Title: Why not Count? Selective State Strategies and
Birth Registration in Senegal.
MA Thesis
Research Master Social Sciences
Graduate School of Social Sciences
University of Amsterdam
Cecile Richetta
Student Number: 11753528
cecile_richetta@hotmail.com
Supervisor: Dr. Imke Harbers
Second reader: Dr. Abbey Steele
Why not Count? Selective State Strategies and Birth Registration
in Senegal.
CECILE RICHETTA
Department of Political Science, University of Amsterdam, Amsterdam NETHERLANDS.
ABSTRACT In 2015 worldwide, one child out of three below the age of five did not have his or her birth registered by the state (AbouZahr et al., 2015, p.1374). While the literature has researched the individual level determinants of registration, Public Health and Development scholars have overlooked the role of states’ presence and strategies at the subnational level. I argue that states in Africa have discriminatory selective incentives for registration, despite its advantages for populations and governments alike. My research theorizes the existence of subnational patterns of state presence and selective strategies, and quantitively illuminates the effect of caring, repressive and gatekeeping strategies on birth registration. My case study of Senegal reveals that the state’s strategies selectively increases or decreases registration rates in different districts. This research is located at the intersection of comparative politics and public health. It contributes to academia by theorizing and modelling the effect of selective strategies in Africa, thus filling a gap in the current literature and helping scholars move forward. It also makes a societal contribution thanks to the policy-implications of my findings. Keywords: civil registration, state presence, state strategies, public goods, Africa, Senegal.
1 Introduction
In 2016, in the regions of West and Central Africa, only 49% of children below the age of five were registered with their certificates as proof, making these regions second to worst in birth registration worldwide (Bhatia et al., 2017, p.2). This evidence comes after decades-long efforts by international and local organizations to improve civil registration and vital statistics (CRVS) services, through awareness campaigns among populations and governments, and improvements of the services for registration. Yet, the results are disappointing, and scholars struggle to understand them.
Birth registration is the first official exchange between a child and the state, a prerequisite for legal identity and the door to civil and political rights. As such, birth registration is a public good that can be solely provided by the state (Hunter & Brill, 2016, p.7). Still, political scientists have neglected the study of civil registration in developing countries and how CRVS relate to matters of state presence. Most of the recent findings on CRVS comes from Global Health and International Development literature and focus on individual level and household level predictors of under-registration. Yet, according to Szreter and Breckenridge (2012), registration is “the raison d’être of modern states”, as they are organizations built on the use of textually mediated information (p.1). It is therefore striking that African states do not appear to prioritize investments in CRVS. At the subnational level, African
states already differ from their European counterparts regarding their territorially uneven institutional capacities, and unequal societal redistributive politics (Boone, 2003; Henn, 2018; Herbst, 2000; Kramon & Posner, 2013; Michalopoulos & Papaioannou, 2013). As it appears that CRVS are not their purpose, the relationship between birth registration and state presence in Africa is ambiguous at best.
This paper contributes to the literature by shifting the focus from individuals to the role of state building and state strategies. I move away from the assumptions that states have the capacity and universal motivation to register all citizens alike. Instead, I argue that African states have selective incentives for investments in public goods, which translates into different strategies that impact individuals’ ability and willingness to register. The paper develops a framework that theorizes state strategies in Africa by examining the process of state building as well as the ‘gatekeeping’ role that African states have been playing since colonialism (Cooper, 2002, p.5). Drawing on three data sources, I will analyse the effect of (1) state-imposed barriers and (2) caring and repressive infrastructures, on the likelihood of registration of Senegalese children. To the best of my knowledge, no one has investigated the relation between state strategies and birth registration in Senegal. It is a low-income country with strong regional variations in terms of development, state presence and local traditional authorities (Boone, 2003), which should help us understand the impact of state strategies on registration. The analysis reveals how effective Senegalese state strategies are. The paper demonstrates that the state has put efficient barriers that include the wealthiest and most urban parts of the population in the CRVS system, while excluding the poor and marginalized individuals. In addition, the Senegalese state has invested differently in its regions, prioritizing caring or repressive infrastructures in some areas, while neglecting others. These subnational strategies in turn directly impact registration rates. With this study, I contribute to the literature by modelling the subnational presence of a state, which should help scholars move away from the traditional assumptions that states have the capacity or a universal incentive to implement CRVS. I also make a societal contribution with the policy-implications of my results: policymakers need to consider how states have already implemented strategies that reflect their state interests and directly impact registration.
The next section introduces the reader to the concept of registration and enumeration, some important findings in the literature and their limitations when it comes to explaining the role of the state. The third section lays out the specific state building process through which African states have gone, the existence of selective state strategies and theorizes their impact on state presence. Section four introduces my case study and the selective strategies in place in Senegal. Section five presents the data, variables and methods used, followed by section six with the results of two distinct analyses. Section seven examines some model diagnostics and robustness checks. The final section discusses the results in relation to my theoretical framework and concludes with the policy-implications of this study and how to move forward.
2 Birth Registration in the Literature
Registration and Enumeration
Birth registration is the act of recording a child’s birth to the state administration, which then delivers a certificate that provides legal identity. Birth certificates are administered by civil registration and vital statistics (CRVS) systems, along with marriage and death certificates. A lack of legal identity can be absolute, if the birth was never registered, or relative, if the person has lost or was never issued his or her certificate (Harbitz & Tamargo, 2009, p.9). Moreover, registration can be (1) timely, which is usually free of charge if the child is registered within the legal timeframe, (2) late, which occurs once this timeframe expires and often requires fees or extra administrative procedures, and (3) absent, in case it did not take place at all (Harbitz & Tamargo, 2009, p.8).
CRVS systems differ from enumeration, the most common form of human accounting that states engage in. Enumeration is a unilateral intervention by state agencies to gather information, which usually takes the form of national censuses or population surveys and produces distinct outcomes for individuals. On the contrary, registration is a bilateral act where both the state and the individual decide to interact with each other and that provides direct benefits to individuals, as a birth certificate gives the “grounding for personhood and human rights” (Szreter & Breckenridge, 2012, p.22). For children, it helps them get access to health and education services, shields them from exploitation, and gives them better prospects for their future (AbouZahr et al., 2015; Bequele, 2005; Lo & Horton, 2015). In addition, only with official identification can citizens claim and exercise their civil and political rights, and more generally participate in modern societies (AbouZahr et al., 2015; Harbitz & Tamargo, 2009; Hunter, 2018; Szreter & Breckenridge, 2012). CRVS are also “essential to the smooth functioning and planning of modern states” (Hunter & Brill, 2016, p.1). Without up-to-date vital statistics, states cannot carefully design policies suiting their populations’ needs.
Considering the importance of registration for both individuals and states, this study investigates the political determinants of under-registration. Instead of looking at the common individual-level demographic predictors, I focus on state presence and state strategies.
