Innovations for Poverty Action Health Microinsurance Education Project Evaluation Northern Region, Ghana Final Endline Report October 2012 1
Contents 1. Executive Summary... 4 2. Introduction... 5 3. Background... 6 Health Insurance in Ghana... 6 4. Description of Study... 9 Health Microinsurance Education... 9 Partnership Roles... 10 Evaluation Design... 10 Data Collection and Analysis... 12 5. Key Results... 14 What types of individuals are more likely to take-up the National Health Insurance Scheme?... 14 Household Financial Status... 17 Attitudes About Insurance... 20 Health Outcomes... 20 What is the impact of the education on consumer knowledge about health microinsurance?... 21 What is the impact of the education on consumer decision making about whether to enroll or not in an available health insurance scheme?... 26 What is the impact of the education on consumer decision making about whether to enroll or re-enroll in the insurance scheme after one year?... 27 Does education affect the ability to appropriately use insurance to access and use covered services?... 28 Which type of education is more effective: education delivered in short sessions every two weeks, or education delivered in one long session in terms of take-up, usage, and re-enrollment?... 29 What is the impact of a refresher training on re-enrollment in the population?... 29 For which types of people does the education have the greatest effect with respect to take-up, re-enrollment, and appropriate use of services?... 30 2
What do households perceive as the benefits of health insurance and what is the impact of the education on these perceptions?... 32 What factors influenced their decision to enroll or not enroll in the health insurance, what are the barriers and did education affect these?... 33 What is the relationship between health insurance and household financial stability and well-being, and the way households accses and finance health treatment? Without health insurance how do households manage health care expenses and what is the effect on consumption, sale of assets and food security?... 37 6. Discussion... 43 Impact of Education on Enrollment... 43 Relatively High Starting Enrollment... 44 Implementation... 45 Change in Registration and Enrollment Over Time... 45 Ongoing Barriers to Enrollment... 49 Implications for external validity... 50 7. Conclusion... 52 7. Appendix A: Statistical Results... 54 8. Appendix B: Stratification... 69 9. References... 71 3
1. Executive Summary National health insurance that provides a comprehensive set of health care services has been available to the formal and informal sectors in Ghana since 2003. However, coverage is far from universal, especially in rural areas. Freedom from Hunger, Sinapi Aba Trust (SAT) and Innovations for Poverty Action partnered to create, implement, and evaluate a program to educate microfinance clients in Ghana s Northern Region about health insurance provided through the National Health Insurance Scheme (NHIS) of Ghana. Designed as a randomized control trial, the evaluation compared outcomes for microfinance clients randomly assigned to receive the education treatment to outcomes for clients randomly assigned to the control group. The study design included two education treatments: a short session treatment that included 6 short sessions given every two weeks, and a consolidated session treatment that covered the same material in one 2-hour session, both administered from October 2010 through March 2011. In addition, half of the individuals in each treatment group were given a reminder session in March 2012. The total sample included 300 microfinance groups; five members from each group were selected to be surveyed. Data collected included a baseline survey conducted in October 2010, a knowledge test administered immediately after the education was completed in March 2011, a midline survey conducted in July 2011, and an endline survey conducted in June 2012, which included a qualitative component. The midline and endline data on insurance enrollment found no significant differences in health insurance enrollment rates between the treatment groups and control group. In addition, there were no significant differences in enrollment rates between those who received reminder sessions and those who did not. The results of the knowledge test administered after the initial treatment sessions suggests that the education did improve knowledge among those who received treatment, however by baseline there were no significant differences in insurance knowledge across treatment and control groups. The education may not have had a large impact because baseline knowledge of insurance was already high, suggesting that knowledge was not a barrier to enrollment. Rather, it appears that convenience of registration and the timing of making the premium payments are more common challenges for individuals who want to enroll. In addition, the treatment was not implemented with all groups assigned to receive it, and enrollment rates were already quite high in the sample at the beginning of the study. These factors made it more difficult for the treatment to have an impact large enough to be statistically observable. The high enrollment, high knowledge about insurance, and incomplete implementation were unique factors that suggest this study may have limited external validity; in environments where knowledge and enrollment are low, educational programs may have more impact. While there were no significant differences in enrollment rates between the treatment and control groups, respondents enrollment increased notably over the course of the study, at a rate that appears to be greater than the increase in enrollment in the general population in the areas where the study was conducted. While activities of the National Health Insurance Scheme aimed at increasing enrollment likely accounts for some of the increase, it is also 4
