Economics Letters 236 (2024) 111571

Contents lists available at ScienceDirect

Economics Letters
journal homepage: www.elsevier.com/locate/ecolet

A note on the measurement of poverty persistence

Antonio Villar
Universidad Pablo de Olavide & ISEAK, Spain

A R T I C L E I N F O A B S T R A C T

JEL classification: This paper introduces a poverty index that incorporates poverty persistence as an integral part of poverty
D31 measurement within a multiperiod framework. Using familiar tools (logarithmic utilities and a utilitarian social
I32 welfare function) we obtain a mathematically straightforward poverty index, which can be interpreted as an
Keywords: estimate of the social cost of poverty. This index can be neatly decomposed into incidence, intensity, and
Poverty persistence inequality, and is additively decomposable by population subgroups. It consists of the log of the geometric mean
Welfare loss
of individual intertemporal utility losses.
Logarithmic utility
Utilitarian welfare function
Decomposability

1. Introduction
The cumulative impact of poverty spells on households is widely
recognized, as it negatively affects some basic aspects of personal and
social life, including health, employment, education, household pros­
pects, and social integration (e.g., Bane and Ellwood, 1986). Yet,
intertemporal poverty has received comparatively less attention, mostly
due to data restrictions (Fusco and van Kerm, 2023). The situation is
changing, as panel data on income distribution are now available in
several countries (e.g., the EU-SILC contains a four-year panel), and
administrative information becomes increasingly accessible and
tractable.
Considering poverty duration is crucial for assessing the welfare cost
of poverty. Indeed, reductions in conventional poverty indices are
compatible with the rising average duration of poverty. This can lead to
overlooking the consolidation of poverty among a part of the popula­
tion. In short, Sen’s (1976) three I’s of poverty (incidence, intensity, and
inequality) need be supplemented by a measure of persistence.
The literature has already dealt with computing past poverty spells
from different perspectives. One perspective refers to the dynamic
analysis of transitions between poverty and non-poverty, focusing on
factors influencing the probability of those transitions (e.g., Huff Ste­
vens, 1998). A second line of research deals with chronic poverty,
defining an additional poverty threshold to identify the “chronically
poor” as those who have been poor during a given number of periods (e.
g., Foster 2009; Calvo and Dercon 2009; Hoy and Zheng 2011; Foster
and Santos 2012; Alkire et al., 2017). The third perspective, focusing on
poverty persistence, aims to incorporate duration into the standard
poverty measurement (e.g., Bossert et al., 2008; Gradín et al., 2012;
Dutta et al., 2013). See Nicholas and Ray (2023) and Fusco and van
Kerm (2023) for a recent review of that literature.1
Our contribution aligns with the third perspective. We present a
straightforward approach to poverty measurement that considers
duration as one of its arguments. Adopting a social welfare approach, we
obtain the welfare cost of poverty by aggregating the disutility of poor
individuals, in line with the proposal in Villar (2023). People are
considered poor when their income falls short of a given money
threshold, the poverty line. The (socially inadmissible) utility loss of a
poor individual, over a given time span, is calculated as the discounted
cumulative effect of the differences between the utility of achieving the
poverty line and the utility of the actual income within each period.
Using a utilitarian social welfare function, we obtain a poverty index
that exhibits an elementary mathematical structure, and it is easy to
interpret. The index can be decomposed into the three I’s of poverty,
incorporates poverty persistence, and is additively decomposable by
populations subgroups.
The social welfare approach to economic problems involves some
form of interpersonal comparability of agents’ utilities. We do so here
assuming that all agents share the same utility function, which is
increasing and concave, and only depends on their individual incomes.

E-mail address: avillar@upo.es.

