Costs, equity, efficiency and feasibility of identifying the poor in Ghana’s National Health Insurance Scheme: empirical analysis of various strategies


Corresponding Author Genevieve Cecilia Aryeetey, Radboud University Nijmegen Medical Center, Nijmegen, The Netherlands. E-mails:;


Objectives  To analyse the costs and evaluate the equity, efficiency and feasibility of four strategies to identify poor households for premium exemptions in Ghana’s National Health Insurance Scheme (NHIS): means testing (MT), proxy means testing (PMT), participatory wealth ranking (PWR) and geographic targeting (GT) in urban, rural and semi-urban settings in Ghana.

Methods  We conducted the study in 145–147 households per setting with MT as our gold standard strategy. We estimated total costs that included costs of household surveys and cost of premiums paid to the poor, efficiency (cost per poor person identified), equity (number of true poor excluded) and the administrative feasibility of implementation.

Results  The cost of exempting one poor individual ranged from US$15.87 to US$95.44; exclusion of the poor ranged between 0% and 73%. MT was most efficient and equitable in rural and urban settings with low-poverty incidence; GT was efficient and equitable in the semi-urban setting with high-poverty incidence. PMT and PWR were less equitable and inefficient although feasible in some settings.

Conclusion  We recommend MT as optimal strategy in low-poverty urban and rural settings and GT as optimal strategy in high-poverty semi-urban setting. The study is relevant to other social and developmental programmes that require identification and exemptions of the poor in low-income countries.


Objectifs:  Analyser les coûts et évaluer l’équité, l’efficacité et la faisabilité, de quatre stratégies pour l’identification des ménages pauvres éligibles pour les plus importantes exonérations dans le système national d’assurance santé au Ghana (SNIS): l’évaluation des moyens (MT), l’évaluation des indicateurs de moyens (PMT), le classement par richesse participative (CRP) et le ciblage géographique (GT), en milieu urbain, rural et semi-urbain au Ghana.

Méthodes:  Nous avons réalisé l’étude sur 145 à 147 ménages par contexte en utilisant MT comme stratégie de référence. Nous avons estimé les coûts totaux, qui comprennent les coûts des enquêtes auprès des ménages et le coût des primes payées aux pauvres, l’efficacité (coût par personne pauvre identifiée), l’équité (nombre de vrais pauvres exclus) et la faisabilité administrative pour l’implémentation.

Résultats:  Le coût pour exempter une personne pauvre variait de 15,87$à 95,44 $ américains; l’exclusion des pauvres variait entre 0% et 73%. MT a été plus efficace et équitable dans les milieux ruraux et urbains avec une faible incidence de pauvreté. GT a été efficace et équitable dans les milieux semi-urbains à forte incidence de pauvreté. PMT et CRP ont été moins équitables et inefficaces, bien que faisables dans certains contextes.

Conclusion:  Nous recommandons MT comme stratégie optimale dans les milieux urbains et ruraux à faible pauvreté et GT comme la stratégie optimale dans les milieux semi-urbains à pauvretéélevée. L’étude est pertinente pour d’autres programmes sociaux et de développement qui nécessitent l’identification des pauvres et les exemptions pour ces derniers dans les pays à faibles revenus.


Objetivos:  Analizar los costes y evaluar la equidad, eficiencia y viabilidad de cuatro estrategias para identificar hogares pobres para recibir exenciones “Premium”, dentro del esquema nacional de seguro médico de Ghana (ENSM): prueba de medios económicos (PME), prueba indirecta (“proxy”) de medios económicos (PIME), clasificación participativa de la riqueza (CPR) y direccionamiento geográfico (DG) en emplazamientos urbanos, rurales y semi-urbanos de Ghana.

Métodos:  Realizamos el estudio en 145-147 hogares por emplazamiento, utilizando la PME como prueba de referencia. Calculamos los costes totales incluyendo los costes de las encuestas en los hogares y el coste de los pagos “premium” para los pobres, eficiencia (coste por persona pobre identificada), equidad (número de pobres reales excluídos) y la viabilidad administrativa de implementarlo.

