Household Health System Contributions and Capacity to Pay: Definitional, Empirical, and Technical Challenges

Transcription

1 Chapter 39 Household Health System Contrbutons and Capacty to Pay: Defntonal, Emprcal, and Techncal Challenges Ke Xu, Jan Klavus, Ke Kawabata, Davd B. Evans, Pya Hanvoravongcha, Juan Pablo Ortz, Radh Zeramdn, Chrstopher J.L. Murray Introducton In addton to mprovng populaton health, an mportant goal of health systems s to ensure that the fnancal burden of payng for health s dstrbuted farly across households (1). Explorng farness n fnancal contrbuton requres the ablty to measure each household s fnancal contrbuton (HFC), defned as the rato of a household s health system contrbutons to ts capacty to pay. Ths chapter, organzed nto fve sectons, ntroduces a method for estmatng the HFC from household survey data. Secton two presents the framework for analyss and the defnton of the numerator and denomnator of HFC. The thrd and fourth sectons descrbe n detal the calculaton of households health system payments through dfferent payment mechansms and the measurement of capacty to pay. In ths context, the data requred for estmaton are also presented. The last secton descrbes some remanng challenges concernng the measurement of capacty to pay, whch are related to the qualty of survey data. Defnng a Household s Fnancal Contrbuton The household fnancal contrbuton (HFC) represents the household s fnancal burden due to health system payments. It s a rato that relates total household expendture for health (HE) through general taxes, socal health nsurance contrbutons, prvate health nsurance premums, and out-of-pocket payments, to the household s capacty to pay (CTP). The capacty to pay of household s essentally ts effectve ncome mnus subsstence expendture requrements (SE): HFC HE = [1] CTP Ideally HFC s defned over a perod of one year, a unt of tme that encompasses many predctable fluctuatons n ncome and expendture. The perod of evaluaton of health expendture and effectve non-subsstence ncome s of theoretcal mportance. Dependng on the avalablty of varous formal and nformal mechansms to borrow and save, households may behave as f they smoothed ther ncome over longer perods of tme. In the extreme, the lfe cycle consumpton hypothess clams that households smooth consumpton over the stream of all future ncome (2). It s possble that n dfferent countres the perod over whch permanent ncome s defned wll vary, beng generally longer n hgh-ncome countres (3). In practce, HFC must be estmated usng data coverng a shorter perod, typcally one month, because surveys seldom nclude questons about expendtures over an entre year, let alone over a longer perod. Household Expendtures for Health The numerator (HE ) ncludes all fnancal contrbutons to the health system attrbutable to the household through taxes, socal securty contrbutons, prvate nsurance, and out-of-pocket payments. These nclude fnancal outlays that the household tself s not necessarly aware of payng, such as the share of sales or value-added taxes that governments then devote to health. As taxes and socal securty contrbutons are rarely earmarked for ther ultmate fnancng purpose, total household payments must be multpled by the share of these revenues that goes to fnance the health system.

2 534 Health Systems Performance Assessment The Concept of Effectve Income To operatonalze the denomnator of HFC, capacty to pay, t s necessary to defne effectve ncome and subsstence expendture. The noton of effectve ncome s meant to reflect household tendences to smooth consumpton over tme, takng nto account expected varatons n ncome, the household s assets (allowng for savng or non-savng), and future earnngs potental. There s a rch economc lterature on dfferent theores of how households make consumpton decsons. For example, n the lfe cycle ncome hypothess (4), households are assumed to smooth ther consumpton over the lfe cycle, so that expected consumpton s equal n all subsequent tme perods. One formulzaton of ths theory of consumpton behavor adapted to the crcumstances of health fnancng s: C 0 = Y + A + Y P δ 0 0 t t = 1 l t 1 + Pt δ t = 1 l t t [2] where C 0 s the consumpton of a household at tme t = 0, gven complete access to consumpton smoothng mechansms and the ablty to consume assets, Y t s ncome at tme t > 0, P t s the probablty of beng alve n each future year, A 0 s the annualzed value of assets (savngs or debts) at tme t = 0, and δ s 1/(1+r), where r s the market nterest or dscount rate. The lfe cycle hypothess s partcularly mportant under three sets of crcumstances: when households face predctable fluctuatons n ncome durng the course of the year; when ther ncome n future years s expected to change; and when they have postve assets (savngs) or negatve assets (debts). In these crcumstances the household s consumpton over the perod of tme that s usually ncorporated nto ncome and expendture surveys e.g. a month could be lower or hgher than ts observed earnngs over that perod. In order for a household to be able to smooth ts consumpton over long