possible that the repeated surveys, along with the treatment activities, might have served as touch points that prompted clients to take action to register or enroll in insurance. Having insurance is not associated with higher likelihood of getting treatment in the case of health events, but those with insurance are more likely to go to doctors, while those without are more likely to go to less highly trained chemical sellers. Those with insurance also incurred somewhat lower out-of-pocket costs for an incident of illness and were more likely to have attended a well-patient visit. Although no causal link can be established, being registered in insurance is associated less likelihood of food insecurity, less likelihood of removing children from school for financial reasons, and less likelihood of selling assets to get money. 2. Introduction Although Ghana introduced a national health insurance program in 2003, enrollment rates of families in the informal sector remain low, particularly in rural areas. In 2010, Freedom from Hunger entered into a partnership with Sinapi Aba Trust (SAT), a Ghanaian microfinance institution (MFI), and Innovations for Poverty Action (IPA), a nongovernmental organization (NGO) specializing in impact evaluation, to design, implement and evaluate a program to teach microfinance clients about health insurance. The key questions of this evaluation are to determine whether the program increases up-take of insurance, and how insurance enrollment affects use of health services, health spending, and indicators of financial security. The timeline for the project was as follows: January 2010-June 2010: Developed, field tested and completed education module (facilitator and trainers guides) March 2010 June 2010: Developed research plan July-September 2010: Assessed data sources, established data collection methods, and completed census survey data. September 2010 November 2010: Baseline study completed September 2010: SAT trainers and staff trained to deliver education. October 2010-March 2011: Education conducted with all treatment groups. March April 2011: Follow-up knowledge surveys (conducted by SAT staff) July-August 2011: Midline study January 2012: SAT staff trained to deliver refresher training February April 2012: Refresher training provided to selected treatment groups April May 2012: Endline Survey and Qualitative Studies This report presents and analyzes final key indicators from the endline survey and qualitative studies and summarizes overall findings and conclusions from the project. We are grateful to the Microinsurance Innovation Facility of the ILO and to an anonymous donor who provided funding for the development of the education module and the test of the 5
impact of the education in a randomized control trial in Ghana. We also are grateful to our microfinance partner Sinapi Aba Trust and the support of their management team and staff. 3. Background Health Insurance in Ghana Ghana s national health insurance program (NHIS) enables individuals in the informal sector to register for health insurance by paying an insurance premium and registration fee (see Table 1), and after a three-month period, receive a comprehensive set of covered health services for no fee 1. Pregnant women, children under age 18 (of registered parents) and persons age 70 and older are not required to pay the annual premium, but may need to pay a small annual registration fee. The health services covered by the NHIS are laid out in the minimum basic benefits package (see Appendix C). The list is fairly extensive and purports to cover 95 percent of all health problems reported in Ghanaian healthcare facilities. A prescribed medicines list is also delineated. Expensive, highly specialized care such as dialysis for chronic renal failure and organ transplants are not covered by the NHIS. Neither are antiretroviral drugs for the treatment of HIV/AIDS, as these drugs are supplied by a separate government program. There is a notable emphasis on female reproductive health in the benefits package. Benefits for maternity care include antenatal care, caesarean sections, and postnatal care for up to six months after birth. Treatment for breast and cervical cancer are included in the package, although treatment for other cancers is not. While the program has dramatically increased access to health care services, there are still a large number of Ghanaians, particularly informal sector workers and the indigent, who are not registered in the health insurance program. At the end of 2009, the Ghanaian National Health Insurance Authority (NHIA) estimated that 62 percent of the population was registered with 48 percent actually having current enrollment. 2 A controversial report published by Oxfam in 2011 suggested that the rate of enrollment was likely to have been much lower and that insurance enrollment rates could be as low as 18 percent. 3 During 2010, the NHIA took a closer look at actively enrolled members as compared to those who had registered but were inactive as a result of non-payment of the annual premium. They determined that 34 percent of the population was actively enrolled at the end of 2010. 4 The NHIA estimates of active membership by region showed considerable variation ranging from a low of 23 percent in the Central Region of Ghana to a high of 53 percent in the Upper West. In the Northern Region, the location of SAT s program and where the study was located, active enrollment was estimated to be 31 percent of the population. 1 National Health Insurance Authority Report 2009 2 National Health Insurance Authority, Annual Report, 2009 3 Oxfam International 2011. Achieving a Shared Goal: Free Universal Health Care in Ghana. http://www.oxfam.org/en/policy/achieving-shared-goal-ghana-healthcare March 9, 2011. 4 National Health Insurance Authority, Annual Report, 2010. 6
The insurance program is run at the district level by local National Health Insurance Scheme (NHIS) offices, and overseen at the regional and national level by the NHIA. The NHIS districts have operated largely as independent franchises, with discretion to set their own registration fees and other policies, but reform of the health insurance program is a current topic of political debate, and it appears that NHIA has made some attempts to take a larger role in coordinating policies across NHIS offices. While NHIS offices can set their own registration fees, which usually range from 2-5 Ghanaian cedis (GHC) (1.32-3.30 USD), NHIA sets annual premiums. Because fees (and sometime premiums) vary by NHIS office, the total cost of registering for insurance varies as well, but is typically around 11-14 GHC (5.57-7.22 USD) for adults in the Northern Region. See Table 1 for a list of premiums and fees charged by the NHIS districts serving the project program participants as of January 2012, as reported to IPA by each district NHIS. Children under 18 are exempt from the premium payment, but usually must pay the registration