Resultados:  El coste de eximir a un individuo pobre estaba entre los US$15.87 y US$95.44; excluir a los pobres estaba en un rango del 0% al 73%. La PME era más eficiente y equitativo en emplazamientos rurales y urbanos con una baja incidencia de pobreza; el DG era eficiente y equitativo en emplazamientos semi-urbanos con una alta incidencia de pobreza. La PIME y la CPR eran menos equitativas e ineficientes, aunque viables en algunos emplazamientos.

Conclusión:  Recomendamos la PME como la estrategia óptima en emplazamientos urbanos y rurales con baja incidencia de pobreza y el el DG como estrategia óptima en emplazamientos semi-urbanos muy pobres. El estudio es relevante para otros programas sociales y de desarrollo que requieren de la identificación y exención de los pobres en países de baja renta.


In 2004, Ghana introduced the National Health Insurance Scheme (NHIS) as part of the nation’s policy objective to minimize out of pocket health expenditure at point of service and to ensure equitable access to health care, particularly for the poor. The NHIS by law exempts certain categories of the population (younger than 18 years and older than 69 years). The policy also stipulates premium exemptions for the core poor (indigents) between the ages of 18 and 69 (Ghana 2003, 2004). Identifying and exempting the indigent have remained a challenge with respect to the strategy to adopt (Stierle et al. 1999; Coady et al. 2003), and the strategy that accurately identifies all poor individuals (maximizing equity) at the lowest cost (maximizing efficiency) is preferable.

We considered four strategies to identify the poor, based on a recent review (Jehu-Appiah et al. 2010), and evaluated these strategies in terms of their equity and efficiency. First, means testing (MT), which identifies poor households or individuals on the basis of an income or expenditure threshold, was recognized as the gold standard strategy for this study as it accurately identifies income poverty and the study is concerned with ability to pay a premium. MT is costly and administratively complex as it requires collection of detailed household consumption expenditure (Grosh 1992; Deaton 1997; Coady & Parker 2005; Lindert 2005). Second, proxy means testing (PMT) identifies the poor based on the indicators that correlate with household socio-economic status (SES) such as education, housing characteristics and ownership of durable assets (Montgomery et al. 2000; Filmer & Pritchett 2001; Ahmed & Bouis 2002; Sahn & Stifel 2003; Johannsen 2006; Vyas & Kumaranayake 2006; Booysen et al. 2008). Third, in participatory wealth ranking (PWR), community representatives identify and rank households into socio-economic categories based on acknowledged indicators in a group discussion (Cambers 1999; Bigman et al. 2000; Simanowitz 2000; Laderchi 2001; Feulfack & Zeller 2005; Van Campenhout 2006; Hargreaves et al. 2007; Collins 2009; Ridde et al. 2010a,b). Fourth, geographic targeting (GT) classifies areas or regions into poverty clusters based on the aggregate poverty indicators (Baker & Grosh 1994; Hentschel et al. 2000; Minot 2000; Coulombe 2005; Elbers et al. 2007).

This study follows up on an empirical study that assessed the effectiveness (inclusion and exclusion errors) of these strategies in urban, rural and semi-urban settings with differing poverty incidence in the central region of Ghana (Aryeetey et al. 2010). The study puts these results in a broader context by assessing total costs (including survey costs and costs of premium paid to the poor and non-poor), efficiency (cost per poor identified) and equity (number of poor excluded) of the strategies. We evaluated the feasibility of implementation of the various strategies reflecting their administrative complexities (required skills and capacity to conduct household surveys, community reception of interviewers, and feasibility of community discussions in the wealth ranking process). The research question is: ‘How do various strategies – MT, PMT, PWR or GT – perform in terms of efficiency and equity to identify the poor for premium exemptions in Ghana’s NHIS, and which strategy is preferable?’


