Sei sulla pagina 1di 16




J. M. H a g e n a a r s




M. S.


P r a a g **

Leyden University, C enter fo r Research in Public Economics A fu n d am en tal p o in t o f discussion in poverty research is w h eth er poverty is an absolute o r a relative concept. If poverty is seen to be a situation o f absolute depriv atio n , a poverty line will usually be defined to be i n d e p e n d e n t o f the general style o f living in society. If poverty is co n sid ered to be a situation o f relative dep riv ation , a poverty line will be defined in relation to the general style o f living in society. T he choice for o n e o f these two ap p ro ach es has im p orta nt co nsequences for social policy, as absolute poverty m ay be red u ced by eco n o m ic grow th, while relative poverty will only decrease w hen incom e inequality decreases. This p a p e r suggests a poverty line definition that is not a priori m eant to be either absolute o r relative, but dep en d s on the perception o f poverty in society. If the poverty line is higher in countries with higher m edian incom e (as an indicator o f general style o f living ) the poverty line is said to be relative; if the poverty line does not vary with m edian income, it is said to be absolute. The poverty line definition suggested ap p ears to be a generalization o f almost all well-known poverty line definitions. Poverty lines th u s defined are estim ated for eight E u ro p ean countries on the basis o f a 1979 survey. The resulting lines a p p e a r to have an elasticity with respect to m ed ian incom e o f 0.51, an d hence can be said to be halfway on the scale betw een ab so lu te an d relative. 1. I n t r o d u c t i o n

In the present situation o f worldwide econom ic recession, poverty research is o f increasing importance. A basic problem encountered in poverty research is the identification o f people living in poverty. This problem is usually solved by the introduction o f a poverty line, an income level that is considered to be the borderline between the poor and the non-poor. Many different poverty line definitions have been proposed, reflecting equally many different views on the nature o f poverty. They vary from a certain fixed level o f purchasing power to decile-definitions o f poverty. The former definition arises from an absolute poverty concept, where poverty is seen as a situation o f insufficient command over resources, in dependent o f the general welfare level in society. The latter definitions arise from a relative poverty concept, where poverty is seen as a situation o f purely relative deprivation. This implies that most poverty research is based on an a priori notion on the nature o f the p h enom en o n to be analyzed. In this paper a general parameterized definition o f a poverty line is proposed, which encom passes all these views for different values o f the parameters. The definition is based on the perception o f poverty in the population, instead o f on the researchers perception o f poverty. It is seen that several well-known poverty line definitions arise as a special case o f the proposed general poverty line
"'This research is part o f the Leyden Incom e E valuation Project. We th an k H uib Van de Stadt for his co m m en ts on an earlier draft. The research described here has benefited from a discussion p ap er by M ichael O 'H igg in s (1980), which co m m ented u p o n earlier research. The responsibility for any rem aining errors is with the authors. This p ap er was presented at the 1983 C o n feren ce o f the International A ssociation for Research on Incom e and Wealth. Part o f the p a p e r was also presented at the Venice C o n feren ce on Poverty, May, 1982. (Van Praag (1982)). **The present affiliation o f the second a u th o r is with the E co n om etric Institute, Erasm us University, Rotterdam . 13 9

definition. The structure o f this paper is as follows. In section 2 a review o f some frequently used poverty line definitions is given, and it is sh own that all can be seen as special cases o f a general principle. In section 3 the so-called Leyden Poverty Line (LPL) is described, and it is shown that the LPL also belongs to the general class. Section 4 serves to compare the different measures. Section 5 gives an illustration o f the method described, while section 6 concludes.




Po verty

L ine

efin itio n s

Poverty is the counterpart o f well-being. Hence poverty may also be defined as lack o f welfare. Let welfare U be assumed to depend on som e vector v describing the aspects o f an individual's position thought to be relevant for the individuals welfare according to som e functional relation U ( y ) . As soon as a critical level 8 on the welfare scale is identified as the poverty border, then the corresponding poverty line y a is defined as the solution o f U ( y s ) = 8. Notice that if _ is a scalar and U y is an increasing function, also y s will be a unique poverty line. If y is a vector, say, income and leisure, there will be a poverty boundary in (income-leisure)-space. If U d epends also on personal characteristics x, for instance family size, then the solution o f U ( y 6, x ) = 8 will be y fi =>'fi(Ar) and we find a poverty line (bou n d ary) differentiated with respect to x. Thus the definition o f the poverty line (boundary) d ep ends on three elements: (a) the choic e o f the relevant variable(s) y. (b) The choic e o f the function U, by its very nature a proxy relationship. (c) The critical level 8. In this paper an economic definition o f poverty will be used (see e.g. Watts (1968)), that is, poverty is considered as a situation where income, representing command over resources, falls below a certain level. That is we define our y to be the scalar income. The value y * will be called the poverty line. In the literature a host o f poverty definitions have been proposed. These notions on the one extreme consider poverty to be a situation o f deprivation o f certain basic goods and services necessary for maintaining physical subsistence, independent o f the level o f well-being in society. Poverty line definitions based on this notion will be called absolute poverty line definitions. At the other extreme poverty is seen to be essentially a situation o f relative deprivation: as Atkinson (1975, page 186) puts it: It is misleading to suggest that poverty may be seen in terms o f an absolute standard which may be applied to all countries and at all times, indepen dent o f the social structure and the level o f development. A poverty line is necessarily defined in relation to social conventions and the contemporary living standards o f a particular society. Definitions o f this type will be called relative poverty line definitions. Most poverty line definitions can be placed on a scale between purely absolute and purely relative; in other words, they will have an income elasticity between zero and one (see e.g. Kilpatrick (1973)). The ch oic e o f a place on this scale has important implications, both for the extent o f poverty that will be measured and for the policy that is needed for a reduction o f poverty.