1
As discussed in Gradín, del Río and Cantó (2012) and commented in Fusco and Kerm (2023), one can distinguish another line of research that considers different
income components, permanent and transitory, and provides measures of chronic versus transitory poverty (e.g., Jalan and Ravallion, 1998).

https://doi.org/10.1016/j.econlet.2024.111571
Received 12 December 2023; Received in revised form 26 January 2024; Accepted 26 January 2024
Available online 28 January 2024
0165-1765/© 2024 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-
nc/4.0/).
A. Villar Economics Letters 236 (2024) 111571

In our interpretation, the utility loss of a poor agent is a normative the time span.
assessment of the contribution to the welfare cost of poverty, rather than We define the welfare cost of poverty as the average value of the
a subjective perception. aggregate utility loss due to poverty. That is, the welfare cost of poverty
relative to a problem P = (Y, z, δ), denoted by C(P), is given by:
2. The model
1 ∑ q ∏ 1
C(P) = μ(yh ) = × ln [̃
Tlñ μ(yh )]q .
nT h∈Q n
Consider a society with n members, N = {1, 2, …, n} at a given point h∈Q

in time, that we call today. Those members, typically households or in­ ∏ 1

Making μ(P), the geometric mean of the geometric
μ(yh )]q = ̃
[̃
dividuals, are referred to as agents. Today’s income distribution is h∈Q
described by a vector x ∈ Rn++ , whereas a scalar z > 0 denotes today’s means of the agent’s utility losses, we have:
poverty threshold. The set of the poor is Q = {h ∈ N /xh < z}, with q
cardinal q. We assume, without loss of generality, that the income of the μ(P)
C(P) = × lñ (2)
n
poor is described by a sub-vector of x denoted by y ∈ Rq++ , i.e., x = (y,
x− y ). The welfare cost of poverty is thus the product of the incidence of
Let T + 1 be the length of the time span considered in our analysis. poverty and the log of the geometric mean of the agents’ utility losses.
We denote by t = 0, 1, 2, …, T the different periods, where t indicates The term ̃ μ(P) combines the intensity and the inequality of poverty,
the distance time from today (that is, t = 1 stands for “yesterday”, t = 2 structurally incorporating the effect of the past poverty periods.
for the “previous day”, and so forth up to t = T). We can think of a time We can easily decompose ̃ μ(P) to obtain separate measures of the
span of four or five years to fix ideas. To save notation, we omit sub­ intensity and the inequality of poverty, by using Atkinson (1970)
indices in today’s values, rather than labelling them with zeroes. For inequality index for ε = 1, which is given by: A (d) = 1 − ̃μ(P) (i.e., one
1 μ(P)
each agent h ∈ Q there is a vector yh = (yh , yh (1), …, yh (T)) describing minus the ratio between the geometric mean and the arithmetic mean of
the income obtained in each of the considered periods. That information the poor’ deviations). Hence,
q×(T+1)
can be summarized by a matrix Y = (y1 , …, yq ) ∈ R++ . As poverty ⎛ ⎞
lines may change over time, we let z = (z, z1 , …, zT ) denote the corre­ q ⎜ ⎟
C(P) = × ⎝1nμ(P) + ln[1 − A1 (P)]⎠ (3)
sponding vector of poverty thresholds. n
⏟⏞⏞⏟ ⏟̅̅̅⏞⏞̅̅̅⏟ ⏟̅̅̅̅̅̅̅̅̅⏞⏞̅̅̅̅̅̅̅̅̅⏟
We measure poverty as a social utility loss, by aggregating individual Incidence
Intensity Inequality

