
0.0025
0.0000
0.0264
0.0006
0.0000
-0.2808 (-0.55) -0.8745 (-1.97)**
-0.4679 (-1.01) -1.0359 (-2.31)**
-1.5663 (-0.57) -7.3257 (-3.10)***
0.6000 (1.08) -1.3970 (-2.58)**
-0.2462 (-0.41) -2.0430 (-3.09)***
0.3908 (0.98) -0.0641 (-0.19) -0.0489 (-0.14)
-0.7660 (-2.12)** -0.6572 (-2.04)** -0.8450 (-2.87)***
-0.3388 (-0.08) -1.4374 (-0.41) -2.5066 (-0.61)
-0.0614 (-0.09) -0.0788 (-0.13) -0.3103 (-0.55)
-0.1363 (-0.21) -0.2723 (-0.46) -0.6098 (-1.08)
School at urban area Education of households, barangay average
Wealth inequality within municipality Involvement in social organizations, barangay average Reading newspapers, barangay average Vote in elections, barangay average Voting affected by ethnicity, barangay average Private schools in the area exists Frequency of audit by central government, school average Accountability index of school officials and municipal DECS officials, school average Autonomy index of school officials and municipal DECS officials, school average Corruption perception of officials and households, municipal average
Log likelihood p-value (significance of model) Corruption perception of officials and households, urban municipalities Corruption perception of officials and households, rural municipalities Corruption5
TOBIT4
1.7404 (1.39) -0.2954 (-0.86) 0.7282 (1.87)* 3.4361 (4.14)*** -4.8499 (-1.60)† -03.7455 (-2.70)*** 2.5304 (1.08) 0.6805 (0.81) 1.6783 (2.09)** -1.1588 (-2.91)*** 1.3840 (1.17) -1.4602 (-1.68)* -0.9728 (-0.43) 2.4405 (2.44)** -3.4059 (-2.18)**
Ownership of durable goods, barangay average
Institutions, Corruption, Decentraliza tion
FIXED EFFECTS3
0.8077 (3.78)*** -0.0612 (-1.06) 0.1021 (1.71)* 0.1180 (0.99) -1.4260 (-4.45)*** -1.1095 (-5.14)*** 0.4249 (1.08) -0.0037 (-0.03) 0.3602 (2.12)** -0.1695 (-1.46) 0.3184 (3.17)*** -0.0288 (-0.20) -0.0010 (-0.03) 0.6787 (5.21)*** -0.7714 (-2.18)**
School size, school average
Voice, Exit, Participatio n
ORDERED PROBIT2
0.5265 (2.27)** -0.0630 (-1.14) 0.1329 (1.83)* 0.4636 (3.45)*** -0.5714 (-1.42) -0.5002 (-2.43)** -0.0803 (-0.19) 0.0653 (0.41) 0.2143 (1.37) -0.1762 (-1.82)* 0.1790 (1.31) -0.2313 (-1.46) 0.1150 (0.25) 0.2403 (1.47) -0.6223 (-1.82)*
School capacity index, school average
School/Com munity Resources
ROBUST REGRESSION1
Corruption perception of officials and households, rich municipalities Corruption perception of officials and households, middle-income municipalities Corruption perception of officials and households, poor municipalities
1
Number of observations with weight less than 0.75: 9 in the 1st column, 6 in the 2nd column. Grouped by regions. Observations are independent across groups (regions) but not necessarily independent within groups. Dependent variable is divided into 8 categories. 3 Grouped by regions. F test that all fixed effects are zero: p=0.03. Hausman test on orthogonality condition: p=0.00. Since orthogonality condition is not satisfied, random effects model is not used. 4 Since results obtained from random effects tobit model are stable, we use region dummies in a tobit model to account for fixed effects. 5 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
42
Table 14: Average NEAT score, average of 1997 and 1998 *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; ROBUST REGRESS. 1
ORDERED PROBIT2
FIXED EFFECTS3
RANDOM EFFECTS4
TOBIT5
