Data

This study is based on a retrospective cross-sectional household survey, conducted in person in four main cities of the country. The data were collected in June–July 2016. The cities included in the survey were Addis Ababa – the country’s primate city, Bahir Dar, the largest city in the Amhara region, Hawassa, the largest city in the SNNP region, and Mekelle, the largest city in the country’s Tigray region. These cities were purposively selected as they accounted for about 20% of the estimated 4.3 million formal sector employees distributed across five categories – civil servants, public sector enterprise employees, private sector workers, NGO workers, pensioners (former civil servants and public sector enterprise workers).

Power calculations (power of 0.8 and significance of 5%) designed to detect the effect of health insurance on the utilization of inpatient care, outpatient care and out-of-pocket (OOP) health expenditures, yielded sample sizes of 3,000 to 4,200 individuals.¹ Based on budgetary considerations, a sample size of 2,100 households which was expected to yield more than 4,200 individuals was targeted. The distribution of the sample across cities and type of formal sector employees was designed to represent the population of formal sector workers in these four cities. Population-level information on the distribution of formal sector workers across different sectors was available from the Ethiopian Statistical Service and the EHIA. This information was used to distribute the sample size across each sector (see Table A1). However, distribution across both cities and sectors was unavailable for all categories of workers. We did, however, have access to information on the distribution of civil servants across each city. This was used as the basis for distributing the sample across the four cities (see Table A2).

Within each city, sample selection was tailored to the sector in question. For instance, regarding civil servants, in each city, a list of ministries, agencies, and bureaus was drawn up and organizations were selected for the survey based on probability proportion to their employment size and within each selected organization, respondents were randomly surveyed (see Table A3). A tricky part of the data collection was gathering information on pensioners. In each city, the survey team identified the locations where pensioners receive payments and allocated the targeted sample size equally to each of these payment locations. For instance, the Addis Ababa survey team visited the Federal main post office which handles payments for pensioners in Addis Ababa. The office provided a list of pensioners and payment centers in ten locations in the city. Five of these locations were randomly selected and an equal number of pensioners were surveyed from each of these locations at the time that they came to receive payments (see Table A4). A similar approach was followed in other cities.

After ensuring that a selected respondent was willing to participate in the survey, enumerators gathered individual and household-level information. The survey contained a household roster which gathered socio-economic information on all household members, their health status and lifestyle choices, outpatient and inpatient healthcare utilization, financing of healthcare, and whether they currently had any form of health insurance. As it turned out, we were able to gather more information and the sample at hand includes 2,749 formal sector employees and their families (6,894 individuals). The key difference (see Table A1) as compared to the plan is a larger share of public sector workers (54% actual versus a plan of 43%) and smaller shares of private sector workers (38% actual versus 43% planned) and pensioners (8% actual versus 14% planned). In terms of the distribution of the sample across the four cities (Addis Ababa – 56% actual versus 54% planned; Bahir Dar – 10.5% planned versus 12.1% actual; Hawassa – 15% actual versus 15.7% planned; Mekelle – 18.5% actual versus 18.2% planned), the differences between the plan and realization are minor (see Table A2). While not perfect, arguably the sample provides a representative picture of HSB and attitudes towards SHI across formal sector workers residing in these four cities.

Methods

The paper relies on descriptive statistics, logit, and multinomial logit (MNL) models to examine its objectives. Logit models are used to examine the probability of seeking treatment conditional on reporting an illness while multinomial logit models are used to examine the choice of healthcare provider. Logit models are also used to explore attitudes towards SHI. In all instances, the outcomes are treated as functions of socio-demographic traits such as household demographics, education of household head, household income, and the share of a household with health insurance. The models also control for regional fixed effects and in some instances, sector of employment.

Ethics clearance

Informed consent was obtained from respondents prior to data collection. Ethics approval (IDPR/LT-0005/2016) was provided by the Research Ethics Committee of the Institute of Development Policy and Research, Addis Ababa University.

Descriptive statistics – socio-economic characteristics and current health insurance status

The sample consists of 6,894 individuals of which 2,749 are current or former employees and fall into one of three categories – public sector workers (civil servants or public sector enterprise workers), private sector workers including NGOs, and (former public sector workers) pensioners. The sample consists of 54% public sector workers, and 38% private sector workers while the remainder (8%) are pensioners. About half the sample is male and 72% are adults. The average household size is 4.3. Since the sample consists of formal sector workers it is not surprising that about 60% have tertiary education. Education levels are highest amongst public sector workers – 77% have tertiary education as opposed to 73% amongst private sector workers. Regarding monthly income, pensioners have the lowest income while private sector employees record the highest income (Table 1).

Table 2 provides information on knowledge of insurance and the current health insurance status of sample respondents. While almost all respondents had heard of health insurance, a much smaller fraction – about 40% – was able to provide correct answers to a set of five questions on the functioning of insurance. There is limited variation across public and private sectors. On average, about 52% of households have at least one household member who has health insurance. This figure varies substantially across sectors with 77% of private sector workers reporting that at least one household member has health insurance while the corresponding figures are 21% among pensioners and about 40% amongst public sector workers. For a majority of current employees (about 75%), their employer pays the insurance premium.