perods of tme effectve captal markets must be avalable. Ths nvolves access to formal or nformal mechansms that allows households to borrow on the bass of the present value of ther future earnngs, or to convert savngs that are n the form of assets nto monetary value. If the household possesses assets, these can be sold and converted nto ncome although temporary problems that mpede the sale and create lqudty problems for the household may exst. A more serous constrant s posed by the fact that n many countres mechansms that allow households to adjust ther consumpton by borrowng may not be readly avalable. Because of mperfectons assocated wth formal and nformal consumpton smoothng mechansms, the ncome that a household s able to consume and would seek to consume gven ts current ncome, assets and access to future earnngs, could dffer from that predcted by the lfe cycle hypothess. Where no mechansms exst to borrow or save, effectve ncome equals ncome at that tme; where mperfect mechansms exst, effectve ncome would be somewhere between current ncome and the expresson gven n equaton [2] (5). The effects of lmted access to a borrowng mechansm can be formulated as: L t Y0 + A0 + Yt Pt δ L t= 1 t C0 = Mn, Y A F Y P F L o t t tδ [3] t t= 1 1+ Pt δ t= 1 The expresson means that a household would lke to consume at the level suggested by the lfe cycle hypothess, but when ts access to borrowng s less than requred, t s forced to consume less. F t s a measure of the access a household has to future earnngs at tme t. When F t s zero, but F 0 > 0, the household cannot draw on future ncome, but s lmted n ts consumpton to ts current ncome and assets. At frst glance, the noton of consumpton smoothng may seem confusng. Fgure 39.1 shows an example of the movements of current ncome (Y), permanent ncome (PY), and effectve ncome (EY) over tme. Current ncome for the hypothetcal household s expected to ncrease rregularly for the next 15 years and then steadly decrease. If the household has access to perfect consumpton smoothng mecha- Fgure 39.1 Current ncome, permanent ncome, and effectve ncome $ EY, effectve PY, permanent Y, current Year

3 Household Health System Contrbutons and Capacty to Pay 535 nsms (and perfect foresght), t would be expected to consume along the dashed lne (PY). Durng the frst 8 10 years the household borrows money (or uses ts savngs), as ndcated by the fact that the PY lne les above the Y lne. Between the 10 and 25-year perod, the household makes savngs, as ts current ncome s hgher than ts expendture expressed by the permanent ncome lne, PY. From the 25th year onwards, the household uses ts savngs (or borrows) to acheve a hgher expendture level than that allowed by ts current ncome. Wth access to partal consumpton smoothng mechansms, the household s effectve consumpton may follow the dotted lne (EY). In the frst thrd of the consumpton cycle, the household does not take full advantage of the consumpton smoothng mechansm, and ts expendture follows n part the current ncome trend. In the mddle part of the cycle, the household saves but not at a rate that corresponds to the level that would be requred by complete consumpton smoothng. In the last years, the household dssaves and takes almost full advantage of the avalable consumpton smoothng devces. Because consderatons of farness n fnancal contrbuton are normatve, the denomnator n HFC needs to be defned n terms of some meanngful and comparable standard across households. In order to reflect the desre of households to smooth consumpton over tme and the lmtatons to consumpton smoothng exstng n many crcumstances, we defne effectve ncome as the level of consumpton that a household s able to acheve, based on a lfe cycle perspectve, and assumng that all households share a standard dscount rate. To avod any ambguty, t s assumed that effectve ncome s as defned n equaton [3] wth the added constrant that all households face the market nterest rate as the dscount rate. Because we defne capacty to pay n terms of effectve ncome, t leads naturally to certan conclusons about what should be ncluded n the denomnator. For example, subsdes rase a household s net ncome and, therefore, ts effectve ncome. Lkewse, tax payments lower the ncome the household receves. Because F t cannot be easly observed, estmatng effectve ncome presents a number of challenges. These are addressed n more detal n secton fve that dscusses mplementaton and ssues assocated wth data collecton and qualty. Subsstence Expendture The second step n defnng capacty to pay s to determne expendture requred for subsstence. There s an extensve lterature on basc needs whch addresses ths queston (6 9). Subsstence s often defned to nclude basc expendtures on food, shelter, and clothng. Ths expendture s subtracted from household total expendture n order to better capture the household s economc resources avalable for health and other spendng. Therefore, HFC can be consdered as a rato of health expendture to the ncome left after expendture necessary to keep the household alve has been subtracted. The choce of a