0.2073 (1.45) -0.0392 (-1.50)† 0.0903 (2.23)** 0.2667 (3.20)*** -0.1168 (-0.54) -0.1202 (-1.06) 0.1691 (0.77) 0.1853 (2.42)** -0.1172 (-1.34) -0.2761 (-5.12)*** -0.0544 (-0.66) -0.3508 (-3.92)*** 03292 (1.27) 0.3045 (2.83)*** -0.6308 (-3.31)**
1.9726 (1.16) -0.4721 (-1.11) 1.5864 (2.12)** 2.8154 (2.17)** -4.6607 (-2.16)** -4.3530 (-2.67)*** -1.0892 (-0.74) 1.5303 (1.25) -0.5381 (-0.50) -2.5994 (-1.87)* -0.9991 (-1.38) -2.4835 (-1.11) 2.9191 (0.88) 4.1548 (3.34)*** -6.5852 (-2.18)**
0.1364 (0.73) -0.0184 (-0.53) 0.0426 (0.97) 0.0715 (0.79) 0.1421 (0.58) -0.4220 (-2.25)** 0.0817 (0.26) 0.1803 (1.96)* -0.0670 (-0.58) -0.1722 (-1.60)† -0.0559 (-0.66) -0.2513 (-2.07)* -0.3631 (-0.80) 0.3281 (3.12)** -1.0156 (-4.32)***
0.2205 (1.66)* -0.0394 (-1.63)† 0.1026 (2.73)*** 0.2209 (2.69)*** -0.1632 (-0.80) -0.2665 (-2.50)** -0.0334 (-0.15) 0.1119 (1.56)† -0.0729 (-0.82) -0.1777 (-3.45)*** -0.0477 (-0.64) -0.2261 (-2.47)** 0.1906 (0.79) 0.2997 (3.01)*** -0.5643 (-3.12)***
0.2300 (1.69)† -0.0391 (-1.57)† 0.1007 (2.61)** 0.2223 (2.63)** -0.1567 (-0.75) -0.2621 (-2.38)** -0.0301 (-0.13) 0.11133 (1.54)† -0.0752 (-0.82) -0.1803 (-3.41)*** -0.0504 (-0.65) -0.2311 (-2.45)** 0.1920 (0.78) 0.2996 (2.93)*** -0.5764 (-3.10)***
36
34, 0.37
34, 0.59
34
34
-
-32.9854
-
41.3887
37.9432
0.0000
0.0000
0.0026
0.0000
0.0000
Corruption perception of officials and households, urban municipalities Corruption perception of officials and households, rural municipalities
-0.4517 (-2.26)** -1.0665 (-4.11)***
-4.1613 (-1.00) -17.7014 (-3.99)***
-0.5476 (-2.68)** -1.6846 (-7.23)***
-0.2878 (-4.18)*** -13676 (-3.36)***
-0.2857 (-1.26) -1.1806 (-4.19)***
Corruption perception of officials and households, rich municipalities Corruption perception of officials and households, middle-income mun. Corruption perception of officials and households, poor municipalities
-1.2739 (-5.28)*** -1.1462 (-5.47)*** -1.1592 (-5.39)***
-12.6123 (-2.01)** -10.0725 (-1.83)8 -10.6090 (-2.06)**
-1.1337 (-2.55)** -1.1427 (-2.96)*** -1.1088 (-2.00)*
-0.9621 (-3.86)*** -0.9349 (-3.72)*** -0.8933 (-3.62)***
-1.0430 (-3.56)*** -0.9975 (-3.88)*** -0.9577 (-3.76)***
VARIABLES School capacity index School size School/Comm unity Resources
School at urban area Education of households, barangay average Ownership of durable goods, barangay average Wealth inequality within municipality Involvement in social organizations, barangay average Reading newspapers, barangay average
Voice, Exit, Participation
Vote in elections, barangay average Voting affected by ethnicity, barangay average Private schools in the area exists
Institutions, Corruption, Decentralizati on
Frequency of audit by central government, school average Accountability index of school officials and mun. DECS officials, school avg. Autonomy index of school officials and municipal DECS officials, school avg. Corruption perception of officials and households, municipal average N, Adj. R2 Log likelihood p-value (significance of model)
Corruption6
1