Seeking outpatient healthcare

Conditional on experiencing an illness in the two months preceding the survey, 85.5% of respondents seek treatment. Amongst those who don’t seek treatment, the majority (more than 50%) expect to recover naturally and hence don’t seek any care. Amongst those who do seek care, almost all (94%), opt for formal care. In terms of sector of care, 56% opt for care at a private health facility while the remainder opt for a public facility (see Table 3). The interaction between sector and type of provider is illustrated in Figure 1. A majority of the care is provided by doctors in private facilities (47%), followed by doctors in public facilities (25%), health workers in private facilities (19%), and health workers in the public sector (9%). There are no sharp differences across types of employment. Despite the higher costs of doctor visits to private facilities, where the cost of outpatient care is 2.66 times the cost of doctor-provided care in public facilities, the bulk of formal sector workers opt for private care due to the perceived capability of staff and the availability of drugs while the cheaper cost of care is the main reason for choosing public sector facilities (see Table 4).

Table 5 provides a multivariate analysis of the probability of seeking outpatient care (logit model) from a formal provider and subsequently the probability of seeking outpatient care from one of four formal healthcare providers (multinomial logit). Two points stand out. First, the role of income, and second, the effect of access to insurance. As compared to the poorest quintile, individuals in the richest income quintile are five percentage points (p-value < 0.10) more likely to seek formal care. As may be expected, access to insurance translates into a higher probability of using formal care. Residing in a household where a member has health insurance is associated with a five percentage point (p-value < 0.05) increase in seeking formal care.

These patterns are emphasized in the results emerging from the MNL model. The two clearest effects are those related to income and access to health insurance. Individuals belonging to the highest income quintile are far more likely to seek care from private sector providers. For instance, as compared to the poorest quintile, individuals belonging to the richest income quintile are 25 percentage points (p-value < 0.01) more likely to seek care from doctors in the private sector and 13 percentage points (p-value < 0.05) less likely to access care from doctors in the public sector. The differences in healthcare seeking behavior between the poorest and other income quintiles is not particularly pronounced. Individuals residing in households with access to health insurance are about nine percentage points (p-value < 0.05) more likely to seek care from doctors at private facilities while avoiding the use of public sector health workers (p-value < 0.05). We also estimated an alternative specification to examine differences in healthcare seeking behavior across employees in different sectors (see Table A5). The results indicate that the sector of employment does not influence HSB as opposed to income which remains salient.

Seeking inpatient healthcare

Approximately two percent of the sample has experienced an episode of hospitalization in the 12 months preceding the survey with an average duration of nine days. In contrast to outpatient care a majority of the sample (62.5%) made use of public facilities (Table 3). On average, the cost of public facilities is about a quarter of that in private facilities, and the main attraction of using public facilities is their lower cost. Private facilities are costlier and staff at such facilities are considered more compassionate and capable (Table 4). A logit analysis (Table 6) of the probability of using public care confirms the effect of income on healthcare choice and shows that households in the higher income quintiles have substantially lower probabilities of using publicly provided care. As compared to the lowest income quintile, respondents in the top three income quintiles are 36 to 56 percentage points (p < 0.001) less likely to use public care. The availability of health insurance reduces the probability of using public healthcare by 26 percentage points (p-value < 0.05). The inclusion of the sector of work does not alter the role of income in influencing the choice of healthcare provider but does wash out the effect of health insurance, suggesting that access to health insurance may be serving as a proxy for working in the private sector.

Attitudes towards social health insurance

The available data provides an opportunity to examine, among other issues, willingness to pay for the scheme. Relevant information on the attitude of respondents towards various aspects of SHI is provided in Table 7. While knowledge about the plans for SHI was already widespread (83% of the sample had heard of SHI) in 2016, knowledge about the SHI premium was not widespread (40% knew the premium). Except for pensioners, all others who indicated that they knew the premium had accurate knowledge of the amount of the SHI premium. After being informed about the premium, about 40% considered it fair and a slightly higher proportion (44%) considered the premium to be affordable. The percentages do not differ considerably across types of employment. Notwithstanding the relatively high fairness and affordability rating, only about 24% were willing to pay 3% or more of their monthly income as a premium. The willingness to pay the 3% premium is similar across employment sectors (23.5% versus 28.7% for the public and private sectors, respectively). Only a minority of respondents were unwilling to pay at all (12.8%). While a majority of the sample (67%) supported the introduction of SHI, the average WTP was 1.6% of monthly income ranging from a low of 1.2% for pensioners to a high of 1.7% for private sector workers. A little more than half the sample (53%) is concerned that after paying for SHI they may not get adequate access to healthcare services. The two main concerns are long waiting times and lack of drugs.

Table 8 presents a series of exploratory regressions, in seriatim, on the link between various socio-economic traits and the perception that the SHI premium is not fair, that it is not affordable, and that SHI should be opposed. The analysis yields several patterns. First, income and the perception that the SHI premium is unfair are positively correlated. For instance, respondents in the two richest quintiles are 10 percentage points (p-value < 0.05) more likely to perceive the premium as unfair as opposed to the lowest income quintile. These two quintiles are not different from the poorest quintile in terms of premium affordability perceptions but still consider the premium unfair. Consistent with the unfair premium perception, the two richest quintiles are 10–13 percentage points (p-value < 0.01) more likely to oppose SHI. In fact, there is greater resistance from all other income groups as compared to the poorest, but the extent of the resistance is higher amongst the two richest quintiles.