Skip to main content
Free

Social Transfers and Growth: Evidence from Luminosity Data

Global Development Institute, University of Manchester

Abstract

The effects of social transfers on growth are still unclear. The limitations of national accounts at subnational levels in developing countries have confined the analysis to the use of simulation models and household-level experimental data. This article contributes to the empirical literature by assessing the effects of Colombia’s Familias en Accion, a human development cash transfer program, on municipal growth rates and per capita growth. The staggered introduction of the program in 2001–4 facilitated an identification strategy based on a difference-in-differences estimation. The lack of subnational GDP data is tackled by using luminosity data generated by satellites orbiting the earth, which have demonstrated to be a suitable proxy for economic growth. The results show that the program generates significant positive effects on municipal growth rates.

I. Introduction

The causal relation between social transfers and economic growth remains controversial. While in OECD countries the evidence tends to show negative effects of social spending on growth, in developing countries some authors have found significant positive effects (Damerau 2011). In the particular case of human development cash transfer programs (popular antipoverty interventions in Latin America), abundant evidence demonstrates relevant positive effects on recipients’ human capital, consumption, and intrahousehold allocation (Rawlings and Rubio 2005). Yet experiments or quasi experiments from which such evidence has emerged have failed to provide reliable support to the relation between transfers and the economic activity beyond beneficiary households. To date, most of the existing empirical literature linking antipoverty programs to economic growth has been based on macroeconomic simulations and impact assessments of the effects of social transfers on individuals residing close to actual recipients. The causal inference under a counterfactual approach has been limited in this context by the individual receipt of antipoverty interventions and the absence of consistent aggregate data for the comparison of treated and untreated units.

Extensive studies in OECD countries have come into view, demonstrating the ambiguous effects of social transfers on growth. The recent integration of welfare benefits with employment services has been supported by time series evidence illustrating that government spending on social assistance might deter growth, while labor activation policies might boost it (Arjona, Ladaique, and Pearson 2003). Conversely, the evidence from developing countries shows opposite figures. General equilibrium simulations have exhibited positive effects in this matter, which are dependent on how programs are funded and on how prices respond to a liquidity injection into a local economy (Coady and Harris 2004; Levy and Sherman 2014).1 In the same way, some other authors have been able to detect evidence of extended economic activity generated by social transfers at the household level. For instance, Angelucci and De Giorgi (2009) and Barrientos and Sabatés-Wheeler (2009) examine the effects of Mexico’s Oportunidades cash transfer program (today known as Prospera) on the ineligible population not participating in the program. An effect of the program on ineligible households would be detectable only if they experienced any economic interaction with the beneficiaries. By comparing ineligible households in treated villages with ineligible households in untreated villages, these authors actually found important increases in consumption in ineligible households who cohabited with the main recipients. Despite these notable efforts aimed at linking social transfers and growth in developing countries, there are two clear drawbacks. On one hand, there is a lack of reliable measurement of national accounts at subnational levels, and, on the other, the strong assumptions in the calculation of the effects of social transfers on growth by simulation are still controversial.

This article intends to generate counterfactual evidence on the relation between social transfers and growth and per capita growth. The initial implementation of Familias en Accion, a cash transfer program from Colombia, offers a unique opportunity to produce unconfounded estimates at the aggregate level beyond beneficiary and nonbeneficiary households. Because of budgetary and logistic conditions, the initial implementation of the program was staggered at the municipal level (known elsewhere as county or district level). While an important number of municipalities were eligible for the intervention, not all of them were covered in the first stage of implementation. The staggered introduction of the cash transfers between 2000 and 2004 created a unique natural experiment to identify a difference-in-differences (DID) treatment effects estimation with a comparison of eligible treated and untreated municipalities. The availability of a panel data set facilitated this identification strategy by testing the parallel-paths assumption that dominates the specification of a DID.

As a proxy for economic growth, the estimations here rely on local economic activity captured by luminosity data recorded by satellites before and after the implementation period of Familias en Accion. The luminosity data have proved to be a consistent predictor of the GDP at national and subnational levels (Henderson, Storeygard, and Weil 2012). The correlation between night lights artificially generated by human activity and conventional measures of production and income has been positively tested in spite of the fact that national accounts and satellite-generated data may be contaminated with measurement errors.2 In a cautious comparison between national accounts and luminosity data, Pinkovskiy and Sala-i-Martin (2016) found that both measurement errors are uncorrelated. Thus, the availability of luminosity data as proxy for GDP and the staggered implementation of Familias en Accion provide an exceptional opportunity to assess the impact of social transfers on the GDP at the local level. The results here demonstrate that the program increased the growth rates of the municipalities where it was initially introduced.