Number of observations with weight less than 0.75: 4. Due to limited sample size, region dummies are not used. Grouped by regions. Observations are independent across groups (regions) but not necessarily independent within groups. Dependent variable is divided into 6 categories. 3 Grouped by regions. F test that all fixed effects are zero: p=0.02. 4 Grouped by regions. Maximum likelihood results. Hausman test on orthogonality condition: p=0.79. Breusch and Pagan Lagrangian multiplier test for random effects (H0: no random effect): p=0.88 .with rho=0.01, where rho is the ratio of variance of random effect component to total variance. 5 Results obtained from random effects tobit model are stable and not reported. Due to limited sample size, region dummies are not used. 6 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
43
Table 15: Log of National Ranking in NEAT, 1997 *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; ROBUST REGRESS. 1
ORDERED PROBIT2
FIXED EFFECTS3
RANDOM EFFECTS TOBIT4
-1.2434 (-1.44) 0.0039 (0.02) 0.5829 (2.61)** -1.6198 (-2.05)* -1.5308 (-1.24) -0.8948 (-1.02) 2.9061 (1.71)† -0.3287 (-0.61) -0.0854 (-0.11) 0.1983 (0.57) 0.7196 (1.91)* 0.3923 (0.69) -1.1515 (-0.54) -0.4721 (-0.62) 2.7714 (2.34)**
-1.4767 (-0.79) 0.1799 (0.59) 0.35544 (1.32) -1.5352 (-0.95) -2.1620 (-1.28) -0.5111 (-0.26) 0.3245 (0.15) -1.4113 (-1.75)* -3.5939 (-1.66)* 1.2880 (2.22)** 0.0304 (0.04) -0.7108 (-0.70) -1.0346 (-0.22) -0.8083 (-0.41) 7.2442 (3.87)***
9.1466 (4.61)*** 0.9866 (2.45)** -0.5986 (-1.95)* 0.7585 (0.70) -6.0045 (-3.77)*** -1.8201 (-1.13) -8.6807 (-1.72)† -0.9921 (-1.29) 2.3508 (1.74)† -0.9591 (-1.21) -0.8752 (-1.73)† 1.5186 (1.54) 11.5248 (2.68)** -3.5853 (-1.98) * 6.1946 (1.90)*
1.2441 (0.93) 0.3941 (1.27) 0.0416 (0.12) -0.2311 (-0.19) -1.6779 (-0.87) -0.2652 (-0.19) 1.2915 (0.49) -1.2055 (1.44)† -1.9027 (-1.56)† 1.2036 (2.22)** 0.2473 (0.42) -0.2190 (-0.25) -2.4318 (-0.73) -1.5388 (-1.30) 3.7036 (2.02)**
35
35, 0.23
35, 0.31
35
-
-41.5582
-
-
0.0446
0.0000
0.0061
0.0083
Corruption perception of officials and households, urban municipalities Corruption perception of officials and households, rural municipalities
1.2982 (0.74) 3.9370 (2.75)**
1.4720 (0.61) 13.1851 (2.85)***
6.4031 (1.65)† 5.9907 (1.55)
1.6554 (0.66) 7.1173 (3.47)***
Corruption perception of officials and households, rich municipalities Corruption perception of officials and households, middle-income municipalities Corruption perception of officials and households, poor municipalities
4.7270 (1.63)† 6.2312 (2.31)** 5.4974 (2.18)**
6.5104 (1.14) 7.6686 (1.64)* 7.8314 (1.97)**
4.9739 (1.31) 5.4210 (1.50) 5.1980 (1.51)
5.9430 (2.27)** 6.8504 (2.82)*** 6.4640 (2.83)***
VARIABLES School capacity index School size School/Com munity Resources
School at urban area Education of households, barangay average Ownership of durable goods, barangay average Wealth inequality within municipality Involvement in social organizations, barangay average Reading newspapers, barangay average
Voice, Exit, Participation
Vote in elections, barangay average Voting affected by ethnicity, barangay average Private schools in the area exists
Institutions, Corruption, Decentralizat ion
Frequency of audit by central government, school average Accountability index of school officials and municipal DECS officials, school average Autonomy index of school officials and municipal DECS officials, school average Corruption perception of officials and households, municipal average N, Adj. R2 Log likelihood p-value (significance of model)
Corruption5
1
