Chronic and Transitory Poverty: Evidence from Egypt, 1997–99

Chronic and Transitory Poverty:

Evidence from Egypt, 1997–99

LAWRENCE HADDAD and AKHTER AHMED *International Food Policy Research Institute, Washington, DC, USA

Summary. — This paper uses a panel data of 347 households in Egypt to measure changes inhousehold consumption during 1997–99 and to identify causes behind the changes. Per capitaconsumption has decreased for the households during this time. The decrease has not beendramatic, but it has occurred at all points along the distribution. Over the two-year period, thenumber of households who have fallen into poverty is over twice as large as the number ofhouseholds who have climbed out of poverty. About two-thirds of overall poverty is chronic(average consumption over time is below the poverty line), and almost half of all poor are alwayspoor. We use quantile regression methods to identify the factors that explain total, chronic, andtransitory poverty. While our analysis is able to document the extent of transitory poverty, itperforms relatively poorly in explaining the determinants of this type of poverty. Thepredominantly chronic nature of poverty in our sample, and our ability to identify associatedcharacteristics, strengthens the case for targeting antipoverty interventions, such as food subsidies.� 2002 Elsevier Science Ltd. All rights reserved.

Key words — panel survey, consumption, chronic and transitory poverty, determinants of poverty

components

1. INTRODUCTION

The injection of the time dimension into thestudy of poverty affords many analytical pos-sibilities. Aspects of poverty that were hithertoobscured come into view. For example, we canidentify the households that are more likely tobe in poverty no matter when or how often wesurvey them. The welfare paths along whichhouseholds move and why they do so also be-come clearer. Why, for example, do somehouseholds move out of poverty over time andwhy do some fall into it? With this dynamicperspective, the study of poverty moves fromthe two-dimensional to the three-dimensional.In the developing world, and in Africa in par-ticular, the study of income mobility and pov-erty dynamics has been stymied by a lack ofdata sets for which there are repeated obser-vations on a given set of households. This hasbegun to change, albeit slowly. In a recentoverview paper, Baulch and Hoddinott (2000)bring together recent studies on poverty dy-namics in the developing world.This paper adds another panel study to the

small set of studies from Africa. It is the firstpanel survey study of poverty dynamics inEgypt and of any country in North Africa. The

analysis rests on a panel data set from eightgovernorates. Specifically the analysis uses datafrom 347 households that were first surveyed inearly 1997 as a part of the national sample of2,450 households (the Egypt IntegratedHousehold Survey or EIHS), 1 and were sur-veyed again in early 1999. The paper decom-poses poverty in Egypt over 1997–99 intochronic and transitory poverty. The paper thenuses regression methods to identify the deter-minants of chronic and transitory poverty.The study of the dynamics of poverty in an

Egyptian context is important for a number ofreasons. First, good macroeconomic perfor-mance over the past five years does not seem tohave made a dent in Egypt�s poverty rate. Percapita GDP at purchasing power parity hasgrown from $2,590 in 1993 to $3,050 in 1997(World Bank, 1999). Yet, a recent review ofpoverty rates in Egypt by El-Laithy and Muk-herjee (1999) highlights the uncertainty as towhether poverty rates have fallen at all. 2 Isthe poverty rate unmoved because a large

World Development Vol. 31, No. 1, pp. 71–85, 2003� 2002 Elsevier Science Ltd. All rights reserved.

Printed in Great Britain0305-750X/02/$ - see front matter

PII: S0305-750X(02)00180-8www.elsevier.com/locate/worlddev

*The authors thank Sanjukta Mukherjee and threeanonymous referees for their useful comments. All er-

rors are ours. Final revision accepted: 22 August 2002.