From Individual-level to State-level Explanations of Under-Registration
Most studies on registration in the developing world come from Global Health and International Development literature, and center on individual and household level predictors. Scholars and professionals found that barriers to registration were geographic, economic, socio-cultural, structural, and discriminatory (AbouZahr et al., 2015; Bhatia et al., 2017; Bowles, 2018; Garenne et al., 2016; Harbitz & Tamargo, 2009; Hunter, 2018; Hunter & Sugiyama, 2017; Jewkes & Wood, 1998; Pelowski et al., 2015). In many cases, the crucial determinant is the child’s parents’ cost-benefit calculation of birth registration. Pelowski et al. (2015) explore the causes of lagging registration in the Kwale County in Kenya and found that rather than fees, knowledge or access, what appeared to be the reason for
under-registration was a “series of small annoyances, coupled with the lack of immediate incentive, that add up to a deliberate decision by a parent that it is not worth the trouble of seeking registration” (p.898). In a multivariate analysis, they found that individual factors like parents’ age, marital status, education, occupation, living environment, number of children, and birth at a hospital influence registration, while neither cost nor distance do (p.896).
As these studies focus on individual-level factors, they illuminate only some of the mechanisms behind under-registration. Scholars acknowledge the central role that the state plays in registration (see for instance: AbouZahr et al., 2007; Garenne et al., 2006; Mahapatra et al., 2007), but fail to model two central state-level factors that affect registration. First, states do not always have the capacity (i.e. power) to implement registration equally throughout their territories. This follows the common categories of strong versus weak states, which are differentiated “by the extent to which states wield authority completely and effectively over their territory” (Steinberg, 2018, p.225). Second, states do not always have a universal incentive to register all people alike. This more recent argument acknowledges that there are conditions under which a state has the capacity but not the incentive to invest in a region and build its presence on the ground (see: Boone, 2012; Herbst, 2000; O’Donnell, 1993; Steinberg, 2018). This “strategic logic behind the territorial distribution of state presence” is not dependent on state capacity but rather on societal pressure on the government to invest in regions that otherwise might not be attractive in terms of economic activity, topography, population distribution, or support for the regime (Steinberg, 2018, p.226). Since the state plays a crucial role in influencing registration, drawing conclusions about the role of the state by solely looking at individuals is problematic.
The necessity to investigate the role of the state is particularly salient in the case of birth certificates because, contrary to other public goods that can be provided by external actors (e.g. vaccination cards), a birth certificate can only be delivered by a state official in a governmental office. Several authors studying public goods provision have already argued that in the developing world, a conceptualization of subnational public good provision is essential due to the variations in range and reach of public goods (Harbers & Steele, 2019), in state involvement and state penetration of its territory (Post, Bronsoler & Salma, 2017), and the existence “areas of limited statehood” where external actors have replaced the state for public good provision (Krasner & Risse, 2017, p.545). Hence, we cannot assume that the state is a passive actor who targets public goods equally throughout its population and territory. The next sections lay out subnational factors and state strategies that influence registration.
3 State-level Explanations of Under-Registration
Subnational State Building and Public Good Provision in Africa
In Europe, the development of CRVS was tightly linked to the development of states’ infrastructural power (i.e. process of building state capacity), as their “institutional capacity […] to penetrate [their] territories and logistically implement decisions” (Mann, 1984, p.113). Scott (1998) argued that states
made their societies ‘legible’ with civil registries to simplify essential functions such as taxation or conscription (see also: D’Arcy & Nistotskaya, 2013; D’Arcy & Nistotskaya, 2017; Zhang & Lee, 2017). Foucault, Faubion and Hurley (1998) developed the concept of the documentary state, according to which governments justify their existence by their ability to control their population. For them, a strong state is one that knows its citizenry, so civil registration is a necessity. In these arguments, states have a universal motivation for registration due to their incentive to control their entire population.
However, these arguments do not hold true for African countries. Historically speaking, colonial states had little if any interest in registering native populations, as they were not citizens of the country (Srzeter & Breckenridge, 2012, p.6).1 They were ‘gatekeeper’ states, limiting channels for
advancement, deciding what and who could enter or leave the country and what regions to invest in (Cooper, 2002, p.5). The process of state building in Africa was thus very different than in Europe. African states deviate from the Weberian ideal of (1) exclusive jurisdiction over the citizenry and all actions taking place on the territory, and (2) the legitimate use of violence within these territorially defined borders (Weber, 1947, p.143). Most of African states do not project their power equally throughout their territory (Bates, 1983; Herbst, 2000; Michalopoulos & Papaioannou, 2013) and share authority with local chiefs (Chabal & Daloz, 1999; Boone, 2003; Boone, 2014; Henn, 2018). Herbst (2000) argued that states in Africa unevenly developed their presence due to the low land-labour ratio and complicated environment, which resulted in a strong urban bias.Boone (2003) focused on rural sub-Saharan Africa and argued that depending on rural elites’ historical strength and interests, central governments implemented different state building strategies. They resulted in “variations in the intrusiveness of the state at the local level, in rulers’ autonomy vis-à-vis local interests, and in the capacity of rural actors to harness state prerogatives and resources” (Boone, 2003, p.11). In a more recent article, Boone (2012) introduces the notion of built state, created from Mann’s (1984) concept of infrastructural power. According to Steinberg (2018), the concept of built state best captures the visibility and immediacy of state presence, and thus the state’s projection of authority in its territory. The concept can be measured by the presence of “outposts for police forces and of the physical offices of a bureaucracy, public clinics and schools, and utilities, in addition to the development of roads” (Steinberg, 2018, p.225). These infrastructures are not only evidences that the state is a relevant institution locally, but in the case of subnational variation, they signal where the state had strategic interests in investments.
In other words, the built state (or state presence) differs between regions within a country, depending on the state’s selective investment strategies. As registration is a public good in which the state can decide to invest in or not, I argue that patterns of under-registration are associated with subnational patterns of state presence. Contingent upon the state’s interests in a region and the people
1 Here, it is crucial to single out the Republic of South Africa. The apartheid state was a heavily bureaucratic
enterprise, one of which aim was to control the movement of the native population from ‘ethnic homelands’ to labour zones (usually mines) with identity cards. But in most of colonial Africa, this was not the case.
living in it, we can expect it to selectively invest in different types of infrastructures that impact registration. In addition, as we see in the next section, a state can also selectively facilitate or complicate the procedure to get a birth certificate and make registration more or less useful to certain groups in the population. These obstacles to registration are state-imposed barriers that operate at the individual level.
Selective State Strategies for Registration.