Number of observations with weight less than 0.75: 4. Grouped by regions. Observations are independent across groups (regions) but not necessarily independent within groups. Dependent variable is divided into 5 categories. 3 Grouped by regions. F test that all fixed effects are zero: p=0.00. Hausman test on orthogonality condition: p=0.00. Since orthogonality condition is not satisfied, random effects model is not used. 4 Grouped by regions. Wald test on existence of random effects (Ho: σu=0): p=1.00 with rho=0.00, where rho is the ratio of variance of random effect component to total variance. 5 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
44
Table 16: Log of standard deviation of NEAT scores in 1997 *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; ROBUST REGRESS. 1
ORDERED PROBIT2
FIXED EFFECTS3
RANDOM EFFECTS4
0.7614 (1.82* 0.0327 (0.35) 0.1324 (1.17) 0.0227 (0.06) 0.2489 (0.38) -0.0232 (-0.06) -0.1485 (-0.17) 0.4127 (1.55)† -0.5821 (-1.45) 0.1902 (1.11) 0.0365 (0.18) 0.2448 (0.86) -0.5729 (-0.53) 0.1969 (-0.52) 1.1222 (1.83)*
5.9546 (3.00)*** 0.6462 (1.49)† 0.4129 (0.74) -0.4464 (-0.25) -0.7805 (-0.27) -2.0533 (-1.54)† -0.1981 (-0.08) 1.6440 (1.53)† -3.0166 (-1.71)* 1.7844 (2.96)*** 0.0015 (0.00) 0.7984 (0.83) -3.3980 (-0.83) -1.0878 (-0.79) 5.4077 (2.48)**
1.1762 (1.98)* -0.0167 (-0.11) -0.0288 (-0.24) 0.4038 (0.98) -0.7995 (-1.09) -0.6193 (-1.28) 1.0868 (0.63) -0.4943 (-1.66)† -0.7330 (-1.67)† 0.2332 (0.84) -0.0874 (-0.39) 0.5364 (1.48) -0.0608 (-0.05) -0.1854 (-0.50) 1.8111 (1.95)*
1.0190 (3.28)*** 0.0927 (1.35) 0.0411 (0.49) 0.2118 (0.72) -0.0936 (-0.19) -0.0394 (-0.13) -0.2524 (-0.40) 0.1485 (0.75) -0.7366 (-2.47)** 0.3117 (2.45)** 0.0057 (0.04) 0.2088 (1.99) -0.2340 (-0.29) -0.1114 (-0.39) 1.1776 (2.59)***
37
37, 0.34
37, 0.28
37
-
-41.3599
-
6.5311
0.0174
0.0206
0.1438
0.0015
Corruption perception of officials and households, urban municipalities Corruption perception of officials and households, rural municipalities
-0.1278 (-0.23) 1.3531 (3.08)***
4.3814 (1.06) 6.1631 (3.81)***
0.7853 (0.78) 3.6989 (2.81)***
0.8196 (1.24) 1.4268 (2.54)**
Corruption perception of officials and households, rich municipalities Corruption perception of officials and households, middle-income municipalities Corruption perception of officials and households, poor municipalities
1.8480 (4.23)*** 1.9356 (4.77)*** 1.6502 (4.34)***
6.1513 (4.13)*** 6.9686 (4.30)*** 6.4466 (4.46)***
3.6842 (4.01)*** 3.4588 (4.06)*** 3.1481 (3.81)***
2.5007 (4.62)*** 2.5179 (5.15)*** 2.3606 (5.08)***
VARIABLES School capacity index School size School/Com munity Resources
School at urban area Education of households, barangay average Ownership of durable goods, barangay average Wealth inequality within municipality Involvement in social organizations, barangay average Reading newspapers, barangay average
Voice, Exit, Participatio n
Vote in elections, barangay average Voting affected by ethnicity, barangay average Private schools in the area exists
Institutions, Corruption, Decentraliza tion
Frequency of audit by central government, school average Accountability index of school officials and municipal DECS officials, school average Autonomy index of school officials and municipal DECS officials, school average Corruption perception of officials and households, municipal average N, Adj. R2 Log likelihood p-value (significance of model)
Corruption5
1