A number o f poverty indices have recently been proposed for measuring the extent o f poverty. (See e.g. Sen (1976), Takayama (1979), Kakwani (1980), Thon (1979,1981)). In this paper we shall restrict ourselves to the simplest index possible, i.e. the percentage o f people with an income lower than the poverty line. We will consider the effect o f changes in the income distribution on this poverty percentage for various poverty definitions. These changes will be either changes in income inequality, to be measured by the standard deviation o f log-incomes o\, or changes in mean-log-income iy, as an indicator o f the average income level. We will describe the income distribution by the distribution function o f log-incomes F (ln y; In the special case that the income distribution is lognormal F equals the normal distribution function with parameters /xv and a r We furthermore assume that all incomes will be incomes per equivalent adult; we will return to the effect o f family size on the poverty lines in section 4.

(a) Basic-needs Approaches Definitions o f this type are based on the ch oic e o f a certain food basket c0, just sufficient to stay alive. In order to transform these costs for food to an income level, which is considered to be the poverty line ys, a certain amount is added to account for other items like clothes and housing. This approach has been the basis for numerous poverty line definitions, starting with Booth (1892) and Rowntree (1901), up to Orshansky (1965, 1968). However, different results arise dep endin g on the way the poverty line in terms o f total expenditures is calculated from the expenditures on food only. ( a l ) Tlte Basic Needs Approach According to Rowntree Rowntree (1901) added a fixed amount o f money, in order to cover distinctly described items like fuel, food and rent. If these other costs are represented by oc'o, the poverty line is simply y B = c0+ o c 0. The general principle U { y s ) : 8 now results in the choice o f a linear function = U ( y ) = y, and 8 is chosen to be c0+ o c 0. This poverty line is obviously absolute; the elasticity o f this line with respect to median income is zero. The percentage o f poor in the population, given this poverty line definition (which we will denote by z, hereafter) is rim, II, = J o o As an illustration we consider the situation where incomes are lognormally distributed, so that F (ln y; (M o-y ) = N ( In y; /xr, a y ). y,

dF(\r\ y \ fLcry ).

W e n o w ha v e

n, . n ( ^ ; 0, l) = N 0 " (Co+ OCo)~ ^ , 0,1

\ crv ) \ (Ty It is seen that this percentage decreases if /jlv increases; it also holds that for In (c0+ oc0) < fjL the poverty percentage decreases if income inequality decreases. y (a2) Basis-Needs Approach According to Orshansky Another poverty line definition based on the costs o f food is given by Orshansky (1965, 1968). Orshansky transformed food costs c0 to an income level by multiplication by an estimate o f the average in com e-food ratio in society. S uppose that the relationship between expenditures on food, c, and family income y may be described by a double logarithmic Engel function (see e.g. Prais and Houthakker (1955), Cramer (1973), Van Praag, Spit, Van de Stadt (1982)), (1) In c = a 0+ a . In y.

The average food-in com e ratio ( c / y ) , used to transform expenditures on food into a poverty line, is chosen to be the geometric average in the popu la tion,1 that is

() 2

(y)=exp'('n y)
= exp Ev(ln c - In y) = exp E ( a 0+ ( a , - 1) In y ) + o -a = exp J oc (<*o+(a i - 1 ) ln > 0 d F (In v; fiy, <rv)

= exp ( a 0 + ( a , 1)/xv) where E v{ ) stands for the mathematical expectation with respect to y. The poverty line corresponding to a specific food basket c0 is now found to be (3) ^8 = C [ ( v ) ] = c e x P

This poverty line may be seen to result from the general principle with U (v)= y and < = c0 5


It appears that the resulting poverty line dep ends on mean log-income in society. The elasticity o f this poverty line with respect to median income, exp (^tr ), equals (1 - a | ) . The percentage o f poor in the population, given this poverty line definition denoted by z2 is

n 2= r

J -C O

1dF (\n

y, iv crv , .).