fee. Table 1: Insurance Premiums and Fees Reported by NHIS Districts Serving Clients of the Tamale, Bole, Salaga, and Walewale SAT Branches All currency in Ghana Cedi (GHC ; exchange rate a.o. August 2012 was 1.94 GHC to 1US Dollar). NHIS District Registration fee for adult Premium for adult Total cost of registration for adult Tolon 5.00 7.20 12.20 Savelugu 4.00 7.20 11.20 Tamale 4.00 7.20 11.20 West Manprusi 4.00 10.00 14.00 Bole 5.00 8.00 13.00 East Gonja 2.00 10.00 12.00 AVERAGE 3.67 7.80 11.47 Once a person registers with NHIS and pays applicable fees and the annual premium, there is a three-month waiting period before the insurance can be used to access health care services, except for pregnant women who can immediately access prenatal and maternity care. By the end of the three-month waiting period, the individual is supposed to receive a health insurance card from NHIS. In some cases, the card arrives late, and if so people are told to obtain a temporary card from NHIS. The insurance remains in effect for one year, at which point the individual must re-enroll and pay the annual premium and applicable registration fees. The annual expiration date is printed on the NHIS card, which is good for five years, and stickers are added to the card at the time of annual re-enrollment to indicate current enrolled status. However, the onus falls on the client to remember to re-enroll; this poses a particular challenge for illiterate clients who cannot read the expiration date on the card, and who may not understand that they need to pay once a year. 7
After the expiration date, covered individuals have a 3-month grace period during which the insurance can be renewed. If an individual fails to re-enroll within that grace period, NHIS policy dictates that the individual must go through another 3-month waiting period. At the start of this study, NHIS offices serving the SAT clients in our sample were not enforcing this rule. Rather, they allowed individuals to access care immediately after re-enrolling, even if the policy had expired. If the insurance had been expired for more than one year, clients were required to pay the premium for every year that they have missed in order to use insurance immediately. In 2011, local NHIS officers reported a change in the enforcement of the expiration policy, indicating that if registrants did not pay the annual premium and fees within 90 days of expiration, that they would lose eligibility for services and be required to wait three months to access services once premiums and fees were paid for the year. When a client s insurance expires at the end of one year, the client is still considered to be registered with NHIS their information is stored in NHIS databases, and if they re-enroll, a new sticker is provided for their membership card that indicates the new expiration date. In order to be considered enrolled or active and eligible for covered services, the client must be current on the premium payment. If the client fails to pay the annual premium, the client may be termed unenrolled, inactive or expired. NHIS offices report that re-enrollment is a particular challenge. While registration rates have increased, many of the registered individuals fail to re-enroll each year. For example, in 2010 the Tolon NHIS office, which serves a rural area near the city of Tamale in Northern Ghana 5, estimated that about half of the population in its district is registered and has a current policy, but another 30 percent has registered but not renewed their insurance, allowing it to expire. This is consistent with findings on our sample at baseline where 70 percent of the respondents report being registered for insurance, but only about 32.6 percent of the total could be either confirmed as currently enrolled (premiums current) from visual inspection of the insurance card, or through extrapolation based on their reported use and ways of paying for health services. There are a number of potential barriers to registration and enrollment in the health insurance program. Individuals may not know about the program, may not understand how insurance works or what is covered, or may not know how to go about registering. Some individuals may also be unable to afford the premium at the time it is due. While an 11-14 GHC payment is not a particularly high amount even in Ghana, a large family may find it a challenge to put together the money to cover every adult household member under age 70, and particularly at a set time each year as there is no flexible payment option. Individuals may also believe that insurance is not a good value for them because of lack of availability of providers, benefit limitations, because they do not think they will need health services, or because they perceive the quality of services available to be low as compared to those who pay for health expenses out-of-pocket, or cash and carry care. Lastly, individuals may have every desire and intention to register, but simply do not get around to doing it. Each of these, with perhaps the exception of lack of knowledge, was observed in our sample either in the quantitative or qualitative surveys and will be discussed more in greater detail. 5 Some of SAT s groups served by its Tamale branch are located in the areas served by the Tolon NHIS office. People may register at any NHIS office, so the Tolon NHIS office possibly serves some people living within the city of Tamale as well. 8
4. Description of Study Health Microinsurance Education For this study we hypothesized that low knowledge about Ghana s health insurance program or about insurance in general was a barrier to registration or re-enrollment, and that education, therefore, may be an effective means of increasing insurance uptake and access to health care services. We theorized that education could be effective in increasing awareness, knowledge, and interest and stimulating greater demand for the health insurance program, pushing those who want to register but have not yet done so, increasing annual re-enrollment, and increasing total active enrollment (those who are current with premium and eligible for benefits) in the sample population. The Health Microinsurance Education (HME) project aimed to provide education about health insurance to clients of the microfinance institution Sinapi Aba Trust in Northern Ghana. The education sessions were designed to be provided at meetings of the clients microfinance groups. Four different