measure of subsstence expendture should also be based on defntons that are comparable across populatons. Ways of estmatng subsstence expendture from survey data are dscussed n the next secton. Measurng Household Health Expendtures As mentoned earler, household contrbutons (10 11) to the health system nclude all drect and ndrect payment sources: general taxes, socal securty contrbutons, prvate nsurance premums, and out-of-pocket payments (12). Ths secton descrbes how these components can be captured from a household survey. Government Spendng on Health Household contrbutons to health that are channelled through government spendng comprse ncome tax, property tax, value-added taxes (VAT), sales tax, excse duty, corporate ncome tax, and other tax sources. In order to dstngush the part of government spendng that s used on health (GHE), each household s tax payments are multpled by the proporton of total government spendng allocated to fnance the health system (ncludng any government subsdes or transfers to socal health nsurance): GHE GHE = nctax vat excse other scalar x ( ) ( ( )) [4] GC [ ] The frst bracketed term s smply the fracton of total government consumpton (GC) that s allocated to health on a natonwde bass. The second bracket ncludes terms used for estmatng the government revenue orgnatng from the household. Usually only ncome tax (nctax ), value-added tax (vat ), and excse dutes (excse ) can be captured from a household survey. These consttute only a part of total taxes pad by the household. The remanng tax (such as corporate ncome, mport dutes or property taxes) and non-tax revenues (fees, fnes, etc.) must be estmated ndrectly. Even for ncome tax, t may be dffcult to obtan

4 536 Health Systems Performance Assessment accurate fgures from a household survey because: (a) people may under- or over-report ther ncome and correspondngly the ncome tax wll be underestmated or overestmated; (b) sometmes the tax treatment of an ncome source cannot be clearly dentfed for partcular countres; (c) even when ncome subject to tax can be clearly dentfed, there may be varous deductons that may be dffcult to capture from the nformaton provded n the survey, despte knowledge of a country s tax laws. The estmaton on VAT/sales tax s more straghtforward than the estmaton of ncome tax because the offcal rates n a country can smply be appled to the reported purchases n the surveys. However, nconsstences may stll exst because of: (a) memory bas assocated wth certan expendture tems n the survey; (b) a complcated structure of applcable tax rates; (c) the fact that excse dutes are sometmes leved on quantty purchased nstead of prce, but the household survey may only record the money value of that tem. Because of the lkely under- and over-reportng n surveys, an adjustment scalar s used to approxmate the total government revenue receved from households. In dong so t s assumed that the dstrbuton of the non-observed tax and non-tax revenue across households s dentcal to the dstrbuton of the observed tax revenue from the survey. The scalar s defned as the rato of expected government revenue (GC e ) to the government revenue (GC s ) estmated from the survey: GCe scalar( x) = [5] GC Expected government revenue shows how much government revenue would be generated by the tax payments of all households ncluded n the survey, f the rato of government revenue to GDP reported n atonal Account estmates appled. For the estmaton of GC e, the GDP mpled by the expendture reported n the survey (gdp s ) must be calculated. It s calculated combnng survey and atonal Accounts nformaton as follows: pc w s (exp ) gdps = = [6] ( PC / GDP) ( PC / GDP) s pc s s prvate consumpton from the survey, PC/GDP s prvate consumpton share of GDP at the natonal level, exp s household consumpton expendture, and w s the household weght from the survey. Survey weghts are appled to account for estmaton bas arsng from samplng desgn and systematc non-response n the sample. Once the survey GDP has been calculated, GC e can be derved from: GC = gdp ( GC / GDP) [7] e s where (GC/GDP) s the government consumpton share of GDP at the natonal level. Government tax revenue from the survey s calculated as: GCs = w( nctax + vat + excse + other ) [8] When formulae [7] and [8] are substtuted nto formula [5], we have: scalar( x) = ( GC / GDP) gdp w nctax + vat + excse + other ( ) s [9] As mentoned earler, the scalar x s desgned to capture the part of government tax revenue that cannot be estmated from the survey data. However, t also ncludes any measurement error from the survey. In order to separate these effects, scalar x could be decomposed nto two parts one comprsng the survey measurement error (scalar x 1 ) and another comprsng the mssng tax nformaton part from the survey (scalar x 2 ). In equatons [10] to [12] natonal level data are denoted by uppercase letters and survey data by lowercase letters: scalar( x ) 1 = [ ] w( nctax + vat + excse + other ) ( ICTAX + VAT + EXCISE + OTHER)/ GDP gdp s [10] Ths equaton represents the extent to whch the household tax contrbutons derved from the survey reflect total natonal recepts from those sources alone, and GC / GDP gdp s scalar( x2) = scalar( x ) w nctax + vat + excse + other 1 ( ) ( ) Substtutng for scalar x 1 n equaton [11] gves: GC scalar( x2 ) = ( VAT + ICTAX + EXCISE + OTHER) [11] [12] that s the extent to whch total government tax revenue s provded by the forms of taxes to whch the households contrbute drectly and whch were captured from the analyss of the surveys. Scalar x 2 equals 1 f the government collects taxes only from households, whle t s less than 1 f t collects revenue from other sources. It should be noted that scalar x = (scalar (x 1 )) (scalar (x 2 )). The advantage of decomposng the scalar s that the two sources of

