- Split View
-
Views
-
Cite
Cite
Ayako Honda, Mandy Ryan, Robert van Niekerk, Diane McIntyre, Improving the public health sector in South Africa: eliciting public preferences using a discrete choice experiment, Health Policy and Planning, Volume 30, Issue 5, June 2015, Pages 600–611, https://doi.org/10.1093/heapol/czu038
- Share Icon Share
Abstract
Background The introduction of national health insurance (NHI), aimed at achieving universal coverage, is the most important issue currently on the South African health policy agenda. Improvement in public sector health-care provision is crucial for the successful implementation of NHI as, regardless of whether health-care services become more affordable and available, if the quality of the services provided is not acceptable, people will not use the services. Although there has been criticism of the quality of public sector health services, limited research is available to identify what communities regard as the greatest problems with the services.
Methods A discrete choice experiment (DCE) was undertaken to elicit public preferences on key dimensions of quality of care when selecting public health facilities in South Africa. Qualitative methods were applied to establish attributes and levels for the DCE. To elicit preferences, interviews with community members were held in two South African provinces: 491 in Western Cape and 499 in Eastern Cape.
Results The availability of necessary medicine at health facilities has the greatest impact on the probability of attending public health facilities. Other clinical quality attributes (i.e. provision of expert advice and provision of a thorough examination) are more valued than non-clinical quality of care attributes (i.e. staff attitude, treatment by doctors or nurses, and waiting time). Treatment by a doctor was less valued than all other attributes.
Conclusion Communities are prepared to tolerate public sector health service characteristics such as a long waiting time, poor staff attitudes and lack of direct access to doctors if they receive the medicine they need, a thorough examination and a clear explanation of the diagnosis and prescribed treatment from health professionals. These findings prioritize issues that the South African government must address in order to meet their commitment to improve public sector health-care service provision.
Using NHI to achieve universal coverage in South Africa requires consideration of community views on health-care provision.
This DCE study revealed that there is a general community preference not to seek care at public health facilities.
Availability of appropriate medicine at health facilities has the greatest impact on health facility choice.
Clinical quality of care attributes (i.e. expert advice and a thorough examination) are more valued than other attributes.
Introduction
Universal coverage (UC) is high on the international health policy agenda, with specific mention in the post Millennium Development Goals and resolutions calling for UC adopted in the World Health Assembly and the United Nations General Assembly during 2012 (McIntyre et al. 2013). The South African government has committed to pursuing National Health Insurance (NHI) in an attempt to achieve UC. The NHI is likely to be funded through a combination of general tax revenue and additional dedicated personal income taxes and payroll taxes on employers, with no out-of-pocket payments required at the point of service delivery. These funds will be pooled in a NHI fund (NHIF) and used to purchase a uniform, relatively comprehensive package of services for all South Africans (Department of Health 2011b).
In August 2011, the Government released a Green Paper outlining the broad approach to the proposed NHI. The Green Paper proposes a number of reforms which are elements of an overall UC reform agenda (Department of Health 2011b). The reforms are critical to the delivery of high quality, accessible public sector health services for the universal health system.
Improvement in the provision of public sector health care is crucial for the successful implementation of NHI as, regardless of whether health-care services become more affordable and available, if the quality of services is unacceptable, people will not use them. The Green Paper recommends that the next five years are devoted to undertaking a number of specific reforms to rebuild the public health sector and improving quality of and access to care, especially at the primary health-care (PHC) level.
Although NHI and the related reforms have significant implications for the entire health system, there has been little or no opportunity for public engagement in the development of the health policy. In addition, in spite of criticism of public sector health service quality, limited research is available on what communities regard as the greatest problems with the services. Consequently, there is an urgent need to incorporate a mechanism for obtaining public opinion in South Africa on health system changes and feed these views into the NHI policy development process.
It is within this context that we undertook a discrete choice experiment (DCE) study with the aim of better understanding public preferences on quality of care and identifying priority areas for reform in the delivery of public health care.
DCEs are a quantitative method to elicit preferences. The technique is based on Lancaster’s theory of value (Lancaster 1966) and assumes that individual decisions about a good or service are determined by the attributes or characteristics of that good or service (Ryan et al. 2008). The theoretical basis for the analysis of DCE data is Random Utility Theory (McFadden 1974, 1986). This theory states that whilst an individual knows the nature of their utility function, the researcher cannot observe this. Thus the utility function is modelled as two components: a systematic (explainable) component and a random (unexplainable) component. The systematic component of the utility function, which the researcher can estimate, is used in this research to provide quantifiable information on the relative importance of dimensions of health-care quality in choice of health facility, the trade-offs between these dimensions (with willingness to pay, a monetary measure of value, being estimated given that a price proxy was included). Given the random component, DCE response data is analysed within a probabilistic framework, and the model can be used to predict how choices change in response to changes in attributes. In this paper we predict the probability of take-up of different defined health facilities. Such information is useful in guiding planning for improved public health-care service provision.
Methods
Establishment of attributes and levels
Qualitative methods have been recognized as useful for defining attributes and levels (Coast et al. 2012), developing questionnaires and improving the validity of DCE studies (De Bekker-Grob et al. 2012). In addition to a review of published and grey literature, focus group discussions (FGDs) were held with community members to identify and define the attributes and levels. Capitalizing on a series of public consultation workshops undertaken in 2010 in South Africa’s nine provinces and involving a range of civil society organizations, FGDs were organized at workshops held in Western Cape, North-West and Mpumalanga. To supplement these results, community-based FGDs were held in Western Cape and Eastern Cape.
A review of literature informed the generation of topics to encourage FGD participants to share their views on health-care provision, particularly factors considered important when accessing public facilities. Group discussions at FGDs were facilitated by a local non-government organization (NGO) staff member with appropriate expertise and local language skills.
The workshop-based FGDs each comprised approximately 10 participants, recruited from the consultation workshops. The community-based FGD in Western Cape was held in Mitchells Plain, a relatively poor socio-economic area, where 11 participants were recruited either from the community health centre or the area adjoining the library. In the Eastern Cape province, the focus group comprised 20 participants recruited from community-based organization representatives attending a workshop organized by a local NGO.
With participant consent, the discussion was tape recorded. The recording was transcribed verbatim into a MS-Word document and transferred to NVivo 7 for coding and categorization. Initially, the transcripts from the five FGDs were analysed independently to examine themes and allow comparison between settings. Subsequently, results were combined to identify common themes across FGD transcripts.
A two-day workshop was held with fieldwork co-ordinators and interviewers to present the results from the five FGDs and determine the final set of attributes and levels. The themes most commonly discussed at the FGDs were chosen for development as attributes. The themes were further grouped as those relating to the clinical aspects of health-care service provision, which involves the act of treating patients (availability of medicine, provision of a thorough examination and expert advice) and non-clinical aspects (staff attitude, treatment by doctors or nurses and waiting time). In addition, as geographical access to health services was raised as a major concern during the FGDs, transportation costs were included as an attribute. Inclusion of this price proxy allowed estimation of monetary measures of value for all the non-monetary attributes (Ryan et al. 2008).
After selection of attributes from the themes identified as important quality factors, independence and inter-relationships in themes were considered along with relevance to the country’s health system reform policy. The final list of attributes and attribute levels are shown in Table 1.