Selection of study setting

We evaluated all strategies in different socio-economic settings (defined by level of poverty and urbanization) as anecdotal evidence suggests that efficiency, equity, and feasibility of strategies might differ across these settings. First, based on the most recent poverty incidence data for Ghana (Coulombe 2005), we selected the poorest district (where 63% of the population live below the income poverty line of GH¢ 370 or US$264 per year per household), the richest district (where 26% of the population live below the income poverty line) in the central Region of Ghana, and the region’s single metropolitan district (where 27% of the population live below the income poverty line). Then, using Ghana’s 2000 population census data classification of rural, urban and semi-urban enumeration areas (EA), we randomly selected a semi-urban EA in the poorest district and a rural EA in the richest district and an urban EA in the metropolitan district. Lastly, we randomly selected 146 households in the urban, 147 households in the rural, and 146 households in the semi-urban settings.

Data source

We conducted the household survey in the central region of Ghana in June 2009. The household questionnaire contained in total 257 questions of which 91 (35%) related to MT and 65 (25%) related to PMT. The rest were questions that related to household health, health insurance enrolment and perceptions of the health insurance scheme. The MT questions covered detailed household monthly consumption (food and non-food) expenditures; PMT questions covered ownership of durable assets, dwelling and housing conditions. The data sources for PWR were based on the indicators of poverty identified and discussed by selected community informants who participated in a wealth ranking exercise. Up to 17 key informants, 10 men and 7 women, participated in the wealth ranking exercise in each setting. We employed interviewers and facilitators who had skills in conducting household surveys and facilitators with experiences in focus group discussions. They were further trained to familiarize themselves with the objectives of our study. By the principle of GT, no surveys were conducted as the entire population was to be exempted from premium payments. For more details on data collection, see Aryeetey et al. (2010).

Data analysis

For MT, we estimated household wealth through monthly consumption expenditures. Following the definitions in the 2005 Ghana Living Standard Survey (GLSS V), we defined households to be poor in case their expenditures are below GH¢ 370 per year (Ghana Statistical Service, 2007). For PMT, we estimated households’ SES index to rank them into poverty quintiles. We first selected household characteristics (such as assets) that were significantly correlated with consumption expenditures, and these were considered as proxies for household wealth. We then used principal component analysis (PCA) to estimate a household SES score. PCA is a statistical procedure to determine weights for a linear index of a set of variables (Filmer & Pritchett 2001; McKenzie 2005; Vyas & Kumaranayake 2006). The household SES score was calculated as the sum of the weight of variables multiplied by their corresponding values. Next, households were ranked into wealth quintiles based on their SES score. We considered the bottom 40% of these households to be poor (for more detail, see Aryeetey et al. 2010). In PWR, we counted how often a certain household was ranked in each wealth category and subsequently classified the household into the wealth quintile it was most frequently ranked in. We repeated this procedure for all households. Households in the two lowest quintiles (‘very poor’ and ‘poor’) were considered as poor. In GT, we defined the number of poor households in each setting based on the estimated district poverty incidence (maps) by Coulombe (2005), using the same monthly per capita expenditure threshold as applied in MT.

Estimates of costs, efficiency and equity

We calculated the time and survey costs of MT, PMT and PWR. As the survey questionnaire contained more questions other than for MT and PMT, we deduced from the questionnaire the specific interview time, travel time and interview days for MT and PMT to estimate their true survey costs. In PWR, we recorded the time spent for community discussions and wealth ranking. The survey costs consisted of salaries for interviewers and facilitators, transport costs, cost of stationery, data entry and other specific costs incurred in each setting (Tables 1 and 2). For GT, survey costs were zero, and cost estimations were related to only the costs of premium exemptions. In addition, we estimated the efficiency of the strategies as the cost per poor person identified or the cost of exempting one poor person from paying insurance premium (Besley & Kanbur 1993; Houssou & Zeller 2010).

Table 1. Time allocation to strategies to identify the poor in three settings
  1. MT, means testing; PMT, proxy means testing; PWR, participatory wealth ranking.

  2. fInterviewers spent 8 h a day to conduct interview and travel (time spent to make appointments, wait for respondents and all other engagements apart from interview).