The relevant contribution of this article focuses on the discussion on the relation between social transfers and growth (Coady and Harris 2004; Levy 2007; Taylor 2012) and the subnational analysis of growth with alternative measures of economic activity. First, as the quantification of the relation between social transfers and growth has been difficult (Barrientos and Scott 2008), the proposed methodology will be able to estimate the effects of Familias en Accion on municipal growth rates, identifying their counterfactual levels in the absence of the program. Unlike previous findings based on simulations or household survey data, the identification strategy here provides an estimation of the average treatment effects on treated units with a panel data set of 732 Colombian municipalities eligible for the program between 1998 and 2004. Thus, the assumptions of the previous simulated evidence are relaxed, while consistent and robust estimations are provided. Second, the limitations of disaggregated data on GDP at subnational levels are tackled by the employment of luminosity data, contributing to a growing literature on the analysis of economic dynamics with alternative measures of national accounts. In addition to the estimation of the effects of postcolonial ethnic institutions on economic development in Africa carried out by Michalopoulos and Papaioannou (2013) and the studies of the economic dynamics of regional political favoritism by Hodler and Raschky (2014) and regional agricultural production and economic growth by Keola, Andersson, and Hall (2015), this article digs deeper into the employment of luminosity data for a counterfactual intervention analysis in a developing country where subnational accounts are yet unavailable.

This study is divided into six sections. After this introduction, Section II reviews the current discussion on the relation of social transfers and growth. Section III details how the introduction of Familias en Accion at the municipal level provides a suitable framework for a DID treatment effects estimation. Section IV describes the data and shows the results of the empirical exercise. Section V examines whether the findings are driven by spillover effects. Finally, Section VI presents the conclusions of the overall approach.

II. Linking Social Transfers and Growth

The term social transfers has been treated as synonymous with social assistance, antipoverty programs, or social safety nets. Barrientos (2013) defines social transfers as transfers of income delivered to households or individuals aimed at alleviating poverty. They are encompassed within social protection along with noncontributory insurance schemes and active or passive labor market policies.

A growing strand of literature based on macroeconomic simulations supports the idea that a significant injection of liquidity into the communities where social transfers traditionally operate can benefit beneficiaries and nonbeneficiaries in the short run (Taylor 2012). For instance, Thome et al. (2013) find that local production increases by 1.14 shillings for each shilling transferred to households participating in the Kenyan Orphan and Vulnerable Children program. Several causality channels have been discussed. Thus, when social transfers are introduced, the demand for goods and services, particularly food, expands (Hoddinott and Skoufias 2004). This leads to augmented savings and investment at the household level leading to a higher demand of capital and labor (Gertler, Martinez, and Rubio-Codina 2012).

Recent analytical contributions explain channels through which social transfers affect growth (Alderman and Yemtsov 2014). For instance, Barrientos (2012) offers a suitable microfounded framework on this relation. He identifies several mediating elements that focus on beneficiary households and on the population as a whole. These mediating elements are wrapped up in what Barrientos (2012) refers to as local economy effects. Social transfers inject liquidity into the communities where they operate, and if the transfers are significantly high, they can boost local investment and consumption. For instance, the incidence of some programs in a single village could reach 80% of the population, while national coverage can be as high as 25% (Soares, Ribas, and Soares 2010). Angelucci and De Giorgi (2009) examined the effects of Oportunidades on ineligible households living in the same villages where the program was put into operation. Given the nature of the intervention, ineligible households were not direct recipients of the cash transfers. However, the authors found significant increases in households’ consumption and asset holding due to the effects the transfers had on the local economy, by using household surveys. The effects of higher human capital or higher household consumption and investment on growth are hard to measure. Thus, local economy effects are relevant to this study to the extent that they can be easily detectable at the aggregate level.

To characterize local economy effects in the short and medium run, figure 1 shows an analytical flow of an economy with the relevant elements affected by a social transfer program. The delivery of the transfers has immediate effects on household consumption, savings, or credit for beneficiary households. Savings and credit imply effects on household-level investment in productive assets or livestock. In fact, a constant flow of income delivered by a social transfer enhances the credit capacity of households in poverty, while the transfer itself works as collateral.3 More consumption and investment translates into a higher local demand, which may lead to increases in local prices and increases in imports of goods. Local demand can also lead to higher local production and trade. The transfers allow households to engage in investment projects in high-risk and more profitable activities that they otherwise would not undertake. The external push from transfers to local production results in a higher demand for factors, including the demand for productive assets by beneficiaries and nonbeneficiaries. The demand for labor also increases at the local level, resulting in higher household consumption even for nonbeneficiary households. Recent empirical findings suggest that antipoverty programs do not discourage labor supply; rather, they may lead to increases in labor supply for some household members (Barrientos and Villa 2015a). If the delivery of the transfers is constant over time, the cycle can create important local economy effects that have been detected by previous counterfactual evaluations.

Figure 1.
Figure 1.

Analytical flow of local economy effects. Source: author

The causal relation between social transfers and growth is one of the hardest to measure. Theoretical approaches provide insights on the channels through which antipoverty programs affect growth. The most relevant transmission channel to this study is the one describing the local economy effects. The empirical evidence has been obtained by general equilibrium models on the basis of simulations and strong assumptions. Other approaches have used household survey data to detect the effects on cohabitant nonbeneficiaries at the local level. None of the studies have quantified the effects of the transfers on growth at an aggregated level using reliable data.