Attribute . | Levels . | Description . |
---|---|---|
Transportation costs | 40 | Transport to and from the health facility costs about R40 |
20 | Transport to and from the health facility costs about R20 | |
10 | Transport to and from the health facility costs about R10 | |
5 | Transport to and from the health facility costs about R5 | |
Staff attitudes | 0 | The staff at the health facility do not treat me with respect |
1 | The staff at the health facility treat me with respect | |
Examination | 0 | The staff at the health facility do not examine me |
1 | The staff at the health facility examine me thoroughly | |
Expert advice | 0 | The staff at the health facility do not explain what is wrong with me or what I need to do to get better |
1 | The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |
Availability of medicine | 0 | When I go to the health facility, they don’t have the medicine I need and I go away without any medicine |
1 | When I go to the health facility, I get the medicine I need | |
Treatment by doctors or nurses | 0 | When I go to the health facility, I first see a nurse who is trained to treat most illnesses and only see a doctor if the nurse cannot treat my illness |
1 | When I go to the facility, I always see a doctor | |
Waiting time | 0 | I spend a whole day in the health facility before I go home |
1 | I spend about half a day in the health facility before I go home |
Attribute . | Levels . | Description . |
---|---|---|
Transportation costs | 40 | Transport to and from the health facility costs about R40 |
20 | Transport to and from the health facility costs about R20 | |
10 | Transport to and from the health facility costs about R10 | |
5 | Transport to and from the health facility costs about R5 | |
Staff attitudes | 0 | The staff at the health facility do not treat me with respect |
1 | The staff at the health facility treat me with respect | |
Examination | 0 | The staff at the health facility do not examine me |
1 | The staff at the health facility examine me thoroughly | |
Expert advice | 0 | The staff at the health facility do not explain what is wrong with me or what I need to do to get better |
1 | The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |
Availability of medicine | 0 | When I go to the health facility, they don’t have the medicine I need and I go away without any medicine |
1 | When I go to the health facility, I get the medicine I need | |
Treatment by doctors or nurses | 0 | When I go to the health facility, I first see a nurse who is trained to treat most illnesses and only see a doctor if the nurse cannot treat my illness |
1 | When I go to the facility, I always see a doctor | |
Waiting time | 0 | I spend a whole day in the health facility before I go home |
1 | I spend about half a day in the health facility before I go home |
Attribute . | Levels . | Description . |
---|---|---|
Transportation costs | 40 | Transport to and from the health facility costs about R40 |
20 | Transport to and from the health facility costs about R20 | |
10 | Transport to and from the health facility costs about R10 | |
5 | Transport to and from the health facility costs about R5 | |
Staff attitudes | 0 | The staff at the health facility do not treat me with respect |
1 | The staff at the health facility treat me with respect | |
Examination | 0 | The staff at the health facility do not examine me |
1 | The staff at the health facility examine me thoroughly | |
Expert advice | 0 | The staff at the health facility do not explain what is wrong with me or what I need to do to get better |
1 | The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |
Availability of medicine | 0 | When I go to the health facility, they don’t have the medicine I need and I go away without any medicine |
1 | When I go to the health facility, I get the medicine I need | |
Treatment by doctors or nurses | 0 | When I go to the health facility, I first see a nurse who is trained to treat most illnesses and only see a doctor if the nurse cannot treat my illness |
1 | When I go to the facility, I always see a doctor | |
Waiting time | 0 | I spend a whole day in the health facility before I go home |
1 | I spend about half a day in the health facility before I go home |
Attribute . | Levels . | Description . |
---|---|---|
Transportation costs | 40 | Transport to and from the health facility costs about R40 |
20 | Transport to and from the health facility costs about R20 | |
10 | Transport to and from the health facility costs about R10 | |
5 | Transport to and from the health facility costs about R5 | |
Staff attitudes | 0 | The staff at the health facility do not treat me with respect |
1 | The staff at the health facility treat me with respect | |
Examination | 0 | The staff at the health facility do not examine me |
1 | The staff at the health facility examine me thoroughly | |
Expert advice | 0 | The staff at the health facility do not explain what is wrong with me or what I need to do to get better |
1 | The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |
Availability of medicine | 0 | When I go to the health facility, they don’t have the medicine I need and I go away without any medicine |
1 | When I go to the health facility, I get the medicine I need | |
Treatment by doctors or nurses | 0 | When I go to the health facility, I first see a nurse who is trained to treat most illnesses and only see a doctor if the nurse cannot treat my illness |
1 | When I go to the facility, I always see a doctor | |
Waiting time | 0 | I spend a whole day in the health facility before I go home |
1 | I spend about half a day in the health facility before I go home |
Design of choice sets
SAS software was used to create a D-efficient fractional factorial design (Kuhfeld 2010), resulting in 16 binary choices. A binary choice format was employed in which, for each choice, respondents were asked: ‘Would you choose the health facility, yes or no?’ This binary format was considered appropriate as a large proportion of the target sample were likely to have limited education and/or survey experience so clarity and simplicity in the survey design would allow respondents to adequately consider the choice tasks. The appropriateness of a binary choice design was further confirmed by experienced fieldworkers who spent considerable time reviewing the choice formats and concluded that the binary choice design was most appropriate to produce a manageable questionnaire. The fieldworkers concluded that a multiple choice design, which asked respondents to answer multiple questions comparing two facilities with seven attributes each, would be difficult given the likely formal education level and survey experience of the sample. In addition, many of the survey areas had limited health facilities, meaning people have little or no experience in choice of health facility, and therefore, the binary choice design better reflected the actual situation for respondents. Consequently, while recognizing that multiple choice design is more commonly used in DCE studies, this study employed a binary choice design.
The choice sets also included two warm-up questions, where respondents were introduced to the choice tasks, and two holdout questions, used to examine whether the model correctly predicted responses. Consequently, respondents were presented with 20 binary choices. Pictorial representations of the choice sets were used to help respondents understand the choice tasks, as demonstrated in Figure 1.
Generation and pre-testing of the questionnaire
In addition to the choice questions, the questionnaire collected information on the demographic and socio-economic characteristics of respondents. The questionnaire was prepared in English and translated into the languages spoken at the study sites (Afrikaans and Xhosa). The translated questionnaires were independently back-translated to validate the translation quality.
The questionnaire was piloted on 20 respondents in Western Cape and 27 respondents in Eastern Cape, in both urban and rural settings. Theoretical validity was confirmed, with all attributes, except treatment by doctors and nurses, being statistically significant, and ‘good’ levels having positive coefficients. Following collection of information on maximum acceptable transportation costs, levels were lowered to adjust to local contexts and ensure the range of levels were broad enough to be relevant to all respondents. Upon completion of the pilot-testing, the interviewers reviewed the selection, definition and levels of attributes, respondents’ understanding of the task, the ease of comprehension of the questions and the manageability of the number of choice sets. The review led to one pictorial representation being revised.
Data collection
Given limited resources a national household survey was not feasible. An attempt was made to select a target sample that reflected the characteristics of South Africans who use public health services. The survey was therefore undertaken in two provinces displaying variations in health status, health-care resources (financial and health workforce), access to basic facilities and socio-economic contexts. In each province, nine sub-districts were selected: three local government administrative centres of a larger district and six selected on the basis of contextual factors such as population size, socio-economic status (SES) and health status and distribution of health facilities.
A sample of approximately 50 randomly-selected households per sub-district was surveyed (approximately 150 households in each district and 450 in each province). This provided a sufficiently large sample to undertake meaningful statistical analysis by sub-group (De Bekker-Grob et al. 2012). Household selection was undertaken using information from sub-district street or aerial maps of households, with the sampling interval calculated according to the number of households in the sub-district. High-income residential areas were excluded from the survey sites as high-income households were assumed to have limited experience in accessing public health-care services, particularly at the PHC level. In South Africa, the use of private sector PHC services is more driven by a households’ SES than private health insurance membership status because private health insurance usually does not cover visits to private health practitioners and those who uses private sector PHC services pay out-of-pocket or from their ‘personal savings account’ that exists within private insurance schemes.
Face-to-face household interview surveys were undertaken in Western Cape and Eastern Cape provinces. Prior to commencing the household surveys, each data collection team received an interview training session that involved a general overview of the study, a detailed review of the questionnaire, instruction on questionnaire administration and role-playing the questionnaire-based interview.
The study was explained to potential respondents using an information sheet and an introductory description of the choice questions in the questionnaire. The information sheet described the overall goal of the study (i.e. to identify areas of the public health sector most in need of improvement) and asked respondents to participate in the survey. The introductory description of the choice sets asked respondents to: imagine that they had been feeling ill for several days (with, for example, a headache, stomach problems, fever, etc.) and that, as their health had not improved, they were thinking of visiting a clinic or hospital to seek help to get better; and then to consider the 20 public health facilities (clinics or hospitals), each with a different mix of characteristics. They were then asked whether or not they would seek care at each of the facilities. At the beginning of the questionnaire, interviewers ensured that respondents were aware that they were making choices within the context of the public health sector by explaining the study was about improving public health service provision. Throughout the questionnaire survey, the interviewers and interviewees discussed experiences in public sector clinics and hospitals. Also, most of the areas visited by the field survey teams were served by public sector health providers and most of the respondents had accessed public sector health providers and/or were very familiar with public sector health providers through discussions with family and neighbours. The interviewers reported hearing many complaints about public sector health services and were often required to extend the time of interviews to accommodate the expression of discontent by interviewees. A small number of those visited by interviewers felt that they did not have sufficient experience in accessing public sector health services (and/or health services in general) and recommended that someone else in the family undertake the questionnaire or responded to the questionnaire by reflecting what they had heard about public health facilities from family members and/or neighbours.
Data were independently double-entered by two locally-trained data capturers and data checked for consistency using EpiInfo. The data was transferred to STATA for analysis.
