Skip to main content
NCBI home page
Bulletin of the World Health Organization logoLink to Bulletin of the World Health Organization
. 2013 May 31;91(7):519–524C. doi: 10.2471/BLT.12.115535

Estimating health expenditure shares from household surveys

Estimer la part des dépenses de santé à partir d'enquêtes sur les ménages

Estimación de los porcentajes de gasto sanitario provenientes de encuestas en los hogares

تقدير حصص الإنفاق الصحي من الدراسات الاستقصائية الأسرية

根据家庭调查估算卫生支出份额

Оценка доли расходов на здравоохранение на основе опросов домашних хозяйств

Rouselle F Lavado a,✉, Benjamin PC Brooks a, Michael Hanlon a
PMCID: PMC3699797  PMID: 23825879

Abstract

Objective

To quantify the effects of household expenditure survey characteristics on the estimated share of a household’s expenditure devoted to health.

Methods

A search was conducted for all country surveys reporting data on health expenditure and total household expenditure. Data on total expenditure and health expenditure were extracted from the surveys to generate the health expenditure share (i.e. fraction of the household expenditure devoted to health). To do this the authors relied on survey microdata or survey reports to calculate the health expenditure share for the particular instrument involved. Health expenditure share was modelled as a function of the survey’s recall period, the number of health expenditure items, the number of total expenditure items, the data collection method and the placement of the health module within the survey. Data exists across space and time, so fixed effects for territory and year were included as well. The model was estimated by means of ordinary least squares regression with clustered standard errors.

Findings

A one-unit increase in the number of health expenditure questions was accompanied by a 1% increase in the estimated health expenditure share. A one-unit increase in the number of non-health expenditure questions resulted in a 0.2% decrease in the estimated share. Increasing the recall period by one month was accompanied by a 6% decrease in the health expenditure share.

Conclusion

The characteristics of a survey instrument examined in the study affect the estimate of the health expenditure share. Those characteristics need to be accounted for when comparing results across surveys within a territory and, ultimately, across territories.

Introduction

Health expenditure share, or the percentage of the household expenditure spent on health care, is an important variable in health financing research.1 This figure is used to determine the number of households incurring catastrophic health expenditures and, in many countries, to derive the estimates of private health expenditure reported in national health accounts.2-5 Studies have shown that health expenditure share estimates derived from household expenditure surveys have problems with “reliability, validity, and comparability”.6,7 For example, two nationally representative surveys conducted in the Philippines in 2003 reported widely different health expenditure shares – 1.3% and 7.7%. One wonders which of the two estimates is a more accurate reflection of reality.

An extensive literature exists on the sources of bias in surveys.8-10 However, few studies have explored how biases affect estimates of out-of-pocket household health expenditure. Lu et al.6 examined how the number of questions on health expenditure and the recall period of a survey affected estimates of household out-of-pocket payments and catastrophic expenditure on health. Their study analysed data from the World Health Surveys (WHS) for 43 countries and from the Living Standards Measurement Survey (LSMS) for three countries. They found that estimates of health spending were lower when the survey had fewer questions and that the estimates were higher when the recall period was shorter. Heijink et al.7 conducted an exhaustive review of the evidence surrounding measurement errors in self-reported household expenditure and health expenditure. They also collected 90 household expenditure surveys from the International Household Survey Network (IHSN). Their findings concurred with those of Lu et al.: households reported higher health expenditures when more questions were asked. The authors reported that the influence of the recall period was unclear, but that the mode of data collection, such as a diary versus face-to-face interviews, did affect the estimates. Most of the studies identified by Heijink et al.7 concluded that diaries yielded lower expenditure figures;11,12 one study showed conflicting results.13 Heijink et al. also suggested that the questionnaire’s structure affects the results.7 In some surveys, health expenditure questions are included within the health module, whereas in others they are placed within the household expenditure module.

As noted, previous studies have identified the direction of the biases inherent in health expenditure share estimates. Our study, however, is the first to quantify the effect of these biases. We analyse multiple surveys per country or territory and show how the estimated share of the household expenditure devoted to health (i.e. health expenditure share) would have varied if survey instruments with different characteristics had been employed. Our contribution makes it possible for analysts to compare health expenditure share estimates across surveys. At the end of the paper we raise some points to be considered when conducting cross-country comparisons of household survey data.