  3. gThe average time spent by interviewer to complete the entire questionnaire.

  4. hThe interview time was estimated as the proportion of questions required by MT/PMT out of the total questionnaire multiplied by the average time spent by the interviewer to complete the entire questionnaire.

  5. iWe maintained the same travel time for MT/PMT, which was calculated from the travel time for the entire survey questionnaire.

  6. jTotal interview time per day was calculated as the number of interviews completed multiplied by the interview time.

  7. kTotal travel time per day also calculated as the number of interviews completed multiplied by travel time per interviewer.

  8. lNumber of interviews completed a day by MT/PMT was estimated as number of hours divided by the sum of interview time and travel time.

  9. mInterview days for MT/PMT were estimated as the total number of household questionnaires divided by number of interviews completed in a day.

  10. nFor MT/PMT, the average time per household interview was calculated based on the interview days and number of households. For PWR, it was the average time spent to rank households based on the number of days spent in the setting by the number of households ranked.

Number of households (a)146146146147147147145145145
Number of interviewers (b)553553553
Number of days spent per community (c)773774774
Total number of questions in household questionnaire (d)257257257257257257
Number of questions required for MT/PMT (e)916591659165
Interviewer’s working hours per day including travel hours (f)8:008:008:008:008:008:008:008:008:00
Interview time per interviewer in hours (all questions) (g)0:500:500:530:530:540:54
Interview time per interviewer (relevant questions) (h) = [(e)/(d)] × (g)0:170:120:190:130:180:13
Travel time per interviewer in hours (i) = (k)/(l)1:051:056:201:061:064:101:071:075:25
Total interview time per interviewer per day (j) = (h) × (l)1:421:181:471:221:451:20
Total travel time per interviewer per day (k) = (i) × (l)6:176:416:126:376:146:39
Number of interviews completed in a day (l) = (f)/[(i) + (h)]
Number of interview days (m) = (a)/(l)25.1823.64926.0724.41225.9124.2912
Time per household interview in hours (n) = (m)/(a)4:083:531:284:153:581:574:174:011:59
Total time for discussion and wealth ranking in hours (PWR only) (o)7:456:507:10
Table 2. Survey costs (US$) of strategies to identify the poor in three settings
Survey costsUrbanRuralSemi-urban
  1. MT, means testing; PMT, proxy means testing; PWR, participatory wealth ranking.

  2. aInterviewers for MT/PMT and facilitators for PWR were paid daily wages of US$30.

  3. bThis included the cost of hiring vehicle, fuel and payment of drivers.

  4. cThe daily wage multiplied by number of interview days (recorded in Table 1).

  5. dDaily cost of transport multiplied by number of interview days.

  6. e,f,gFor MT/PMT, this was 35%/25% of the total cost of stationery, printing of questionnaire and data entry, respectively. For PWR, we recorded the actual cost of purchase of stationery materials for the three settings.

  7. hCost of community entry included tokens given to community for permission to undertake survey or wealth ranking in the community, payment for announcements and appreciation to a community volunteer for his/her assistance whenever necessary.

  8. iKey informants who participated in the PWR exercise were appreciated with some snacks, soaps, etc., for their time spent.

Daily wage (a)30.0030.0030.0030.0030.0030.0030.0030.0030.00
Daily cost of transport (b)70.3970.3970.3970.3970.3970.3970.3970.3970.39
Interviewer’s/facilitator’s salary (c)755.30709.14270.00782.05731.85360.00777.26728.67360.00
Transport (d)354.44332.78211.17366.99343.43211.17364.74341.94281.56
Stationery (e)11.258.0437.9011.258.0437.9011.258.0437.90
Printing of questionnaires (f)116.6683.33 116.6683.33 116.6683.33 
Data entry (g)100.0071.4035.00100.0071.4035.00100.0071.4035.00
Community entry (h)11.7011.7011.7011.4011.4011.405.705.705.70
Participants appreciation (i)  35.70  35.70  42.90
Total Survey Cost (j)1349.341216.39601.471388.341249.45691.171375.611239.07763.06

We defined the equity profile of the various strategies in comparison with MT and calculated its indicator by the proportion of poor persons exempted out of the total number of poor (as defined by MT) eligible for exemption. This calculation was facilitated by constructing Venn diagrams to identify households that were consistently identified as being poor across strategies (Aryeetey et al. 2010).