In the case of registration, state officials can decide to (1) include or (2) exclude specific regions of the CRVS system. Inclusion and exclusion are two sides of the same phenomenon, as by selectively including certain regions or individuals in the registration system, the state automatically excludes others. However, the extent to which the state purposefully excludes some regions differ. For instance, Bowles (2018) show that during the 1966 reform of the CRVS system in Tanzania, which made registration compulsory, there was a sequencing process: the state first selected wealthy districts for the reform, before making it universal (p.23). This exemplifies strategic inclusion and exclusion, as governments benefit most from registering individuals in wealthy districts for taxation purposes. Moreover, to further limit registration, the government set up financial barriers to obtain identity documents (e.g. a relatively high price for getting a birth certificate) and required proof of identification “for activities that differentially benefit the rich” such as receiving higher education or accessing the formal financial sector (Bowles, 2018, p.8). By combining high barriers with selective gatekeeping, that is which services require legal identity, the Tanzanian state implemented a strategy of “repressive gatekeeping” (Bowles, 2018, p.3-7). Tanzania and other African countries implemented the repressive gatekeeping strategy to prevent an adverse selection problem, where the poor register to benefit from state services without participating to its fiscal capacity, and the rich do not register to avoid taxation (Bowles, 2018, p.8; Ferguson, 1999). Hence, for Bowles (2018) an “exclusionary political logic lies at the heart of [the] failure to count the world’s poor” (p.2). The example of Tanzania also highlights how states’ strategies can be situated at two levels: at the level of the region, they translate into sequencing or different investments in infrastructures, while at the level of the individuals, they are embodied by the presence of administrative barriers to registration that individuals encounter.
Strategies of inclusion and exclusion are central to explaining under-registration, as “rapid progress is made when governments reach an administrative barrier to achieving some other objective, be it one oriented toward inclusion or control” (Hunter & Brill, 2016, p.3). While Hunter and Brill (2016) make a distinction between inclusion and control, I argue that it is first about inclusion or exclusion, and then there can be two distinct motivations for strategic inclusion in the CRVS system. Either the state wants to (1) care for its population, or (2) repress its population.2
2 For Foucault, Faubion and Hurley (1998), caring and repressive strategies have the same end goal: control of the
population. African states have incentives to care for the rich, so they accept to register and be taxed. While this is a form of control, building caring infrastructures means building a specific kind of state presence that does not have the same effect on the individuals as a repressive state presence does. They also do not have the same policy-implications. For these reasons, I argue that it is essential to differentiate between caring and repressive strategies.
Caring and Repressive Subnational Strategies of Inclusion
A state can decide to act in a caring manner towards parts of its population and push for registration by building or improving public services. In the Global Health literature, this caring strategy echoes with scholars’ policy recommendations to improve registration by upgrading the services available (AbouZahr et al., 2007; Mahapatra et al., 2007). For instance, in South Africa after the end of apartheid, part of the success of the CRVS system is attributed to governmental investments in caring infrastructures, like hospitals and registration centers (see: Plagerson et al., 2012; Schreiber, 2014; Gibbs et al., 2018). In turn, satisfaction with services increases trust in the government, which can also boost registration (AbouZhar et al., 2007; Bratton, 2007; Lavallee et al., 2008). Moreover, if these caring infrastructures (hospitals, schools, government offices) require legal identity, individuals will register to access them—following the state’s strategy of selective gatekeeping. Due to these mechanisms, the presence of caring infrastructures is positively associated with registration.
However, in the case of African countries, there is a chance that states also imposed barriers to registration. As Bowles (2018) showed, the Tanzanian state was most interested in registering wealthy individuals to be able to tax them, which led to a bias in the selection of districts for compulsory registration. In exchange for registration, wealthy individuals gained privileged access to high-end public goods, such as passports, universities, and the best hospitals in the country. Poor and marginalized individuals were excluded from the registration system because of the costly administrative barriers put up by the state. Seeing the state as a repressive gatekeeper could help us understand why, even after infrastructural improvements, registration levels remain low. This would indicate that there is more exclusion than inclusion, and that the state maintains high barriers to registration that prevent people from getting a birth certificate.
Another possible scenario is when the central government faces rebellion in part of its territory, and to control the local population, heavily registers them—this follows Scott’s (1998) argument on legibility and Foucault et al. (1998) argument about state power and registration, but in this case the state selectively picks which groups it needs to control. From a comparative politics perspective, in these regions, we expect to have higher levels of coercive state capacity due to the state’s incentive to repress (parts of) the local population. Coercive capacity was defined by Mann (1984) as the degree of autonomy that political leaders enjoy in regards to their civil society and is measured by the size of coercive institutions, that is the number and territorial spread of police forces, army, or prisons (Fortin-Rittberger, 2014, p.1251). If the strategy of ‘enforced registration’ is successful, areas with numerous repressive infrastructures such as prisons, or police stations, should have higher levels of registration.
The following diagram visualizes my theoretical argument. I argue that instead of a universal motivation for registration, the state has a selective incentive that translates into strategies of inclusion or exclusion. These strategies can only be seen if we look subnationally, as they vary depending on the regions and population groups. Inclusive strategies, be it for caring or repressive reasons, will have a positive effect on registration, while exclusion strategies will negatively affect registration.
Figure 1. Path Diagram: The Effect of State’s Selective Strategies on Birth Registration
4 Case Study: State Presence and State Strategies in Senegal.
The issue of under-registration of births in Senegal is salient, as in 2016 UNICEF estimated that only 62% of Senegalese children had been registered with their certificates as proof (Bhatia et al., 2017, p.5). To explain why registration levels have remained low, I argue that the state is (1) maintaining barriers to exclude the poor and marginalized people and include wealthy individuals, and (2) has caring and repressive strategies, reflected by the type of infrastructures, that have different impacts on registration.
Inclusive and Exclusive Strategies: Administrative Barriers to Registration in Senegal
In Senegal the system of birth registration designed by the government shows some costly barriers for the most vulnerable part of its population. Currently, there are two types of registration centers: (1) principal centers, which are at the municipality (commune) level, and (2) secondary centers, which are in the rural municipalities (commune rurale). While timely registration is free, the issuance of a paper certificate costs between 200 and 500 CFA in urban municipalities and between 75 and 150 CFA in rural municipalities (Diop, 2012, p.29). The parents must also present a certificate of childbirth from a doctor or a midwife, or two witnesses of the childbirth for the registration of the child. For mothers giving birth outside of a hospital, the necessity of a certificate delivered by a health professional is a barrier. Hence, in the case of Senegal, factors such as wealth or birth in a hospital are not purely individual determinants but instead politicized by the state’s gatekeeping policies. In addition, if parents want to register their child after a year, they must get the approval of the district justice court. The civil registration in Senegal is troubled by other issues, such as a lack of coordination between the different institutional actors responsible for administrating the CRVS system, non-accessibility to certain registration centers in rural areas, lack of registers and papers, poor conservation of registers, lack of alphabetical registers, functional offices or qualified personnel (UNSTATS, 2017).
Wealth and birth at a hospital are the focus of my first analysis evaluating their impact on registration. Both factors are politicized by state-imposed barriers, so I expect rich individuals and the individuals who can give birth in a hospital to be included in the registration system, and poor individuals and the ones who cannot give birth in a hospital to be excluded from it.