III. Introduction of Familias en Accion as a DID Setting

An economic crisis in the late 1990s hit most Latin American countries. In 1999 the Colombian GDP shrank 4.9% after a decade of positive growth. According to official documents, the signs of the economic crisis manifested through high unemployment rates, malnutrition of young children, and increased school dropouts. This situation especially affected households in the lowest income quintile (DNP 2001). The Colombian government responded through the introduction of a human development conditional cash transfer program, known as Familias en Accion, whose features had evidenced relevant positive outcomes in Mexico.4

The program was established in 2000, with a pilot phase beginning in December and ending in April 2001 in two municipalities. As no transfers were delivered in 2000, this year is considered to be part of the preprogram period. At the geographic level, during 2001–4 the intervention aimed at covering municipalities with a population below 100,000 inhabitants and with the availability of a private or public bank. The geographic eligibility criteria remained unchanged until 2005. Under these parameters, 732 municipalities were identified for the operation of the program. The expansion of Familias en Accion started in 2001, with the aim of registering 340,000 households in extreme poverty. In 2002 the new government suspended the operation of the program for 8 months. In 2003 the Familias en Accion was restored and expanded, and in 2004 the program started consolidating, and no additional households were registered (Accion Social 2010).

The initial introduction of Familias en Accion offers an opportunity for the identification of treatment effects with a DID setting. Table 1 shows that the gradual registration of municipalities defined a cumulative number of treated and comparison units. From the total 732 eligible municipalities, two were registered in 2000 (but no transfers were delivered in this year), 360 in 2001, 244 in 2002, and six in 2003. As a consequence, by 2003 and 2004 the number of comparison municipalities was 120, while the number of treated reached 612. The gradual introduction of the program was made on a nonrandom basis. Municipalities that were registered earlier or later between 2000 and 2003 were chosen by the administration of the program according to unobserved institutional factors. A single comparison with the resulting composition of treated comparison municipalities would be contaminated by administrative bias. In fact, 732 municipalities share a similar geographic selection criterion; nevertheless, the single difference in the selected outcomes between the 612 treated and the final 120 comparison municipalities would be biased unless an identification strategy is taken into consideration.

Table 1.

Expansion of Familias en Accion and Difference-in-Differences Setting in 2000–2004

 MunicipalitiesHouseholds (in Registered Municipalities)
YearEligibleRegisteredComparisonTreatedEligibleRegisteredTransfers Delivered
(Nominal US$)
2000732273021,184993
2001732360370362320,145236,9013,340,729
2002732244126606247,257164,09845,228,673
200373261206126,6724,46694,262,460
200473201206120095,408,507
Total732612120612575,258406,458238,240,368

Source. Familias en Accion (2004).

View Table Image

The initial selection of participating households was straightforward, given the fact that in 1993 the Colombian government had introduced a socioeconomic classification system known as Sisben. The Sisben would consist of a proxy means test assigning a score between 0 and 100 to every individual household, with 0 being the poorest and 100 the wealthiest. Still,. the program obtains the Sisben information and selects those households with a score below 11 and 17.5 in urban and rural areas, respectively. The program would transfer an average of US$25 per month to registered households with children under age 18. Transfers are conditional on regular school attendance and health checkups of participating children. The initial goal of the program was to register 340,000 families; however, by 2003 the number of registered households reached 406,458, with a total transferred amount of US$238 million (at nominal prices) in 612 municipalities (see table 1).5

The DID setting is suitable when before-and-after intervention data are available for treated and comparison groups. It accounts for unobserved time-invariant selection bias (Abadie 2005). In this particular case, the DID treatment effects are obtained by estimating a fixed effects model with several pre- and posttreatment years over 1998–2004, with the intervention starting in 2001 (Imbens and Wooldridge 2009). The specification here consists of the identification of the DID effects of the program on growth between 2001 and 2004. The panel structure of the data facilitated by the gradual introduction of the cash transfers provides an opportunity to specify a linear fixed effect analysis. Bearing this in mind, the following stochastic equation determines the estimation of the DID treatment effects between 2001 and 2004:

(1)yit=νi+eit+t=20012004β1tDitYt+t=19982004β2tYt+β3Xit,
where yit denotes the outcome variable (growth and per capita growth), νi is the time-invariant municipal effects, and eit is defined as the idiosyncratic effect. The estimand β1t is the effect of the program in each year, t. These effects are generated by the interaction between Dit (a binary variable indicating 1 when the municipality, i, is treated by the program and 0 otherwise in year t), and a binary variable, Yt, indicating each year in the posttreatment period. The estimand β2t denotes the level of the outcome variable for the comparison group resulting from the inclusion of the time trend, Yt, into the equation. Finally, β3 represents the coefficient of an additional time-varying control covariate.

Several assumptions and tests are considered within this setting. First, the DID is assumed to account for parallel paths of the outcome variable before the intervention for treated and comparison groups. Despite the fact that the parallel-paths assumption can be visually checked, further tests are provided in the next section. Second, the standard errors obtained by the fixed effects model could be serially correlated, increasing the significance of the estimand (Bertrand, Duflo, and Mullainathan 2004). Therefore, a test of serial correlation is conducted as evidence of the efficiency of the effect estimations.

IV. Data and Results

Several data sources were considered for the estimations of the DID effects. The final working panel data set is composed of 5,124 observations corresponding to the 732 eligible municipalities over 7 years. Part of the information was provided from the administrative and public records of Familias en Accion during its initial stage of implementation (2001–4). Some additional covariates related to the public transfers received by the eligible municipalities and regions from the national government are also obtained from open data sources. Finally, the outcome variables, the growth and per capita growth rates, are measured from the luminosity data.