Number of observations with weight less than 0.75: 2. Grouped by regions. Observations are independent across groups (regions) but not necessarily independent within groups. Dependent variable is divided into 6 categories. 3 Grouped by regions. F test that all fixed effects are zero: p=0.17. 4 Grouped by regions. Maximum likelihood results. Hausman test on orthogonality condition: p=0.42. Breusch and Pagan Lagrangian multiplier test for random effects (H0: no random effect): p=0.87, with rho=0.00, where rho is the ratio of variance of random effect component to total variance. Random effects tobit gives the same as random-effects (MLE). 6 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
45
Table 17: Coefficient of variation in NEAT scores, 1997 Dependent variable is the standard deviation of NEAT scores in 1997 divided by mean score in 1997 *** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; ROBUST REGRESS. 1 0.2577 (2.76)** -0.0069 (-0.34) 0.0173 (0.69) 0.0252 (0.29) 0.0015 (0.01) -0.0072 (-0.08) 0.0264 (0.14) -0.0992 (-1.67)† -0.2777 (-3.09)*** 0.1171 (3.05)*** 0.0286 (0.64) -0.0064 (-0.10) 0.0313 (0.13) -0.1040 (-1.20) 0.5903 (4.32)***
ORDERED PROBIT2 3.4977 (1.81)* 0.5728 (1.09) 0.4721 (0.75) -1.2675 (-0.90) 0.8797 (0.32) -0.8544 (-1.51)† -0.9055 (-0.36) -0.6722 (-0.53) -3.4358 (-2.00)** 1.4914 (2.50)** 0.0606 (0.10) 0.7914 (1.38) -4.0755 (-1.03) -0.8667 (-0.98) 6.2884 (2.48)**
FIXED EFFECTS3 0.3404 (2.35)** 0.0193 (0.50) 0.0068 (0.23) -0.0091 (-0.09) -0.2498 (-1.40) -0.1707 (-1.53) -0.0699 (-0.17) -0.1591 (-2.20)* -0.1325 (-1.24) 0.0460 (0.68) -0.0467 (-0.87) 0.1531 (1.73)† 0.1956 (0.69) -0.0974 (-1.07) 0.4188 (1.85)*
RANDOM EFFECTS4 0.1452 (1.88)* 0.0321 (1.89)* 0.0299 (1.44) -0.0121 (-0.17) -0.0322 (-0.26) -0.0021 (-0.03) -0.0239 (-0.15) -0.0430 (-0.87) -0.2100 (-2.84)*** 0.0802 (2.54)** -0.0035 (-0.10) 0.0431 (0.82) -0.0502 (-0.25) -0.0156 (-0.22) 0.3682 (3.26)***
37
37, 0.29
37, 0.35
37
-
-38.3304
0.0052
0.0000
0.0722
0.0010
Corruption perception of officials and households, urban municipalities Corruption perception of officials and households, rural municipalities
0.0584 (0.22) 0.4808 (2.23)**
1.13214 (0.39) 10.5521 (3.37)***
0.2976 (1.08) 0.6417 (1.78)*
0.1668 (1.05) 0.5085 (3.76)***
Corruption perception of officials and households, rich municipalities Corruption perception of officials and households, middleincome municipalities Corruption perception of officials and households, poor municipalities
0.5903 (3.86)*** 0.6061 (4.26)*** 0.6013 (4.52)***
11.5878 (2.82)*** 12.0001 (3.35)*** 12.3725 (3.65)***
0.7887 (2.87)** 0.7378 (2.89)** 0.7114 (2.87)**
0.5644 (4.07)*** 0.5876 (4.68)*** 0.5936 (4.98)***
VARIABLES School capacity index School size School/Comm unity Resources
School at urban area Education of households, barangay average Ownership of durable goods, barangay average Wealth inequality within municipality Involvement in social organizations, barangay average Reading newspapers, barangay average
Voice, Exit, Participation
Vote in elections, barangay average Voting affected by ethnicity, barangay average Private schools in the area exists Frequency of audit by central government, school average
Institutions, Corruption, Decentralizati on
Accountability index of school officials and municipal DECS officials, school average Autonomy index of school officials and municipal DECS officials, school average Corruption perception of officials and households, municipal average N, Adj. R2 Log likelihood p-value (significance of model)