Methods

We conducted an exhaustive search of all surveys that reported information on health expenditure and total household expenditure. First we identified the data sources used by the World Health Organization to estimate out-of-pocket expenditure in its Global Health Expenditure Database. We identified a total of 719 household expenditure surveys. However, to conduct our analysis, we needed both the survey questionnaire and microdata (or a report) illustrating how to calculate the health expenditure share for that particular instrument. To obtain this information for the 719 surveys, we looked up the questionnaires, documentation reports and microdata in the IHSN, the Global Health Data Exchange and the web sites of statistical offices and the health ministry of each country or territory. These supplementary sources were available for 214 surveys. The final sample therefore consisted of 214 survey years across 78 territories, as presented in Table 1 (available at: http://www.who.int/bulletin/volumes/91/7/12-115535). A flowchart showing our search strategy is presented in Fig. 1.

Table 1. List of surveys with data on total household expenditure and household health expenditure, by territory and year.

Country or territory Survey Years
High income
Croatia Household Budget Survey 2003, 2004
World Health Survey 2003
Czech Republic World Health Survey 2002
Estonia Household Budget Survey 2000–2004
World Health Survey 2003
Hungary World Health Survey 2003
Singapore Household Expenditure Survey 2007
Slovakia World Health Survey 2003
Slovenia World Health Survey 2003
Household Budget Survey 2004
Spain World Health Survey 2002
Taiwan, China Family Income and Expenditure Survey 1980–2009
United Arab Emirates World Health Survey 2003
Upper-middle income
Azerbaijan Quarterly Household Survey 2002–2005
Belarus Household Living Standard Survey 2009
Bosnia and Herzegovina Bosnia and Herzegovina Household Survey Panel Series 2004
World Health Survey 2003
Botswana Household Income and Expenditure Survey 1993, 2002
Brazil World Health Survey 2003
Bulgaria Multi-topic Household Survey 2003
Integrated Household Survey 1995, 2001
China World Health Survey 2002
Dominican Republic Encuesta Nacional de Ingresos y Gastos de los Hogares 2007
World Health Survey 2003
Ecuador World Health Survey 2003
Kazakhstan World Health Survey 2002
Latvia World Health Survey 2003
Malaysia World Health Survey 2003
Maldives Household Income and Expenditure Survey 2002
Mauritius Mauritius Household Budget Survey 2001, 2006
World Health Survey 2003
Mexico Encuesta Nacional de Ingresos y Gastos de los Hogares 1984, 1989, 1992, 1994, 1996, 1998, 2000, 2002, 2004–2006, 2008
World Health Survey 2002
Namibia World Health Survey 2003
Russian Federation Russian Federation Longitudinal Monitoring Survey 1992–1995, 2000–2002
World Health Survey 2003
Saint Lucia Saint Lucia Survey of Living Conditions and Household Budget 2005
Serbia Living Standard Measurement Survey 2002, 2003, 2007
South Africa Income and Expenditure Survey 1995, 2000
World Health Survey 2002
National Income Dynamics Survey 2008
Tunisia World Health Survey 2003
Turkey World Health Survey 2003
Uruguay World Health Survey 2002
Lower-middle income
Bhutan Bhutan Living Standard Survey 2003, 2007
Congo World Health Survey 2003
Côte d'Ivoire Côte d'Ivoire Living Standard Survey 1985, 1986
World Health Survey 2003
Federated States of Micronesia Household Income and Expenditure Survey 2005
Georgia World Health Survey 2003
Ghana Ghana Living Standards Survey 1991, 1998, 2005
World Health Survey 2003
Guatemala World Health Survey 2003
India World Health Survey 2003
Indonesia Survei Sosial Ekonomi Nasional 1992, 1994–2005, 2007
Iraq Household Socio-Economic Survey in Iraq 2006
Lao People's Democratic Republic Lao People's Democratic Republic Expenditure and Consumption Survey 2002, 2007
World Health Survey 2003
Morocco World Health Survey 2003
Nicaragua Encuesta Nacional de Hogares sobre Medición de Nivel de Vida 2009
Pakistan Pakistan Social and Living Standards Measurement Survey 1999, 2005, 2006, 2008
World Health Survey 2003
Paraguay World Health Survey 2002
Philippines Family Income and Expenditure Survey 1991, 1994, 1997, 2003, 2006, 2009
World Health Survey 2003
Annual Poverty Indicator Survey 2004, 2007
Senegal Enquête senegalaise auprès des ménages 1994, 2001
Enquête de suivi de la pauvreté au Senegal 2005
World Health Survey 2003
Solomon Islands Household Income and Expenditure Survey 2005 2005
Sri Lanka Household Income and Expenditure Survey 1995, 2005, 2006
World Health Survey 2003
Sudan Sudan National Baseline Household Survey 2009
Swaziland Household Income and Expenditure Survey 1995
World Health Survey 2003
Ukraine World Health Survey 2002
Viet Nam Viet Nam Living Standards Survey 1995
Viet Nam Household Living Standards Survey 2002
World Health Survey 2002
Zambia Living Conditions Monitoring Survey 2002, 2004, 2006, 2010
World Health Survey 2003
Low income
Bangladesh World Health Survey 2003
Benin Enquête 1–2–3 2003
Burkina Faso Enquête prioritaire étude sur les conditions de vie des ménages 1994
Enquête sur les dépenses des ménages de Ouagadougou 1998
Enquête nationale 2003
World Health Survey 2002
Cambodia Cambodia Socio-Economic Survey 1997, 1999, 2004, 2007
Chad World Health Survey 2003
Comoros Enquête intégrale auprès des ménages 2004
World Health Survey 2003
Ethiopia Urban/Rural Household Consumption Survey 1995, 1999, 2004
World Health Survey 2003
Gambia Household Economic Survey 1992
Household Education and Health Survey 1993
Kenya World Health Survey 2004
Madagascar Enquête nationale auprès des ménages 1993, 1997, 2001
Malawi World Health Survey 2003
Mali World Health Survey 2003
Mauritania World Health Survey 2003
Mozambique Inquérito ao Orçamento Familiar 2008
Inquérito aos Agregados Familiares sobre Orçamento Familiar 2002
Myanmar World Health Survey 2003
Nepal World Health Survey 2003
Niger Enquête permanente de conjoncture économique et sociale 1995
Core Welfare Indicator Questionnaire 2005
Tajikistan Living Standards Survey in the Republic of Tajikistan 1999, 2003, 2007, 2009
United Republic of Tanzania Household Budget Survey 2000, 2007
Uganda Uganda National Household Survey 2002 2002, 2005
Zimbabwe World Health Survey 2003
Open in a new tab