One of the main limitations to the assessment of the effects of social transfers on growth is the lack of reliable data at the subnational levels in which these interventions were introduced. National accounts and household surveys do not disaggregate at the municipal level in Colombia. An alternative is the use of luminosity data. Artificial and human-generated night lights are captured by satellites that observe and store the luminosity information from any spot on the earth between the latitudes of 65 degrees north and south. Although the information has been primarily generated by the US Air Force Defense Meteorological Satellite Program since 1992, the National Oceanic and Atmospheric Administration (NOAA) collects and provides the luminosity data.6 Potential contamination from lunar cycles, fires, and the northern lights is removed in order to prevent confounded results. The images are generated globally between 20:00 and 22:00 hours, with a recording capacity of 3,000 kilometers over 14 orbits per day (Elvidge et al. 2009). A reliable reading of night lights results from a 30 arc product (0.86 square kilometers from the equator), and thus the available image resolution does not provide information from small areas. In the stable version, the intensity of the lights is provided in the form of an index between 1 and 63, with 1 indicating an absence of luminosity, and 63 the brightest spots. The sum of light pixels is obtained for Colombian municipalities with a raster calculator tool, filtering and cleaning the data with the Q-GIS open system.

The brightest areas will reflect higher economic activity with a consistent correlation between the measure of luminosity and national accounts (Chen and Nordhaus 2011). The resulting change in the emission of lights generated at each municipality is attributed to changes in the economic activity. Although most activities are developed during daytime, it does not mean that the change in night lights is literally the reflection of every economic transaction. Instead, the night lights are the outcome of the economic activity, reflecting the results of the observed levels of consumption and investment (Henderson et al. 2012).

The use of luminosity information for the assessment of economic growth at subnational levels has been recently scrutinized. In spite of the fact that recordings from satellites around the earth can contain measurement error, they have proved to be a reliable approach for measuring economic activity (Henderson, Storeygard, and Weil 2011, 2012). Regarding the comparison with national account and survey data, Pinkovskiy and Sala-i-Martin (2016) demonstrate that the measurement errors emerging from luminosity data are uncorrelated with those from official accounts. In particular, the generating process of luminosity data is evidently independent from the one generating official aggregated data and household surveys. The fact that measurement errors are uncorrelated implies that revised correlations between GDP and luminosity data are not confounded. Therefore, it is reliable to use luminosity as proxy for economic activity under an unbiased validation.

Henderson et al. (2011) demonstrated that Colombia is among other countries showing a significant correlation between growth rates obtained from luminosity data and from national accounts. Figure 2 shows a scatter plot comparing growth rates from both sources at the regional level.7 Similar to the estimated correlation rate among countries at the global level, the correlation of growth rates from luminosity data and national accounts is 0.58.

Figure 2.
Figure 2.

Regional growth rates: luminosity and national accounts. Sources: DANE (2014) and NOAA (2014)

To look into the behavior of luminosity growth as a proxy for GDP growth, figures 3 and 4 show their evolution over the period of analysis (per capita growth will also be considered later on). Figure 3 compares the GDP growth measured by the Colombian official statistics office (DANE 2014) and the one generated by the luminosity data provided by NOAA (2014). What is perhaps most interesting in these dynamics is that both measures of growth behave roughly the same. For instance, the luminosity data predicted accurately the economic crisis in the late 1990s, while after 2001 both measures do not seem to be highly correlated. Apparently the luminosity data reveal a faster recovery of the country after the crisis.

Figure 3.
Figure 3.

Country-level GDP growth rates. Sources: DANE (2014) and NOAA (2014)

Figure 4 shows the municipal-level growth rates measured by the luminosity data for treated and comparison groups defined for municipalities in the program by 2003–4. Unlike the aggregated data, these municipalities did not follow the same trend or experience a similar cycle. Instead of following the same country-level trend, their growth rate plummeted in 2003 and sharply recovered in 2004. Given the fact that the absence of statistics is a dominant characteristic of these municipalities, lack of documentation makes it difficult to clarify what drove the proxy for economic activity to plummet in 2003.8 Nonetheless, the most striking observation that emerges from this figure is that before the program was introduced, treated and comparison groups notably experienced similar trends. After 2001 the growth rates of both groups started to behave differently, with a higher growth rate for the treated group in 2004 (precisely when the program was consolidated and no new municipalities were covered). In a nutshell, the fact that both groups experience similar trends before the treatment to some extent is an important visual asset in the identification strategy of the DID setting.

Figure 4.
Figure 4.

Municipal-level growth rates. Sources: DANE (2014) and NOAA (2014)

A complementary exercise was done in an attempt to understand the relevance of the transfer on municipal economic activity. The light indexes of the country were summed to obtain the participation of each municipality in total emission of lights at night. The resulting share was multiplied by the value of Colombian nominal GDP according to the national accounts, with the aim of obtaining an estimate of the value of municipal GDP in 2004. The proxy for GDP calculated for treated municipalities totaled US$245 million at nominal prices, that is, only 3% of Colombian total GDP. Since the transfers were almost US$95 million, the proportion of the transfers in reference to a proxy for local economic activity could reach 38.6%. This proportion makes it difficult to ignore that the expected impact of the program on growth is relevant to the treated municipalities whose economic activity, under these circumstances, might be highly sensitive to the social transfers.