' i f a n o th e r indicator o f the average food-incom e ratio is used, (2) will becom e slightly more com plicated. For any specific functional form o f the incom e distribution, however, the poverty line may again be derived.


In case o f a lognormal incom e distribution this amounts to n 2 = N ( 'n ^ ~ Mv; 0 , l )

It may be seen that unless ! equals zero, this percentage decreases if /xy increases. Furthermore if In c0 < a 0+ i in other words if the minimum food basket is smaller than the geometric mean food basket in society, the poverty percentage also decreases if income inequality decreases. (b) Food-Ratio M ethod One might alternatively derive a poverty line from the Engel function by setting a maxim um value y0 for the ratio o f food expenditures to total income; if so m e o n e s food-in com e ratio c / y is higher than y0, this person is called poor; if c / y is lower than y0, this person is called non-poor. This method has been used by e.g. Love and Oja (1975) on Canadian data. If once again equation (1) describes the relationship between food and income in society, the poverty line corresponding to y0 is the solution o f (5) yielding / a 0 - l n y 0\ In y0 = ln = a 0+ ( a , 1) In y 8


= e x p n

In this poverty line definition the general principle holds with U ( y ) = exp {(l or|) In y - a ()} and 6 = l / y (). If y > does not change with xy or a 2, this < v method thus results in an absolute poverty line that will be denoted by z3. This assumption is not very realistic; in practice researchers or politicians will almost always relate the value o f y0 to the income distribution in society. This has led T ow nsend (1979) to the observation that an absolute poverty line is not only undesirable, but almost impossible. The corresponding poverty percentage under the assumption that y0 is exogenous will be f lnI3 n 3=
J -00

d F ( \ n y \ / j L a y ),

which in case o f lognormal incomes equals


n 3= r " d N (In

- N ( a


- (l ~ a 'K : 0 , 1Y

If y0 is constant and chosen to be larger than the geometric mean food-in com e ratio, both a reduction in incom e inequality and an increase in fiy will result in a lower poverty percentage.

(c) Fraction o f Median Income Approach Another well-known poverty line is defined as a certain percentage o f an index o f average income in society. (See e.g. O E C D (1976).) If instead o f a percentage o f average income, the poverty line is chosen as a fraction o f median incom e,2 this poverty line can be denoted by (8) y K = r0 exp {fiy ) (0 To< 1).

The welfare proxy used in the derivation o f this line is the ratio o f actual income to median incom e in society, yielding U ( y ) = v / e x p (/x ,) and 8 = r0. This poverty line, denoted by z4, is obviously relative. The corresponding poverty percentage is

IL =
or, in case o f lognormal incomes:

d F ( \ n y ; n-y, ary ),


n 4=

d N ( \ n y \ f x a v) = N -00 \ Cy T

lY )

Hence the poverty percentage depends on income inequality only; if income inequality decreases, the poverty percentages decreases as well. Econom ic growth does not, however, reduce poverty. (d) Percentile o f the Income Distribution Approach Finally, a poverty line may be defined as the borderline o f a certain percentile fr(, o f the income distribution: (10) I d F ( \ n y \ /ijno-y ) =

If for instance the incom e distribution is lognormal, the poverty line equals (11) v6 = exp(7V '( e 0) a y + n y ).

The general principle results in U { y ) = F (\n y , <7V and 8 = e0. In this defini ) tion the welfare proxy used is som eo n es relative position in the income distribu tion. This poverty line dep ends both on average in come and on income inequality; for values o f e() smaller than 0.5, the poverty line increases if income inequality decreases. D ue to the poverty line definition den oted by z5, the percentage o f poor is (12)


Hence neither econom ic growth nor changes in income inequality will reduce poverty.
2This choice is m ade in o rd e r to simplify the r e m ain d er o f the discussion; if average incom e is chosen, form ula (8) will be slightly m ore com plicated, involving crv as well. 144

3. T




L ine


An alternative poverty line definition has been introduced by Goedhart, Halberstadt, Kapteyn and Van Praag (1977). That poverty line has been called Leyden Poverty Line (LPL) after its place o f origination. This definition, elabor ated upon in e.g. Van Praag, Goedhart and Kapteyn (1980) and Van Praag, Hagenaars and Van Weeren (1982) is based on the relationship U ( y ) between welfare U and incom e y, as derived from a specific set o f attitude questions in a survey. In a similar way as before the poverty line is specified as the income level y St for which holds U ( y ii) = 8. The cardinal utility function chosen in that context is the Individual Welfare Function o f Income (W FI), introduced by Van Praag (1968, 1971) and elaborated upon in e.g. Kapteyn (1977), Van Praag (1981), Kapteyn and Wansbeek (1982), Van Praag and Spit (1982). In Van Praag (1968) a theoretical framework has been developed suggesting that the WFI can be approximately described by a lognormal distribution function (13) U ( y ) = A ( y ; n, a ) = N ( \ n y \ f x , cr),