5 Household Health System Contrbutons and Capacty to Pay 537 under- or overestmaton could be more clearly dentfed. In the analyss, total household consumpton s assumed to equal total prvate consumpton. Strctly speakng, ths s not the case snce prvate consumpton s the market value of all goods and servces purchased, or receved n knd, by households and non-proft nsttutons. If the non-proft component s large, the scalar wll be underestmated as a result of underestmaton of survey GDP. Whenever ncome tax (nctax ) s not avalable drectly from a survey, t s estmated from reported ncome and the country s tax schedule nformaton. Reported ncome ncludes salares and non-salary earnngs from all employment actvtes. on-salary earnngs nclude n knd benefts. Employment ncludes self-employment as well as a second job when relevant. The ncome tax pad by each ndvdual n the household s then aggregated to a monthly value at the household level. The queston arses as to whch ndvduals are subject to ncome tax. In many countres, partcularly the poorer ones, only formal sector employees pay ncome taxes. The way to dentfy formal sector workers vares by country. Usually answers to the job classfcaton questons n the surveys wll ndcate whether an ndvdual works n the prvate, publc or nformal sector. Other methods of dentfcaton can be used n more dffcult cases. Sales tax or value-added tax (vat ) and excse dutes can be mputed from household expendtures on varous categores of goods and servces. Ths nvolves applyng the tax rates derved from offcal tax documents to household expendtures on the correspondng commodtes reported n the survey. The nformaton on other taxes (other ), such as those pad on real estate, can often be obtaned drectly from the household survey. Otherwse, the value of property owned by the household can be estmated from the questonnare. Tax rates obtaned from the tax schedule of each country are then appled for the calculaton. Socal Health Insurance Contrbutons The second component of HE that needs to be estmated s the total socal health nsurance contrbuton of the household (SSH ), whch can be formulated as follows: SSH = soc ( scalar( y)) [13] Household socal securty contrbutons (soc ) are computed smlarly to ncome tax. If socal securty contrbutons are provded drectly by the survey n the form of a specfc queston on payments or contrbutons, ths nformaton s used. When ths s not the case, the offcal contrbuton rate s appled to the salary from the prmary job of the ndvdual (after determnng whether the earnngs are pre-tax or post-tax). The assumpton s that only formal sector employees, or full-tme permanent workers, contrbute to socal securty. Although contrbuton rates may vary wth respect to level of ncome, and sometmes the sector of the economy n whch the ndvdual works, t s assumed that the employer s contrbuton share s borne by the employee n the form of reduced net salares. For the computaton, ths mples that employers contrbutons should be added to those of the employee. Ths assumpton s generally appled n the tax ncdence lterature and t smplfes the analyss and comparson across countres (13). As wth ncome tax, the socal securty contrbutons by ndvduals are summed to obtan the contrbutons at the household level. Because only a share of socal securty payments are used n the health sector, the scalar y s ntroduced. It can be formalzed as: gdps( SSH / GDP) scalar( y) = w ( soc ) [14] The numerator reflects the expected socal securty contrbutons to health for the GDP observed n the survey the survey GDP (gdp s ) multpled by the share of health socal securty payments n GDP at the natonal level. The denomnator s the weghted sum of the household contrbutons to socal nsurance of all forms. Ths scalar also can be decomposed nto two parts. The frst s due to measurement error of socal securty contrbutons n the survey n some household surveys, socal securty contrbutons reported at the household level can, n sum, dffer from the value stated n the atonal Accounts. These dscrepances are essentally the result of under- or over-reportng socal securty contrbutons n the household survey. Ths can be expressed as: gdps( SOC / GDP) scalar( y1 ) = w( soc) [15] or the extent to whch reported household contrbutons sum to the contrbutons expected from natonal aggregates. Ths scalar could be hgher or lower than unty.