Unit Incentive Strategy Implementation Effect
State Selective Inclusion Exclusion Care Repression BR BR
+
-
+
BR Hospitals, schools, offices, etc. Gatekeeping. Prisons, police stations, etc. No infrastructures. Administrative barriers.Inclusive Strategies: Caring and Repressive Infrastructures in Senegal.
Senegal has been depicted as the archetypical centralized state, with rulers in Dakar obsessed with top-down control and territorial uniformity, a legacy inherited from the French colonial state (see: Wunch & Owulu, 1990). However, some other articles have shed light on very important subnational strategies of state development. Bernier (1976) studied the territorial formation and colonial governance in French Senegal and illuminated two distinct strategies: assimilation and association. In the communes of Saint Louis, Goree, Rufisque and Dakar, the cities where the French state first implemented itself, a strategy of “association” was applied where the Senegalese inhabitants of these communes were made French citizens (Bernier, 1976, p.464). Civil registration centers were created in these four communes between 1730 and 1875. The rest of the population was under the assimilation policy and strict colonial rule, with neither rights nor legal identity. Hence, during colonial rule, the registration of local population already varied subnationally with a bias favoring the four communes. As for state presence (i.e. built state), Boone (2003) argued that through colonial and post-colonial times, the central state has implemented three different strategies in its territory (outside of the four communes): (1) a strategy of power-sharing with strong rural notables in the Wolof groundnut basin and in the River Valley, (2) a strategy of administrative occupation in the rebellious Casamance, and (3) a strategy of administrative neglect in Central and East Senegal (p.39; see: Figure 1 Appendix 1).3
Due to these biases, we can expect regions where the state had greater economic interests and where notables could bargain with the central government, such as the Groundnut Basin and River Valley, to have better caring infrastructures and higher levels of registration. On the other hand, the region in the South of Senegal below The Gambia, the Casamance, has had a history of resistance to the central government. During French colonialism, the Casamance was put under strict military rule and to control the local population, the government resorted to census, classification, conscription and administration of tribes (Dieng & Ehemba, 2013, p.348). A strategy of ‘divide and rule’ was implemented by the colonial state and then taken over by the post-colonial government. The current administrative boundaries in the Casamance align with the “colonial mapping of instability”: for instance in 1984 the Casamance was cut in two regions, a strategy to control the resistance movement that was born around that time (Dieng & Ehemba, 2013, p.354-356). Since 1982, a conflict opposing the independentist Mouvements des Forces Democratiques de la Casamance (MFDC) to the central government has been taking place (De Jong & Gasser, 2005, p.217). Casamance people suffer from marginalization in employment, poor investments in infrastructures, “internal colonialism” where administrative positions are filled with people from Northern Senegal, and are victims of land reform and land grabs by the government (De Jong & Gasser, 2005, p.175). Hence, in this region we can expect low levels of caring infrastructures and high levels of repressive infrastructures, and if the state’s
3 The Wolof Groundnut Basin is roughly located in the regions of Thies, Diourbel, Fatick, Kaolack and Kaffrine.
The North River Valley is in the region of Saint Louis. The Casamance comprises the three regions of Zinguichor, Sedhiou, and Kolda. For a map with all districts and regions, see: Figure 2, Appendix 1.
strategy of control is successful, relatively high levels of registration even though the region is poorly developed. Finally, in regions where the government had no inclusion interest, there should be few infrastructures and low levels of registration.
The two maps below show some preliminary support for my expectations. The left map visualizes the number of caring infrastructures for every 100,000 individuals and the percentage of individuals who have a birth certificate in each of the 45 districts of Senegal, according to the 2013 National Census. The right map visualizes the number of repressive infrastructures for every 100,000 individuals in each district, and the percentage of individuals who have a birth certificate or not.
Figure 2. Number of Caring (left) and Repressive (right) Infrastructures by 100,000 individuals and Percentage of Population with Birth Certificates in Senegalese Districts
Source: National Census of Senegal 2013, National Geographic Agency HGIS.
Note: There are 45 Districts in Senegal, they are the administrative unit below the Region. For a map with all districts and regions, see Figure 2 Appendix 1. Legend Classification: Natural Breaks (Jenks). Caring infrastructures include numbers of schools, government offices (not registration offices), hospitals, fire stations and post offices, divided by total population of district and multiplied by 100,000. Repressive infrastructures include absolute numbers of police stations, prisons and military camps, divided by total population of district and multiplied by 100,000. The absolute number of (1) caring and (2) repressive infrastructures are used for my independent variables. For full descriptive statistics, see Table 2 Appendix 2.
These maps show that there is an uneven spatial spread of caring and repressive infrastructures within the territory, even when we account for the population of the district. As predicted by Boone (2003), there is a concentration of caring infrastructures around Dakar and in the North River Valley, and the rest of the coast and the Southeast have relatively high numbers of caring infrastructures. These regions also have high registration rates. The center of Senegal, which has been neglected by the central government, has few to no infrastructures, and the lowest registration rates of the country. The Southern region of Casamance has also been neglected in terms of caring infrastructures, but the map on the right shows that the South in general has numerous repressive infrastructures. Central Senegal is all together neglected with no repressive infrastructures either, while Western Senegal has numerous repressive infrastructures. Taken together, the Western and Northern regions have the highest numbers of infrastructures and show comparatively high levels of registration. This supports my argument that the
state has invested differently in its regions, and this seems to correlate with registration. The next section explains my data sources, the operationalization of my variables, and my methodology.
5 Operationalization, Sample and Method
In order to test my arguments that (1) state-imposed barriers include or exclude certain parts of the population, and (2) state strategies of care or repression impact registration rates, I use one dependent variable and five independent variables, in two different analyses. In the first analysis, I evaluate the impact of assistance during childbirth by a medical professional and wealth on whether a child below the age of five is registered and has a certificate, using logistic multivariate regression. In the second analysis, I evaluate the impact of caring and repressive infrastructures at the district level on whether a child has been registered and has a certificate, using linear random effects multilevel modelling.
Outcome Variable: Birth Registration and Birth Certificate
For my individual level variables, I use the Demographics and Health Survey (DHS) conducted by USAID between 2015 and 2016 in Senegal. In the DHS 2015-2016, interviewers asked in each household with a child below the age of five whether the child was registered and had a birth certificate as proof. Thus, the variable differentiates between absolute and relative lack of identity (Harbitz & Tamargo, 2009, p.9). My dependent variable has 3 categories: (1) not registered, (2) registered without a certificate, and (3) registered with a certificate. The following histogram visualizes the percentage of respondents in each category in the overall sample. Most children below the age of five are registered without a certificate (42% of the sample), the second biggest category is no registration at all (38%) and finally 20% of the children in the sample are registered with their certificates as proof. This distribution points to the fact that having to pay for a paper certificate is an effective barrier, and that most people only do what is free, that is register without a certificate.