Even though visually treated and comparison groups showed similar growth trends before the intervention, the parallel-paths assumption was tested following Mora and Reggio (2012). This assumption indicates that the growth rates of treated and comparison groups would follow the same trend in absence of the program. In other words, it implies that the differences in the growth level before and after the intervention are time invariant if no intervention was introduced. Mora and Reggio (2012) propose a method and a complementary Stata code that focuses on the behavior of the double difference of the outcome variable to test a parallel-growth assumption as alternative to the parallel-paths assumption. It is based on the analysis of q baseline periods before the treatment and s periods thereafter. If the parallel-growth assumption is met, then the comparison group is a suitable counterfactual for the treated group, equivalent to the parallel-paths assumption. The latter is due to the expected change in the growth rate among treated municipalities if being left untreated equalizes the expected change in the growth rate among comparison municipalities.

Formally, Mora and Reggio (2014) propose a fully flexible model to test the parallel-growth assumption by comparing treated and control trends. This could be applied to the estimation of the expected effects of Familias en Accion on growth as follows:

(2)E[yit|Dit]=β0+t=19992004β1tYt+β2Dit+t=19992004β3tYtDit.
After several iterations for each s and q years, if a lag operator, L, defines Δq(1Lq), then the parallel-growth assumption is tested by evaluating the null hypothesis H0:Δq1β3t*=0, with t* indicating the last year or period before the introduction of the treatment (year 2000 in our case).

Table 2 shows the results of the test based on Mora and Reggio (2014). For each baseline period the hypothesis of parallel growth is not rejected, while the overall test of the hypothesis of common predynamics is also not rejected. Thus, the DID identification strategy that suggests that comparison municipalities are a suitable counterfactual for the treated group cannot be rejected. Similarly, the individual cells in table 2 confirm that parallel paths are still valid regardless of the number of municipalities in the treated and comparison groups. In other words, in spite of the fact that the gradual introduction of the program was not random, the parallel-paths assumption holds for treated and comparison municipalities in 2001 as well as those in 2003–4.

Table 2.

Parallel-Paths Assumption Test

 s = 1s = 2s = 3H0: q = q − 1H0: s = s − 1
q = 1−.007−.0013−.004 29.47
 (.003)(.003)(.003) [.000]
q = 2−.011−.022−.018.04735.08
 (.006)(.008)(.011)[.115][.000]
q = 3−.018−.042−.057.06623.99
 (.009)(.023)(.041)[.219][.000]

Source. NOAA (2014).

Note. Parallel-paths assumption test based on Mora and Reggio (2012) and output table from Mora and Reggio (2014). Posttreatment = s; pretreatment = q; sample period = 1998–2004; treatment period = 2001–4. Robust standard errors clustered at regional level in parentheses; p-values in brackets. H0: common predynamics = 2.482, p = .2890.

View Table Image

Apart from testing the parallel-paths assumption, additional time-varying covariates were taken into consideration. Table 3 shows the averages for the selected covariates for treated and comparison groups as defined by the operation of the program. These covariates are described in their pretreatment status in 1999. Instead of showing a test of equality of means between groups, the relevant condition for this analysis is also the test for the parallel-paths assumption on the covariates. Thus, the Mora and Reggio (2014) test is presented in the last three columns. The first two covariates are pertinent to control for the dynamics of luminosity data. The proportion of households having electricity and the real prices of electricity were thus considered with no important differences between treated and comparison groups.9 Similarly, the population of the municipalities was also taken into consideration as the main administrative criterion in the enrollment of the treated municipalities into the program. Tax transfers from the national government and royalty revenues were also considered, as they can externally affect growth and per capita growth of the municipalities and alter program participation through higher local public spending and investment. Although national transfers and royalties could be endogenously correlated with local economic effects, they are determined exogenously by the Colombian constitution and the mineral endowment of each territory. They can account for up to 90% of local fiscal revenues for these municipalities, given that the private economic sector is limited (Chaparro, Smart, and Zapata 2004). Transfers from the national government to regional-level public administration were included in the regression analysis as additional controls. More important, the test for the parallel-paths assumption reveals that, on average, the selected covariates follow similar trends in the absence of Familias en Accion.

Table 3.

Preprogram Descriptive Statistics and Test for Parallel-Paths Assumption

 MeanParallel PathsDifferences in Control Group
CovariateComparisonTreatedCommon Predynamicsp2001 versus 20022001 versus 20032002 versus 2003
Preprogram levels (1998–2000):       
 Electricity coverage81.1181.70.139.710−.188−.142−.113
 [19.96][14.21]  (2.274)(2.390)(.359)
 Electricity price (kW/h)157.7159.6.386.534−.386.502.349
 [25.26][24.80]  (2.367)(2.770)(.615)
 Population (×1,000)28,19220,071.002.966−1,814−1,399−1,602.8
 [23,624][16,240]  (1,537)(1,711)(1,481)
 National transfers.234.244.031.861.021−.001−.003
 [.269][.254]  (.016)(.029)(.004)
 Royalties.176.191.027.869−.055−.021−.005
 [1.202][.966]  (.077)(.144)(.006)
 Regional transfers1.1871.0151.413.235−.031−.063.080
 [2.234][1.875]  (.297)(.147)(.065)
 Regional royalties8.29210.980.834.3612.6332.5632.160
 [15.06][16.94]  (1.725)(3.536)(.120)
Number of municipalities120612732496490246

Sources. DNP (2013) and UPME (2014).