education treatments were tested as follows: (Table 2): Table 2: Treatment Groups Treatment Technical Learning Conversations (TLCs) Technical Learning Conversations (TLCs) plus a reminder session Consolidated Sessions Consolidated Sessions plus a reminder session Control Group Description Six 30 minutes sessions administered every two weeks. Six 30 minutes sessions administered every two weeks, plus an additional 30-minute session one year later reminding clients they must re-enroll to prevent their insurance from expiring. One two hour session with same content as TLCs, administered once. One two hour session, with same content as TLCs, administered once, plus an additional 30- minute session one year later reminding clients they must re-enroll to prevent their insurance from expiring. No education sessions at any time The education sessions were delivered by the loan officers who serve the microfinance clients. After completing the education program, loan officers were to arrange for an NHIS agent to visit the group to provide an opportunity for clients to register or re-enroll in health insurance. The education modules were then evaluated using a range of evaluation tools (post-test survey of knowledge change, quantitative baseline, midline, and endline client surveys, and qualitative studies) to assess and understand their impact on changes in health insurance knowledge and take-up rates of national health insurance. The education sessions began in October 2010. Although scheduled to end in early January 2011, challenges with scheduling meetings with groups delayed completion of education for 9
some groups until early March. The additional 30 minute reminder sessions, took place in February and March 2012. Partnership Roles This study involved a collaboration of three organizations: Freedom from Hunger (FFH), a U.S.-based NGO; Innovations for Poverty Action (IPA), a U.S.-based research NGO, and; Sinapi Aba Trust (SAT), a Ghanaian microfinance institution. The health insurance education materials were designed by FFH. The education materials include a trainer s guide, facilitator s guide, relevant resource materials, and supervision and monitoring tools. FFH also trained SAT branch managers and financial service officers (FSOs) to deliver the training to clients. FFH reimbursed the related costs of training to SAT and provided technical support as well as funds for SAT to provide a small incentive for the FSOs to complete the education as scheduled. SAT selected branches for education delivery, identified active groups for randomization, provided logistical support for training staff, and implemented the education with its clients in four of its branches in the Northern Region. SAT also worked closely with IPA and FFH to plan the evaluation and to assure compliance with research protocols, to coordinate with the National Health Insurance Scheme districts to assure that all information provided to client was correct and to invite insurance marketers to visit the client groups in the sample to offer insurance enrollment. SAT also collected data for the knowledge survey post-test. IPA has worked closely with FFH and SAT to design and plan a program implementation and research design that adheres to the randomized design. In addition, with the guidance of academic researchers Raymond Guiteras PhD of University of Maryland and Harounan Kazianga PhD of Oklahoma State University, IPA designed and conducted the data collection surveys used to determine program effect on client health insurance knowledge, health insurance take-up rates, and reported use of and spending for health services. IPA did limited monitoring of the program implementation. Evaluation Design A randomized control trial (RCT) is a type of impact evaluation that randomly assigns some individuals to participate in a program (the treatment group), and some individuals to not participate (the control group), and to compare the outcomes for the two groups. Randomized control trials have the advantage that, with a large enough sample, the treatment and control groups are statistically identical; on average the only difference between them is that one group gets the treatment and one does not. Therefore, any differences in outcomes can be attributed with certainty to the treatment, provided that the randomization has been successful. The HME Project Evaluation employed an RCT to answer the following research questions: 10
1. What types of individuals are more likely to take-up the National Health Insurance Scheme? 2. What is the impact of the education on consumer knowledge about health microinsurance? 3. What is the impact of the education on consumer decision making about whether to enroll or not in an available health insurance scheme? 4. Does education influence decision-making about enrollment and re-enrollment in the insurance scheme after one year? 5. Does the education affect the ability to appropriately use insurance to access and use covered services? 6. Which type of education is more effective: education delivered in the TLC format that includes six 30 minute sessions provided every two weeks, or education delivered in one consolidated session in terms of take-up, usage, and re-enrollment? 7. What is the impact of a refresher training (provided one year after the initial education) on re-enrollment in the population? 8. Did the education have different effects on different types of people with respect to take-up, re-enrollment, and appropriate use of services? 9. What do households perceive as the benefits of health insurance and what is the impact of the education on these perceptions? 10. What factors influenced their decision to enroll or not enroll in the health insurance, what are the perceived barriers and did education affect these? 11. What is the relationship between health insurance and household financial stability and well-being, and the way households access and finance health treatment? Without health insurance how do households manage health care expenses and what is the effect on consumption, sale of assets and food security? In order to answer these questions, the HME Project evaluation used four treatment groups, one for each of the four education approaches (described in Table 2 above), and one control group. Since education sessions are given to an entire credit group at once, randomization was done at the level of the credit group; that is, all clients in the same credit group were assigned to the same treatment group or to the control group. The sample for the evaluation comprised credit groups that were believed to be active at the time of baseline in four SAT branches in the Northern Region: the Tamale branch, the Walewale branch, the Salaga branch, and the Bole branch. Active credit groups are those that are currently meeting on a regular basis. Credit groups may become inactive if they 11