Corruption5
1
58.0959
Number of observations with weight less than 0.75: 2. Grouped by regions. Observations are independent across groups (regions) but not necessarily independent within groups. Dependent variable is divided into 5 categories. 3 Grouped by regions. F test that all fixed effects are zero: p=0.14. 3 Grouped by regions. Maximum likelihood results. Hausman test on orthogonality condition: p=0.08. Breusch and Pagan Lagrangian multiplier test for random effects (H0: no random effect): p=0.74 with rho=0.00, where rho is the ratio of variance of random effect component to total variance. Random effects tobit gives the same as random-effects (MLE). 4 Corruption variable in the original model is divided based on, first, the location (urban vs. rural) and then the prosperity of municipalities (rich, middle income, and poor). To determine the prosperity of a municipality we use the official guidelines provided by the Government of Philippines that divides municipalities into six categories based on average income in the last three years. Coefficients of other variables are not reported. 2
46
Table 18: Robustness Test Elastisicities around the mean are reported . t-statistics are in parenthesis*** significant at 1%; ** significant at 5%; * significant at 10%; † significant at 15%; (1) Random effects results (if not applicable, then fixed effects results are reported) (2) Robust regression results (if not applicable, then Heckman’s selection model results are reported) (3) Perceptions on local corruption are “cleaned” against pessimism/optimism using perceptions on national corruption. (4) IV approach: Instruments for corruption are ethnic fragmentation and ethnic voting in local elections. (5) IV approach: Instruments for corruption are square of exogenous variables.
BASE MODEL (ORIGINAL RESULTS) DEPENDENT VARIABLES Immunization rate (municipal average) Immunization rate (households) Delay in Receiving Immunization (households) Choosing Public Health Facilities for Immunization Waiting Time in Public Health Clinics Denied Vaccine from Public Health Clinics Satisfaction with Public Health Clinics Satisfaction with Public Schools Percentage of students passed NEAT Average NEAT score National Ranking in NEAT Standard deviation of NEAT scores Coefficient of variation in NEAT scores
(1) -0.36 *** (2.66) -1.62 *** (-2.31) 0.07 ** (2.24) -0.07 * (-1.91) 0.21 * (1.87) 0.46 *** (2.58) -0.48 *** (-2.95) -0.52 ** (-2.22) -0.27 ** (-2.38) -0.15 *** (-3.12) 1.01 ** (2.02) 0.31** (2.59) 0.82 *** (3.26)
(2) -0.38 *** (-3.10) 0.20 * (1.81) 0.44 ** (2.02) -0.59 ** (-2.33) -0.51 * (-1.85) -0.21 ** (-2.18) -0.17 *** (-3.31) 0.76 ** (2.34) 0.32 * (1.83) 0.51 *** (4.32)
47
SENSITIVITY ANALYSIS (3) -0.22 ** (-2.25) -0.78 ** (-2.18) -0.05 ** (-1.99) -0.03 (-1.45) 0.14 (1.57) 0.30 ** (2.17) -0.22 * (-1.89) -0.14 (-1.12) -0.23 ** (-2.20) -0.13 ** (-2.58) 0.98 ** (2.10) 0.27 ** (2.32) 0.71 ** (2.49)
(4) -0.28 ** (-2.32) -1.33 ** (-2.45) -0.08 ** (-2.21) -0.04 (-1.51) 0.20 * (1.69) 0.42 *** (2.71) -0.39 ** (-2.11) -0.36 * (-1.70) -0.25 ** (-2.31) -0.12 ** (-2.45) 1.05 * (2.38) 0.28 ** (2.37) 0.80 *** (2.99)
(5) -0.30 ** (-2.40) -1.58 *** (-2.70) -0.09 ** (-2.33) -0.06 * (-1.87) 0.20 ** (1.71) 0.36 ** (2.33) -0.44 *** (-2.29) -0.42 ** (-1.99) -0.26 ** (-2.35) -0.13 ** (-2.50) 0.95 ** (2.29) 0.24 ** (2.19) 0.75 ** (2.87)