Note. Treated and comparison are as defined by the program selection in 2003–4. Standard deviations in brackets; robust standard errors at the regional level in parentheses. Linear regression preprogram differences in the control group are as defined by the gradual introduction of the program in each year between 2001 and 2003. Electricity price and transfers with 2004 = 100; common predynamics test according to Mora and Reggio (2014).

View Table Image

Although the parallel-paths assumption was tested for each posttreatment year, the exercise presented in table 3 compares the means of the selected covariates for the different samples of the comparison group. Since the introduction of the program was gradual on a nonrandom basis, the last three columns of the table show the preprogram differences in the covariates for the control group in 2001, 2002, and 2003. This exercise is revealing in the sense that it cannot be accepted that there are observable differences in the control group. Therefore, potential selection bias might be explained by unobservable administrative factors that are expected to be ruled out by the fixed effects model specified in equation (1).

A final consideration is focused on the estimation of the standard errors. Bertrand et al. (2004) demonstrate that autocorrelated residuals can lead to underestimated standard errors and wrong significant coefficients. The autocorrelation test by Wooldridge (2001) for panel data was conducted following Drukker (2003). Two outcome variables were considered: first, the growth measured with the luminosity data shows that autocorrelation cannot be accepted (F(1, 732) = 1.445; Pr > .229), and, second, the per capita growth also suggests absence of autocorrelation (F(1, 732) = 1.432; Pr > .231). Since autocorrelation cannot be accepted, the standard errors obtained from the fixed effects estimation are reliable under current settings.

Turning now to the results, table 4 presents the estimates for equation (1), in which the year of gradual introduction of Familias en Accion is explicit in the analysis. Columns 1 and 4 show the fixed-effects estimation with no consideration of additional covariates. The coefficients of interest are those that result from the interaction of the treatment status and year. Strong and robust evidence of the positive effects of the program on growth was found, in particular for 2004, when the program was consolidated after its initial introduction stage between 2001 and 2003 (recall that US$238 million had been transferred to nearly 400,000 families). This effect translated into a 0.7 percentage points higher growth rate for treated municipalities in 2004. Effects in 2003 are also significant with a magnitude of 0.61 percentage points, but such significance is not robust to the inclusion of other covariates.

Table 4.

Fixed Effects Linear Regression Results

 Growth RatePer Capita Growth Rate
Variable(1)(2)(3)(4)(5)(6)
D2004 × Y2004.0076***.0077***.0075**.0075***.0073***.0074**
 (.0022)(.0025)(.0028)(.0019)(.0020)(.0029)
D2003 × Y2003.0061*.0061*.0057.0059*.0059*.0056
 (.0036)(.0037)(.0048)(.0033)(.0035)(.0038)
D2002 × Y2002.0004.0010.0009.0009.0017.0015
 (.0024)(.0027)(.0027)(.0026)(.0029)(.0028)
D2001 × Y2001.0018.0021.0020−.0065−.0061−.0062
 (.0028)(.0030)(.0030)(.0056)(.0061)(.0059)
Y1998−.0008 .0000−.0008 .0000
 (.0039) (.0000)(.0039) (.0000)
Y1999−.0034−.0026−.0025−.0034−.0026−.0025
 (.0047)(.0020)(.0022)(.0048)(.0020)(.0021)
Y2000.0145***.0151***.0152***.0141***.0148***.0147***
 (.0050)(.0034)(.0037)(.0051)(.0033)(.0036)
Y2001.0477***.0482***.0483***−.0075*−.0070***−.0070**
 (.0049)(.0030)(.0032)(.0037)(.0024)(.0026)
Y2002.0056.0057.0057.0045.0045.0045
 (.0040)(.0035)(.0036)(.0042)(.0033)(.0034)
Y2003−.0293***−.292***−.0292***−.0297***−.0296***−.0296***
 (.0040)(.0042)(.0043)(.0039)(.0038)(.0039)
Y2004.0167***.0173***.0186***.0168***.0173***.0182***
 (.0037)(.0029)(.0049)(.0038)(.0030)(.0045)
Electricity coverage  .0000  .0000
   (.0001)  (.0001)
Electricity price (kW/h)  .0000  .0000
   (.0000)  (.0000)
Population  −.0000  −.0000
   (.0000)  (.0000)
National transfers  −.0003  −.0014
   (.0016)  (.0016)
Royalties  .0014***  .0012**
   (.0004)  (.0005)
Regional transfers  −.0002  −.0002
   (.0003)  (.0003)
Regional royalties  −.0000*  −.0000
   (.0000)  (.0000)
Constant.0019.0012.0035.0019.0012.0053
 (.0031)(.0014)(.0107)(.0030)(.0012)(.0105)
Observations5,0374,3205,0375,0374,3205,037
R2.475.524.525.336.385.386

Sources. DNP (2013), NOAA (2014), and UPME (2014).

Note. Fixed-effects model with robust standard errors clustered at regional level (in parentheses); 1% of each side of the growth distribution was trimmed for this analysis. Electricity prices and transfers with 2004 = 100. Number of municipalities = 732.