6 538 Health Systems Performance Assessment The second part of scalar y s related to the fact that only part of socal securty payments go to health. Ths component of the scalar y can be expressed as: gdps( SSH / GDP) scalar( y2) =. [16] scalar( y ) w ( soc ) 1 After substtutng for scalar y 1 n equaton [16], scalar y 2 can be expressed as SSH/SOC, or the proporton of socal securty payments gong to health. ote that, as n the case of the x scalar, scalar y = (scalar (y 1 )) (scalar (y 2 )). Prvate Health Insurance The thrd component of HE s the prvate health nsurance premums pad by households (PRV ). In most cases, data are avalable drectly from the household survey. In some countres, employers contrbute to prvate health nsurance on behalf of ther employees (we contnue to assume that the employer s contrbuton to prvate health nsurance s de facto part of the employee s ncome). Hence the employers contrbutons should also be ncluded n the analyss. It s lkely that excludng the employers part wll lead to underestmaton of ths component of prepayment n countres where prvate health nsurance plays a domnant role n health system fnancng, and employers subsdze employees prvate health nsurance premums. In the case of socal health nsurance and tax contrbutons, a scalar was used to adjust the level of household spendng to natonally reported fgures. Ths procedure, however, cannot be undertaken n the case of prvate nsurance premums snce relable sources of the employers contrbuton share at the natonal level are usually not avalable. The same apples to out-of-pocket payments. To avod an upward bas n the estmaton of prvate health nsurance contrbutons, premum refunds or credts granted from the prvate nsurance company, for example for not usng the servces n a prevous perod, must be deducted from the declared level of household prvate health nsurance premums. Out-of-Pocket Payments Out-of-pocket payments (OOP ) nclude all categores of health-related expenses pad drectly by the household at the tme the household receves the health servce. Typcally these nclude doctor s consultaton fees, purchases of medcaton, and hosptal blls. Spendng on alternatve and/or tradtonal medcne s ncluded n out-of-pocket spendng, whereas expendture on health-related transportaton s excluded. It s mportant to note that some people may be payng out-of-pocket for health care, but recevng a rembursement later from socal and/or prvate health nsurance schemes. To avod ntroducng an upward bas n health expendtures, rembursements are deducted from household gross out-of-pocket payments. Detals of these rembursements are usually gven n the surveys. Total Household Health Expendtures Puttng the components together, household s health expendture comprses ts payments to government that are channelled to health, as well as ts health-related socal nsurance contrbutons, prvate health nsurance contrbutons, and out-of-pocket payments: HE = GHE + SSH + PRV + OOP [17] Measurng Household Capacty to Pay As mentoned n secton two, household capacty to pay s defned as effectve ncome net of subsstence expendture. Ths secton descrbes how to estmate effectve ncome and subsstence expendture from survey data. Effectve Income It s not possble to try to estmate permanent ncome over the lfe cycle n a mult-country context wth the surveys avalable, whch would requre nformaton on lkely future ncome, assets, and the potental to borrow or lend. To reduce the short-term fluctuatons observed n ncome data as reported n surveys, however, household consumpton expendture s used as the proxy for effectve ncome. Ths choce s based on two consderatons. Frst, the varance of current expendture s smaller than the varance of current ncome over tme. Income data reflect random shocks whle expendture data conform better to the noton of effectve ncome. In defnng capacty to pay, t s mportant to try to elmnate the effect of random shocks on ncome to the greatest extent possble. Second, n most of the household surveys expendture data are more relable than ncome data. Ths s partcularly true n developng countres where the nformal sector s typcally relatvely large and survey

7 Household Health System Contrbutons and Capacty to Pay 539 respondents may not wsh to reveal ther true ncome for varous reasons (14 15). Subsstence Expendture It was argued earler that household capacty to pay for health servces should not be determned wth respect to ts total effectve ncome. The reason s that unless households frst meet basc subsstence needs, they wll not be n a poston to fnance health servces. In The World Health Report 2000, actual food expendture was used as a proxy for household subsstence expendture (1). However, food expendture may not capture the actual subsstence expendture of the household, even though effort was made to lmt ths expendture to basc food only. A rch famly may spend a greater absolute amount on food than a poor famly although the food expendture share of total household consumpton expendture stll follows Engel s law.e. the share of food to total ncome falls wth ncreases n ncome (16). For ths reason, the approach has been modfed. In order to elmnate spendng on non-essental food and to mprove nternatonal comparablty, one possblty would be to use the nternatonal poverty lne as a measure of subsstence expendture. Ths poverty lne s