Menu
Menu
Menu
Menu
Introduction Birth registration remains limited in most low and middle-income countries. We investigated which characteristics of birth registration facilities might determine caregivers’ decisions to register children in Ethiopia. Methods We conducted a discrete choice experiment in randomly selected households in Addis Ababa and the Southern Nations, Nationalities, and People’s Region. We interviewed caregivers of children 0-5 years old. We asked participants to make eight choices between pairs of hypothetical registration facilities. These facilities were characterised by six attributes selected through a literature review and consultations with local stakeholders. Levels of these attributes were assigned at random using a fractional design. We analysed the choice data using mixed logit models that account for heterogeneity in preferences across respondents. We calculated respondents’ willingness to pay to access registration facilities with specific attributes. We analysed all data separately by place of residence (urban vs rural). Results Seven hundred and five respondents made 5614 choices. They exhibited preferences for registration facilities that charged lower fees for birth certificates, that required shorter waiting time to complete procedures and that were located closer to their residence. Respondents preferred registration facilities that were open on weekends, and where they could complete procedures in a single visit. In urban areas, respondents also favoured registration facilities that remained open for extended hours on weekdays, and where the presence of only one of the parents was required for registration. There was significant heterogeneity between respondents in the utility derived from several attributes of registration facilities. Willingness to pay for access to registration facilities with particular attributes was larger in urban than rural areas. Conclusion In these regions of Ethiopia, changes to the operating schedule of registration facilities and to application procedures might help improve registration rates. Discrete choice experiments can help orient initiatives aimed at improving birth registration.
This study is part of Performance Monitoring for Action (PMA), a multicountry project that collects survey data on key health indicators.24 We worked in Ethiopia, a country of more than 105 million inhabitants in East Africa (figure 1). Ethiopia is a low-income country: in 2018, its gross domestic product was US$772.3 per capita, according to World Bank estimates. Ethiopia has one of the lowest birth registration rates in Eastern and Southern Africa, with approximately 3% of children under age 5 registered in.15 By comparison, a third of under-5 children are registered in nearby Uganda,25 and more than two-thirds are registered in neighbouring Kenya.26 PMA has conducted nationally representative surveys in Ethiopia since 2013, with a focus on family planning, maternal/newborn health and water/sanitation.27 28 Map of the regions included in the birth registration study. SNNPR, Southern Nations, Nationalities, and People’s Region. Several months after the sixth round of PMA data collection (‘R6 survey’ thereafter), we conducted a follow-up study of birth registration (‘Birth registration study’ thereafter) in two regions: Addis Ababa and Southern Nations, Nationalities, and People’s Region (SNNPR; figure 1). Addis Ababa is an urban region with 3.2 million inhabitants, according to projections based on data from the 2007 census. SNNPR is one of the most populous regions, with 17.9 million inhabitants. It borders Kenya and South Sudan to the south and west, respectively. It is predominantly rural, but it also includes several large cities of >100 000 inhabitants. According to the most recent Demographic and Health Survey (2016), 24% of children under-5 had their birth registered in Addis Ababa, the highest registration rate in the country. In comparison, 3% of children under-5 in SNNPR had their birth registered, on par with the national average.15 In Ethiopia, the Vital Events Registration and National Identification proclamation of 2012 (revised in 2017) is the law that regulates the administrative process of birth registration. The federal Vital Events Registration Agency (VERA) was established in 2014 to oversee this process. The implementation of civil registration (including births) under the new law began nationwide in 2016.29 Health facilities, as well as health extension workers who routinely visit households, are expected to produce notification forms for births. These forms contain information about the child (name, date of birth). They do not replace the forms and certificates that must be obtained from the registration offices located in each kebele, that is, the lowest administrative unit in the country. There are more than 18 000 kebeles in Ethiopia, the large majority of which now offer birth registration services.30 Kebele offices are often accessible for most of the population they serve, particularly in urban areas. In some of the rural and mountainous parts of the SNNPR, however, households may be located several hours away from their kebele office.31 The level of staffing and equipment of kebele offices also varies between urban and rural areas: in Addis Ababa and other urban areas of the country, civil registration officers in a growing proportion of kebele offices use computers to register births or issue certificates, whereas paper forms remain the norm in virtually all rural areas. Birth registration is free in Ethiopia, but families might be charged a fee to