*. Significant at 10%.

**. Significant at 5%.

***. Significant at 1%.

View Table Image

Several checks were conducted in order to confirm the robustness of these results, in particular for 2004. Columns 2 and 5 show the estimations with a different baseline, by ruling out the sample from 1998. In both cases—growth rate and per capita growth—the results are still robust to this change despite a slight difference in the coefficients.10 The inclusion of additional covariates also confirms the robustness of the initial results. Indeed, electricity coverage and prices were not statistically significant, and their coefficients are close to zero. Columns 3 and 6 reveal that, accounting for exogenous factors that may affect local growth and program coverage (such as population and revenues from the central government), the results from columns 1 and 4 are still robust.

The effects of the program on the local economy are captured by the estimations of the fixed effects model in 2004. Recall that the share of the program in the imputed GDP at treated municipalities is approximately 38%. Bearing this in mind, it does not seem that an effect of 0.7 percentage points on growth rates is significantly high. Eligible municipalities where Familias en Accion operated in 2001–4 are dominated by high levels of extreme poverty and low economic dynamics. The poor economic performance of these municipalities makes them highly sensitive to the injection of an important amount of resources, especially in the form of cash transfers contingent on the human capital formation of beneficiary children.

Another explanation of these results has to do with the implementation of the program. The scaling up of the transfers was relevant in 2001, but in 2002 the operation of the program was suspended as a consequence of the coming to power of a new government. The first stage of the program was completely consolidated by the end of the period of analysis. Indeed, the suspension was complete until 2004, when the program had been incorporated in the government’s plan and the transfers were delivered without budgetary restrictions.

In sum, the luminosity data provide a reliable proxy for GDP growth and per capita growth of Colombian municipalities initially selected by Familias en Accion. This offered an opportunity to identify a treated and comparison group to assess the effect of Familias en Accion on growth rates under a DID setting. The identification strategy included the confirmation of the parallel-paths assumption and the absence of autocorrelation. The results revealed that the program caused a positive effect of 0.7 percentage points in the growth rate and growth rate per capita on treated municipalities in 2004, when it was completely consolidated during its first stage. These results are robust to a change in the baseline and the inclusion of additional exogenous covariates.

V. Checking for Spillover Effects

To check the consistency of the estimation results, this section focuses on whether there were spillover effects among treated and comparison municipalities, particularly in 2004. For instance, investment plans from comparison municipalities might be reoriented toward treated municipalities where better demand prospects could have been identified by local entrepreneurs. These dynamics would lead to the overestimation of the effects of the program. To test the existence of spillover effects, this section employs a method developed by Miguel and Kremer (2004), who propose an econometric approach based on the distance from treated units to every unit in the sample. In this case, particular attention is paid to the number of treated municipalities within a radius of 50 kilometers (31 miles) for all municipalities. The number of nearby treated municipalities is an explicit indication of proximity to places where the program operates.

Based on a before-and-after analysis, the specification here compares the pre- and posttreatment status of the outcome variable in 2004. The following equation is estimated by linear regression:

(3)yi=β0+β1Pi+β2Di+β3PiDi+β4PiNiT+β5NiT+ei,
where, for each municipality i, Pi is a binary variable with values of 1 in the posttreatment period and 0 in the pretreatment period; NiT indicates the number of treated municipalities within a radius of 50 kilometers. On average, each municipality is surrounded by 13 treated nearby municipalities, with a minimum of 0 and a maximum of 63. According to Miguel and Kremer (2004), β4 is assumed to capture the spillover effects of the transfers in 2004. If this estimand is significant, then the spillover effects cannot be rejected. Similarly, β5 captures time-invariant characteristics, such as isolation, that might be correlated with the distance to treated municipalities.

Table 5 shows the results of the estimation of equation (3) for the outcomes of growth and per capita growth. The results show that the presence of spillover effects of the program cannot be accepted. Indeed, the coefficients associated with the number of municipalities within a radius of 50 kilometers are very close to zero. In other words, according to this estimation there is no evidence that the effects of Familias en Accion on growth rates in participating municipalities are reflected on nearby municipalities. Therefore, the positive effects of the program on growth rates in 2004 reported in table 4 are not necessarily driven by spillover effects.

Table 5.

Spillover Effects Estimation (Relevant Coefficients Only)

VariableGrowth RateGrowth Rate per Capita
No. of treated neighbors × Y2004−.0000−.0000
 (.0001)(.0001)
No. of treated neighbors−.0001−.0001
 (.0001)(.0001)
R2.131.124

Sources. DNP (2013), NOAA (2014), and UPME (2014).

Note. Linear regression with robust standard errors clustered at regional level (in parentheses); 1% of each side of the growth distribution was trimmed for this analysis. Number of municipalities = 732; number of observations = 2,718.

View Table Image

VI. Conclusion

This article has contributed to the existing evidence on the effects of social transfers on economic growth. The lack of reliable data and the limitations of household surveys have restricted the scope of the study of this relation. To date, most of the empirical evidence has emerged from simulations and the comparison of treated and control households in the impact evaluation literature. This article has set out to fill the gap of the effects of transfers on growth by using an alternative measure of economic activity at the municipal level in Colombia.