set at one nternatonal dollar per day per person n 1985 currency terms, frst used by the World Bank n the World Development Report 1990 (17). The one nternatonal dollar subsstence level s based on a study of absolute poverty lnes n 33 countres (18). Ths alternatve was explored by adjustng the 1985 level to nomnal unts correspondng to the year of the household survey usng an approprate prce deflator. To account for dfferences n consumpton patterns and prces, food PPPs (Purchasng Power Partes) rather than general GDP PPPs, were used to convert the nternatonal poverty lne expressed n nternatonal dollars nto local currency unts. Ths converson was made to express subsstence expendtures n the same unts as data collected n the surveys. Fnally, an adjustment for household sze was made to brng the poverty lne, defned at the ndvdual level, to the household level. The ntroducton of the food poverty lne elmnates the problem that actual food expendture ncludes spendng on non-essental food for many households. It also mproves nternatonal comparablty. Snce the food poverty lne stays the same as ncome ncreases, more progressvty s bult nto the dstrbuton of HFC compared to the stuaton where actual food expendture s used. Fgure 39.2 shows the share of actual food expendture and the share of the food poverty lne n total household consumpton expendture across expendture decles. The food poverty level share of total expendture declnes more rapdly wth ncreases n expendture (ncome) than does actual food expendture. Ths ndcates that the capacty to pay of rcher households would be underestmated by usng actual food expendture. The problem wth the nternatonal poverty lne s that several causes of uncertanty are unavodably bult nto estmaton. These nclude varous problems assocated wth the constructon of food PPP converson factors. For ths reason, we explored an approach that partly resembles the assessment of natonal poverty lnes. Usng the observaton that food expendture as a proporton of total expendture ncreases wth ncreasng poverty, a food share based poverty lne for each country was estmated. The poverty lne for a gven survey was set equal to the average food expendture of households whose food share of total expendture was n the 45 to 55 percentle range (used n preference to the sngle household at the 50th percentle). Ths was adjusted for household sze usng an equvalence scale of eqsze = hhsze β. The adjustment factor β was obtaned from household survey data from all 59 countres usng the followng fxed-effects regresson: n 1 ln food = lnk + β ln hhsze + γ country + ε = 1 Fgure 39.2 Subsstence expendture as a share of total consumpton expendture, Bangladesh % Expendture decle Actual food Poverty lne & adjusted total expendture [18]

8 540 Health Systems Performance Assessment The estmated value of β was wth confdence nterval The estmaton of food poverty lne s explaned further n Xu et al. (19). Ths approach has the advantage that t does not requre the estmaton of PPPs for countres where prce observatons have not been made, nor does t requre as many ntermedate calculaton steps that can ntroduce uncertanty n the analyss. In addton, t s an ndcator estmated drectly from the survey data. The food share based measure stll elmnates luxurous food spendng and ntroduces more progressvty to the underlyng expendture dstrbuton than the orgnal measure based on actual food expendture. Household Capacty to Pay The numerator of the HFC comprses all health expendtures by the household, ncludng those deducted at source (e.g. tax and socal securty payments). The denomnator of HFC, household s capacty to pay (CTP ), s a measure of the non-subsstence effectve ncome of the household (effectve ncome mnus subsstence expendture). Total household expendture s used as the proxy for effectve ncome. However, the expendture reported n household surveys does not nclude the health expendtures deducted from ncome at source, whch are ncluded n the numerator. In order, then, to mantan consstency between the numerator and the denomnator, tax and socal securty contrbutons deducted at source must be added to the denomnator whch becomes: CTP = EXP SE + GHE ndtax ( GHE / GC) + SSH [19] The expresson GHE ndtax (GHE/GC) represents that part of household tax contrbutons whch s deducted at source ndrect taxes (ndtax ) are ncluded n EXP and are not deducted at source. Household expendture nformaton s avalable drectly from the household survey and converted nto a monthly value. Total household consumpton expendture (EXP ) ncludes both monetary and n knd payments on all goods and servces, as well as the money value of the consumpton of householdmade products. Household consumpton expendture ncludes ndrect taxes such as the VAT/sales tax, excse dutes, as these taxes are vewed as part of the household s capacty to pay. As a means of qualty control, responses from households reportng zero expendture or zero food expendture were consdered to be reportng errors and were excluded from the analyss. Addtonal Data Requrements and the Reference Perod The way the household fnancal contrbuton (HFC) can be calculated usng nformaton from natonal household ncome and expendture surveys was the focus of the last two sectons. It was shown, however, that addtonal nformaton was requred ncludng detaled government taxaton