Figure 3. Histogram, Dependent Variable “Registration & Certificate”, Overall Sample
Source: DHS 2015-2016, own calculations. Note: N = 11,650. 0 1 0 2 0 3 0 4 0 P er ce nt T ot al S am pl e NO BR BR NO BC BR & BC
Independent Variables Analysis I: Assistance during Childbirth and Wealth Index
For my first analysis, which investigates the effect of state-created barriers, my two independent variables measure (1) whether the mother was helped during childbirth by a medical professional or not, and (2) the wealth of the family. The presence of a medical professional during childbirth helps parents acquire a certificate of childbirth, which is a prerequisite for birth registration. The first independent variable is a two-category variable of (0) no official medical help (includes traditional and trained birth attendant, another person and no one), and (1) official medical help (includes doctor, nurse and midwife).4 Second, since acquiring a certificate costs between 75 and 200 CFA, the wealth of the
family is a politicized factor influencing registration. I use the DHS variable Wealth Index with its five categories (1) poorest, (2) poorer, (3) middle, (4) richer and (5) richest.
The following figure visualizes the mean score on the dependent variable (registration), by wealth group and whether the mother received medical assistance during childbirth. This figure hints at a strong effect of both medical assistance and wealth on registration: poor individuals have a lower registration mean than middle wealth and rich individuals, and within each wealth category, individuals who did not give birth with official medical help have lower registration means. The effect of medical assistance during childbirth is most pronounced for poor individuals, so individuals in the lowest wealth index category are most sensitive to the administrative barrier of childbirth certificate. This is consistent with the theoretical argument that states are most interested in registering the rich and set up barriers that hurt the individuals who need registration the most to protect themselves and their children. Figure 4. Bar Graph with Confidence Intervals, Dependent Variable “Registration & Certificate”, by Medical Assistance during Childbirth and Wealth Index Group
Source: DHS 2015-2016, own calculations.
Note: Wealth Index recoded for visualization purposes: poorest/poorer and richer/richest categories combined; N = 11,650.
4 The DHS has 7 distinct variables that ask whether the mother received birth assistance by (1) a doctor, (2) a
midwife, (3) a nurse, (4) a traditional birth attendant, (5) a trained birth attendant, (6) another person, and (7) no one. I recoded the variable into a two-category one.
.4 .6 .8 1 1. 2 M e an R e gi st ra tio n
Poor Middle Rich
Wealth Index
No Medical Professional Medical Professional Confidence Interval
Independent Variables Analysis II: Density of Infrastructures
For the second part of my analysis investigating the effect of subnational state presence on registration, I use three independent variables (1) overall density of infrastructures, (2) density of caring infrastructures, and (3) density of repressive infrastructures. These variables operationalize Mann’s (1964) concept of infrastructural power, using Soifer’s (2008) approach of infrastructural power as a state’s reach, that is both the range (i.e. number and type) and subnational spread of public institutions in a territory (p.236). Infrastructural power is also very similar to Boone’s (2012) concept of built state, which according to Steinberg (2018) can be measured by the presence of “outposts for police forces and of the physical offices of a bureaucracy, public clinics and schools, and utilities, in addition to the development of roads” (p.225). In other words, these variables measure the extent of a state presence (range and reach), and the kind of state presence (caring or repressive).
The variables are at the district level. I used four ArcGIS layers of data on infrastructures collected by the National Geospatial Agency and projected them onto a map of Senegalese districts. For the caring infrastructures, I included: schools, government offices (not registration offices), hospitals, fire stations and post offices. In my sample, the number of caring infrastructures by district ranges from 0 to 62. For the repressive infrastructures, I included: prisons, police stations, and military installments. The number of repressive infrastructures by district ranges from 0 to 8. Then I divided the added count of overall, caring and repressive infrastructures in each district by its area and multiplied it by 100. My three independent variables are therefore measuring density of infrastructures by the 100km2. Finally,
to ease the interpretation, I standardized all three variables.
Control Variables
In addition to my dependent and independent variables, I include various controls following the previous findings in the literature. At the individual level, I control for: child’s sex, as I suspect that females will be less likely to be registered, even if this is debated in the literature (AbouZahr et al., 2015; Harbitz & Tarmargo, 2009); child’s age, as an older child is more likely to be registered; birth index (how many siblings before the child) as usually the first child is more likely to be registered (Pelowski et al., 2015); mother’s education and mother’s age when she first gave birth were found to be strongly associated with registration (Garenne et al., 2016; Pelowski et al., 2015); rurality and travel time to the closest urban centers were both found to be negatively associated with registration (Harbitz & Tamargo, 2009). In my main models, I do not control for ethnicity or religion. This is because there is a strong social integration between ethnic groups thanks to the cross-cutting effects of Muslim brotherhoods, a phenomenon commonly called cousinage in Senegal (Hartmann, 2010, p.775). I will only include them in a robustness check model. At the district level, I collected data from the 2013 Census so I have District Population, as the higher the population the more likely the state will have invested in some infrastructures, and District Education, which captures the level of development of the district with the percentage of people who were enrolled in school (regardless of the grade achieved)
in the district. Districts with higher level of development are likely to have more registration. Details of the operationalization of all my variables can be found in Table 1 and Table 2, Appendix 2.
Sources and Sample
I use three different sources of quantitative data. For my individual level variables, I use the Demographics and Health Survey (DHS) conducted by USAID between 2015 and 2016 in Senegal. The DHS has been used several times in the political science literature (see for instance: Kramon & Posner, 2013; Henn, 2018) and presents strong advantages. It includes a variety of questions on health and demographic characteristics of the respondents (children and parents), and characteristics of the household. The data is adequately sampled and is representative at the regional level. Respondents are nested in small clusters (i.e. villages or neighborhoods), which have geographic coordinates available upon request (Perez-Heydrich et al., 2013). The geographic locations of the clusters are randomly displaced, however the displacement is contained within the second subnational level, which is the district level and my level of analysis.5 This means that district level variables are appropriate context
variables, although the sample is unfortunately not representative at this level.
For the context-level variables, I first use the 2013 National Census of Senegal conducted by the National Agency of Statistics (NAS) of Senegal. The NAS made available the regional census reports, in which they detailed some variables by district, including population and education enrollment. Second, I use data collected by the American National Geographic Agency (NGA) for the Digital Globe's Human Geography Information Survey (HGIS) and made available on ArcGIS Online. In the format of ArcGIS layers, the NGA geographically coded the position of various infrastructures in Senegal, from 2011 to 2016. I used these layers to create variables operationalizing the number and spread of infrastructures in Senegalese territory. This data was made available recently and, to the best of my knowledge, has not yet been used for academic research.
As mentioned previously, there are four communes in Senegal (Saint Louis, Goree, Rufisque and Dakar) where the colonial state implemented a CRVS system to give the French nationality to the local inhabitants (Bernier, 1976, p.464). Due to this historical bias, I decided to exclude the districts with these cities, as they represent historical outliers and could bias my results. I thus leave out three districts (Dakar, Rufisque, Saint Louis) which represents 637 respondents in the sample. I also excluded the respondents who were not Senegalese, which represents 404 respondents.