Still, some authors are reluctant to acknowledge the influence of social assistance on growth. As for conditional cash transfer programs in Latin America, the expectations are centered around the effects of these interventions on growth in the long run, when children equipped with higher levels of human capital join the labor markets. Nonetheless, on the basis of the framework provided by Barrientos (2012), it is feasible to believe in short-term effects at the local level. Previous findings have shown that an injection of liquidity into the economy, in the form of cash transfers given to households in extreme poverty, can benefit nonbeneficiaries by means of local economic transactions. Local economy effects cannot be neglected as a potential source of growth.

This article has shown that social transfers in the form of cash transfers generate positive effects on growth and per capita growth rates. The initial introduction of Familias en Accion in Colombia between 2000 and 2004 offered a suitable natural experiment to set up a DID approach, identifying treated and comparison groups of municipalities. Replicating former geographic eligibility criteria, 732 municipalities were considered for the empirical exercise. Since national accounts do not disaggregate the GDP at the municipal level, this article has relied on luminosity data captured by satellites orbiting the earth. The use of luminosity data has provided a reliable proxy for GDP at the national or subnational levels. After testing the DID setting against potential sources of confoundedness, the results showed that the program generated an impact of 0.7 percentage points on municipal growth and per capita growth rates in 2004. Robustness checks confirmed the strength of this finding.

The results from this article have made a relevant contribution to the current literature. This is the first counterfactual approach to determine the sign and magnitude of the effects of social transfers on growth and per capita growth. The theoretical approaches based on the local economy effects are certainly confirmed on the basis of the employment of rich data and the fulfillment of the underlying assumptions in the DID setting. This is also the first approach to the use of luminosity data for the assessment of social transfer interventions, especially at the subnational level. Unlike previous simulations and the detection of indirect treatment effects, the findings here provide direct assessments without making strong assumptions.

Finally, it must be acknowledged that spillover effects may arise among treated and comparison municipalities. However, complementary estimations suggest that spillover effects cannot be accepted. Despite the effects of social transfers on growth that have been estimated, more research is needed to determine whether the growth generated by these interventions through local economy effects is inclusive in the short run. Although previous impact evaluations have detected significant effects on human capital formation of children in Latin America, it is still necessary to assess to what extent adult beneficiaries are obtaining economic benefits from the additional growth generated by social transfers.

Notes

The author is grateful to Anthony Mveyange from the World Bank for his comments and support. Armando Barrientos and Serena Masino from the University of Manchester provided a valuable review. This research is reproduced here with acknowledgment of UNU-WIDER in Helsinki. The author is grateful to the attendees at the UNU-WIDER’s internal seminar, especially to Miguel Niño-Zarazua and Channing Arndt. Errors, omission, and opinions are the author’s.

1 . A recent introduction of a cash transfer program in Indonesia motivated the simulation of its effects on gross domestic product (GDP) growth (Yusuf 2013). Using a general equilibrium model, Yusuf demonstrates that the introduction of the program hindered economic growth if it was funded with value-added taxes. However, this effect could be reduced if the program was funded by dismantling fuel subsidies.

2 . Official statistics agencies may rely on inaccurate surveys and formal transactions that over- or underestimate the calculations of the GDP. Likewise, atmospheric conditions, such as humidity, snow, and temperature may affect how satellites capture night lights.

3 . Angelucci, Attanasio, and Di Maro (2012) find that Oportunidades in Mexico boosts savings and investment in beneficiary households. Gertler et al. (2012) also find that beneficiary households invest in productive assets and livestock, which represents higher household consumption in the medium run.

4 . Familias en Accion was initially funded by loans from the World Bank and the Inter-American Development Bank. These multilateral organizations also facilitated the adoption of these social transfers in Colombia (Barrientos and Villa 2015b).

5 . According to the impact evaluation report of the program in 2001–4 commissioned to IFS-Econometría-SEI (2006), school attendance increased by 5.1 and 7.2 percentage points in urban and rural areas for children between 12 and 17 years of age, respectively. The program reduced the number of repeated school years by 0.12 for children between 14 and 17 years. As for labor markets, child labor was almost eradicated by a reduction of 5.5 percentage points in rural areas, while the number of weekly worked hours by adults remained unaffected. There were no significant impacts on household incomes, while consumption was increased by 5 percentage points in rural areas. Similarly, food consumption increased 15% in rural areas (especially cereals and proteins). Thus, the program reduced food poverty exclusively in rural areas by 12.6%, while multidimensional living condition indexes were unmodified.

6 . See http://ngdc.noaa.gov/eog/download.html.

7 . National accounts in Colombia disaggregate at the regional level since the 1990s.

8 . This coincides with the fact that in 2002 a new government came into power with the promise to undertake a war on drugs and Marxist guerrilla groups. The strategy, known as Plan Colombia, was implemented in the poorest municipalities of the country, with international aid from the United States and European Union. By deploying military troops and eradicating illegal crops, the plan changed the dynamics of economic activity in rural areas and small villages (Petras 2000).

9 . Information on coverage and prices was obtained from the Unidad de Planeacion Minero Energetica (Unit for Mining and Energy Planning; UPME 2014).

10 . Externally, it was proved that the confidence intervals of these coefficients overlap (95% CI: 0.045–0.183 and 95% CI: 0.040–0.183).

References