obtain their child’s birth certificate, with the amount of the fee set by each administrative region. Birth registration is mandatory and should be accomplished within 90 days of birth. After that delay, penalties might be incurred but are rarely enforced in practice. The parents of a child must both be present at the registration office, and they must show their identification card to register the birth of their child. If one or both parents cannot be present to register a birth, additional procedures (eg, affidavits, sworn statements) are required to allow the available parent, or a guardian, to carry out birth registration. For the birth registration study that included the DCE, we selected a subset of households that had participated in the R6 survey. The R6 survey was conducted in June and July 2018. It used a two-stage cluster design, with urban-rural and administrative regions as strata. In the first stage, 44 EAs were selected in the SNNPR, and 22 EAs in Addis Ababa. Urban EAs were oversampled in the SNNPR. In the second stage, 35 households were selected at random within each EA. In total, 1617 households in the SNNPR, and 761 households in Addis Ababa, participated in the R6 survey. Households were eligible for the birth registration study if they had a child aged 0–5 years among their members. We revisited selected households between December 2018 and March 2019. At that time, we confirmed the presence of children 0–5 years old using a household roster. We determined which household member was the primary caregiver of each listed child, that is, the parent or legal guardian. We selected study participants among adult caregivers. If there was only one primary caregiver in the household, he/she was automatically selected. If there were multiple primary caregivers in a household, we selected one at random. Households in three rural EAs in SNNPR could not be included due to security reasons. The birth registration study consisted of a face-to-face interview with selected caregivers. In addition to the DCE module, the questionnaire ascertained a caregiver’s demographic characteristics, his/her knowledge of birth registration, the registration status of the children he/she cares for and exposure to messages stressing the need to register births. The DCE was designed to estimate the relative value that caregivers assign to attributes of birth registration facilities in considering whether and where to register their child(ren). We first conducted a review of the literature on the barriers to birth registration in LMICs. The protocol of this review is provided in online supplementary file 1. Based on review results, we identified several barriers that are characteristics of the registration process or the facilities that carry out this process. We determined which of these barriers were relevant to the Ethiopian context through a review of legislative documents (ie, the 2012 and 2017 proclamations), and consultation with VERA officials. bmjgh-2019-002209supp001.pdf This process yielded six DCE attributes (table 1): (1) the cost of obtaining a birth certificate, (2) the time to wait for service at the registration facility, (3) the number of visits required to register a birth and obtain the birth certificate, (4) the opening schedule of the registration facility, (5) the distance to the registration facility from the caregiver’s residence, and (6) whether the presence of one or both parents is required to complete the registration. For each attribute, we selected two to three levels that were either representative of the situation of birth registration in Addis Ababa and SNNPR or constituted desirable alternatives. We piloted the DCE design with stakeholders, potential data collectors and participants. Based on feedback, we refined the definition and levels of each attribute, and we developed training instructions. Attributes and levels of registration facilities used in the discrete choice experiment, Addis Ababa and SNNPR of Ethiopia 2018/2019 SNNPR, Southern Nations, Nationalities, and People’s Region. There were 324 potential combinations of the attributes and levels described in table 1. Respondents could not evaluate each of these combinations. Instead, we asked them to make eight choices between two randomly selected hypothetical registration facilities. The alternatives in each of the eight choice sets were formed using the DCREATE module in Stata, which creates efficient fractional designs.32 This approach allows assessing preferences for each level of the attributes. Our DCE was unlabelled,33 with alternatives presented to respondents under the headings of ‘facility A’ and ‘facility B’. In each of the choice sets, we also gave respondents the option to select neither facility. This ‘opt-out’ option helps increase the external validity of DCE data because respondents are not forced to choose between two (possibly unrealistic) alternatives.34 35 In addition, the DCE included two practice choice sets, during which interviewers demonstrated DCE procedures, verified respondents’ understanding of DCE procedures and addressed questions. We also added a choice set that contained a ‘dominant’ alternative, that is, one of the two hypothetical facilities was preferable to the other facility on all attributes.36 This choice set was inserted to evaluate the respondents’ comprehension of DCE procedures. Based on feedback obtained during the pilot, we randomly placed it in the sequence of choice sets to avoid instances where interviewers would select the dominant choice themselves to save time, instead of asking respondents to make