Analysis of DCE data
All labels are defined in Table 3 and Ui is the utility, or benefit derived from attending a defined (i) public health facility; α is a constant term representing the general preference to attend a public health facility (compared with the alternatives of either attending a private facility or not seeking medical care). β2–β7 show the relative importance of the best levels for all categorical attributes of quality of care for public health facilities. Given that effects coding was used to analyse these attributes, the impact of the worst level is equal to the negative of the best level. It was hypothesized that all ‘good’ levels of the attributes have positive coefficients. β1, the coefficient for the transportation cost attribute (price proxy), was entered as a continuous variable and was expected to be negative, indicating that the respondents prefer lower transportation costs when accessing health-care services.
To estimate the trade-offs that respondents are willing to make between attributes, willingness to pay (WTP) for marginal improvements in attributes, and associated confidence intervals, were estimated for all attributes (Ryan et al. 2012). Willingness to pay indicates the monetary value of marginal improvements in quality attributes. WTP estimates were calculated as the ratio of the coefficient of interest to the negative of the coefficient on the cost attribute (i.e. transportation cost).
Sub-group analyses were undertaken to investigate the influence of demographic characteristics (age, gender and languages group), province (Western Cape and Eastern Cape) and SES on preferences. Wald statistics was used to test for differences across sub-groups.
The SES of respondent households was assessed using an index produced from principal component analysis of the 2008 National Income Dynamics Study (NIDS) data to establish a factor score for each asset indicator and using 24 proxy indicators: three relating to household ownership of consumer durables and 21 relating to household dwelling conditions (Filmer and Pritchett 2001; Vyas and Kumaranayake 2006). The factor scores established from the NIDS data were applied to the study sample to construct a SES score for each household in the study sample. The cut off points for four equal-sized groups constructed in the NIDS sample were then applied to categorize the study sample households into lower, lower-medium, upper-medium and higher SES groups. (The details of the assessment are available from the first author.)
Validity issues
Given that DCEs rely on responses to hypothetical scenarios, it is crucial to examine the validity of responses. This was assessed in a number of ways. Holdout questions were included in the choice sets to determine whether the model correctly predicted responses. Theoretical validity was examined by looking at whether the coefficients moved as anticipated. To assess the validity of the results, debriefing sessions were held with the provincial data collection teams to establish whether the key findings from the analysis resonated with the interviewers’ perceptions of the importance of attributes based on their engagement with respondents.
Results
Respondents
The characteristics of respondents are shown in Table 2. The survey included 990 respondents, with over two-thirds being female. Two factors may contribute to the skewed gender distribution: firstly, according to the data collectors, females were, in general, more willing to take part in the survey than males; secondly, in some areas it was not easy to find male respondents at home during daytime hours. However, it was necessary to schedule household visits during the daytime for security reasons. The language group distribution approximately reflects the actual distribution of language groups in the population in the two provinces, i.e. the majority of the Western Cape population speak Afrikaans and in Eastern Cape the majority speak Xhosa. The proportion of lower SES respondents was considerably lower than those of medium and higher SES respondents. This SES distribution pattern may have occurred because, for safety reasons, conducting the household survey in high-risk or extremely remote areas, where a high proportion of lower SES households tend to reside, was avoided. In addition, as mentioned in the methods section, the SES score was constructed by applying national survey data to the study sample. The use of national data to determine the upper and lower boundary values for each quintile may have contributed to the skewed SES distribution towards higher income groups. It should be noted that the higher SES group are not high income households in the absolute term. The SES classification indicates that they belong to the 1st quintile of the income groups present in South Africa.
Variables . | Western Cape . | Eastern Cape . | Total . |
---|---|---|---|
N | 491 | 499 | 990 |
District (%) | |||
Cape Town 169 (34) | Amathole 162 (32) | ||
Overberg 162 (33) | Cacadu/NMB 177 (35) | ||
Central Karoo 160 (33) | Ukhahlamba 160 (32) | ||
Sex (%) | |||
Male | 142 (29) | 165 (33) | 307 (31) |
Female | 349 (71) | 334 (67) | 683 (69) |
Age group (%) | |||
18–34 years old | 193 (39) | 181 (36) | 374 (38) |
35–49 years old | 150 (31) | 144 (29) | 294 (30) |
≥50 years old | 148 (30) | 174 (35) | 322 (33) |
Home language (%) | |||
IsiXhosa | 131 (27) | 382 (77) | 513 (52) |
Afrikaans | 343 (70) | 102 (20) | 445 (45) |
Other | 17 (3) | 15 (3) | 32 (3) |
SES (%) | |||
Higher | 222 (45) | 126 (25) | 348 (35) |
Upper medium | 241 (49) | 194 (39) | 435 (44) |
Lower medium | 28 (6) | 135 (27) | 163 (17) |
Lower | 0 (0) | 44 (9) | 44 (4) |
Variables . | Western Cape . | Eastern Cape . | Total . |
---|---|---|---|
N | 491 | 499 | 990 |
District (%) | |||
Cape Town 169 (34) | Amathole 162 (32) | ||
Overberg 162 (33) | Cacadu/NMB 177 (35) | ||
Central Karoo 160 (33) | Ukhahlamba 160 (32) | ||
Sex (%) | |||
Male | 142 (29) | 165 (33) | 307 (31) |
Female | 349 (71) | 334 (67) | 683 (69) |
Age group (%) | |||
18–34 years old | 193 (39) | 181 (36) | 374 (38) |
35–49 years old | 150 (31) | 144 (29) | 294 (30) |
≥50 years old | 148 (30) | 174 (35) | 322 (33) |
Home language (%) | |||
IsiXhosa | 131 (27) | 382 (77) | 513 (52) |
Afrikaans | 343 (70) | 102 (20) | 445 (45) |
Other | 17 (3) | 15 (3) | 32 (3) |
SES (%) | |||
Higher | 222 (45) | 126 (25) | 348 (35) |
Upper medium | 241 (49) | 194 (39) | 435 (44) |
Lower medium | 28 (6) | 135 (27) | 163 (17) |
Lower | 0 (0) | 44 (9) | 44 (4) |
Note: The SES of respondent households was assessed using an index produced from principal component analysis using proxy indicators relating to household ownership of consumer durables and household dwelling conditions. NMB, Nelson Mandela Bay.
Variables . | Western Cape . | Eastern Cape . | Total . |
---|---|---|---|
N | 491 | 499 | 990 |
District (%) | |||
Cape Town 169 (34) | Amathole 162 (32) | ||
Overberg 162 (33) | Cacadu/NMB 177 (35) | ||
Central Karoo 160 (33) | Ukhahlamba 160 (32) | ||
Sex (%) | |||
Male | 142 (29) | 165 (33) | 307 (31) |
Female | 349 (71) | 334 (67) | 683 (69) |
Age group (%) | |||
18–34 years old | 193 (39) | 181 (36) | 374 (38) |
35–49 years old | 150 (31) | 144 (29) | 294 (30) |
≥50 years old | 148 (30) | 174 (35) | 322 (33) |
Home language (%) | |||
IsiXhosa | 131 (27) | 382 (77) | 513 (52) |
Afrikaans | 343 (70) | 102 (20) | 445 (45) |
Other | 17 (3) | 15 (3) | 32 (3) |
SES (%) | |||
Higher | 222 (45) | 126 (25) | 348 (35) |
Upper medium | 241 (49) | 194 (39) | 435 (44) |
Lower medium | 28 (6) | 135 (27) | 163 (17) |
Lower | 0 (0) | 44 (9) | 44 (4) |
Variables . | Western Cape . | Eastern Cape . | Total . |
---|---|---|---|
N | 491 | 499 | 990 |
District (%) | |||
Cape Town 169 (34) | Amathole 162 (32) | ||
Overberg 162 (33) | Cacadu/NMB 177 (35) | ||
Central Karoo 160 (33) | Ukhahlamba 160 (32) | ||
Sex (%) | |||
Male | 142 (29) | 165 (33) | 307 (31) |
Female | 349 (71) | 334 (67) | 683 (69) |
Age group (%) | |||
18–34 years old | 193 (39) | 181 (36) | 374 (38) |
35–49 years old | 150 (31) | 144 (29) | 294 (30) |
≥50 years old | 148 (30) | 174 (35) | 322 (33) |
Home language (%) | |||
IsiXhosa | 131 (27) | 382 (77) | 513 (52) |
Afrikaans | 343 (70) | 102 (20) | 445 (45) |
Other | 17 (3) | 15 (3) | 32 (3) |
SES (%) | |||
Higher | 222 (45) | 126 (25) | 348 (35) |
Upper medium | 241 (49) | 194 (39) | 435 (44) |
Lower medium | 28 (6) | 135 (27) | 163 (17) |
Lower | 0 (0) | 44 (9) | 44 (4) |
Note: The SES of respondent households was assessed using an index produced from principal component analysis using proxy indicators relating to household ownership of consumer durables and household dwelling conditions. NMB, Nelson Mandela Bay.