Fig. 1.

Flowchart showing strategy followed in searching for surveys

WHO, World Health Organization.

Fig. 1

Open in a new tab

Surveys vary in how they report expenditures and, as a result, analysts have to employ different methods to calculate total household expenditure. For example, some LSMS surveys (such as the Viet Nam Living Standards Survey) place expenditure on food and on other recurrent daily and annual expenses and expenditures on health, education and housing in separate categories. To calculate the number of questions on total expenditure, we summed all expenditure questions corresponding to the first 12 categories of the Classification of Individual Consumption According to Purpose (COICOP), which are food and non-alcoholic beverages; alcoholic beverages, tobacco and narcotics; clothing and footwear; housing, water, electricity, gas and other fuels; furnishing, household equipment and routine household maintenance; health; transport; communication; recreation and culture; education; restaurants and hotels, and miscellaneous goods and services. All questions contributed to total expenditure regardless of their location in the survey instrument.

For surveys in which respondents were interviewed, we counted all the questions that pertained to expenditures. For surveys with screening or conditional (“skip”) questions, we assumed that the respondent replied affirmatively and counted all the questions that followed the screening question. The process was more complicated for surveys in which respondents recorded results in a diary. If the survey report contained a detailed disaggregation of expenditure items, we counted all the items that were listed. If the report contained only aggregated totals (22 surveys), we used the number of disaggregated categories in the COICOP. For surveys in which a combination of interviews and diaries was used we relied on the data generated from the interview.

In our sample, the number of questions on health expenditure ranged from one (WHS surveys) to 274 (Dominican Republic 2007 Encuesta Nacional de Ingresos y Gastos de los Hogares [ENIGH]). Hence, the WHS asked a single question about the household’s total amount of health expenditure; other surveys asked multiple questions, each of them focused on a specific type of health expenditure. The number of questions on total expenditure ranged from one (WHS surveys) to 2431 (Dominican Republic 2007 ENIGH). Again, as the number of questions increased, the questions became more specific. WHS surveys had single questions for health expenditure and a single question for total expenditure, and they also had eight questions on disaggregated categories of out-of-pocket health spending and six questions on total expenditure. Given the focus of our analysis, we calculated four different iterations of health expenditure shares based on the aggregation of health and total expenditure questions. The survey with the largest number of questions – the Dominican Republic 2007 ENIGH survey – was conducted through the use of diaries. When we state that the survey had 2431 health expenditure questions, we are referring to the number of expenditure items enumerated in the survey report. Since the diary method allows respondents to record their purchases in detail, this survey reports on a greater number of items than other surveys. For example, instead of reporting overall expenditure on medicines, the Dominican Republic survey reported expenditures disaggregated by different types of medicines, such as antihistamines, anti-depressants and analgesics.