stop working with SAT entirely, choose not to take loans during a particular loan cycle, or if they fall behind on their payments. Active credit groups were identified by taking lists of groups from SAT, and conducting a census interview with each of the groups. The census interview ascertained that the group was actually active, collected basic information about the group members including enrollment status, and recorded contact information so that the group could be contacted for future survey interviews. While groups that were inactive at the beginning of the program were screened out, some groups became inactive between the census and the beginning of the education sessions. Additional groups became inactive over the course of the study, creating a challenge for both implementation and evaluation of the project. The sample size was set at a total 300 credit groups, to ensure enough power to measure the impact of the intervention. In order to have 300 groups, groups from four different SAT branches were included in the sample. In Walewale, Salaga and Bole, 60 active groups were selected for inclusion in the sample from among the branches active groups. The Tamale branch has the largest number of credit groups, so the sample from Tamale was twice the size as the samples in the other branches, with 120 active groups included in the sample. In Salaga and Bole, more than 60 active groups were identified; in those cases, the 60 groups to be included in the sample were selected randomly from the total number of groups. Five members in each credit group were randomly selected to be surveyed. 6 After the credit groups to be included in the sample were selected, the credit groups in the sample were randomly assigned to treatment and control groups. Of the sample credit groups, 40 percent were assigned to the control group, while 15 percent were assigned to each of the four treatment groups. Assignment was random, stratified on branch, urban or rural, and high or low enrollment, based on information collected through a census of credit groups. Stratification means that randomization is done separately for each of the combinations stratification variables. Enrollment was defined as being current on an NHIS insurance policy. For further detail on the stratification groups, see Appendix B. Data Collection and Analysis The impact of the program was assessed using data from several sources: a baseline survey, a post-education knowledge test, a midline survey, an endline survey, and a qualitative study The baseline survey was administered from September 2010 to November 2010. The survey was administered at each respondent s home, unless the respondent requested an alternative location, such as the place of business or SAT microfinance group meeting location. The survey sample included 1609 respondents; the additional respondents over 1500 came from extra client groups that were included in the survey in case one of the 6 Credit groups that had fewer than 5 members were randomly paired with another credit group with fewer than 5 members to create a new credit group with at least 5 members. These pairs are treated as one credit group in the research design; both credit groups assigned to the pair are placed in the same treatment group or in the control group. 12
original client groups dropped out in between the census and the implementation of the treatments. 1505 respondents were successfully tracked and consented to be interviewed. The survey was conducted in paper form, by enumerators hired and trained by IPA. The survey was 40 pages long and took approximately 1-2 hours to complete. Data entry for the baseline survey was done by IPA s in-house data entry team in Accra, using double-double data entry. Data entry was conducted from January 2011 to February 2011. A post-education knowledge test was administered after the last education session was completed to assess the education s impact on client knowledge of insurance. The posteducation knowledge test was made up of health insurance knowledge questions (Section E) from the baseline survey. The knowledge test was administered by the SAT loan officer conducting the education sessions. The last knowledge tests were administered in March 2011. A number of issues arose with the collection of the knowledge tests which could have implications for data quality, and consequently confidence in the results of the data analysis. First, the SAT loan officers did not survey the entire subsample. IPA received 155 surveys out of 600 respondents who were randomly selected to receive the knowledge test. Second, the SAT loan officers did not follow the randomization. A much higher than expected number of respondents were alternates, suggesting that the SAT loan officers may have conducted the survey with any of the three respondents listed on their roster, rather than focusing on the first two before using the alternate. In some cases, the respondent interviewed was not among the three selected in the randomization at all. Because of the small total number of surveys received, all knowledge tests were included in the analysis, regardless of whether the respondent was among the list of three respondents given to SAT to survey. Lastly, it is important to remember that the SAT loan officers conducted the knowledge tests themselves. Best practice would be to have independent evaluators with no stake in the findings. Loan officers may have felt pressure to get results showing improvement as a result of their efforts. To minimize the risk of this, we tried to emphasize to both the loan officers and the respondents that the knowledge test was intended to test the program, not the people implementing it or receiving it. The midline take-up survey was conducted with the same respondents from the baseline survey sample at SAT microfinance group meetings in July 2011 by surveyors hired and trained by IPA. The survey was much shorter than the baseline survey and covered the enrollment information in the Household Roster section of the baseline survey. The endline take-up survey was conducted in April and May 2012 with the same respondents from the baseline and midline survey sample, after the reminder sessions were completed for the groups randomly selected to receive these. Data entry was done by the IPA team in May and June of 2012. The endline survey collected all of the same data as the midline survey, plus some additional information about household finances, reasons for enrolling or not enrolling in insurance, and how households dealt with health events. A qualitative study was conducted at the same time as the endline survey. Focus groups were conducted with a random selection of respondents. The interviews, conducted with groups of respondents who had the same registration status, asked what respondents knew about insurance, what attributes of the program they thought were good and what attributes were bad, and what attributes were most important to their decision to enroll. 