where A ( ; fx, a ) and N ( ; /x, a ) are the lognormal and normal distribution function, respectively. For each individual the location and shape o f this function, determined by its parameters /x and cr2, is estimated from the following com posite survey question, called the Income Evaluation Question (IEQ): Please try to indicate what you consider to be an appropriate amount o f m oney for each o f the following cases? Under my (our) conditions I would call an after-tax income per w e e k / m o n t h / y e a r 1 of: about about about about about about .............. ....very bad .............. ....bad .............. ....insufficient .............. ....sufficient .............. ....good .............. ... very g o o d

'Please encircle the appropriate period On the assumption that people try to maximize the information given by their responses to the stimuli, offered by the verbal qualifications very b ad , bad, etc. up to very g o o d , the answers are set equal to the means o f equal quantiles o f the finite interval [0, 1]. This method is comparable to the procedure used by Jasso and Rossi (1977) in a similar context. The answers o f a hypothetical respondent /, denoted by y u, . . . to y 6l are depicted in Figure 1. The equal quantile assumption amounts to

u,(y,) = ' r

( / = l , . . . , 6).

U (y )

Figure 1. The Welfare F unction o f Income.

Making use o f the lognormal specification we find:

; _A

(14) yielding (15) where u, is defined as

A (^ if; / (Tt ) = ^ o

( / = ! , . . . , 6)

In y it = /a, + orMi

( / = 1 , . . . , 6)


JV(u,)= ^

i 5

( i = l ........6),

where N ( ) stands for N ( ; 0, 1). After adding an i.i.d.errorterm e to equation (15) fx, and <7, can be estimated for each individual t by ordinary least squares. Obviously the identification o f the theoretical concept with the response on the IEQ is a step which one must be willing to accept. The procedure described above is tantamount to the operationalization o f a theoretical notion by the definition o f a measurement procedure. At many places both the philosophical and the practical value o f this operationalization has been thoroughly studied (Van Praag (1971, Kapteyn (1977)). The equal quantile assumption has been tested empirically by Buyze (1982) with a positive result. Attempts to explain the variation in n and or, thus estimated, between individuals have been made, among others, by Van Praag (1971), Van Praag and Kapteyn (1973), Kapteyn (1977), Kapteyn, Wansbeek and Buyze (1980), Van Praag (1981) and Van Praag and Spit (1982). Their results culminate in the followin g hypothesis: The Welfare Function o f Income is approximately equal to a per ceived Income Distribution Function. A similar theory was described by Duesenberry as early as 1949 and more recently by Layard (1980) in his formulation of a Status Ranking Principle (see also Jasso (1980) for similar ideas). If the Income Distribution Function perceived by the

individual equalled the actual income distribution function, this statement would imply that an income level is evaluated by 0.6, if 60 per cent o f the total population earns less. The evaluation o f an income level would be solely determined by its relative ranking on the income scale. How ever, the perceived distribution o f incomes differs from the actual distribution, for the followin g reasons. First, the individuals perception o f the income distribution will depend on his own relative position in this distribution, which seems adequately described by his ow n income. Second, the income distribution o f his peers, say his social reference group, will influence his standards. Thirdly, s o m e o n e s perception o f his relative position in the income distribution may depend on his income and his social reference group in the past and in the future. In what follows we will restrict ourselves to the contemporaneous factors; for a first exploration o f the dynamic aspects o f this theory we refer to Van de Stadt, Kapteyn and Van der Geer (1985) and Van Praag and Van Weeren (1983). If only terms pertaining to the present are used, the parameter n, and a, o f the WFI are known to be fairly well explained by the equations (17) (18) /u ^ /S o + 0 , In y, + p 2m t o-, = p 3s,

where y, stands for the incom e o f individual t, m, stands for the log-income in / s reference group, and s, stands for the standard deviation o f log-incomes in t's reference group (see Kapteyn, Van Praag, Van Herwaarden (1978)). If we assume that everyone in society has the same reference group, i.e. all other individuals in society, we h a ve1 m, = p y and s, = cry yielding (cf. Van Praag, Goedhart, Kapteyn (1980)) (19) (20) fi, = /30 + /3| In y, + p 2fJiy a, = j33crr

For each individual t the incom e level y Sj corresponding to a welfare level 8 can now be derived by solving the equation (21) N ((In y 6 t - /a,)/cr( ; 0, 1) = 5

Yielding after substitution o f (17) and (18) (22) In y s% = /30 + j3, In y s.t + /32/av + f c a yu& l

where us is implicitly defined by N ( u s ) = 8. If y, > y Sti we have U ( y , ) > u { y s , t) = 8 and if y, < > we have U ( y , ) < U ( y s , ) = 8. Solution o f (22) with respect to y Sl , yields the corresponding definition o f a national poverty line z6: (23) In z6 = 1 - - (/30 +ffa/Ay-h&ovus).
U P i)

3If cross-sectional d ata o f o n e society are used, the param eter /32 can n o t be estim ated, but will be included in the intercept. If how ev er panel data are available, o r international d ata, as we will use in section 5, /32 may be identified.