utility losses of those who are currently poor.2 To do so we adopt three

This expression makes it clear that the welfare cost of poverty is an
conventional assumptions:
increasing function of each of its components, as ln[1 − A1 (P)] < 0. It
also shows that poverty reduction policies will be more effective when
(i) All agents share the same utility function, which only depends on
focussed on those agents with lower incomes and larger poverty spells.
their own incomes within each period. That is, uth (y(t)) = ut (yh (t))
Equation [2] is also additively decomposable by population sub­
, t = 0, 1, …, T. The index t in the utility function indicates that groups. To see that, suppose that the population can be divided into G
utility may be time dependent. different population subgroups, g = 1, 2, …, G, according to socio-
(ii) ut (yh (t)) = δt u(yh (t)), where δ ∈ [0, 1] is a discount factor that demographic characteristics. The set of the poor can be described as Q =
gives us today’s value of the utility of yh (t) at time t (we can think G
Ug=1 Qg , where Qg is the set of the poor in population subgroup g, with
1
of this parameter as a function of the “interest rate”, e.g., δ = 1+r ).
cardinal q(g). We assume that the parameter δ is common for all popu­
(iii) The utility function is increasing and concave, and specifically, its lation subgroups, for the sake of simplicity (introducing different values
changes are inversely proportional to its levels. That is: u(yh (t)) = is trivial).
lnyh (t). According to equation [2] the social cost of poverty for population
subgroup g is given by:
Let P = (Y, z, δ) denote a poverty evaluation problem, or simply a
problem. Note that the length of the time span and the number of the q(g)
Cg (P) = μ(P(g)).
× lñ
poor are given implicitly by the dimension of matrix Y. The utility loss of n
individual h along the (T +1) periods, with a discount factor δ, is given ∑
Since C(P) = Gg=1 Cg (P), we can write:
by:
q ∑G
q(g)
∑ (4)
T
t C(P) = × μ(P(g))
× lñ
dh (P) = max{0, δ (lnzt − lnyh (t)) }. n g=1
q
t=0

T{ ( )1δt } Letting the geometric mean of population subgroup g be given by:

∏ zt T
Let ̃
μ(yh ) = max 1, stand for the geometric mean of
t=0
yh (t) ∏
G p(q)

( )δt μ(g) =
̃ μ(P(g))] Gp
[̃
the agent’s time-adjusted deviations, yhz(t)
t
, during the entire time g=1

span. Then, we can rewrite the expression above as: we can alternatively write:
μ(yh )
dh (P) = Tlñ (1) p
C(P) = μ(g)
× G × lñ (5)
n
That is, the utility loss of agent h corresponds to the log of the geo­
metric mean of the time-adjusted deviations, multiplied by the length of
3. Final remarks

We have presented a poverty index that incorporates persistence as

2
Our aim is measuring the utility loss of those who are poor today, consid­
ering past periods of poverty, rather than measuring the overall poverty in the
period. Hence, a poor agent at t > 0 that is not poor today does not enter into
the computation. And, symmetrically, a new poor agent enters with all his/her
past poverty history up to T.
an integral part of its definition. The formula is obtained in two steps. In
the first step, we evaluate each agent’s welfare loss along the time span
considered, as the aggregate intertemporal utility loss. Assuming loga­
rithmic utilities, the resulting formula is a variant of Watts (1968)
poverty index (an index already characterized in Zheng 1993). We