13
5. Key Results What types of individuals are more likely to take-up the National Health Insurance Scheme? The baseline survey looked extensively at the characteristics associated with enrollment and registration rates providing a comprehensive picture of the sample and the characteristics of clients and families who were registered in the insurance. Basic Demographics for Adult Registration and Enrollment Females were significantly more likely to be registered in insurance than males, by approximately 14 percentage points, on average (Appendix Table 31)). Older individuals were also significantly more likely to be registered than younger individuals, with those over 45 on average about 10 percentage points more likely to be registered than adults between the ages of 18 and 30. Educational attainment is also a significant indicator of insurance registration status with the likelihood of registration going up with education. Compared to those with no schooling, those who achieved primary-level schooling are 6 percentage points more likely to be registered and this continues upward with levels of schooling with individuals with tertiary schooling 36 percentage points more likely to be registered than those with no education at all. Being married, on average, leads to a small but significant increase in the likelihood of being registered in insurance; having once been married (and being currently divorced or widowed) has no significant relationship. A number of ethnic and religious variables were significantly associated with higher or lower registration rates. These variables likely serve as proxies for living in geographical areas more specific than the four regions included as controls. The geographical variables included were also significant. Being located in a rural area was associated with a 6 percentage decline in the likelihood of being registered. Being located in Walewale, Bole, or Salaga was all positively associated with likelihood of being registered, compared with Tamale. Spouses of household heads are most likely to be registered for insurance; on average, they are 13 percentage points more likely to be registered than household members who are neither the head of the household, spouse of the household or child of the household (appendix Table 32). Heads of households are also more likely to be registered than other household members. There was no significant difference between adult children of household heads and other household members. 14
In general, demographic factors were more closely associated with registration status than enrollment status (appendix 15
Table 33). Demographic variables explained about 8 percent of variation in registration status, but only about 4 percent of variation in enrollment status. (It should be noted that in this context, we use the term explain to mean that the correlations between the demographic variables we are looking at and registration status can predict about 8 percent of the variation in registration status. However, this does not necessarily imply that these cause the variation in registration status.) There are several possible explanations for this. First, it could be that who is enrolled depends a lot just on chance. For example, if most people re-enroll when they get sick, then those who have been sick recently are more likely to be enrolled. Second, it could be that extra variation is introduced through the data collection methods. If who has access to their card is fairly random, then this will create random variation in the enrollment status variable that would not be explained by demographics. Lastly, it is possible that while demographics are highly correlated with registration status, other non-random variables that are not correlated with demographics are more important in explaining enrollment status. When looking at demographic traits associated with enrollment, fewer variables are significantly correlated with enrollment status than were correlated with registration status, and most of the correlation sizes are smaller. As with registration, women are more likely to be currently enrolled. The correlation size is about half of what it was for registration, with women 5 percentage points more likely to be currently enrolled than men, on average. Older adults were more likely to be currently enrolled than adults 18 to 30. No education variables were significantly correlated with enrollment status. This is surprising given that all of the education variables were significantly correlated with registration, with quite large correlation sizes in some cases. This could be explained by the hypothesis that enrollment status is more random than registration status. The only geographic variable that was significantly related to enrollment was living in Walewale. Walewale residents were 9 percentage points more likely to be enrolled than residents of Tamale. There was no significant difference between residents of Bole or Salaga and Tamale, and residents of rural areas were no more or less likely to be enrolled than urban residents. Basic Demographics for Children s Registration and Enrollment Status We also looked at attributes associated with registration and enrollment for children. Children are slightly more likely to be registered than adults, likely reflecting the fact that there is no premium payment for children. Gender is not a statistically significant predictor of the likelihood a child will be registered. Children over age 7 are significantly more likely to be registered than younger children, with children in the older age group 6 percentage points more likely to be registered, on average. For full results, see AppendixTable 34: Predictors of Child Individual Insurance) 16