The shortest recall period was 10 days (2006 Household Socio-Economic Survey in Iraq) and the longest was 12 months (surveys from Bulgaria, Côte d’Ivoire, the Federated States of Micronesia, Gambia, Ghana, Madagascar, Mauritius and Saint Lucia). We derived health expenditure shares from microdata and validated those estimates by comparing them with the data from survey reports. If microdata were not available, we used the health expenditure shares from survey reports. Health expenditure shares ranged from 0.1% (Gambia 1992 Household Economic Survey) to 27.4% (1999 Cambodia Socio-Economic Survey). Descriptive statistics for these variables are presented in Table 2.

Table 2. Descriptive statistics for independent variables.

Variable Mean SD Min Median Max
Health expenditure share (%) 6.28 4.07 0.11 5.24 27.42
Number of health questions 8.54 18.10 1 5 274
Number of total expenditure questions 96.58 202.28 1 6 2431
Recall period (months) 3.00 4.01 0.30 1 12
Data collection modea 0.11 – – – –
Health module placementb 0.04 – – – –
Open in a new tab

SD, standard deviation.

a The indicator was assigned a value of 1 if the data were obtained by survey and a value of 0 if obtained through interview.

b The indicator was assigned a value of 1 if in a separate module and a value of 0 otherwise.

The dependent variable was the share of household expenditure spent on health. We modelled it as a function of the recall period, the number of health expenditure questions and the number of total expenditure questions. We also included binary indicators to represent the data collection method and the placement of the health module within the survey. We included the number of total expenditure questions because we hypothesized that one additional question pertaining to non-health expenditure would marginally increase the estimate of total expenditure without affecting the estimate of health expenditure. Therefore, we anticipated that the total number of expenditure questions would be inversely related to the health expenditure share. We assigned a value of 1 to the indicator for the data collection method if the data had been collected through a diary. We assigned a value of 1 to the indicator for the placement of the question within the health module if health expenditure questions and total expenditure questions were placed separately.

We included in the model gross domestic product (GDP) per capita and average years of education, as well as fixed effects for territory and World Bank income categorization, to control for unobservable characteristics at the territory level. We allowed the territory fixed effects to interact with year indicators to generate a unique time trend for every country. The model was estimated using ordinary least squares regression. We found 57 unique survey types and we clustered our observations based on these types. To account for potential heteroskedasticity, we report heteroskedasticity-robust standard errors.

To illustrate how health expenditure shares are influenced by survey characteristics, we defined three types of survey instruments: “minimalist”, “typical” and “extensive”. A “minimalist” instrument would have one expenditure question, one health expenditure question and a two-week recall period. These thresholds represent the minimal value of those variables in our sample. A “typical” instrument would have six expenditure questions, five health expenditure questions and a one-month recall period. These thresholds represent the median value of those variables in our sample. An “extensive” instrument would have 2431 expenditure questions, 274 health expenditure questions and a 12-month recall period. These thresholds represent the maximum value of those variables in our sample. We used the point estimates to predict counterfactual values for each of the surveys in our sample, such that each observation has three counterfactual values (one for each type of instrument). For this exercise, we assumed that the surveys were collected through an interview and that the module on health expenditure was nested in the expenditure module. To generate confidence intervals we drew 1000 random samples from normal multivariate distributions based on regression coefficient point estimates and the variance-covariance matrix obtained from our main model and used them to generate 1000 estimates of health expenditure share. The middle 95% of these estimates are presented as our confidence intervals in Fig. 2.

Fig. 2.

Counterfactual scenarios of health expenditure shares by survey characteristic, South Africa, 1995–2008

Note: The x-axis represents time and the y-axis represents health expenditure share. The points in the figure represent actual data from surveys in South Africa that were included in our sample. The lower line represents our predicted values if those surveys had all employed an extensive instrument. The middle line represents our predicted values if those surveys had all employed a typical instrument. The upper line represents our predicted values if those surveys had all employed a minimalist instrument.