71

percentage of households below the povertyline remain stuck there or does the rate maskroughly equal numbers of households movinginto and out of poverty? If the answer issomewhere toward the former end of thespectrum, then a set of policies that facilitatesasset accumulation would seem a high priority.If the answer lies toward the latter end of thespectrum, then a set of policies to stimulateinsurance policies and safety nets might be ahigher priority.Second, for those who have escaped poverty,

the analysis should shed some light as to themain policy factors that are responsible for theexit. If education is key, then reform in the ed-ucation sector should focus more on access. Ifeducation seems to play no role, is this a signalthat education quality needs to be dramaticallyimproved? If access to land is important, thenwhat can be done to increase access? If access toland is unimportant, does this signal that a moreconcerted effort should be placed on increasingnonfarm income sources?Third, the results will shed light on a debate

that is taking place throughout the North Africaregion on how to target food price subsidies andother poverty interventions (Tuck & Lindert,1996). Food price subsidies for cereals are cur-rently universally available in many countries ofthe region. Several countries have made explicitattempts to reduce the attractiveness of univer-sally subsidized cereals and make them moreself-targeted. There is less scope for expandingthe self-targeting nature of food subsidies inEgypt, given the relatively low income inequal-ity (World Bank, 1999) and that the incomeelasticities of subsidized foods are close to zero(Alderman & Lindert, 1998). Future efforts atreducing leakage to the nonpoor will need toinvolve administrative targeting via some proxymeans test. If most of the poverty in Egypt istransitory, it will be more difficult to undertakesuch targeting, because it is more difficult topredict income using proxies. If most of thepoverty is chronic, the prediction of incomeusing proxies will be easier and more durable.Finally, the study will illustrate the potential

and limitations of panel studies. 3 The gov-ernments in the region are much less prone toundertake panel data collection than theircounterparts in Asia. The benefits of the recentSouth African panel data set from KwaZulu-Natal (May, Carter, Haddad, & Maluccio,2000) have, for example, persuaded the na-tional statistical authorities to begin the con-struction of a national panel (Stats SA, 2000).

Section 2 of the paper describes the data setcollected in 1997 and 1999 and uses the 1997EIHS data to analyze the ways in which these347 households are different (if at all) from therepresentative sample of households drawnfrom 20 governorates in 1997. In Section 3 wedescribe the changes in welfare experienced bythe 347 households in the past two years and weclassify them according to the permanence ofthe poverty they experience. In Section 4 we usequantile regression methods to identify thefactors that explain total, chronic, and transi-tory poverty. Section 5 summarizes the resultsand discusses their implications.

2. THE 1997 DATA AND THE 1999SUBSAMPLE

In this section we describe the EIHS datacollected in 1997 and the subsample collected in1999. We investigate the representativeness ofthe follow-up subsample by comparing it withthe main sample using 1997 data. We also as-sess the seriousness of any shortfall or attritionin the planned number of subsample interviews.Many of the results from these analyses arepresented in appendix tables.

(a) EIHS 1997

The 1997 EIHS sampled 2,450 householdsselected from 20 governorates (out of a total of26 Egyptian governorates), using a two-stage,stratified selection process. The first stage of theselection process randomly selected 125 pri-mary sampling units (PSUs) with probabilityproportional to estimated size. In the secondstage, 20 households were randomly selectedfrom each PSU. The design also stratified se-lection on the following five regions of Egypt:Metropolitan, Lower Urban, Lower Rural,Upper Urban, and Upper Rural. The totalpopulation of these 20 governorates constitutesabout 98% of the total population in Egypt.The sample frame used for the selection processwas supplied by the Central Agency for PublicMobilization and Statistics (CAPMAS). Thesample is representative of the 20 governorates.

(b) The 1999 EIHS subsample

For the 1999 EIHS subsample, budget con-straints restricted data collection to 19 PSUs.We selected PSUs from all five regions in pro-portion to their frequency per region in the

WORLD DEVELOPMENT72

1997 EIHS. Roughly one in six PSUs weretargeted per region. The PSUs were not selectedrandomly, however. Given the desire to selectone in six PSUs from every region of thecountry, we chose PSUs such that we sampledas many governorates as possible (eight), givenbudget limitations. The selection of govern-orates at random is inappropriate for such asmall number selection (eight from 20) and soselection was purposive to get representation interms of regions and poverty levels in 1997. Theselected PSUs and their location are given inAppendix Table 10. Appendix Table 11 com-pares the EIHS 1997 and EIHS 1999 subsampledistribution of PSUs by region and by gov-ernorate. We explore the consequences of ourchoices in terms of the representativeness of the1999 sample in the next subsection.As in 1997, the goal in 1999 was to resurvey

20 households per PSU to give a total subsam-ple of 380 households. Thirty-two households(8.4% of the target households) could not beresurveyed, however, because members of somehouseholds were absent when the enumeratorsvisited those households, and some householdshad moved away from the original PSU. Onlytwo households refused to participate in theresurvey. We explore the consequences of thisattrition in the following section.