the choice. Finally, we randomly varied between respondents the order in which attributes were listed in each choice set. This allowed assessing whether respondents made decisions based on the value of the attributes that were listed first. Prior studies have used similar checks to establish the reliability of DCE data.19 20 37 We determined the sample size of the birth registration study to estimate indicators of CRVS coverage with a desired level of precision. For analyses of DCE data, given large differences between urban and rural areas in (A) characteristics of respondents, and (B) accessibility and equipment of kebele offices, we sought to elicit caregivers’ preferences separately by place of residence. According to formulas of the statistical power of DCEs,38 sample sizes from the birth registration study in these sampling strata were sufficient to estimate the main effects of each facility attribute on respondents’ choices. We administered the questionnaire with Open Data Kit, a data collection platform frequently used in LMICs.39 We translated study instruments into Amharic. We trained data collectors for 5 days on study procedures. Interviewers first read a script explaining DCE procedures to respondents. Then, they stated the levels of the attributes of each hypothetical facility included in a choice set. They repeated these attributes if necessary, and encouraged respondents to take their time in making each DCE choice. We built automated quality checks into the DCE module. We flagged instances where a respondent opted out (ie, selecting neither facility), or selected the same facility (eg, facility A), in each choice set. Field supervisors were alerted to those occurrences and were asked to provide feedback to interviewers. In some cases, they revisited respondents for verification and corrections, if needed. We tabulated the characteristics of caregivers, by place of residence (urban vs rural). These included descriptions of their gender, age group, educational level, marital status and religion. We also included an assessment of their household wealth based on ownership of assets. This variable was constructed from R6 survey data, with methods similar to those used in Demographic and Health Surveys.40 It allowed classifying household in wealth quintiles. We also reported the proportion of caregivers who had ever heard messages (from any source) about the need to register births. We tested for differences in the distribution of these characteristics between urban and rural areas using χ2 tests. Our analyses of DCE data relied on the assumption that caregivers are rational actors, who make choices that maximise their individual utility.41 The utility U that a DCE respondent r derives from selecting alternative i in a choice set t was specified as: where Xi,t is a vector of variables describing the attributes of an alternative; βr is a vector of coefficients that represent the marginal utility that respondents derive from each level of these attributes (their ‘preferences’); and εr, i, t is an unobserved error term that is assumed to be independent of individual preferences and attribute levels.42 Given a respondent’s preferences, the probability of selecting alternative i among a set of J alternatives in a choice set is described by a logit model42 43: DCE data have often been analysed using conditional logit models,44 45 which assume that (A) there is no heterogeneity in preferences across respondents, and (B) there is no correlation among the multiple choices made by the same individual. In this paper, we relaxed these strong assumptions. We used mixed logit models, in which the parameter estimates can be written as the sum of their population average, b, and a term ηr that represents individual deviations from this average,42 so that: In our models, the Xi,t vector included all the attributes listed in table 1. We treated costs, distances and waiting times as continuous variables, expressed in birr, walking time and hours, respectively. Other attributes were treated as categorical variables and were dummy coded,43 that is, with a reference category taking value 0. We also included an opt-out constant, that is, a dummy variable taking value 1 if the alternative was not to select any of the two randomly selected facilities included in each choice set,34 35 and 0 otherwise. We used the mixlogit command in Stata46 to estimate mean coefficients (b) and their SDs, along with 95% CIs. We also tested the null hypothesis that all SDs were jointly equal to 0 (likelihood ratio test). Despite the stratified sampling scheme of the R6 survey, we report unweighted analyses of DCE data. We do so because our analyses were stratified by urban versus rural place of residence, which were the main domains for which estimates were sought in the R6 survey. In addition, the survey weights are not related to the dependent variable in our mixed logit models (ie, DCE choices). Unweighted estimates are thus unbiased and more efficient than weighted estimates.47 To further understand respondents’ preferences for various attributes of registration facilities, we conducted a willingness-to-pay (WTP) analysis. We divided the coefficient of each variable obtained using mixed logit models by minus one times the coefficient associated with registration costs.48 This allowed standardising the relative utility derived from registering a birth at a facility with a given level of an attribute against costs. All WTP estimates were computed in birr (ie, the local currency in Ethiopia), and translated into US$ using the exchange rate on 1 January 2019.
N/A