This poverty line is partly relative; its elasticity with respect to median income exp (fxy ) dep ends on the values o f the coefficients (32 and (3t. The poverty line furthermore depends on o\, unless the welfare level 8 is chosen to be 0.5, for then would equal zero. The percentage o f poor in case o f a lognormal Income Distribution can be calculated as ru= d N (In y ; ay) = N + ( \ + i ~ \ ) v + }crvus (1 ~ \) ( T v


An increase in fxy will decrease the poverty percentage if (/3, + /S2) < 1. If /3, + /32 = 1 the poverty percentage dep ends merely on income inequality, represented by cry. If in addition to a possible finding that /3| + /32= 1 we should also have /3 = 0, >31 = 0 and / 3 , = 1, the poverty percentage would be constant and equal to the welfare level 8, chosen to be the poverty threshold. The conclusion o f this paper so far is that any poverty line definition suggested is a specific result o f a general principle and that the various specifications correspond to different ch oices o f the function U and the welfare level S. In the next section this will be formally summarized. In section 5 the equations underly ing the Leyden Poverty Line will be empirically estimated, yielding a poverty line corresponding to the opinion o f a representative sample.




om parison


iffe r e n t

efin itio n s

The poverty lines and poverty percentages corresponding to the various definitions are summarized in Table 1. For the calculation o f the poverty percentages it has been assumed that incomes are lognormally distributed. In order to distinguish

T ABLE 1 S ix P o v e r t y L i n e s Definition (1) Basic needs, Rowntree (2) Basic needs, O rsh ansk y (3) F ood ratio
an d

P erc



c o r d in g to

Va r i o u s D

efin itio n s o f

an d

Poverty Line In z i = In (c0 + oc0)

Poverty Percentage In ( + o c ) - u _ !c0---;0 . ij _ o f \

In Zi = l c n 0-0+ ( -,)/iy l

n 2= N ^ln c - a * '

0 ,

In z, =

p-ln y0 1 -<*,

( - d a y

(4) Percentage m edian incom e (5) Percentile (6) Leyden m eth o d

In z4 = In t0 +

n . = N ( ^ ;o , . ) n. - co
|[ ^ N (

In z5 = N '( e0)crv + /t,, ln z6 = o + l ^ y + i a vu& I-/3,

(I \)& y

l )


parameters that are chosen a priori and parameters that are estimated in a (budget) survey the latter are denoted with hats. The poverty lines vary from absolute (basic needs according to Rowntree and food ratio) to merely relative (percentage o f median income and percentile) methods. The other poverty line definitions (basic needs method according to Orshansky and Leyden method) may be anywhere on this scale, dep endin g on the values o f the estimated coefficients. If c?i = l, implying that the elasticity o f food with respect to income equals 1, the basic needs line according to Orshansky becom es absolute; if c i i = 0 , implying that expenditures on food are inelastic with respect to in come, the basic needs approach yields a com pletely relative poverty line. A nalogously the LPL becom es an absolute poverty line if /32 = 0, and a completely relative concept if & = 1 -/3 Actually, the LPL definition appears to be a generalization o f all other poverty line definitions: If 0 2 = 1 - 0 , and 0 o = O, we have a percentile definition with p > N ~ '( 8 )
N ( o) = 1 ~ P\

If >32 = 1 >3, and 8 = 0.5, we have a percentage-of-median-income-definition, with In r0 = 1-/3, If 0 2 = O and 5 = 0 . 5 , we have a food-ratio-definition where o ~ l n y 0_ 1-. 0O ^ 1-0, .