Method I: Multivariate Logistic Analysis
To evaluate the effects of state-imposed barriers, I conduct a multivariate logistic analysis at the individual level. As my outcome variable is categorical, I use logistic regression rather than linear. In
5 Random displacement between 0 and 2 km for urban clusters and between 0 and 5 km for rural clusters, with
1% of the sample displaced by 10km (Perez-Heydrich et al., 2013). For this reason, the clusters cannot be used to calculate distance between the respondents and a feature.
addition, as the individuals in the sample are nested into districts, the error terms are not independent from each other, so I clustered the standard errors by district. The aim is to evaluate the percentage of the variance in the dependent variable explained by my independent variables, and then add individual characteristics of the child, mother and household, to net out the effects of my independent variables.
Method II: Random Effects Multilevel Modelling
To evaluate the effects of state presence and strategies on registration, I conduct a multilevel analysis. My level 1 observations are individuals from the DHS, and my level 2 is the district (arrondissement). As mentioned before, once I exclude the four big communes, I have 42 districts, which is enough to perform a multilevel analysis. I use this model because individuals are nested within their districts. Nested data violate several key assumptions of OLS: (1) independent observations, (2) independent error terms, and (3) homoscedastic errors, normal distribution of errors (Rabe-Hesketh & Skrondal, 2012). Thanks to multilevel modelling, dependencies within clusters become the subject of study and the correlation within clusters is not only expected but explicitly modelled. In addition, I use random effects and not fixed effects since my independent variables are invariant across individuals in their districts. Figure 3 below visualizes the existence of both between and within cluster (districts) variation, which support my choice of multilevel modelling.
Figure 5. Histograms, Dependent Variable “Registration & Certificate”, by District
Source: DHS 2015-2016, own calculations. Note: N = 11,650; n = 42. 0 2 0 4 0 6 0 8 0 0 2 0 4 0 6 0 8 0 0 2 0 4 0 6 0 8 0 0 2 0 4 0 6 0 8 0 0 2 0 4 0 6 0 8 0 0 2 0 4 0 6 0 8 0 NO BR BR NO NC BC NO BR BR NO NC BC NO BR BR NO NC BC NO BR BR NO NC BC NO BR BR NO NC BC NO BR BR NO NC BC NO BR BR NO NC BC
Bakel Bambey Bignona Birkelane Bounkiling Dagana Diourbel
Fatick Foundioungne Gossas Goudiry Goudoump Guediawaye Guinguineo
Kaffrine Kanel Kaolack Kebemer Kedougou Kolda Koungheul
Koupentoum Linguere Louga M'Bour Malem Hoddar Matam Mbacke
Medina Yoro Foulah Nioro du Rip Oussouye Pikine Podor Ranerou-Ferlo Salemata
Saraya Sedhiou Tambacounda Thies Tivaouane Velingara Ziguinchor
P
e
rc
e
6 Results
Analysis I: State Sponsored Barriers and Birth Registration.
I run four different models. Model 1 and Model 2 include each independent variable separately. Model 3 includes both independent variables, and Model 4 includes all variables, independent and controls. The collinearity scores of the variables can be found in Table 1 (Appendix 3) and show no signs of multicollinearity. Table 1 below shows the results.
Table 1. Results Multivariate Logistic Regression; Dependent Variable: Registration & Certificate Model (1) Model (2) Model (3) Model (4) Odds Ratio Odds Ratio Odds Ratio Odds Ratio Assistance Childbirth (REF: NO
MEDICAL PROF.)
Medical Professional 3.354*** (.332)
2.624*** (.240) 2.394*** (.213) Wealth Index (REF: POOREST)
Poorer 2.050*** (.187) 1.851*** (.170) 1.596*** (.147) Middle 5.027*** (.881) 4.268*** (.737) 2.812*** (.503) Richer 5.961*** (1.431) 4.923*** (1.141) 2.516*** (.587) Richest 14.78*** (3.397) 11.45*** (2.345) 4.897*** (1.055) Sex (REF: MALE)
Female .894* (.041)
Age Child (Year) 1.043* (.020)
Birth Index .884* (.053)
Mother Education (REF: NO ED.)
Incomplete Primary 1.636*** (.099)
Complete Primary 1.174 (.181)
Incomplete Secondary 1.926*** (.208)
Complete Sec. & Higher 8.246** (5.950)
Mother’s Age 1st Child 1.042*** (.010)
Travel Urban Center (min) .997*** (.001)
Rurality (REF: URBAN CLUSTER)
Rural Cluster .804 (.134)
R2 .054 .095 .125 .149
Log lik. -7262.2 -6947.6 -6717.5 -6533.9
N 11650 11650 11650 11650
Source: DHS 2015-2016, own calculations
Note: Exponentiated coefficients; Standard errors in parentheses; * p < .05, ** p < .01, *** p < .001; Standard errors clustered by District (n = 42). Likelihood ratio test impossible with clustered standard errors.
In Model 1, the main independent variable Assistance during Childbirth explains by itself 5.4% of the variance in the dependent variable. The odds ratio coefficient is strong, significant, and in the
expected direction. In Model 2, the independent variable Wealth Index explains 9.5% of the variance in the outcome variable. The odds ratio coefficient is strong, significant, and in the expected direction. In Model 3, the coefficients are significant and in the same direction as in Model 1 and 2, although they are slightly weaker. Together, the two independent variables explain 12.5% of the variance in the dependent variable. This indicates that we can move on to the full model with controls, in which the interpretation of coefficients is most precise.
In Model 4, both independent variables still have strong and significant coefficients that are in the expected direction according to my theory. Children whose mothers have given birth with professional medical help have their proportional odds of being in the category “registered without a certificate” or “registered with a certificate” increase by 139.4%, compared to children whose mothers did not have professional medical help during childbirth, holding all other variables constant. Children with a wealth index “poorer” have their proportional odds of being in the category “registered without a certificate” or higher increase by 59,6%, compared to children with a wealth index “poorest”. In comparison, children with the wealth index “richest” have their proportional odds of being in the category “registered without a certificate” or “registered with a certificate” increase by 389,7%, compared to children with a wealth index “poorest”, holding all other variables constant. Therefore, the exclusion effect of the administrative barrier grows stronger as we move down the wealth index categories. These results support my argument that in Senegal, state-imposed barriers work effectively to keep certain groups from registering while encouraging others to do so. There are mechanisms of inclusion of the richer individuals and the ones who have access to professional medical help, while the individuals who do not have such access and are poorer, are excluded from civil registration and have more difficulties obtaining birth certificates. Therefore, factors like birth in a hospital and wealth are politicized by state gatekeeping policies and cannot be treated solely as demographic characteristics.