Global analysis
The results from the regression analysis are shown in Table 3. All attributes were statistically significant, indicating that they impact on the probability of choosing a health facility. In addition, all ‘good’ levels of the attributes have positive coefficients, implying that they positively impact on the decision to use a health facility. As expected, the coefficient for the transportation cost attribute is significant and negative, indicating that higher transportation costs negatively impact on the decisions to use a health facility.
. | Beta (β) Variables . | Model 1 . | |
---|---|---|---|
Coefficients . | Standard Error . | ||
Transportation costs | Transport (β1) | −0.03*** | 0.00 |
Staff attitude | Attitude_good (β2) | 0.20*** | 0.02 |
The staff at the health facility treat me with respect | |||
Examination | Exam_given (β3) | 0.44*** | 0.02 |
The staff at the health facility examine me thoroughly | |||
Expert advice | Advice_given (β4) | 0.47*** | 0.02 |
The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |||
Availability of medicine | Med_avail (β5) | 1.20*** | 0.02 |
When I go to the health facility, I get the medicine I need | |||
Treatment by doctors or nurses | Tx_dr (β6) | 0.12*** | 0.02 |
When I go to the health facility, I always see a doctor | |||
Waiting time | Wait_short (β7) | 0.29*** | 0.02 |
I spend about half a day in the health facility before I go home | |||
Constant | Const (β8) | −0.15*** | 0.04 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −8072.2448 | ||
Wald chi-square | 2884.29 | ||
Prob > chi-square | 0.0000 |
. | Beta (β) Variables . | Model 1 . | |
---|---|---|---|
Coefficients . | Standard Error . | ||
Transportation costs | Transport (β1) | −0.03*** | 0.00 |
Staff attitude | Attitude_good (β2) | 0.20*** | 0.02 |
The staff at the health facility treat me with respect | |||
Examination | Exam_given (β3) | 0.44*** | 0.02 |
The staff at the health facility examine me thoroughly | |||
Expert advice | Advice_given (β4) | 0.47*** | 0.02 |
The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |||
Availability of medicine | Med_avail (β5) | 1.20*** | 0.02 |
When I go to the health facility, I get the medicine I need | |||
Treatment by doctors or nurses | Tx_dr (β6) | 0.12*** | 0.02 |
When I go to the health facility, I always see a doctor | |||
Waiting time | Wait_short (β7) | 0.29*** | 0.02 |
I spend about half a day in the health facility before I go home | |||
Constant | Const (β8) | −0.15*** | 0.04 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −8072.2448 | ||
Wald chi-square | 2884.29 | ||
Prob > chi-square | 0.0000 |
Note: Three choice questions were not answered and consequently removed from the analysis. Accordingly, there are only 15837 observations.
*** P < 0.001.
. | Beta (β) Variables . | Model 1 . | |
---|---|---|---|
Coefficients . | Standard Error . | ||
Transportation costs | Transport (β1) | −0.03*** | 0.00 |
Staff attitude | Attitude_good (β2) | 0.20*** | 0.02 |
The staff at the health facility treat me with respect | |||
Examination | Exam_given (β3) | 0.44*** | 0.02 |
The staff at the health facility examine me thoroughly | |||
Expert advice | Advice_given (β4) | 0.47*** | 0.02 |
The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |||
Availability of medicine | Med_avail (β5) | 1.20*** | 0.02 |
When I go to the health facility, I get the medicine I need | |||
Treatment by doctors or nurses | Tx_dr (β6) | 0.12*** | 0.02 |
When I go to the health facility, I always see a doctor | |||
Waiting time | Wait_short (β7) | 0.29*** | 0.02 |
I spend about half a day in the health facility before I go home | |||
Constant | Const (β8) | −0.15*** | 0.04 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −8072.2448 | ||
Wald chi-square | 2884.29 | ||
Prob > chi-square | 0.0000 |
. | Beta (β) Variables . | Model 1 . | |
---|---|---|---|
Coefficients . | Standard Error . | ||
Transportation costs | Transport (β1) | −0.03*** | 0.00 |
Staff attitude | Attitude_good (β2) | 0.20*** | 0.02 |
The staff at the health facility treat me with respect | |||
Examination | Exam_given (β3) | 0.44*** | 0.02 |
The staff at the health facility examine me thoroughly | |||
Expert advice | Advice_given (β4) | 0.47*** | 0.02 |
The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |||
Availability of medicine | Med_avail (β5) | 1.20*** | 0.02 |
When I go to the health facility, I get the medicine I need | |||
Treatment by doctors or nurses | Tx_dr (β6) | 0.12*** | 0.02 |
When I go to the health facility, I always see a doctor | |||
Waiting time | Wait_short (β7) | 0.29*** | 0.02 |
I spend about half a day in the health facility before I go home | |||
Constant | Const (β8) | −0.15*** | 0.04 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −8072.2448 | ||
Wald chi-square | 2884.29 | ||
Prob > chi-square | 0.0000 |
Note: Three choice questions were not answered and consequently removed from the analysis. Accordingly, there are only 15837 observations.
*** P < 0.001.
The constant term was statistically significant with a negative coefficient, suggesting a general preference not to seek care at public health facilities (but for either seeking care at private facilities or not seeking facility-based care).1
Sub-group analysis
Sub-group analysis for provinces and SES are shown in Table 4 and Table 5. Analysis by age, gender and language did not vary from the global analysis so are not reported (but are available from the authors on request).
. | Model 2 . | ||
---|---|---|---|
Coefficient . | SE . | P>|z| . | |
Transportation costs | −0.03 | 0.00 | 0.000 |
Transportation costs × Eastern Cape | −0.01 | 0.00 | 0.004 |
Staff attitude | 0.31 | 0.03 | 0.000 |
Staff attitude × Eastern Cape | −0.18 | 0.04 | 0.000 |
Examination | 0.57 | 0.32 | 0.000 |
Examination × Eastern Cape | −0.24 | 0.04 | 0.000 |
Expert advice | 0.84 | 0.03 | 0.000 |
Expert advice × Eastern Cape | −0.68 | 0.04 | 0.000 |
Availability of medicine | 1.15 | 0.03 | 0.000 |
Availability of medicine × Eastern Cape | 0.16 | 0.05 | 0.001 |
Treatment by doctors or nurses | 0.10 | 0.03 | 0.001 |
Treatment by doctors or nurses × Eastern Cape | 0.07 | 0.04 | 0.074 |
Waiting time | 0.39 | 0.03 | 0.000 |
Waiting time × Eastern Cape | −0.19 | 0.04 | 0.000 |
Constant | −0.38 | 0.07 | 0.000 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −7879.4008 | ||
Wald chi-square | 3044.23 | ||
Prob > chi-square | 0.0000 |
. | Model 2 . | ||
---|---|---|---|
Coefficient . | SE . | P>|z| . | |
Transportation costs | −0.03 | 0.00 | 0.000 |
Transportation costs × Eastern Cape | −0.01 | 0.00 | 0.004 |
Staff attitude | 0.31 | 0.03 | 0.000 |
Staff attitude × Eastern Cape | −0.18 | 0.04 | 0.000 |
Examination | 0.57 | 0.32 | 0.000 |
Examination × Eastern Cape | −0.24 | 0.04 | 0.000 |
Expert advice | 0.84 | 0.03 | 0.000 |
Expert advice × Eastern Cape | −0.68 | 0.04 | 0.000 |
Availability of medicine | 1.15 | 0.03 | 0.000 |