or a basic needs definition according to Rowntree, with In (c0+oco) = 0o


If yS2 < 1 /3| and 8 = 0.5, we have a basic-needs definition according to Orshansky, with

i In c() a 0~

/3u a n 1-0,

/I * \ ^2 ( 1 a , ) = 1-0

This is summarized in Table 1. The nature o f the poverty line, either relative or absolute, thus arises from the estimation o f the parameter vector 0 on survey data, instead o f being implicitly im posed by the researcher by the choice o f a certain poverty line. If the parameters are estimated on data from a representative sample o f the population, the resulting line may be seen as a reflection o f the prevailing views on poverty in society. The LPL definition, derived from survey data and thus based on the vox populi, appears therefore in a sense an improvement upon definitions that depend

merely on the researchers opinion on the nature o f poverty, like the percentile method or the fraction o f median-income method. The basic-needs and food-ratio definitions, although based on an a priori poverty concept, depend on survey data as well, albeit o f a different type, viz., budget surveys. As budget surveys require a considerable effort o f the respondents, the response is usually relatively low and ridden by selection bias; the resulting data set may very well be not representative o f the society as a whole. The LPL can be seen as an improvement over budget survey definitions as it is based on a few direct survey questions on the welfare-income relationship rather than the more circumstantial evidence provided by a less com prehensive welfare proxy, like expenditures on food. It may be assumed that the response under ceteris paribus conditions may be larger and more representative than that for a comprehensive budget survey. (See for a further comparison Van Praag, Spit and Van de Stadt (1981)). Up to now we have assumed that all incom e and welfare measurements are family-size independent, e.g. by using equivalent-adult scales. The derivation and use o f equivalence scales will introduce another difference between the various poverty line definitions. The poverty line definitions based on budget survey data may be easily adapted to yield a family size effect in the Engel Function (see e.g. Deaton and Muellbauer (1980) for a review o f the possibilities). Both the percentage o f m edian-in com e and the percentile method will need some exogenously determined equivalence scale, to be applied for the calculation o f the parameters o f the family-size corrected income distribution and for the differentiation o f the poverty line itself. However, most equivalence scales face a problem, mentioned among others by Abel-Smith (1982): Moreover, the chosen way o f life becom es very different for families with and without children. The process o f having children seems to change expenditure preferences. ( ......... ) What is really meant by equiv ale nce when ways o f life are so different? The LPL may be differentiated according to family size by the inclusion o f family size in equation (19). As this equation attempts to measure welfare directly, the effects o f family size on welfare may be found in a natural way, such as to incorporate changing preferences as well (Kapteyn and Van Praag (1976) and Van Praag, Spit and Van de Stadt (1982)). These welfare-neutral equivalence scales may also be derived for various other characteristics, like degree o f urbanization, occupation, etc. (see Van Praag, Hagenaars and Van Weeren (1982)).

5. A

m pirica l

Ill u str a tio n

Once data on the welfare parameters fi and a from a representative sample are available, equations (19) and (20) may be estimated after adding an i.i.d. disturbance term to each equation. The resulting parameters /30, (3U (i2, and will yield information on the extent to which poverty is perceived as a situation o f relative deprivation. As mentioned in section 3, time series, panel data or international data are needed for an estimation o f these equations.

In order to illustrate the method described, the parameters have been esti mated using cross-sectional data o f 8 European countries.4 The data were collected in October 1979 for a research program carried out at the Center for Research in Public Econom ic s in Leyden, with financial support o f the European Communities. For a description o f the data and the results by country we refer to Van Praag, Hagenaars and Van Weeren (1982). In this paper we restrict ourselves to the estimation o f the parameters P, using a transformed version o f equation (21), where (19) and (20) are substituted for ju and a,. For each respondnet we have ., an estimate o f the welfare parameter x, and a his actual incom e y the logmedian in com e o f the country he is living in, /x and the standard deviation o f log-incomes, ay These data enable us to calculate for each individual the value the income level that is evaluated by 8, and to use this as the dep endent variable in the equation (21') 1n y ^ t = p n+ p , In y, + p 2/xy + p'3(ry + p 4 In f s + e

where p'^ p yus and where In f s stands for log (family size), in order to allow for family-size differentiated poverty lines. Moreover, the data for different welfare levels have been pooled, assuming that the values o f the parameters do not change over welfare levels, and allowing for a direct effect o f <jy on In v: (21)" In y a = 1.410 + 0.574 In y + 0.224 + 0 . 7 12 a v + 0.807 cr.u, + 0.085 In f s (0.102) (0.003) ( 0 . 0 1 0 ) (0.026) (0.013) (0.003) R 2 = 0.626 N = 39,396.5

It is seen that p { is about 0.57 and p 2 equals 0.22, implying an elasticity o f the poverty line with respect to median incom e o f p 2/ { \ - P ,) = 0.51. This may be interpreted as the extent to which the poverty line is relative. It is furthermore seen that cry has a significant positive effect on the poverty line. The level o f inequality in a country has apparently per se an increasing influence on the poverty line o f a country, whatever the welfare level considered. This result may be explained theoretically by the fact that the preference drift P u the parameter that reflects the change in evaluation when family income changes, is probably related to the income inequality in a country. This can be seen from the estimates o f an alternative specification: (28) In y 8 = l.?63 + 0.536 In y + 0.072 cr In y + 0.226 xy + 0.807 (t v u s + 0.085 In f s (0.093) (0.003) (0.003) (0.010) (0 .0 1 3 ) (0.003) R 2 = 0.626 N = 39,396.