Tables 1–3 illustrate the successive steps undertaken in estimating costs, equity and efficiency of our strategies in the analysis. The variables used in calculating MT and PMT were collected from the same household questionnaire. Thus, to obtain the separate cost for each strategy, we deduced the time allocations (Table 1) in terms of the number of hours required to complete a questionnaire and the interview days if it would only include questions related to MT or PMT. The number of hours spent for PWR was recorded directly from the community meetings. In the urban setting, for example, the number of interview days with the employment of five interviewers was estimated at 25.18 and 23.64 for MT and PMT, respectively. Further details of the procedure of time allocation are provided at the bottom of Table 1.

Table 3. Coverage, costs and indicators of efficiency and equity of strategies to identify the poor in three settings
SettingStrategyPopulation (a)Population eligible for exemptions (b)Poverty incidence (c)Error of exclusion (d)Error of inclusion (e)Number of poor individuals exempted (f) = (a) × (b) × (c) × (1−d)Number of non-poor individuals exempted (g) = (a) × (b) × (e)Total number of individuals exempted (h) = (f) + (g)Number of poor excluded (equity indicator) (i) = (a) × (b) × (c) × (d)
Costs US$
SettingStrategySurvey costs (j)Cost of premium exemptions (k) = (h) × ϕTotal Cost (l) = (j) + (k)Cost per poor person identified (efficiency indicator) (m) = (l)/(f)
  1. MT, means testing; PMT, proxy means testing; PWR, participatory wealth ranking; GT, geographic targeting.

  2. aThe total number of individuals recorded in the household survey represented the population in each setting.

  3. bThe percentage of population eligible for exemptions were individuals between the ages of 18 and 69.

  4. cWe used the district-based poverty incidence estimated by Coulombe (2005), that related to each setting.

  5. d,eThe errors of inclusion and exclusion were those estimated from our previous empirical study of accuracy of strategies for the same study setting.

  6. kThe cost of health insurance premium is US$10, denoted as ϕ, based on the minimum amount that the poor pay as premium.


Table 2 reports the total survey costs for each strategy and setting. They comprised the interviewers and facilitators’ salaries, transport, print and stationary, data entry and other relevant costs incurred during the survey. Total survey costs across the three settings for MT ranged between US$1349 and US$1388. For PMT, total survey costs were between US$1216 and US$1249. PWR recorded the lowest costs between US$601 and US$763.

Table 3 shows the results of the relationship between equity and efficiency estimates. The top half of the table reports the results of equity analysis. In our analyses, we excluded persons below the ages of 18 years and above 70 years as these are exempted by law (Ghana 2003). Based on the poverty incidence, errors of exclusion and inclusion per setting, we estimated the total number of poor individuals identified by each strategy for exemptions and the numbers of poor excluded and included (columns f, g and h, respectively). Our equity indicator (column i) is calculated as the proportion of the number of poor identified for premium exemption out of the total of number of poor eligible for exemption. The bottom half of Table 3 reports the total cost of exemptions (column l) which is the sum of survey cost and cost of premium (equal to US$10). The cost of premium exemptions is then calculated as the total number of individuals exempted multiplied by the premium (column k). Our efficiency indicator, the cost per poor person identified, is calculated as the total cost divided by number of poor individuals exempted (column f).