2
A. Villar
introduce a discount factor to make utilities homogeneous so that they
can be added. The individual poverty indicator is a multidimensional
poverty index, with time periods as dimensions. In the second step, we
obtain the social welfare loss due to poverty by recurring to the familiar
utilitarian social welfare function (characterized in D’Aspremont and
Gevers, 1977).
Our index satisfies all the usual properties, including normalization
(the index is zero when there are no poor people), continuity (the index
is a continuous function on its domain), focus (changes in the income of
the non-poor do affect the index), symmetry (permuting individual in­
come profiles does not alter the index), monotonicity (increasing the
income of the poor reduces the index), scale invariance (multiplying
incomes and thresholds by a positive constant does not alter the mea­
surement), replication invariance (combining two identical populations
does not alter de index), principle of transfers (a transfer from a richer to
a poorer person does not increase the index), decomposability by com­
ponents, and additive decomposability by population subgroups (prop­
erties already discussed in Section 2).3
What we call the welfare cost of poverty refers to the intertemporal
cost of those who are currently poor, expressed in per capita, per period
terms. Trivially, T × C(P) is a measure of the per capita utility loss for
the entire time span, and n × T × C(P) gives us the aggregate utility loss.
We might have chosen a different modelling option, focusing on the
utility losses of all those who have been poor for some period during the
time span considered, even if they are not poor today. The formula ap­
plies to this case, mutatis mutandis, even though the interpretation of
the measure would be slightly different.
Dutta et al. (2013) discuss how to deal with the impact of
non-poverty spells, when measuring poverty persistence. They propose
different alternatives that consider the time structure and the intensity
of those spells. We do not address this question in our model. Yet,
introducing time discounting implies that the older the period of
poverty, the smaller the impact on the poverty index. By the same token,
older non-poverty spells reduce the poverty index less than more recent
ones. That is, even though all non-poverty spells add nothing to the
index (an application of the focus principle), their time structure affect
poverty measurement via the discount coefficients of the complemen­
tary poverty periods.
Let us conclude by emphasizing that computing duration in poverty
measurement enhances the relevance of those policies aimed at reducing
poverty. From this perspective, minimal income schemes may be rein­
forced, as those agents who leave the set of the poor carry with them the
history of past poverty periods.
3
Zheng (1993) shows that the key properties to characterize the Watts
poverty index are scale invariance, continuity, monotonicity, and additive
decomposability by population subgroups.
Economics Letters 236 (2024) 111571
Data availability
No data was used for the research described in the article.
Acknowledgment
Thanks are due to an anonymous referee for helpful comments.
Financial support from the Spanish Ministerio de Ciencia e Innovación,
under project PID2019-104452RB-I00 is gratefully acknowledged.
Funding for open access publishing: Universidad Pablo de Olavide.
References
Alkire, S., Apablaza, M., Chakravarty, S., Yalonetzky, G., 2017. Measuring chronic
multidimensional poverty. J. Policy. Model. 39, 983–1006.
Atkinson, A.B., 1970. On the measurement of inequality. J. Econ. Theory 2, 244–263.
Bane, M.J., Ellwood, D.T., 1986. Slipping into and out of poverty: the dynamics of spells.
J. Hum. Resour. 21, 1–23.
Bossert, W., Chakravarty, S., D’Ambrosio, C., 2012. Poverty and time. J. Econ. Inequal.
10, 145–162.
Calvo, C., Dercon, S., 2009. Chronic poverty and all that: the measurement of poverty
over time. T. Addison, D. Hulme, and R. Kanbur (eds), Poverty Dynamics:
Interdisciplinary Perspectives. Oxford University Press, Oxford, p. 2009.
D’Aspremont, J.C., Gevers, L., 1977. Equity and the informational basis of collective
choice. Rev. Econ. Stud. 44, 199–209.
Dutta, I., Roope, L., Zank, H., 2013. On intertemporal poverty measures: the role of
affluence and want. Soc. Choice Welfare 41 (4), 741–762.
Foster, J.E., 2009. A class of chronic poverty measures. Addison T., Hulme D., Kanbur R.
(Eds.). Poverty Dynamics: Interdisciplinary Perspectives. Oxford University Press,
Oxford.
Foster, J.E., Santos, M.E., 2012. Measuring Chronic Poverty. OPHI working paper, p. 52
n◦ .
Fusco, A., van Kerm, P., Silber, J., 2023. Measuring poverty persistence, Ch. 18. Research
Handbook on Measuring Poverty and Deprivation. Edward Elgar Publishing,
p. 2023.
Gradín, C., Del Río, C., Cantó, O., 2012. Measuring poverty accounting for time. Review
of Income and Wealth 58, 330–354.
Hoy & Deng Hoy, M., Zheng, B., 2006. Measuring lifetime poverty. J. Econ. Theory 146,
2544–2562.
Huff Stevens, A., 1998. Climbing out of poverty, falling back in measuring the persistence
of poverty over multiple spells. J. Hum. Resour. 34, 557–588.
Jalan, J., Ravallion, M., 1998. Transient poverty in postreform rural China. J. Comp.
Econ. 26, 338–357.
17 in Nicholas, A., Ray, R., Silber, J., 2023. Ch. 17 Chronic poverty measures, Ch.
Research Handbook on Measuring Poverty and Deprivation. Edward Elgar
Publishing, p. 2023.
Sen, A., 1976. Poverty: an ordinal approach to measurement. Econometrica 219–231.
Villar, A., 2023. Welfare poverty measurement. J. Human. Dev. Capabil. 24, 1–16.
Watts, H., Moynihan, D.P., 1968. An economic definition of poverty. On Understanding
Poverty. Basic Books, New York.
Zheng, B., 1993. An axiomatic characterization of the Watts poverty index. Econ. Lett.
42, 81–86.