Availability of medicine × Eastern Cape | 0.16 | 0.05 | 0.001 |
Treatment by doctors or nurses | 0.10 | 0.03 | 0.001 |
Treatment by doctors or nurses × Eastern Cape | 0.07 | 0.04 | 0.074 |
Waiting time | 0.39 | 0.03 | 0.000 |
Waiting time × Eastern Cape | −0.19 | 0.04 | 0.000 |
Constant | −0.38 | 0.07 | 0.000 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −7879.4008 | ||
Wald chi-square | 3044.23 | ||
Prob > chi-square | 0.0000 |
. | Model 2 . | ||
---|---|---|---|
Coefficient . | SE . | P>|z| . | |
Transportation costs | −0.03 | 0.00 | 0.000 |
Transportation costs × Eastern Cape | −0.01 | 0.00 | 0.004 |
Staff attitude | 0.31 | 0.03 | 0.000 |
Staff attitude × Eastern Cape | −0.18 | 0.04 | 0.000 |
Examination | 0.57 | 0.32 | 0.000 |
Examination × Eastern Cape | −0.24 | 0.04 | 0.000 |
Expert advice | 0.84 | 0.03 | 0.000 |
Expert advice × Eastern Cape | −0.68 | 0.04 | 0.000 |
Availability of medicine | 1.15 | 0.03 | 0.000 |
Availability of medicine × Eastern Cape | 0.16 | 0.05 | 0.001 |
Treatment by doctors or nurses | 0.10 | 0.03 | 0.001 |
Treatment by doctors or nurses × Eastern Cape | 0.07 | 0.04 | 0.074 |
Waiting time | 0.39 | 0.03 | 0.000 |
Waiting time × Eastern Cape | −0.19 | 0.04 | 0.000 |
Constant | −0.38 | 0.07 | 0.000 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −7879.4008 | ||
Wald chi-square | 3044.23 | ||
Prob > chi-square | 0.0000 |
. | Model 2 . | ||
---|---|---|---|
Coefficient . | SE . | P>|z| . | |
Transportation costs | −0.03 | 0.00 | 0.000 |
Transportation costs × Eastern Cape | −0.01 | 0.00 | 0.004 |
Staff attitude | 0.31 | 0.03 | 0.000 |
Staff attitude × Eastern Cape | −0.18 | 0.04 | 0.000 |
Examination | 0.57 | 0.32 | 0.000 |
Examination × Eastern Cape | −0.24 | 0.04 | 0.000 |
Expert advice | 0.84 | 0.03 | 0.000 |
Expert advice × Eastern Cape | −0.68 | 0.04 | 0.000 |
Availability of medicine | 1.15 | 0.03 | 0.000 |
Availability of medicine × Eastern Cape | 0.16 | 0.05 | 0.001 |
Treatment by doctors or nurses | 0.10 | 0.03 | 0.001 |
Treatment by doctors or nurses × Eastern Cape | 0.07 | 0.04 | 0.074 |
Waiting time | 0.39 | 0.03 | 0.000 |
Waiting time × Eastern Cape | −0.19 | 0.04 | 0.000 |
Constant | −0.38 | 0.07 | 0.000 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −7879.4008 | ||
Wald chi-square | 3044.23 | ||
Prob > chi-square | 0.0000 |
. | Model 3 . | ||
---|---|---|---|
Coefficient . | SE . | P > |z| . | |
Transportation costs | −0.05 | 0.01 | 0.000 |
Transportation costs × lower medium | 0.01 | 0.01 | 0.370 |
Transportation costs × upper medium | 0.02 | 0.01 | 0.079 |
Transportation costs × higher | 0.02 | 0.01 | 0.085 |
Staff attitude | 0.19 | 0.11 | 0.076 |
Staff attitude × lower medium | − 0.00 | 0.12 | 0.982 |
Staff attitude × upper medium | 0.01 | 0.11 | 0.908 |
Staff attitude × higher | 0.03 | 0.11 | 0.773 |
Examination | 0.57 | 0.11 | 0.000 |
Examination × lower medium | −0.15 | 0.12 | 0.218 |
Examination × upper medium | −0.10 | 0.11 | 0.371 |
Examination × higher | −0.18 | 0.12 | 0.126 |
Expert advice | 0.27 | 0.11 | 0.010 |
Expert advice × lower medium | 0.02 | 0.12 | 0.849 |
Expert advice × upper medium | 0.24 | 0.11 | 0.032 |
Expert advice × higher | 0.27 | 0.11 | 0.017 |
Availability of medicine | 1.55 | 1.12 | 0.000 |
Availability of medicine × lower medium | −0.08 | 0.13 | 0.571 |
Availability of medicine × upper medium | −0.37 | 0.12 | 0.003 |
Availability of medicine × higher | −0.45 | 0.13 | 0.000 |
Treatment by doctors or nurses | 0.08 | 0.10 | 0.420 |
Treatment by doctors or nurses × lower medium | 0.17 | 0.12 | 0.137 |
Treatment by doctors or nurses × upper medium | 0.01 | 0.11 | 0.932 |
Treatment by doctors or nurses × higher | 0.03 | 0.11 | 0.808 |
Waiting time | 0.43 | 0.11 | 0.000 |
Waiting time × lower medium | −0.05 | 0.12 | 0.662 |
Waiting time × upper medium | −0.18 | 0.11 | 0.107 |
Waiting time × higher | −0.15 | 0.11 | 0.166 |
Constant | 0.25 | 0.22 | 0.247 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −8017.4191 | ||
Wald chi-square | 2903.49 | ||
Prob > chi-square | 0.0000 |
. | Model 3 . | ||
---|---|---|---|
Coefficient . | SE . | P > |z| . | |
Transportation costs | −0.05 | 0.01 | 0.000 |
Transportation costs × lower medium | 0.01 | 0.01 | 0.370 |
Transportation costs × upper medium | 0.02 | 0.01 | 0.079 |
Transportation costs × higher | 0.02 | 0.01 | 0.085 |
Staff attitude | 0.19 | 0.11 | 0.076 |
Staff attitude × lower medium | − 0.00 | 0.12 | 0.982 |
Staff attitude × upper medium | 0.01 | 0.11 | 0.908 |
Staff attitude × higher | 0.03 | 0.11 | 0.773 |
Examination | 0.57 | 0.11 | 0.000 |
Examination × lower medium | −0.15 | 0.12 | 0.218 |
Examination × upper medium | −0.10 | 0.11 | 0.371 |
Examination × higher | −0.18 | 0.12 | 0.126 |
Expert advice | 0.27 | 0.11 | 0.010 |
Expert advice × lower medium | 0.02 | 0.12 | 0.849 |
Expert advice × upper medium | 0.24 | 0.11 | 0.032 |
Expert advice × higher | 0.27 | 0.11 | 0.017 |
Availability of medicine | 1.55 | 1.12 | 0.000 |
Availability of medicine × lower medium | −0.08 | 0.13 | 0.571 |
Availability of medicine × upper medium | −0.37 | 0.12 | 0.003 |
Availability of medicine × higher | −0.45 | 0.13 | 0.000 |
Treatment by doctors or nurses | 0.08 | 0.10 | 0.420 |
Treatment by doctors or nurses × lower medium | 0.17 | 0.12 | 0.137 |
Treatment by doctors or nurses × upper medium | 0.01 | 0.11 | 0.932 |
Treatment by doctors or nurses × higher | 0.03 | 0.11 | 0.808 |
Waiting time | 0.43 | 0.11 | 0.000 |
Waiting time × lower medium | −0.05 | 0.12 | 0.662 |
Waiting time × upper medium | −0.18 | 0.11 | 0.107 |
Waiting time × higher | −0.15 | 0.11 | 0.166 |
Constant | 0.25 | 0.22 | 0.247 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −8017.4191 | ||
Wald chi-square | 2903.49 | ||
Prob > chi-square | 0.0000 |
. | Model 3 . | ||
---|---|---|---|
Coefficient . | SE . | P > |z| . | |
Transportation costs | −0.05 | 0.01 | 0.000 |
Transportation costs × lower medium | 0.01 | 0.01 | 0.370 |
Transportation costs × upper medium | 0.02 | 0.01 | 0.079 |
Transportation costs × higher | 0.02 | 0.01 | 0.085 |
Staff attitude | 0.19 | 0.11 | 0.076 |
Staff attitude × lower medium | − 0.00 | 0.12 | 0.982 |
Staff attitude × upper medium | 0.01 | 0.11 | 0.908 |
Staff attitude × higher | 0.03 | 0.11 | 0.773 |
Examination | 0.57 | 0.11 | 0.000 |
Examination × lower medium | −0.15 | 0.12 | 0.218 |
Examination × upper medium | −0.10 | 0.11 | 0.371 |
Examination × higher | −0.18 | 0.12 | 0.126 |
Expert advice | 0.27 | 0.11 | 0.010 |
Expert advice × lower medium | 0.02 | 0.12 | 0.849 |
Expert advice × upper medium | 0.24 | 0.11 | 0.032 |
Expert advice × higher | 0.27 | 0.11 | 0.017 |
Availability of medicine | 1.55 | 1.12 | 0.000 |
Availability of medicine × lower medium | −0.08 | 0.13 | 0.571 |
Availability of medicine × upper medium | −0.37 | 0.12 | 0.003 |
Availability of medicine × higher | −0.45 | 0.13 | 0.000 |
Treatment by doctors or nurses | 0.08 | 0.10 | 0.420 |
Treatment by doctors or nurses × lower medium | 0.17 | 0.12 | 0.137 |