documents, and natonal health accounts fgures. The man sources of ths addtonal nformaton are: Government taxaton documents, ncludng nformaton on the systems of ncome tax, sales tax, value-added tax, excse tax, and property tax. atonal Health Accounts (HA) fgures. WHO provdes yearly estmates of varous components of health expendture from prvate and publc sources for all ts Member States, and these were used to gve a reference for checkng the relablty of the survey data (1). Socal securty and health nsurance laws that provde nformaton on premums and other contrbutons to the health system. Ideally the HFC would be measured over a long enough perod to allow smoothng of consumpton. In practce, ths s not possble where data must be taken from exstng surveys, whch ask about short recall perods, usually less than a year. For ths reason, all the varables needed for computng HFC are converted nto monthly fgures regardless of the recall perods used n the ndvdual surveys. If the recall perod n any survey was greater than one month and the nflaton rate dffered between the months for whch nformaton was sought, expendtures were deflated to a common month usng the local consumer prce ndex. Summary and Conclusons In summary, the household fnancal contrbuton, or HFC, s defned as a rato of health payments made by the household to ts capacty to pay. Household health payments consst of four sources: general taxaton, socal securty contrbutons, prvate health nsurance premums, and out-of-pocket payments. Household capacty to pay s measured as non-subsstence effectve ncome, whch s n practce calculated as total household expendture net of the food poverty lne. Most of the requred nformaton can be obtaned

9 Household Health System Contrbutons and Capacty to Pay 541 from household survey data combned wth knowledge of the tax and socal securty systems n the dfferent countres. However, a number of challenges assocated wth mprovng the qualty of the data and comparablty across countres reman. Measurement error n the context of ncome or expendture data derved from surveys s a well known problem (20 21). Measurement error could be ntroduced at any stage of the survey: desgn of the survey nstrument, data collecton or data entry. The typcal problem wth ncome survey data s underestmaton. Ths s due to the desgn of the questons and the tendency of respondents to understate ther true earnngs. The more detaled the ncome queston, for example the more ncome categores used, the more accurate the fgures generated. Another problem assocated wth the constructon of ncome questonnares s whether the respondents understand the desred ncome concept correctly and n the same way. If the questons and the mplementaton of the survey are not carefully controlled, stuatons may arse where some of the respondents report ther ncome gross of taxes whle others are reportng ther ncome net of taxes. These problems also complcate comparson of ncome-based measures between countres. Whle expendture may be reported more accurately by household book-keepng, certan expendture tems may be dffcult to capture, or they may be only partly captured by the survey. It may, for example, be dffcult to record and mpute correctly the value of home producton that s consumed n the home rather than sold. If consumpton from own-producton s substantal, t s possble to produce per capta expendture fgures that are hgher than the correspondng GDP fgures derved from natonal accounts. To the extent that these dfferences are real and they are due to the fact that home-producton s not adequately mputed n the calculaton of the country GDP, ths dscrepancy s acceptable. However, n some cases the value of ownproducton s clearly overestmated n surveys and t s then dffcult to know how to treat the survey data. The tme perod over whch households are asked to recall ther expendtures s not dentcal across surveys or countres. Ths has been a concern n the analyss of catastrophc expendtures. A short recall perod wll have a smaller memory bas than a long recall perod, whle the latter may capture catastrophc expendtures better than the former. The drecton of bases generated by dfferent recall perods s not self-evdent. In estmatng government spendng on health, t s assumed that government revenue comes from general taxes, whch s true for the majorty of countres. When revenue from other sources, however, such as the sale of ol or other natonal assets s substantal, the queston arses as how or whether to assgn such revenue to households. As there are no common gudelnes on how best to deal wth revenue accrung from natonally owned assets, t was decded n ths context to exclude t from the calculaton of HFC. Ths decson was based on the consderaton that revenues arsng from the sales of natonal assets do not represent ether a fnancng burden to the household or an ncrease n capacty to pay that s freely at ts dsposal. The effect of ths knd of government revenue wll be reflected n the HFC snce t wll reduce health servce prces n publc facltes or ncrease health funds n publc nsurance, thereby reducng the out-of-pocket payments of households. The applcaton of scalars to adjust for the unobserved part of government (or socal health nsurance) revenue s necessary for obtanng estmates that are consstent wth macro-level nformaton. If, for some reason, the households