The coefficient o f In y equals 0.536 + 0.072 cr,, reflecting the fact that changes in income will give rise to larger changes in needs, if the income inequality in a population is larger. This relationship follows indeed from Van Praag and Spit
4These countries are Belgium, D enm ark, France, G erm an y , Ireland, Italy, The N etherlands and T h e U nited Kingdom . sThis n u m b e r is c o m p o sed o f 13.132 observations, for each o f which three welfare levels, < = 0.4 ? an d 0.6, are used.


(1982) and Van Praag (1981), where the variables explaining the Individual Welfare Function o f Income are derived theoretically according to social filter theory. In future research this relationship will be further explored. This implies that an increase in the income inequality increases the number o f poor for two reasons: firstly, because more p eople will have an income below the poverty line as defined before the change in income inequality, and secondly because the poverty line itself will shift upwards, causing an additional increase in the number o f poor. All coefficients are significantly different from zero, and the R ~'s are very high for cross-sectional individual data. However, these figures should be inter preted with som e caution: it is o f course possible that the perception o f poverty and welfare differs over countries for more reasons than the on es mentioned here, for instance, because o f differences in income history, climatological differ ences and cultural differences. The use o f this method might be better illustrated if panel data were available, in order that som e o f these factors may be accounted for. Nevertheless, the illustration given above sh ows that the method described in this paper, which is based on direct measurement o f welfare, may yield family size differentiated poverty lines which reflect the perception o f poverty in the society, and hence are related to median incom e and income inequality in society. Their specific form, however, is not a priori postulated, but estimated from survey data. For the calculation o f poverty lines proper more detailed data are required. For these figures we refer to Van Praag, Hagenaars and Van Weeren (1982).

6. C

o n c l u s io n

It has been shown that most well-known poverty line definitions are based on an assumption about the nature o f poverty, viz. whether poverty is absolute, relative or som ew here in between. The Leyden Poverty Line has the advantage that this choic e is not made by the researcher but as a result o f the perception o f poverty by a representative sample o f the population. All poverty line definitions considered may be seen as special cases o f the Leyden definition, dependin g on the value o f the estimated parameters, and hence on the perceived nature o f poverty. The Leyden Poverty Line may be differentiated for family size and other characteristics in a welfare-neutral way. An empirical illustration showed that the Leyden Poverty Line is dependent on both median income and income inequality in society: the elasticity o f the poverty line with respect to median income appeared to be about 0.51. The effect o f increasing income inequality is to shift the poverty line upwards.

R eferences
Abel-Smith, B. (1982), The Study an d Definition of Poverty: values an d social aims, p a p e r presented at the Venice C o n feren ce on Poverty, Ju n e 4-5, 1982, published in G. Sarpellon (1984). A tkinson, A. B. (1975), The Econom ics o f Inequality, C la re n d o n Press, Oxford. Booth, C. (1982), L ife a n d L abour o f the People in London, M acM illan, L ondon. C ram er, J. S. (1973), Empirical Econometrics, North H o llan d Publishing C o m p an y , Am sterdam .