It is also interesting to notice several things about the control variables. First, the sex of the child is significantly associated with registration: female children have their proportional odds of being in the category “registered without a certificate” or “registered with a certificate” decrease by 10,6%, compared to males, holding all other variables constant. This runs against some findings in the Global Health literature: cross-country comparisons have shown no relation between the sex of the child and registration (see: Bhatia et al., 2017; Garenne et al., 2016), but it seems that for Senegal, there is a negative effect of sex. Second, birth index is negatively associated with registration, which means that the oldest child gets priority for registration. From an individual perspective, this makes sense, especially in terms of inheritance. Regarding the mother’s characteristics, the more educated she is, the more likely the child will be registered. Interestingly, the older the mother is for her first child, the more likely the child will be registered. Note that in my sample, some mothers gave birth to their first child at the age of 12. This correlates with Garenne et al. (2016) finding that, in the Agincourt community in South Africa, adolescent mothers were systematically excluded of the registration system, due to stigma
and fear of retribution. Finally, the longer the travel time to an urban center, the less likely a child is to be registered, while rurality is not significantly associated with registration but in the expected direction.
Analysis II: Caring Infrastructures, Repressive Infrastructures and Registration.
For the second part of my analysis, I first run 5 basic models. Model 1 is the empty model and indicates the intraclass correlation, which is the percentage of the variance of my dependent variable that resides at the cluster level. Model 2 is a simple bivariate regression between All Infrastructures and Registration & Certificate to evaluate the general impact of infrastructures on registration. Model 3 is a bivariate regression between Caring Infrastructures and Registration & Certificate to evaluate the impact of caring infrastructures on registration. Model 4 does the same but with Repressive Infrastructures, and finally Model 5 combines both Repressive Infrastructures and Caring Infrastructures to evaluate their impact together on birth certificate acquisition. However, a multicollinearity test between my variables indicate that my two independent variables are collinear (see: Table 2, Appendix 3). The following table sums up the results.
Table 2. Results RE-Multilevel Regression; Dependent Variable: Registration & Certificate
Model (1) Model (2) Model (3) Model (4) Model (5) All Infrastructures (density 100km2) .229***
(.040)
Caring Infrastructures (density 100km2) .221*** (.038)
-.171 (.122) Repressive Infrastructures (density
100km2) .338*** (.038) .558** (.191) Intercept .859*** (.033) .894*** (.029) .893*** (.029) .911*** (.027) .919*** (.028) Between-cluster Variance (𝜎𝑢2) .041 (.007) .036 (.006) .036 (.006) .034 (.006) .034 (.006) Within-cluster Variance (𝜎𝑒2) .522 (.015) .522 (0.15) .522 (.015) .522 (0.15) .522 (.015) ICC .073 .064 .064 .061 .061
Source: NGA-HGIS 2011-216, own calculations
Note: Standard errors in parentheses; Robust standard errors; * p < .05, ** p < .01, *** p < .001; N = 11650; n = 42. According to Model 1, 7.3% of the variance of the outcome variable resides at the district level, which supports the use of multilevel modelling. Overall, all three independent variables have a positive significant impact on registration and birth certificate acquisition. When combined into Model 5, the coefficient for caring infrastructures is negative and insignificant, due to multicollinearity. Therefore, in the next models I separate the independent variables. Model 6 includes Caring Infrastructures with the individual and district level controls. Model 7 adds the two independent variables of Analysis I to Model 6. Model 8 includes Repressive Infrastructures with individual and district level controls. Model 9 adds the two independent variables of Analysis I to Model 8.
Table 3. Results RE-Multilevel OLS Regression; Dependent Variable: Registration & Certificate Model (6) Model (7) Model (8) Model (9) Caring Infrastructures (density
100km2)
.101*** (.024) .078** (.026) Repressive Infrastructures
(density 100km2)
.159*** (.038) .129** (.041) Wealth Index (REF: POOREST)
Poorer .142*** (.033) .142*** (.033)
Middle .244*** (.053) .244*** (.053)
Richer .247*** (.053) .247*** (.053)
Richest .259*** (.064) .258*** (.064)
Assistance Childbirth (REF: NO HEALTH PROFESSIONAL)
Health Professional .186*** (.026) .186*** (.026)
Sex Child (REF: MALE)
Female -.040** (.015) -.038** (.014) -.040** (.014) -.038** (.014) Age Child (Year) -.000 (.005) -.001 (.005) -.000 (.005) -.001 (.005) Education Mother .049*** (.009) .027** (.008) . 049*** (.009) .027** (.008) Age Mother 1st Birth .012*** (.003) .008** (.003) .012*** (.003) .008** (.003)
Rurality (REF: URBAN CLUSTER)
Rural Cluster -.114* (.055) -.011 (.050) -.113* (.055) -.010 (.050) Travel Time Urban Center (min) -.002*** (.000) -.001*** (.000) -.002*** (.000) -.001*** (.000)
District Population (Logged) -.055 (.039) -.059 (.042) -.060 (.039) -.063 (.041) District Education -.001 (.002) -.001 (.002) -.001 (.002) -.001 (.002) Intercept 1.574** (.481) 1.397** (.507) 1.662*** (.486) 1.472** (.510) Between-cluster Variance (𝜎𝑢2) .023 (.005) .020 (.004) .023 (.005) .019 (.004) Within-cluster Variance (𝜎𝑒2) .501 (.014) .487 (.014) .502 (.014) .487 (.014) ICC .044 .039 .044 .038
Source: DHS 2015-2016, NGA-HGIS, NAS 2013, own calculations
Note: Standard errors in parentheses; Robust standard errors; * p < .05, ** p < .01, *** p < .001; N = 11650; n = 42. In Model 6, the independent variable Caring Infrastructures is in the expected direction of a positive relationship and significantly associated with registration. When Density of Caring Infrastructures increases by one standard deviation (0.17), it leads to an increase of .101 in the dependent variable Registration & Certificate, which ranges from 0 to 2. In Model 7 the coefficient is still significant and in the expected direct, however it is slightly smaller than in Model 6. When the variable Density of Caring Infrastructures increases by one standard deviation (0.17), it leads to an increase of .078 in the dependent variable Registration & Certificate. This coefficient supports my theoretical expectation that improvements in caring infrastructures have a positive impact on registration, as they signal an inclusive and caring strategy on the part of the state. From the individual’s perspective, having more caring infrastructures can increase satisfaction with services, trust in the state, and expectations that registration will be useful, most notably to access these services. However, the
fact that there is a reduction of the coefficient of Caring Infrastructures between Model 6 and Model 7 indicates that once we include the state-imposed barriers to the model, the effect of infrastructural improvements is reduced. This in turn means that state-created obstacles work effectively to keep people away from registration, even with improved infrastructures. The individual-level coefficients are similar to the first analysis, except for the age of the child, which is no longer significant, and neither District Population nor District Education are significantly associated with registration. The percentage of the variance at the district level (ICC) was reduced by 3.4 percentage points between the empty model and Model 7, so Model 7 explains approximately half of the variance at the district level.