Treatment by doctors or nurses × upper medium | 0.01 | 0.11 | 0.932 |
Treatment by doctors or nurses × higher | 0.03 | 0.11 | 0.808 |
Waiting time | 0.43 | 0.11 | 0.000 |
Waiting time × lower medium | −0.05 | 0.12 | 0.662 |
Waiting time × upper medium | −0.18 | 0.11 | 0.107 |
Waiting time × higher | −0.15 | 0.11 | 0.166 |
Constant | 0.25 | 0.22 | 0.247 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −8017.4191 | ||
Wald chi-square | 2903.49 | ||
Prob > chi-square | 0.0000 |
. | Model 3 . | ||
---|---|---|---|
Coefficient . | SE . | P > |z| . | |
Transportation costs | −0.05 | 0.01 | 0.000 |
Transportation costs × lower medium | 0.01 | 0.01 | 0.370 |
Transportation costs × upper medium | 0.02 | 0.01 | 0.079 |
Transportation costs × higher | 0.02 | 0.01 | 0.085 |
Staff attitude | 0.19 | 0.11 | 0.076 |
Staff attitude × lower medium | − 0.00 | 0.12 | 0.982 |
Staff attitude × upper medium | 0.01 | 0.11 | 0.908 |
Staff attitude × higher | 0.03 | 0.11 | 0.773 |
Examination | 0.57 | 0.11 | 0.000 |
Examination × lower medium | −0.15 | 0.12 | 0.218 |
Examination × upper medium | −0.10 | 0.11 | 0.371 |
Examination × higher | −0.18 | 0.12 | 0.126 |
Expert advice | 0.27 | 0.11 | 0.010 |
Expert advice × lower medium | 0.02 | 0.12 | 0.849 |
Expert advice × upper medium | 0.24 | 0.11 | 0.032 |
Expert advice × higher | 0.27 | 0.11 | 0.017 |
Availability of medicine | 1.55 | 1.12 | 0.000 |
Availability of medicine × lower medium | −0.08 | 0.13 | 0.571 |
Availability of medicine × upper medium | −0.37 | 0.12 | 0.003 |
Availability of medicine × higher | −0.45 | 0.13 | 0.000 |
Treatment by doctors or nurses | 0.08 | 0.10 | 0.420 |
Treatment by doctors or nurses × lower medium | 0.17 | 0.12 | 0.137 |
Treatment by doctors or nurses × upper medium | 0.01 | 0.11 | 0.932 |
Treatment by doctors or nurses × higher | 0.03 | 0.11 | 0.808 |
Waiting time | 0.43 | 0.11 | 0.000 |
Waiting time × lower medium | −0.05 | 0.12 | 0.662 |
Waiting time × upper medium | −0.18 | 0.11 | 0.107 |
Waiting time × higher | −0.15 | 0.11 | 0.166 |
Constant | 0.25 | 0.22 | 0.247 |
Number of groups | 990 | ||
Number of observations | 15837 | ||
Log Likelihood | −8017.4191 | ||
Wald chi-square | 2903.49 | ||
Prob > chi-square | 0.0000 |
Table 4 shows the results by province, with the excluded group being Western Cape. The interactions are positive for availability of medicine and treatment by doctors or nurses, indicating that people in Eastern Cape value drug availability and treatment by doctors more than those in Western Cape.
While the constant term was statistically significant with a negative coefficient in Western Cape, in Eastern Cape, it was insignificant. This can be interpreted as those in the Eastern Cape having no preference for facility-type when seeking care.2
The SES interaction terms are shown in Table 5 with the excluded group being the lower SES group. The interactions are positive for expert advice and negative for availability of medicine, indicating that medium and higher SES groups value expert advice more than lower SES groups. However, drug availability is more valued by lower SES groups than medium and higher SES groups. In addition, the interactions are positive for an increase in transportation costs for the upper medium and higher SES groups, indicating that higher transportation costs more negatively impact on lower SES groups in accessing a public health facility.
For the higher SES group, the constant term was statistically significant with a negative coefficient, whereas for the lower and medium (both upper and lower) SES groups the constant term was insignificant. These results suggests that people in the higher SES group would prefer not to seek health care at public health facilities while people in the medium or lower SES groups have no preference for the type of facility in which they seek care.
Willingness-to-pay
Willingness-to-pay (WTP) for a marginal improvement in all significant attributes (Table 6) revealed: (1) the availability of necessary medicine at health facility increases WTP by R35.5; (2) the provision of expert advice increases WTP by R14.0; (3) the provision of a thorough examination by health facility staff increases WTP by R12.8; (4) spending only half a day waiting at the health facility increases WTP by R8.5; (5) health facility staff treating patients with respect increases WTP by R6.0; and (6) treatment by a medical doctor increases WTP by R3.6. Thus, the marginal change in the availability of necessary medicine at health facilities has the greatest impact on the probability of attending public health facilities. Generally, clinical quality attributes (i.e. availability of necessary medicine, provision of expert advice and provision of a thorough examination) were more valued than non-clinical quality of care attributes (i.e. staff attitude, treatment by doctors or nurses, and waiting time).
. | Global . | Western Cape . | Eastern Cape . |
---|---|---|---|
Staff attitude | 6.0 (4.8–7.2) | 10.4 (8.1–12.7) | 3.4 (2.0–4.9) |
The staff at the health facility treat me with respect | |||
Examination | 12.8 (11.3–14.4) | 19.0 (15.8–22.1) | 8.3 (6.7–9.9) |
The staff at the health facility examine me thoroughly | |||
Expert advice | 14.0 (12.4–15.6) | 27.8 (23.4–32.2) | 3.8 (2.4–5.3) |
The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |||
Availability of medicine | 35.5 (32.2–38.9) | 38.3 (32.5–44.1) | 32.8 (29.1–36.5) |
When I go to the health facility, I get the medicine I need | |||
Treatment by doctors or nurses | 3.6 (2.3–4.9) | 3.2 (1.1–5.2) | 4.3 (2.7–5.8) |
When I go to the health facility, I always see a doctor | |||
Waiting time | 8.5 (7.1–9.9) | 13.0 (10.4–15.6) | 5.1 (3.6–6.6) |
I spend about half a day in the health facility before I go home | |||
Constant | 6.0 (4.8–7.2) | 10.4 (8.1–12.7) | 3.4 (2.0–4.9) |
. | Global . | Western Cape . | Eastern Cape . |
---|---|---|---|
Staff attitude | 6.0 (4.8–7.2) | 10.4 (8.1–12.7) | 3.4 (2.0–4.9) |
The staff at the health facility treat me with respect | |||
Examination | 12.8 (11.3–14.4) | 19.0 (15.8–22.1) | 8.3 (6.7–9.9) |
The staff at the health facility examine me thoroughly | |||
Expert advice | 14.0 (12.4–15.6) | 27.8 (23.4–32.2) | 3.8 (2.4–5.3) |
The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |||
Availability of medicine | 35.5 (32.2–38.9) | 38.3 (32.5–44.1) | 32.8 (29.1–36.5) |
When I go to the health facility, I get the medicine I need | |||
Treatment by doctors or nurses | 3.6 (2.3–4.9) | 3.2 (1.1–5.2) | 4.3 (2.7–5.8) |
When I go to the health facility, I always see a doctor | |||
Waiting time | 8.5 (7.1–9.9) | 13.0 (10.4–15.6) | 5.1 (3.6–6.6) |
I spend about half a day in the health facility before I go home | |||
Constant | 6.0 (4.8–7.2) | 10.4 (8.1–12.7) | 3.4 (2.0–4.9) |
Note: WTP estimates are calculated using results of the random effects logit models (Table 3). Hole’s WTP command for STATA was used to estimate confidence intervals for WTP estimates within a DCE.