D eato n , A. a n d M uellbauer, J. (1980), Economics an d C onsum er Behavior, C am b rid g e University Press, C am b rid g e. D uesenberry, J. S. (1949), Incom e, Saving and the Theory o f C onsum er Behavior, Princeton University Press, Princeton. G o e d h a r t, Th., H alberstadt, V., Kapteyn, A., an d Van Praag, B. M. S. (1977), The Poverty Line: C o n c e p t an d M easu rem en t, Journal o f H um an Resources, Vol. 12, pp. 503-520. Jasso, G. an d Rossi, P. H. (1977), Distributive Justice an d E arned Income, A m erican Sociological Review, vol. 42, pp. 639-651. Jasso, G. (1978), On the Justice o f Earnings: A New Specification o f the Justice E valuation F unction, The A m erican Journal o f Sociology, Vol. 83, pp. 1398-1419. Kakw ani, N. (1980), On a Class o f Poverty Measures, Econometrica, Vol. 48, No. 2, pp. 437-446. Kapteyn, A. (1977), A Theory o f Preference Formation, Ph.D. thesis, Leyden University. K apteyn, A., Van Praag, B. M. S. an d Van Herw aarden, F. G. (1978), Individual W elfare F unctions an d Social Reference Spaces, Economics Letters, Vol. I, pp. 173-178. Kapteyn, A. an d Van Praag, B. M. S. (1980), Family C o m p o sitio n an d Family Welfare, in: Simon, J. an d J. DaV anzo (eds), Research in Population Economics, II, JA J-Press, G reen w ich , pp. 77-97. K apteyn, A. a n d W ansbeek, T. J. ( 19 8 2 ) ,T he Individual Welfare F unction: M easurem ent, E x planation an d Policy Applications, Statistical Studies, Vol. 32, Staatsuitgeverij, The Hague. Kilpatrick, R. W. (1973), The Incom e Elasticity of the Poverty Line, The Review o f Economics and Statistics, Vol. 55, pp. 327-332. L ayard, R. (1980), H u m an Satisfactions an d Public Policy, The Economic Journal, Vol. 90, pp. 737-750. Love, R. a n d Oja, G. (1975), Low Incom e in C a n ad a, R eview o f Incom e and Wealth, Series 25, No. 1. O E C D (1976), Public E x p en d itu res on Incom e M aintenance Programs, Studies in Resource Allocation, Vol. 3, Paris. O Higgins, M. (1980), Poverty in Europe The Subjective A ssessm en t o f Poverty Lines: A n Evaluation, University o f Bath. O rshansky, M. (1965), C o u n tin g the Poor: A nother Look at the Poverty Profile, Social Security Bulletin, Vol. 28. O rshansky, M. (1968), The S hap e o f Poverty in 1966, Social Security Bulletin, Vol. 31, pp. 3-31. Prais, S. J. a n d H o u th ak k er H. S. (1955), The A nalysis o f Fam ily Budgets, C a m b rid g e University Press, C am bridge. Rainw ater, L. (1974), W hat M o n ey Buys, Inequality a n d the Social M eanings o f Income, Basic Books Inc., N ew York. Rowntree, S. (1901), P overty: A S tu d y o f Town Life, M acm illan , London. Sarpellon, G ., ed. (1984), Understanding Poverty, Angeli, Milano. Sen, A. (1976), Poverty: An O rd in al A p pro ach to M easurem ent, Econometrica, Vol. 44, pp. 219-231. T ak ayam a, N. (1979), Poverty, Incom e Inequality an d Their Measures: Professor S en s axiom atic a p p r o a c h reconsidered, Econometrica, Vol. 47, No. 3. T h o n , D. (1979), O n M easu rin g Poverty, Review o f Incom e a n d Wealth, Vol. 25, pp. 429-439. T h o n , D. (1981), In co m e Inequality an d Poverty: some problem s, R eview o f Incom e a n d Wealth, Vol. 27. pp. 207-210. T o w n sen d , P. (1979), Poverty in the U nited Kingdom, A Survey o f Household Resources a n d Standards o f Living, Penguin Books. Van de Stadt, H., K ap tey n , A., an d Van de G eer, S. A. (1985), The Relativity o f Utility: Evidence from Panel D ata, Fo rth co m in g in The R eview o f Economics and Statistics. Van Praag, B. M. S. (1968), Individual Welfare Functions a n d C onsum er Behavior, N o rth -H o llan d P ublishing C o m p a n y , Am sterdam . Van Praag, B. M. S. (1 9 7 1), T h e Welfare F unction of In co m e in Belgium: An E mpirical Investigation, European Econom ic Review, Vol. 2, pp. 337-369. Van Praag, B. M. S. an d K aptey n , A. (1973), Further Evidence on the Individual W elfare F unction o f Incom e: An Em pirical Investigation in The N eth erlan d s, European Econom ic Review, Vol. 4, pp. 33-62. Van Praag, B. M. S. (1980), A Social Filter A p p ro ach to the Individual W elfare Fu n ctio n , Report 80.12, Leyden University, C e n te r for Research in Public Econom ics. Van Praag, B. M. S., G o e d h a r t, Th. and Kapteyn, A. (1980), T he Poverty Line. A Pilot Survey in E urop e, The Review o f Econom ics a nd Statistics, Vol. 62, pp. 461-465. Van Praag, B. M. S. (1981), Reflections on the Theory o f Individual Welfare Functions, published in the Proceedings o f the Am erican Statistical A ssociation, Detroit, 1981. Van Praag, B. M. S. (1982), A C o m p a riso n o f Objective a n d Subjective M easures o f Poverty, published in G. S arpellon (1984).


Van Praag, B. M. S., H ag enaars, A. J. M. an d Van W eeren, H. (1982), Poverty in E u ro p e, The Review o f Incom e a n d Wealth, Vol. 28, pp. 345-359. Van Praag, B. M. S., Spit, J. S. an d Van de Stadt, H. (1982), A co m p ariso n between the Food Ratio Poverty Line a n d the Levden Poverty Line, The Review o f Economics an d Statistics, Vol. 64, pp. 691-694. Van Praag, B. M. S. an d Spit, J. S. (1982), The Social Filter Process and Incom e E valuation An Empirical Study in the Social Reference M echanism . Report 82.08, Leyden University. Center for R esearch in Public Economics. Watts, H. (1968), An Econom ic Definition o f Poverty, in: D. P. M o ynihan, On Understanding Poverty, Basic Books, Inc. New York.