Fig. 2

Open in a new tab

Results

Regression results are reported in Table 3. The results are consistent with qualitative conclusions from the literature: The greater the number of health expenditure questions, the greater the health expenditure share. Other factors held constant, a one-unit increase in the number of health questions was accompanied by a 1% increase in health expenditure share. A one-unit increase in the number of total expenditure questions (while holding the number of health expenditure questions constant) was accompanied by a 0.2% decrease in health expenditure share. A one-month increase in the recall period was accompanied by a 6% reduction in health expenditure share. Surveys that employed a diary generated lower health expenditure shares. Country income classification, GDP and education were not found to be significantly related to health expenditure share and removing them from the model did not alter the statistical significance of the other independent variables namely, number of health questions, number of expenditure questions, recall period, and survey type.

Table 3. Regression results with the natural log of the health expenditure share as the dependent variable and survey characteristics as independent variables.

Survey characteristic Regression model 1
 Regression model 2
Coefficient (SE) P Coefficient (SE) P
Number of health expenditure questions 0.011 (0.004) 0.005 0.011 (0.004) 0.005
Number of total expenditure questions −0.002 (0.000) 0.000 −0.002 (0.000) 0.000
Recall period (month) −0.057 (0.012) 0.000 −0.065 (0.014) 0.000
Data collection mode −0.559 (0.160) 0.001 −0.514 (0.174) 0.005
Health module placement −0.273 (0.174) 0.123 −0.206 (0.214) 0.339
GDP per capita 0.193 (0.562) 0.733 0.310 (0.676) 0.649
Education (in years) 1.063 (0.991) 0.288 1.748 (1.140) 0.131
WB low-income country NA NA −0.209 (0.129) 0.112
WB lower-middle-income country NA NA −0.466 (0.279) 0.101
WB upper-middle-income country NA NA −0.513 (0.209) 0.017
WB high-income country NA NA 0.044 (0.052) 0.399
Open in a new tab

GDP, Gross Domestic Product; NA, not applicable; SE, standard error; WB, World Bank.

Note: The regression model also includes territory fixed effects, year trends and territory–year interaction terms.

Fig. 2 illustrates the counterfactual estimates for South Africa, which fielded surveys of all three types: minimalist, typical and extensive. As is evident in the figure, the instrument’s characteristics affect the estimated household health expenditure share.

The results yielded by minimalist, typical and extensive instruments differ dramatically. In most cases, the minimalist instrument results in health expenditure shares that are twice as high as those derived from the extensive instrument. This is problematic because unadjusted health expenditure shares (i.e. shares calculated without regard for the influence of the survey instrument) are routinely used to estimate two important metrics: the level of out-of-pocket expenditure reported in national health accounts and the level of catastrophic health expenditure across countries. For example, in the Philippines in 2003, catastrophic health expenditure was incurred by 8.3% of the households according to WHS data,14 but by only 0.8% according to data from the Family Income and Expenditure Survey.15 For both surveys, catastrophic health expenditure was based on the same threshold – 25% of household income. This dramatic discrepancy in the estimates generates mixed, confusing messages that policy-makers cannot properly interpret.

Discussion

Policy-makers need to rely on accurate and reliable out-of-pocket expenditure estimates. Household expenditure surveys were originally designed to measure consumer price index, living standards and household consumption for the national accounts, but not to measure out-of-pocket expenditure. Because of this limitation, the manual A system of health accounts: 2011 edition advocates an “integrative approach” to estimating private expenditure that involves making use of all available data sources, such as provider tax returns, pharmaceutical sales databases and household surveys.16 This approach would triangulate flows from these different channels to generate an accurate estimate.4 Although this approach is ideal, it is also impractical, especially in the near term for low-income countries. An interim solution would be to rigorously track the flow of funds at selected validation sites, as is done for the Medical Expenditure Panel Survey of the United States of America. This exercise would capture expenditure outflows from households to all health-care platforms in the community, including hospitals, clinics and pharmacies, and would provide a “gold standard” estimate of out-of-pocket expenditure that could then be used to adjust existing household survey data. Analysts will be able to systematically, reliably and accurately estimate out-of-pocket expenditure only if these validated estimates exist. Efforts should be made to ensure that policy-makers have access to data that capture reality rather than the idiosyncrasies of survey design.

Acknowledgements

We thank Eduardo Banzon, Joseph Dieleman, Gabriel Angelo Domingo, Karen Eggleston, Emmanuela Gakidou, Lizheng Shi, Tessa Tan Torres-Edejer and Madeleine Valera, as well as all panel participants at the 4th Biennial American Society of Health Economists and at the 2nd Symposium on Health Systems Research for their helpful comments. We also thank Brian Childress, Casey M Graves, Marissa Ianarrone, Annie Haakenstad, Katherine Leach-Kemon and Abby McLain for their generous assistance.