. | Global . | Western Cape . | Eastern Cape . |
---|---|---|---|
Staff attitude | 6.0 (4.8–7.2) | 10.4 (8.1–12.7) | 3.4 (2.0–4.9) |
The staff at the health facility treat me with respect | |||
Examination | 12.8 (11.3–14.4) | 19.0 (15.8–22.1) | 8.3 (6.7–9.9) |
The staff at the health facility examine me thoroughly | |||
Expert advice | 14.0 (12.4–15.6) | 27.8 (23.4–32.2) | 3.8 (2.4–5.3) |
The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |||
Availability of medicine | 35.5 (32.2–38.9) | 38.3 (32.5–44.1) | 32.8 (29.1–36.5) |
When I go to the health facility, I get the medicine I need | |||
Treatment by doctors or nurses | 3.6 (2.3–4.9) | 3.2 (1.1–5.2) | 4.3 (2.7–5.8) |
When I go to the health facility, I always see a doctor | |||
Waiting time | 8.5 (7.1–9.9) | 13.0 (10.4–15.6) | 5.1 (3.6–6.6) |
I spend about half a day in the health facility before I go home | |||
Constant | 6.0 (4.8–7.2) | 10.4 (8.1–12.7) | 3.4 (2.0–4.9) |
. | Global . | Western Cape . | Eastern Cape . |
---|---|---|---|
Staff attitude | 6.0 (4.8–7.2) | 10.4 (8.1–12.7) | 3.4 (2.0–4.9) |
The staff at the health facility treat me with respect | |||
Examination | 12.8 (11.3–14.4) | 19.0 (15.8–22.1) | 8.3 (6.7–9.9) |
The staff at the health facility examine me thoroughly | |||
Expert advice | 14.0 (12.4–15.6) | 27.8 (23.4–32.2) | 3.8 (2.4–5.3) |
The staff at the health facility explain what is wrong with me and give me advice about what I need to do to get better | |||
Availability of medicine | 35.5 (32.2–38.9) | 38.3 (32.5–44.1) | 32.8 (29.1–36.5) |
When I go to the health facility, I get the medicine I need | |||
Treatment by doctors or nurses | 3.6 (2.3–4.9) | 3.2 (1.1–5.2) | 4.3 (2.7–5.8) |
When I go to the health facility, I always see a doctor | |||
Waiting time | 8.5 (7.1–9.9) | 13.0 (10.4–15.6) | 5.1 (3.6–6.6) |
I spend about half a day in the health facility before I go home | |||
Constant | 6.0 (4.8–7.2) | 10.4 (8.1–12.7) | 3.4 (2.0–4.9) |
Note: WTP estimates are calculated using results of the random effects logit models (Table 3). Hole’s WTP command for STATA was used to estimate confidence intervals for WTP estimates within a DCE.
Individuals are more likely to attend a health facility if the defined facility results in a positive WTP. In fact, if clinical attributes are at ‘good’ levels, even if non-clinical attributes are at a ‘worse’ level, people are still likely to attend the health facility. For example, although patients may have to wait for significant time at the health facility (WTP R−8.5); the health facility staff do not treat patients with respect (WTP R−6.0); and a nurse treats a patient (WTP R−3.6), if the necessary medicine is available at health facility (WTP R35.5), expert advice is provided (WTP R14.0); and thorough examination is provided by health facility staff (WTP R12.8), the overall WTP is positive (R44.2), indicating that people would seek care at the public health facility (as they would be better off than not attending the public health facility).
Treatment by a medical doctor had the smallest marginal WTP of all the quality of care attributes. Consequently, even if a medical doctor, rather than a nurse, always sees a patient in a facility, the overall WTP becomes negative (R−73.2) if the other attributes are at ‘worse’ levels, indicating that people would not seek care at the public health facility (since if they did they would be worse off).
Comparing uptake rate of choosing a health facility with clinical and non-clinical quality attributes
Estimation of the probability of choosing a defined public health facility confirmed the result that clinical quality attributes are valued more highly. The uptake rate for a health facility with better clinical quality (in terms of availability of necessary medicine, provision of expert advice and provision of a thorough examination when transportation costs R40) is predicted to be 69.4%, whereas the uptake rate for a health facility with better non-clinical quality (in terms of staff attitude, treatment by doctors or nurses, and waiting time when transportation costs R40) is estimated to be 10.2%. In fact, the uptake rate of a health facility with better availability of medicine alone (with no other attributes having ‘better’ levels and R40 as transportation costs) is predicted to be 47.5%.
Validity of responses
Regarding the holdout questions, the observed choices (the number of respondents who answered yes to the holdout questions) were 11.1% and 50.0% compared with the predicted choices (the probability of ‘yes’ responses to the holdout questions) of 10.7% and 55.0% respectively. The results indicate that the model correctly predicts responses with only a slight deviation from the expected outcome.
The theoretical validity of the model is evidenced by all ‘good’ levels of attributes having positive coefficients, as anticipated, and the transportation cost attributes having a negative coefficient.
Debriefing sessions with the provincial data collection teams indicated that the key findings from the analysis corresponded with the interviewers’ perceptions of the importance of attributes and provided further support for the validity of responses.
Discussion
The broad approach that the South African government will take to implementing the proposed NHI is outlined in a 2011 Green Paper (Department of Health 2011b). This clearly indicates that the first five-year phase of the implementation will be devoted to improving public sector health services. In order to achieve this, it is critical that policy-makers are well informed of the population’s views on public sector service quality. In particular, awareness of the improvements in public sector services desired by the community can inform the prioritization of health system changes.
Regarding the preferences of public health-care users, a key finding is that the most important issue in the provision of public health facilities is drug availability. This has been recognized as a serious problem in South Africa with key difficulties relating to drug distribution caused by inadequate facility-level capacity for effective re-ordering and poor distribution from provincial depots to individual facilities (Gray and Suleman 1999). Recently, problems have arisen in the procurement of medicine, with drug budgets sometimes being fully expended before the financial year-end and suppliers subsequently halting deliveries. Consequently, it is critical to more accurately forecast annual drug and budget requirements and develop drug distribution alternatives. An efficient network of private drug distributors currently services private pharmacies and hospitals. Accessing such services in areas facing distribution problems could be considered by public facilities. The key feasibility constraint in this approach is the capacity of private distributors, as the volume of drugs provided through public sector facilities is far greater than that of private pharmacies and hospitals. It may be appropriate to conduct a pilot programme with private distributors in one province to assess the relative benefits of public and private distribution, ability to scale up capacity and impact on drug availability.
Survey respondents also considered the explanation of illnesses and appropriate treatment based on a thorough examination by clinical staff to be important. In fact, the WTP results indicate that people are likely to seek public sector health services if they receive the medicine they need, a thorough examination and a clear explanation of the diagnosis and prescribed treatment from a health professional, despite a long waiting time, poor staff attitudes and lack of direct access to doctors. Inadequate staffing levels and high patient loads influence the extent to which expert advice is provided and thorough examination undertaken at public sector facilities (Day and Gray 2010). Low staffing levels translate to medical staff having limited time with each patient and lengthy waiting times at medical facilities. In addition, there is considerable evidence of poor morale and a lack of caring ethos by health centre staff when insufficient staffing levels and a heavy workload exist (McIntyre and Klugman 2003; Walker and Gilson 2004). Improving staffing levels would not only allow health facility staff more time to undertake examinations and explain health problems, it would also decrease waiting times and possibly address poor staff attitudes, attributes valued by respondents (although not as highly as clinical attributes). While it is not feasible to improve staffing levels dramatically in the short term due to the long lead-time in training health professionals, the South African government is committed to increasing the supply of health workers (Department of Health 2011a). The budget includes financial resources to increase capacity to train health workers. There is also an emphasis on exploring task-shifting, wherein less well-trained staff take on routine tasks, allowing better skilled staff to undertake more complex clinical activities (Department of Health 2011a). Nonetheless, long-term efforts are required to improve staffing in health facilities.
In the context of current health policy debate, a finding that is of interest is the low value given to seeing a doctor as the first-line provider (in the global analysis and in Western Cape). The WTP results suggest that despite treatment by a nurse, people are likely to visit a public health facility if other aspects of quality of care are ensured at a facility. At present in South Africa, primary care services in the public sector are very much ‘nurse-led’: in most facilities, patients are first seen by a nurse, who diagnoses and treats patients unless the nurse feels the patient needs to be referred to a doctor. While some feel that PHC services should be more doctor-intensive, a middle-income country such as South Africa must carefully evaluate the affordability and sustainability of such a change. The communities surveyed do not see this issue as a key priority for public sector improvement. At the policy level, improvements in public service quality should not be made unless the benefits outweigh the costs. However, in a public health-care system where resources are limited, the rule that benefits outweigh costs is not sufficient as too many improvements may achieve this threshold. Policy makers should ensure improvements are targeted at attributes of quality of public services where the benefit/cost ratio is maximized. So, for optimal decisions, costing of the various improvements is necessary.
