To examine the causal effects of aid on economic growth, the study uses district level data collected for the period 1999 to 2013. The analysis includes all 28 administrative districts
in Malawi,27 however the analysis is in most specifications reduced to 24 districts because
some districts were recently formed (by splitting them from other districts).28
Nighttime light data is used to proxy for economic activity at the district level. Research has shown that luminosity reflects human economic activity such as private consumption,
27Table 4.B.1 lists all the districts in Malawi
28Neno and Likoma districts were formed after splitting from Mwanza and Nkhatabay districts respec- tively. For these new districts, some data on most of the variables is missing not because they are not necessarily reported, but rather because for most of the years under study they were still being reported as part of the districts they were split from. Thus they are entirely excluded but they are subsumed as part of the parent districts. Further the two major cities of Lilongwe (the capital city) and Blantyre are also left out in most specifications.
production and government expenditures (Hodler and Raschky (2014)). Geographers (Elvidge et al. (1997), Sutton et al. (2007)) and ecologists (Doll et al. (2006)) first used light density to study urbanization. Economists Chen and Nordhaus (2011) and Hen- derson et al. (2012) subsequently showed that light intensity at night is a good proxy for economic activity that was traditionally measured by GDP. Recent examples of the use of luminosity include Michalopoulos and Papaioannou (2013) which studies develop- ment in Africa. The use of nighttime data in empirical growth studies has since gained prominence.
There are two key advantages of using luminosity data over the most commonly used GDP-based measures of economic activity; firstly, since it is available in the same quality
for all countries,29 it is a useful measure particularly for aid-recipient developing countries
whose data is often considered of low quality30 (Chapter 3 discusses this aspect in more
detail). Secondly, it is available in the same quality at regional or lower (localized) levels, which is very novel for use when conducting a within country analysis. GDP data is simply not available at sub-national level.
Satellite data on nighttime images have been recorded every night by weather satellites at the United States Air Force (USAF) Defense Meteorological Satellite Program (DMSP) using their Operational Linescan System (OLS) sensors since the 1970’s but a digital archive only begun being stored in 1992. Once recorded, the original nighttime light readings are re-calibrated by scientists at the National Oceanic and Atmospheric Ad- ministration’s (NOAA) National Geophysical Data Center (NGDC) to leave mostly only
man-made light31 (Henderson et al. (2012)). A satellite-year dataset is then produced by
averaging all data from all orbits of a given satellite in a given year over all valid nights. Figure 4.3.1 depicts luminosity at the pixel level for Malawi in 1999 and 2010 against the
district borders.32 For analysis in this paper, we calculate average light density at the
district level (average light intensity per square kilometer) in each year over the period
29Apart from countries very close to the North or South Poles
30This is due to poor data collection capacity and sometimes as a result of manipulation of official statistics by rent-seeking governments with poor institutions.
31This includes accounting for variations in sensor settings over time, removing natural anomalies such as the effect of the lunar cycle and auroral activity and removing obscures from cloud cover
32The light dataset is available for free at the National Geophysical Data Center’s website: http: //ngdc.noaa.gov/eog/dmsp/downloadV4composites.html while maps for administrative districts are downloaded from DIVA-GIS, available athttp://www.diva-gis.org/gdata
1999 to 2013.
Figure 4.3.1: Nighttime images for Malawi in 1999 (left) and 2010 (right)
The compiled dataset reports light intensity as a digital number, an integer on a scale that ranges from 0 to 63 where 0 implies no light recorded and higher digital number values imply higher light intensity recorded. Figure 4.A.4 shows an example of a zoomed- in nighttime image of Lilongwe in 2013, showing the digital number values for various locations within a pixel. Using ArcGIS software, average nighttime light data is calculated for all 28 districts from 1999 to 2013, creating a panel data of light data for 28 districts over 15 years. This dataset allows us to examine causal effects of aid on economic activities in Malawi.
It should be noted that ‘growth’ as measured by nighttime light data needs to be in- terpreted cautiously, especially when measuring at the sub-national level for low income countries. While nighttime emissions are correlated with economic activity at the ground level, it is a relatively stronger measure of population density Mellander et al. (2015). For low income countries, it is reasonable to assume that much of the aid in rural areas may not be used in light-generating activities, but could lead to a reallocation of popu- lation and activity across space. In our case, we control for this possibility by including population density as one of the district controls.