tax outlays are overestmated or underestmated, ths wll have a correspondng mpact on HFC and any summary measure of the dstrbuton calculated on the bass of these ratos. Dscrepances between survey and natonal data can occur f substantal parts of government revenue accrue from state-owned enterprses, non-tax revenue, or external donatons rather than from households. Bas may also be generated f the ncome data that are used for estmatng ncome taxes, and n some cases the socal health nsurance contrbutons, are of poor qualty. The adjustment scalar s crtcal to ensure that the level of the revenues s correct, although the dstrbuton of ths revenue s not known. The fact that the unobserved revenue s beng contrbuted to households n the same proportons as the observed revenue, could n some cases lead to msleadng estmates of the true dstrbuton of the HFCs. As explaned above, the household fnancal contrbuton to the health system should nclude all the components pad drectly or ndrectly by the household. In prncpal, prvate health nsurance premums pad by both employer and employee should be ncluded n the calculaton. Informaton on employer s contrbuton to prvate health nsurance s very dffcult to obtan from household surveys. In countres where prvate nsurance s substantal and employers partcpate n ts fundng, lke n the Unted States, household health expendtures and consequently HFC could be underestmated. Because of the tme lag between the recorded outof-pocket payment and the nsurance rembursement

10 542 Health Systems Performance Assessment receved at a later date, negatve out-of-pocket payments could be estmated for some households. For the same reason, negatve drect taxes mght occur from ncome regster data. Two possble solutons have been suggested: one s to delete the observatons wth negatve values and the other s to set the negatve value at zero. The currently followed practce n HFC calculaton conforms to the latter approach. Household surveys are a rch source of data that allows for a detaled mcro-level analyss on a relatvely comparatve bass across countres. evertheless, dfferences n survey desgn and qualty exst. In addton, some of the avalable surveys are not very recent and may not be sutable for analysng health system farness f substantal health fnancng reforms have taken place snce they were undertaken. It s mportant, therefore, to contnually try to mprove survey desgn and conduct so that the problems assocated wth data qualty and comparablty are reduced. Acknowledgements The authors are grateful to Guy Carrn and Wllam D. Savedoff for comments on an earler draft, to Felca Knaul for her contrbuton to the early stage of the development of the methodology, to Ana Mylena Agular-Rvera for formattng the tables and graphs, and to athale Etomba and Anna Moore for data and reference searches. References (1) World Health Organzaton. The World Health Report Health Systems: Improvng Performance. Geneva, World Health Organzaton, (2) Ando A, Modglan F. The lfe cycle hypothess of savng: aggregate mplcaton and tests. Amercan Economc Revew, 1963, 53: (3) Fredman M. A theory of the consumpton functon. Prnceton, Prnceton Unversty Press, (4) Romer DH. Advanced macroeconomcs. ew York, McGraw-Hll, (5) Behrman P. Health sector reform n developng countres: makng health development sustanable. Boston, Harvard Unversty Press, (6) Sen A. Poverty and famnes: an essay on enttlement and deprvaton. Oxford, Clarendon Press, (7) Sen A. Resources, values and development. Oxford, Blackwell, (8) Sen A. The standard of lvng. ew York, Cambrdge Unversty Press, (9) Streeten P et al. Frst thngs frst: meetng basc human needs n the developng countres, ew York, Oxford Unversty Press, (10) Fuchs VR. Who shall lve? Health, economcs and socal choce. Stanford, Stanford Unversty, (11) Iglehart JK. The Amercan health care system - expendtures. The ew England Journal of Medcne, 1999, 340: (12) Murray CJL et al. Defnng and measurng farness n fnancal contrbuton to the health system. EIP Dscusson Paper o. 24. Geneva, World Health Organzaton, URL: papers/dscusson_papers.cfm# (13) Wagstaff A et al. Equty n the fnance of health care: some further nternatonal comparsons. Journal of Health Economcs, 1999, 18: (14) Deaton A. Understandng consumpton. Oxford, Oxford Unversty Press, (15) Bous HE. The effect of ncome on demand for food n poor countres: are our food consumpton databases gvng us relable estmates? Journal of Development Economcs, 1994, 44(1): (16) Mas-Colell A, Whnston MD, Green JR. Mcroeconomc theory. ew York, Oxford Unversty Press, (17) World Bank. World development report: poverty. ew York, Oxford Unversty Press, (18) Ravllon M et al. Quantfyng the magntude and severty of absolute poverty n the developng world n the md-1980s. Pre Workng Paper Seres o Washngton, DC, World Bank, (19) Xu K et al. Understandng household catastrophc health expendtures: a mult-country analyss. In: Murray CJL, Evans DB, eds. Health systems performance assessment: debates, methods and emprcsm. Geneva, World Health Organzaton, (20) Pauln GD, Ferraro DL. Imputng ncome n the consumer expendture survey. Monthly Labor Revew, 1994, 117(12): (21) McGregor P, achane D. Identfyng the poor: a comparson of ncome and expendture ndcators usng the 1985 Famly Expendture Survey. Oxford Bulletn of Economcs & Statstcs, 1995, 57(1):