The results suggest there may be a general preference not to utilize public health facilities, and that public health facilities need to improve to encourage use of the services that they offer. However, the study design does not allow determination of whether the ‘no’ option means that the respondent would prefer to attend a private facility, or simply not to seek health care. Future work should explore the interpretation of the negatively significant constant.
Whilst the sub-group analysis should be treated with caution, given the relatively small sample sizes of the lower SES groups and the differences in respondent characteristics between provinces, there were some important findings. Despite the limitations in the interpretation of the constant term, whilst Western Cape residents may have preference not to use public sector services, the respondents in Eastern Cape appeared not to have a preference for the type of health-care provider from whom they received treatment. Furthermore, the strength of preferences for service quality attributes, except the availability of medicine and treatment by doctors or nurses, was greater in Western Cape than Eastern Cape. In addition to the difference in SES between the two provinces, there are fewer alternatives to public health facilities in Eastern Cape, as supported by the fact that in 2008, while 64% of outpatient visits were to a public provider in Western Cape, nearly 82% of visits took place at a public facility in Eastern Cape (Alaba and McIntyre 2012). Given that individuals’ responses are influenced by their own experience of health services, it is unsurprising that a stronger preference for public sector service improvement was recorded in Western Cape – the people in Western Cape, in general, have an opportunity to compare the health facility options available whereas the limited awareness of different types of health facilities in Eastern Cape may have led the people to accept the services currently available in the community. If public sector health service quality does not improve, there is a chance that, in Western Cape, those who can afford to pay for private health services will not use public sector health services even when the NHI is introduced, and that, in Eastern Cape, inequity in the ability to access quality health services will be high compared with other, better off provinces in South Africa.
A significant negative constant term for the higher SES group, who can afford a wider range of alternatives than other SES groups, implies a preference for not using public sector providers. A stronger preference for expert advice was expressed by the higher SES groups (including medium and higher SES groups) than lower SES groups but drug availability was more valued by lower SES groups than higher SES groups. The latter result is not surprising given that the lower SES group do not have to purchase medicine outside a public health facility if the public facility has the necessary medicine in stock. In addition, the positive interaction terms for higher transportation costs among the higher SES groups, conversely indicates that higher transportation costs have stronger negative consequences for accessing public health facilities for the lower SES group.
This study not only contributes evidence that can inform decision-making within South Africa, but it also contributes to limited literature on determining community preferences in low- and middle-income countries (LMIC) using DCE. A number of studies have used DCE to elicit job choice preferences for health workers in LMICs (Ryan et al. 2012) or examine the relative importance of different criteria for priority setting by policy makers, health administrators or those involved with HIV/AIDS interventions (Baltussen et al. 2006, 2007; Jehu-Appiah et al., 2008; Youngkong et al., 2010). However, only a few studies examining preferences for certain types or levels of care have administered DCE to the general population (Hanson et al. 2005; Kruk et al. 2010). While these studies were confined to specific geographical areas (i.e. one specific town or district), our study was undertaken in three districts within each of two provinces. While our study cannot claim to be nationally representative, it does reflect a wide range of geographic and socio-economic contexts.
There are a number of methodological limitations in this study. As discussed in the results section, the methodological challenges experienced in administering the household level DCE, including safety concerns which made it difficult to elicit the preferences of males (who were often at work during daylight hours when the DCE was undertaken) and lower SES groups (who were often in high risk or remote areas), resulted in a skewed distribution of gender and SES groups. Future work should attempt to employ study designs that allow equal representation of such groups. In addition, the relatively small sample sizes for lower SES groups may have limited the statistical power of the analysis to detect differences between SES groups.
This DCE was administered in Western Cape and Eastern Cape, two of South Africa’s nine provinces. Due to resource constraints, the household DCE could not be undertaken in all provinces in South Africa, which may limit the generalizability of the study results. However, in order to cover variations across the country, the two provinces were chosen to demonstrate contrasting contexts in terms of health status, health-care resources, access to basic facilities and SES.
Although the study aim and hypothetical scenario were carefully explained to respondents, the study may have not clearly specified what a ‘no’ response meant—whether it indicated seeking care at a private facility or not seeking facility-based care—which may have implications for the interpretation of the constant terms. Future research could explore this issue in more detail by defining what the ‘neither’ option means in terms of alternatives to using public health facilities.
While the WTP, a commonly used output from DCEs, provides useful information in determining the relative importance of attributes and the trade-offs between attributes (De Bekker-Grob et al. 2012), the study recognizes that the levels of the cost attribute can also affect the estimates (Drummond et al. 2005). Consequently, our study paid careful attention when developing the attribute levels for transportation costs and adjusted the levels based on the results from the pilot testing of the questionnaire.
The importance of validating the DCE results using subsequent monitoring and evaluation of policies has been recognized (De Bekker-Grob et al. 2012). Given that DCEs rely on responses to hypothetical choices, it is crucial to examine the external validity of the responses, that is, do respondents actually behave in reality as they say they will in a hypothetical context. We attempted to investigate this by asking interviewers to comment on whether they thought the regression results reflected the thoughts of the interviewees. Whilst the results were positive, indicating that the key findings from the analysis corresponded with the interviewers’ perceptions of the importance of attributes, we recognize the limitations of this approach and suggest that future research should include a qualitative study specifically designed to investigate validity.
While this study focused on the preferences of users of public health facilities, it would be interesting to elicit the preferences of those that currently utilize private health facilities, with a view to determining what would encourage this group to use public health facilities. It would also be interesting to establish the attributes that would influence a switch to public facilities, and whether the issues of concern to public health facility users are the same as those of private facility users.
Conclusions
This study indicates that communities are prepared to tolerate public sector health service characteristics such as a long waiting time, poor staff attitudes and the lack of direct access to doctors if they receive the medicines they need, a thorough examination and a clear explanation of the diagnosis and prescribed treatment from health professionals. These findings point to priorities for the South African Government to address in order to meet its commitment to ‘shed the image of poor quality services’ (Department of Health 2011b). The strong preference for improved medicine availability highlights that focusing on this aspect of health-care service delivery would be a ‘quick win’ for the Government, and could dramatically increase the uptake of public sector health services.
However, attention should also be given to interventions that enable clinicians to undertake adequate patient examinations and explain the nature of illnesses and what is required for treatment. This is likely to require improved staffing levels. Interventions that take more time to produce effects, such as changing staff attitudes, should not be neglected. Our study found these attributes were statistically significant, indicating that despite having a lower level of preference, they do influence decisions on whether public sector services are considered acceptable.
At a time when many countries are pursuing UC, evidence, such as is presented in this paper, is critical to inform policy-makers of the community’s views on interventions to promote access to quality health care for all. Given the limitations in this study wherein resource constraints resulted in the survey only being undertaken in two of the South African provinces, it is hoped that the study will be replicated in other provinces throughout South Africa to allow the views of the wider population to be reflected in the process of the NHI reform development. DCE studies can be used as a tool to enable citizens to have a greater role in prioritizing the health interventions that make a difference to their lives. Despite a number of methodological challenges in designing and undertaking the DCE study, our study highlights the value of the information that DCEs can provide by eliciting community views on health systems changes in the LMIC context.
Acknowledgements
The authors are extremely indebted to the provincial data collection teams in the Western Cape and Eastern Cape. Special thanks to the co-ordinators of the provincial data collection teams, Vanessa Daries and Bulelwa Mafu. They are also grateful to Gouwah Samuels and Marsha Orgill for their invaluable contribution to the FGDs.
Funding
This study was undertaken as part of a larger project on public engagement in health systems change in South Africa (“Eliciting public preferences in relation to National Health Insurance in South Africa”) funded by the Atlantic Philanthropies. The field data collection was partly funded by the Department of Science and Technology and National Research Foundation through the South African Research Chairs Initiative. M.R. is funded by the University of Aberdeen. The University of Aberdeen is a charity registered in Scotland, No SC013683.
Conflict of interest statement. None declared.
1 Alternatively this statistically significant constant can be interpreted as respondents systematically bringing additional information to the choice tasks which resulted in them not choosing the public health facility on offer. However, it is also important to consider that the study team undertook extensive qualitative research to establish the attributes and the attribute levels and tried to ensure that all important attributes were included in the choice questions by carefully reflecting qualitative research results and feedback from pilot testing of the questionnaire.
2 The statistically insignificant constant could also mean that, when choosing a public health facility, the attributes of the service drove the choice and that respondents in the Eastern Cape prefer to choose a facility than not choose a facility.