In both urban and rural settings with low-poverty incidence, GT was the most expensive strategy as it exempted all individuals and hence incurred large costs of premium exemptions. GT was equitable as all poor individuals were exempted. PMT and PWR incurred significant survey costs and were costly in terms of paying premium to the non-poor because of relatively large errors of inclusion (in the urban setting, PMT: 36%, PWR: 50%; in the rural setting, PMT: 21%, PWR: 17%) – both strategies were therefore inefficient. Because both strategies also had relatively large errors of exclusion (in the urban setting, PMT: 36%, PWR: 50%; in the rural setting, PMT: 53%, PWR: 73%), they excluded many poor, rendering both strategies not equitable. MT, although having highest survey costs, was the least costly strategy because it incurred no premium exemptions for the non-poor, rendering it the most efficient strategy. It was also equitable as no poor individual was excluded.

In the semi-urban setting with high-poverty incidence, GT included relatively few non-poor (37% compared to 73% in the urban and 74% in the rural setting) rendering it more efficient. Again, PMT and PWR faced survey costs and had large errors of inclusion (27% and 60%, respectively, increasing the cost of inclusion of the non-poor) and therefore high cost per poor person identified. PMT excluded many poor (46%) rendering it inequitable. In contrast, PWR excluded few poor (3%) and was equitable. MT incurred large survey costs, but – by definition – neither excluded poor nor included non-poor individuals. It thus performed well in equity and efficiency.

Table 3 illustrates that there is no equity–efficiency trade-off: in urban, rural and semi-urban settings, both MT and GT are most equitable and MT is most efficient – MT is then the strategy of choice if only equity and efficiency considerations are taken into account. In the semi-urban setting, both MT and GT are most equitable, and GT is most efficient – GT is then the strategy of choice if only equity and efficiency considerations are taken into account.

With reference to feasibility of implementation, experience from our field work revealed that in general, interviewers were welcomed into the communities without difficulty. However, in the urban setting, PWR facilitators reported that some of the informants were reluctant to participate while others found it difficult to rank their fellow households, possibly because of the low level of social capital making it difficult to know the SES of all households.


For decades, the development of targeting mechanisms to identify potential beneficiaries, particularly the poor, in social welfare programmes has received much attention in policies of developing countries. The argument for targeting is that resources are best allocated to those who need them most. Our study evaluates different strategies to identify the poor for premium exemptions in Ghana’s NHIS, in different socio-economic settings. We propose a simple decision framework on the choice of optimal strategy, based on three criteria: equity (numbers of poor excluded), efficiency (cost per poor person identified) and feasibility of implementation (Table 4). We interpret its findings by the poverty incidence per setting.

Table 4. Decision framework on optimal strategy to identify the poor on the basis of efficiency, equity and feasibility
SettingPoverty incidenceStrategyEfficiencyEquityFeasibilityRecommended strategy
  1. MT, means testing; PMT, proxy means testing; PWR, participatory wealth ranking; GT, geographic targeting.

  2. −, worst performance; 0, weak performance; +, good performance; ++, excellent performance.


We recognize MT in relatively low-poverty areas being the most efficient and equitable in our two low-poverty settings. MT has been applied in many studies as a strategy to target the poor for various social programmes (Grosh 1992; Willis & Leighton 1995; Coady et al. 2003; Hernandez et al. 2007). In some MT programmes such as the Colombia Student Loan Program and Honduras Food Stamps for Female Headed Households, for example, the annual cost per identified beneficiary was US$700 and US$40 within a beneficiary population of 48 000 and 125 000, respectively (Betancur-Mejia 1990; Franklin 1990; Ballenger & Courtney 1991), which is similar to what we found in our study. The potential drawback of MT is that it requires highly skilled administrative capacity to ensure that accurate data are obtained – a capacity that is insufficient in many developing countries including Ghana and makes the feasibility of MT a challenge. PMT may be considered as an alternative to MT in settings where administrative capacity is limited. Many social development programmes that require identification of poor beneficiaries have applied PMT as the targeting mechanism notwithstanding the reported high exclusion errors. In a comparative study of five Latin American countries on various social targeting programmes, the exclusion errors ranged between 26% and 84%. (Castaneda 2005; Castaneda et al. 2005; De la Briere & Lindert 2005; Dutrey 2007). Proponents of PMT argue that PMT is relatively simple to administer and does not require huge administrative burden and skilled capacity in comparison with MT (Montgomery et al. 2000; Filmer & Pritchett 2001; McKenzie 2005).