Geographic variables were significant. Living in a rural area was associated with a 15 percentage point decrease in likelihood of registration for the model for children of all ages, and a 13 percentage point decrease in the model for school-age children. Children in Bole and Salaga were significantly more likely to be registered than children in Tamale. For school-age children, school enrollment status was a large and significant determinant of insurance registration status. A child enrolled in school was, on average, 28 percentage points more likely to be registered than a school-age child not enrolled in school. Children of household heads were significantly more likely to be registered than children more distantly related to the household head; a child of the household head was 5 percentage points more likely to be registered than the child of another household member. (Appendix Error! Reference source not found.) Children ages 7 to 17 are less likely to be currently enrolled than younger children, despite the fact that they are more likely to be registered (full results are reported in Table in the appendix.) This is probably because although, because they are older, their parents have had more time to register them, it is also more likely that more time has elapsed since their registration, so their insurance is more likely to be expired. Gender is not significantly correlated with enrollment status. Children living in Tamale were significantly less likely to be currently enrolled compared with children living in any other location. Living in a rural location was not significantly related to enrollment status, despite the fact that there was a large negative correlation between living in a rural location and being registered. There was no significant relationship between enrollment in school and being currently enrolled in health insurance, despite the fact that there was a very large correlation between being enrolled in school and being reported as being registered for health insurance (appendix Table 37: Household Position and Child Individual Insurance Enrollment). Household Financial Status The baseline study also looked at relationship between financial attributes and household registration rates; some of these measures were also collected in the endline survey. In general, the baseline data showed that insurance registration and enrollment status was not closely associated with household income or spending measures. Table 3 reports results for regressions of household insurance registration rates on different measures of household income and consumption, including weekly income from the respondent s SAT business, annual income from the farm harvest, weekly income from other sources besides the SAT business and farming, and a measure of annual consumption per household. (Note that this is not a comprehensive measure of household consumption, but rather reflects only household consumption for the expenditure categories covered in our survey.) 17
None of the income variables are significantly related to the household insurance registration rate. While this could indicate that household income is not correlated with insurance registration, or it could be due to the limits of our income measures. While financial measures were collected in both the baseline and midline, with similar results, neither survey used methods such as diaries to comprehensively capture income or expenditures. Revenues from the SAT business cover account only for sales volume, but do not take into account profitability, so actual household income is uncertain. Farm income is the most comprehensively measured, since it is based on estimates for earnings from individual crops, but more than half of all respondents live in non-farming households. The measure of income from sources other than the SAT business and farming may have limited accuracy because of the broadness of this category. It is easy for respondents to accidentally omit income sources such as earnings from seasonal labor that they are not currently engaged in, or income that family members earn and spend without first contributing to the household s communal resource pool. Table 3: Income and Household Registration Rates, Baseline Survey Regression Coefficient Standard P-Value N R-Squared Error (1) Share of household that 0.0039 0.0032 0.22 1505 0.0010 is registered on sales revenue from SAT business (in 1000 GHC) (2) Share of household that is registered on income from farming (in 1000 GHC) 0.0377 0.0459 0.41 1505 0.0004 (3) Share of household that is registered on income from other sources (in 1000 GHC) (4) Share of household that is registered on annual household consumption per household member (in 1000 GHC) 0.0004 0.0006 0.47 1505 0.0004 0.8 0.2 0.00* 1505 0.0099 Annual consumption per household member is positively and significantly correlated with the household registration rate. However, the size of the correlation is not large: 100 GHC in additional consumption per household member is associated with a 0.5 percentage point increase in the household registration rate. Also, the R-squared associated with this regression is only 0.0099, meaning that differences in annual consumption per household member explain only 1 percent of the variation in registration rates. The lack of a statistically significant correlation between income and registration rates and the small size of the coefficient for consumption as a predictor of insurance registration suggest that there is not a strong relationship between financial resources and insurance 18