The results are very similar for Model 8 and Model 9. The independent variable Repressive Infrastructures is significant in Model 8 and in the direction predicted by my theoretical expectations. When Density of Repressive Infrastructures increases by one standard deviation (0.15), it leads to an increase of .159 in the dependent variable. In Model 9, the independent variable is still significant, and the coefficient is stronger than for the Caring Infrastructures variable. When the variable Repressive Infrastructures increases by one standard deviation (0.15), it leads to an increase of .129 in the outcome variable. This coefficient supports my theoretical expectation that in some regions, the state has an interest to control parts of the population for repression purposes. This interest is reflected in the presence of repressive infrastructures, and the strong positive association between repressive infrastructures and registration means that the strategy of ‘enforced registration’ is successful, and individuals in districts with high coercive capacity register more. At the state level, this indicates a mechanism of legibility according to which the state tries to register population in areas with security issues to better control them. Interestingly, there is also a reduction of the coefficient of Repressive Infrastructures between Model 8 and Model 9. This means that when we consider state-sponsored barriers, the effect of infrastructures is reduced because the barriers are stronger than the incentive to register. Finally, the ICC is reduced by 3.5 percentage points between Model 1 and Model 9, so Model 9 explains almost half of the district level variance, like Model 7.
7 Model Diagnostics and Robustness Checks
For each best model of my various analyses, I checked whether the data was (1) normally distributed, and (2) homoscedastic. The various resulting figures are in Appendix 4. For Model 4 (Analysis I) the data is non-linear and heteroskedastic, which biases my standard errors. The use of logistic regression exacerbates the issue of heteroskedasticity. For Model 7, Analysis II, the level 2 residuals are linear and normally distributed, while the level 1 residuals are not perfectly linear nor normally distributed. In addition, the data also shows some heteroskedasticity. The results are similar for Model 9, Analysis II.
To investigate how the relationships found hold in different settings, I perform 4 four distinct robustness checks: (1) Analysis I Model 4 with ethnicity and religion of the mother, (2) Analysis I with all districts, (3) Analysis II with a logistic model rather than a linear one, and (4) Analysis II with all
districts. All results can be found in the tables in Appendix 5. Analysis I is robust across both checks. If I add Ethnicity and Religion to the preferred model of Analysis I, I explain an additional 5 percentage point of variance in the dependent variable, however neither variables have significant coefficients. So, as predicted, neither religion nor ethnicity are significant predictors of registration in Senegal (see: Table 1, Appendix 5). If I run the 4 models with all districts (n = 45), the coefficients are similar to the analysis with 42 districts (see: Table 2, Appendix 5). This means that at the individual level, the historical effect of early registration does not impact the likelihood of a child being registered. Analysis II is on the other hand not confirmed in the robustness checks, but this is statistically sound. Using logistic regression, especially in multilevel random effects modelling, greatly exacerbates the problem of heteroskedasticity, which means that the standard errors are biased. This should explain why in Table Table 3 and 4 (Appendix 5) the coefficients for Caring Infrastructures and Repressive Infrastructures are no longer significant. The coefficients also lose significance and are almost null when we include all 45 districts (see Table 5 and 6, Appendix 5). This could indicate a bias effect of the three districts with the communes, who have higher levels of infrastructures and registration (for instance, Dakar has 277 caring infrastructures and 26 repressive infrastructures, and a registration rate of over 90 percent).
8 Discussion and Conclusion
This paper has shown that there are mechanisms explaining under-registration that are not situated at the individual level, but rather at the state level. Scholars need to carefully think about the motivations and actions of states when studying problems of CRVS in the developing world. The case study of Senegal has revealed that, through its state building process, the Senegalese state has different presences on its territory, so scholars need to consider the legacy of state building when studying birth registration. In Senegal, regions of interests to the state and with strong rural elites were privileged with investments in caring infrastructures, which resulted in higher levels of registration. The southern Casamance has been under strict control since colonial times, and strategic investments resulted in numerous repressive infrastructures and few caring ones. These infrastructures also positively affect registration levels, but for a matter of repression, and not welfare. Consequently, high registration rates are not always evidencing a state that is present to take care of its population. Scholars and professionals alike should keep in mind that registration is a historical tool of repression and control, and so can be an indicator that a coercive state is present in an area (Srzeter & Breckenridge, 2012, p.7). In addition, this confirms the need to differentiate between caring and repressive infrastructures and strategies, even if authors like Scott (1998) or Foucault et al. (1998) argued that they are all mechanisms of control.
In Senegal, state-sponsored barriers filter the individuals that the state wants to register; the necessity of a childbirth certificate for registration perpetuates absolute lack of identity, as marginalized groups cannot register, while the compulsory fees for a birth certificate keep poorer individuals in a state of relative lack of identity (Harbitz & Tamargo, 2009, p.9). For policymakers, this means that
improvements in caring infrastructures can have a positive impact on registration, so long as the administrative barriers set up by the state are low enough for individuals to be able to register. Altogether, these findings confirm what Hunter and Brill (2016) claimed, which is that “rapid progress is made when governments reach an administrative barrier to achieving some other objective”, whether it is care or repression (p.3).
This paper’s contribution is threefold. First, by theorizing the existence of subnational state presence in Africa and modelling it in Senegal, I filled a gap in the literature as most scholars only used individual level predictors in their studies. I successfully showed that to understand the issue of under-registration, we need to include the role of state strategies of gatekeeping, care and repression. Second, the findings of my analyses help explain why “even following targeted infrastructural […] the number of children registered still remains stubbornly low in many of these same developing regions” (Pelowski et al., 2015, p.882). The existence of gatekeeping mechanisms put up in place by governments mitigates the effect of improved infrastructures, because it is not in their interests to register all inhabitants alike. Finally, this study should also serve as a caution: high registration rates may be the result of coercive capacity and repressive policies, and therefore not always indicative of an efficient welfare state.
This study also has some limitations that need to be considered. First, I focused solely on the state, due to the present gap in the literature and the necessity to theorize and model the impact of state presence and strategies on registration. Consequently, this study is a story about structures in place rather than agency. However, individuals respond to these structures in various ways that have not been modelled here. Second, the quality of the data shows some problems, especially regarding the heteroskedasticity of the residuals. The standard errors in my analysis could be biased. Thirdly, the DHS sample is unfortunately not representative at the district level, so the generalization of my findings is limited. Finally, the data collected by the National Geospatial Agency needs to be cross-referenced with other existing datasets to ensure its quality.
Moving forward, scholars now need to investigate the reactions of individuals to various state presence and institutional structures. Academics can explore the impact of caring infrastructures on political trust, and its impact on registration, as well as the impact of repressive infrastructures on migration or voting intent, two other political behaviors that could have an impact on an individual’s willingness to register. For instance, from the Senegalese state’s perspective, the high rates of registration in the South are the result of a successful legibility campaign implemented by the government; from a political behavior perspective, these high rates of registration could be the result of the inhabitants’ desire to safely migrate or vote against the government. In conclusion, scholars need to stop studying registration solely as a public health issue and start seeing it as a political phenomenon that involves both state level decisions and individual level decisions.