Funding:

This research was supported by core funding from the Bill & Melinda Gates Foundation. The funders had no role in study design, data collection and analysis, interpretation of data, decision to publish, or preparation of the manuscript. The corresponding author had full access to all data analyzed and had final responsibility for the decision to submit this original research paper for publication

Competing interests:

None declared.

References

  • 1.Xu K, Ravndal F, Evans DB, Carrin G. Assessing the reliability of household expenditure data: results of the World Health Survey. Health Policy. 2009;91:297–305. doi: 10.1016/j.healthpol.2009.01.002. [DOI] [PubMed] [Google Scholar]
  • 2.World Health Organization [Internet]. General statistical procedures used to construct WHO health expenditure database. Geneva: World Health Organization; 2012. Available from: http://apps.who.int/nha/database/ResourcesPage.aspx [accessed 11 April 2013].
  • 3.Guide to producing national health accounts: with special applications to low-income and middle-income countries Geneva: World Health Organization; 2003. Available from: http://www.who.int/nha/docs/English_PG.pdf [accessed 11 April 2013].
  • 4.Rannan-Eliya R, Lorenzoni L. Guidelines for improving the comparability and availability of private health expenditures under the system of health accounts framework Paris: OECD Publishing; 2010 (OECD Health Working Paper No. 52). Available from: http://dx.doi.org/10.1787/5kmbrcg0clvc-en [accessed 11 April 2013].
  • 5.National Statistical Coordination Board [Internet]. Philippine National Health Accounts (PNHA): private sources, out-of-pocket. Manila: NSCB; 2012. Available from: http://www.nscb.gov.ph/technotes/pnha/private.asp [accessed 11 April 2013].
  • 6.Lu C, Chin B, Li G, Murray CJL. Limitations of methods for measuring out-of-pocket and catastrophic private health expenditures. Bull World Health Organ. 2009;87:238–44, 244A-244D. doi: 10.2471/BLT.08.054379. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Heijink R, Xu K, Saksena P, Evans D. Validity and comparability of out-of-pocket health expenditure from household surveys: a review of the literature and current survey instruments Geneva: World Health Organization; 2011 (Discussion Paper No. 1-2011). Available from: http://www.who.int/health_financing/documents/dp_e_11_01-oop_errors.pdf [accessed 11 April 2013].
  • 8.Moore J, Stinson L, Welniak E. Income measurement error in surveys: a review. J Off Stat. 2000;16:331–61. [Google Scholar]
  • 9.Winter J. Response bias in survey‐based measures of household consumption. Econ Bull. 2004;3:1–12. [Google Scholar]
  • 10.Clarke PM, Fiebig DG, Gerdtham UG. Optimal recall length in survey design. J Health Econ. 2008;27:1275–84. doi: 10.1016/j.jhealeco.2008.05.012. [DOI] [PubMed] [Google Scholar]
  • 11.Silberstein AR. Response performance in the consumer expenditure diary survey Washington: Bureau of Labor Statistics; 1991. Available from: www.amstat.org/sections/srms/proceedings/.../1991_055.pdf [accessed 11 April 2013].
  • 12.Deaton A, Grosh M. Consumption. In: Grosh M, Glewwe P, editors. Designing household survey questionnaires for developing countries Washington: World Bank; 2000. [Google Scholar]
  • 13.Battistin E. Errors in survey reports of consumption expenditures London: Institute of Fiscal Studies; 2002 (IFS Working Paper). Available from: http://www.ifs.org.uk/publications/1873 [accessed 11 April 2013].
  • 14.The World Bank [Internet]. Health equity and financial protection datasheet: Philippines (English). Washington: WB; 2012. Available from: http://documents.worldbank.org/curated/en/2012/08/16715988/health-equity-financial-protection-datasheet-philippines [accessed 11 April 2013].
  • 15.The World Bank. Philippine health sector review: transforming the Philippine health sector, challenges and future directions Washington: WB; 2011.
  • 16.Organisation for Economic Co-Operation and Development. Eurostat & World Health Organization. A system of health accounts: 2011 edition Paris: OECD Publishing; 2011. Available from: http://www.oecd.org/els/health-systems/asystemofhealthaccounts2011.htm [accessed 11 April 2013]. [Google Scholar]

Articles from Bulletin of the World Health Organization are provided here courtesy of World Health Organization

RESOURCES