We recognize GT as an optimal strategy in relatively high-poverty incidence settings, because the cost of including the non-poor is then less than survey costs of MT, PMT or PWR. GT is only feasible in settings where accurate poverty incidence data have been estimated. At the present poverty incidence of 63% in the semi-urban setting, GT is more efficient than MT. In general, the definition of the poverty incidence ‘threshold’ above which GT is the most efficient strategy in any particular setting depends on population numbers, survey costs of the other strategies and their inclusion and exclusion errors and can be mathematically calculated. GT has been widely used to design poverty maps for Ghana and to target the poor in other countries in sub-Saharan Africa (Bigman et al. 2000; Fofack 2000; Simler & Nhate 2003; Kraybill & Bashaasha 2006).

For the purposes of large-scale policy intervention, as in Ghana’s NHIS, where the focus is on exemption from payment of a fee rather than the more complex issues of stigma, social exclusion and marginalization were relativity matters; PWR may not be a useful strategy to adopt in comparison with the other strategies. PWR is rather subjective because definitions and perceptions of poverty are community specific even though poverty indicators are sometimes similar. Nonetheless, through the application of various participatory poverty assessment tools, the subjectivity of the poverty ranking results of key informants can be minimized, thereby curtailing PWRs disadvantages. The application of PWR is best in rural communities where people are likely to know the SES of their fellow community members. In Burkina Faso, for example, community-based targeting has been applied, and its feasibility tested to exempt the worst off from user fees in some selected rural communities (Ridde et al. 2010a,b; Soures et al. 2010). One study reported that ‘the community-based process minimized inclusion biases, as the people selected were poorer and more vulnerable than the rest of the population. However, there were significant exclusion biases; the selection was very restrictive because the waivers had to be endogenously funded’. They also identified the procedure’s emphasis on local solidarity and democracy that may limit its possible scale-up (Ridde et al. 2011, p. 6). Our study draws similar conclusions particularly on the limitation of using community-based targeting mechanism for scale-up to regional or national level.

A number of issues are important in the interpretation of the results. First, the study was conducted in one of the 10 regions of the country, which may not be representative of Ghana’s population. However, it is possible to repeat the procedure with data from a representative sample of the population. The GLSS, which is carried out on regular basis, includes the essential data requirements to estimate poverty incidence needed for GT and equity and efficiency needed for PMT. Second, our results were limited to analysis of low-poverty incidence urban and rural settings and a high-poverty semi-urban setting. The results might differ if high-poverty incidence urban and rural and low-poverty incidence semi-urban were included in the analysis. Third, we did not include community time costs in our analysis because of the difficulties in measuring such costs (Dutrey 2007). Because MT, PMT and PWR rely on community time inputs, including community costs would render these strategies more expensive in comparison with GT. Fourth, in our costs estimates, we also excluded the psychological and social cost associated with applying for and receiving state support, economic losses because of disincentive effects and any loss of political support for the programme (Grosh 1994; Gwatkin 2000; Smith & Subbarao 2003). These costs are difficult to quantify and are best considered qualitatively.

In conclusion, the equity, efficiency and feasibility of different strategies to identify the poor for exemption from fee payments vary somewhat depending on the socio-economic setting. The ability to administer is also an important consideration. Generally, where the incidence of poverty is high, GT is likely to be the best approach. In lower-poverty incidence settings, MT may be the best approach with the caveat that the ability to administer this more technically challenging approach must be there. Failing this, PMT will be a better strategy. PWR is of limited value where the reason for identifying the poor is for a clear straightforward decision related to exemptions from fee payment. However, where more complex programmes that require consideration of poverty from a relativist angle because of issues of marginalization, stigma, etc., PWR is worth considering.

This study was carried out as a follow-up on review and empirical studies on strategies to identify the poor for premium exemptions in Ghana’s NHIS. It holds relevance to other social and developmental programmes that require identification and exemptions of the poor in low-income countries.