(AMP), managed at Malawi’s Ministry of Finance (MoF) in Lilongwe. The AMP is the government’s main tool for tracking and reporting progress of aid-funded activities
in Malawi. The bulk of the data in the AMP comes from AidData33 which conducted
the Malawi Geocoding project, a joint venture between Development Gateway,34 and the
Robert S. Strauss Center’s Climate Change and African Political Stability (CCAPS).35
The Malawi Geocoding project was the first effort to compile comprehensive geocoded data of all donor activities in a single recipient country in Africa. Figure 4.3.2 shows a map of Malawi with locations of geocoded projects.
Figure 4.3.2: Location of geocoded projects in Malawi
Based on information reported by both donors and the Malawi Government, AidData compiled a dataset that contains projects that started as far back as 1996 up to those that started in 2011 when the geocoding exercise took place. The dataset has geocoded
data on projects from over 30 donor agencies for 548 projects, representing $5.3 billion
in total commitments (approximately 80% of total foreign aid to Malawi between 2000 and 2011). The AMP dataset includes projects that started after the geocoding exercise in 2012 and 2013. Thus in total the AMP data has about 623 projects representing over
$7.1 billion of aid received during the period of study (about 82% of total aid during the
33AidData is a joint initiative of Development Gateway, the College of William and Mary and Brigham Young University providing a portal to access on over a million aid financed activities of over 90 donors from 1964 to 2011 (http://www.aiddata.org)
34http://www.developmentgateway.org/about/Case-Studies/Geocoding-in-Malawi
35CCAPS is a joint research program among the University of Texas at Austin, the University of North Texas, the College of William and Mary, and Trinity College Dublin that seeks to provide research based practical guidance for policy makers.
period).
The AMP data disaggregates cumulative project totals into annual commitments and actual disbursements of each project in a particular district, hence actual disbursements are used. A limitation of the AMP data is that it contains some projects that have not
been allocated to any districts.36 Since the empirical estimation is based entirely on a
district-year units, the projects without district locations have been excluded, reducing the number of projects used in this study to 593 projects.
The study uses additional district-level controls. Data on local public expenditures was provided by the National Local Government Finance Committee (NLGFC) through Dis- tricts’ Annual Budget Estimates, which contains estimates of all available financing at the district level (excluding aid since aid is managed by central government Ministries). Pop- ulation data (including population of major ethnic groups in a district) was collected from the National Statistical Office’s (NSO) population census reports. In particular, reports from 1998 Population and Housing Census and 2008 Population and Housing Census were used.
Data on party affiliations of Members of Parliament, as well as list of Cabinet Ministers, was collected from Parliamentary Hansards found at the Malawi National Assembly li- brary. Rainfall data (in millimeters) was obtained from meteorological reports provided by the 22 meteorological stations that form the weather network in Malawi. Poverty rates, infant mortality, life expectancy and rate of food insecurity were all collected from various
reports from the NSO.37 Data on maize production was collected from Annual Crop Yield
estimates complied by the Ministry of Agriculture and Food Security (MoAFS) while education sector data (on gross primary enrolment and number of primary school class- room buildings) is from the Ministry of Education, Science and Technology’s (MoEST) Education Management Information System (EMIS) annual reports. Table 4.B.2 gives a summary of the data used in the analysis and their sources while Table 4.B.3 shows descriptive statistics of the variables in the baseline sample.
36They have not been allocated any geographic coordinates and are therefore classified district unal- locable
37Particularly, reports from three rounds of Living Standards Measurement Surveys (LSMS) namely Integrated Household Survey (IHS1 in 1997-98, IHS2 in 2004-05 and IHS3 in 2010-11); another three rounds of Malawi Demographic and Health Surveys (2000, 2004 and 2010); other isolated reports such as Malawi-An Atlas of Social Statistics (2002), annual NSO’s Welfare Monitoring Surveys reports (2005- 2011) were used.