registration for our sample. If this is correct, then it is possible that the cost of the insurance premium is not an important constraint to registration for our sample. There is no significant correlation between our income or consumption measures and household enrollment rate, despite the fact that maintaining current enrollment requires more financial resources than simply registering once (Table 4). This suggests that the cost of paying premiums is not a significant barrier to either registering for insurance or maintaining current enrollment. Table 4: Income and Household Enrollment Rates, Baseline Survey Regression Coefficient Standard P-Value N R-Squared Error (1) Share of household that 0.0004 0.0019 0.85 1505 0.0000 is enrolled on sales revenue from SAT business (in 1000 GHC) (2) Share of household that is enrolled on income from farming (in 1000 GHC) 0.0013 0.0278 0.41 1505 0.0000 (3) Share of household that is enrolled on income from other sources (in 1000 GHC) (4) Share of household that is enrolled on annual household consumption per household member (in 1000 GHC) -0.0002 0.0004 0.50 1505 0.0003 0.2 0.1 0.12 1505 0.0016 The endline survey also collected some information about household finances, including average daily household food consumption, average weekly income, and how many phones the household owns. As with the baseline survey, there was very little relationship between the financial measures and the likelihood that a respondent would be registered or enrolled in health insurance. (See Appendix A Table 38 and Table 39.) 19
Attitudes About Insurance Evidence that more positive attitudes about insurance are associated with higher registration and enrollment was mixed. There were two questions on the baseline that dealt directly with attitudes. Respondents were asked whether they agreed or disagreed with the statements I would rather risk having to pay for health expenses using cash and carry than pay for health insurance and Health insurance is not a good value for the money. In each case, a response of disagree indicated a more positive attitude towards insurance. We regress insurance registration on a variable equal to 1 if the respondent showed a positive attitude towards insurance, for each of the two questions, and on a number of demographic variables included as controls. Results are reported in Table T in the Appendix. A response of Disagree to the first question was positively associated with insurance registration. There was no significant relationship between enrollment and answers to this question. There was also no significant relationship between responses to the second question and either registration or enrollment rates. (See Appendix A Table 40.) 7 Again, causality could run in either direction. It may be that respondents with more positive views of insurance are more likely to register for it. It could also be that respondents with insurance have good experiences with it, and therefore have more favorable attitudes towards it. Health Outcomes Health insurance registration is significantly related to a higher likelihood of reporting a health event in the past month, but the correlation size is small. Being registered for insurance is correlated with a 2 percentage point increase in the likelihood that an individual reported experiencing a health event (See Appendix 7 Key hypotheses were tested using both linear and logistic regression models. Results were similar; no variables were identified as significant using the logistical regression that were no significant for the linear model. For ease of interpretation, we report the linear regression results. 20
Table 41 ). This may be because individuals who report health events are more aware that they are at higher risk, and thus are more likely to register for insurance. Second, it may be that individuals with health insurance are more likely to seek treatment for less severe health events, and as discussed above, may be more likely to report these health events because the act of seeking treatment contributes to the respondents perception of the event as a health event. Current health insurance enrollment had an even larger correlation with reporting a health event: being currently enrolled was associated with a 12 percentage point increase in the probability that an individual would have reported having had a health event in the past month. As with registration, it is possible that those who are most likely to experience a health event are aware of this, and are more conscientious about keeping their policies current. However, it is also possible that much of the relationship is due to causality in the other direction. Prior to the change in NHIS local policy enforcement in 2011, it was possible for individuals to discover that their insurance was expired when they experienced a health event and attempted to use their insurance to pay for treatment. In cases where the enrollment was expired, the individual (or someone on the individual s behalf) could pay the premium due and provide immediate access to covered health services. As a result, at baseline someone who has had a health event in the past month was likely to either have had a current policy at the time of the health event, or to have re-enrolled at the time they needed services. Despite having health events at a higher rate, individuals who are registered for health insurance rate their health significantly higher than those who are not registered, when asked to rank their health on a 10-point scale, with higher numbers indicating better health. In the endline survey, respondents who reported they were registered for insurance had an average self-perceived health ranking of 7.31 out of 10, compared with 7.05 for those who were not registered, a statistically significant difference. There was no statistically significant difference between those who were confirmed enrolled and those who were not. It could be that individuals who take care of their health in general are more likely to register in insurance and have better health outcomes. These findings would also be consistent with insurance leading to better health outcomes through higher quality treatment, or leading to more peace of mind relating to health. What is the impact of the education on consumer knowledge about health microinsurance? To test the impact of education on SAT clients knowledge and attitudes about health insurance, clients at baseline were quizzed about their knowledge and their attitudes regarding health insurance. The same set of questions were used to develop a posteducation knowledge test that was administered a second time to a sub-sample of the baseline clients, immediately after the education was administered. The results immediately below (Table 5, Table 6, and Table 7) are based on the responses from clients who both participated in the baseline survey and the post-education knowledge test. As noted in the methods section, the post-education knowledge test participants were supposed to be a randomly-selected sub-set of clients who had participated in the baseline. Thus, the 21