(c) The representativeness of the 1999 subsample

To make the subsequent results more gener-alizable, we would like the subsample of 380households to be representative of the largersample of 2,450 households. The previous sec-tion of the paper indicates that this is notguaranteed, given the purposive samplingstrategy. But how serious is the nonrepresen-tativeness? This section seeks to address that

question by examining the 1997 values of realper capita consumption and other variables forthe subsample versus the entire sample, usingsimple two-way tables and regression analysis.Table 1 presents the differences in mean real

per capita consumption in 1997 between thesubsample (n ¼ 374 when 1997 missing valuesare dropped) in column 1 and the remaininghouseholds (n ¼ 2,075) for rural and urbanhouseholds in column 2. The mean levels areapproximately 10% lower for the sample of 374households, although the differences are notsignificantly different at the 5% level. But, only347 of the 374 households could be resurveyed,due to attrition losses. How different are the347 from the 27 households that could not beresurveyed? A comparison of columns 3 and 4in Table 1 shows that the rural households thatcould not be resurveyed in 1999 were slightlypoorer, on average, but that the urban group isless poor, on average. Again none of the dif-ferences in real per capita consumption aresignificant at the 5% level.Table 1 focuses only on differences in real per

capita expenditure and does not control for anumber of potentially confounding factors.Therefore we further explore representativenessand attrition in the context of multiple regres-sion. Column 1 of Appendix Table 12 presentslogit regression results that seek to distinguishbetween the subsample (n ¼ 374) and remain-ing households (n ¼ 2,075), using the 1997data. The results show that the two groups arenot different in terms of real per capita con-sumption in 1997, thus confirming the result inTable 1. Only two of the variables are signifi-cantly different between the two groups at the5% level. Controlling for the other observablefactors, the subsample households are lesslikely to be employed in the construction or

Table 1. Real per capita consumption in 1997 for subsample households, broken down by whether we were able toresurvey them in 1999, compared to the full sample

Real per capita consumption in 1997

(1) EIHS 1997householdsexcluded from1999 subsample

(2) EIHS 1997householdsin 1999subsample

(3) EIHS 1997households

in 1999 subsample:surveyed

(4) EIHS 1997households in 1999subsample: couldnot be surveyed

(5) Totalhouseholds

(from 1997 EIHS)

Rural 230.0 206.6 207.0 198.3 226.6N 1,130 196 186 10 1,326

Urban 307.6 286.9 278.6 365.7 304.3N 945 178 161 17 1,123

Total 265.3 244.8 240.2 303.7 262.2N 2,075 374 347 27 2,449

CHRONIC AND TRANSITORY POVERTY 73

manufacturing sectors (relative to the omittedcategory of agriculture). In terms of attrition(the households that were selected for sub-sampling, but we were unable to locate or thatrefused to be interviewed), column 2 of Ap-pendix Table 12 uses logit regression to lookfor variables that might be significantly differ-ent between the two groups (n ¼ 347 andn ¼ 27) but finds none at the 5% level. Thus weconclude that the subsample is not too differentfrom the larger sample. We also conclude thatattrition bias cannot be found in terms ofmeasurable characteristics.

3. CHANGES IN WELFARE BETWEEN1997 AND 1999

This section now focuses exclusively on thesubsample of 347 households and compares thewelfare of these households in 1997 and 1999.Table 2 indicates that during 1997–99, averageper capita consumption declined for the totalsample (240.2–212.7), and that the decline isobserved for rural and urban households, butmore strongly for urban households. Never-theless, the declines are not strong enough to besignificantly different from zero at the 5% level.The mean is just one aspect of a distribution.

How different are the distributions of per capitaconsumption in 1997 and 1999 for these 347households? Table 3 describes the various per-

centile points in the distribution for the twovariables. The table indicates that there hasbeen a downward shift in per capita expendi-tures for the sample at regular points through-out the distribution, not just in terms of mean(or median). Nevertheless, Table 3 tells us noth-ing about movement within the distribution. Tobegin describing the upward or downwardmobility of households based on their consump-tion levels in 1997 and 1999, a transition matrixis presented in Table 4.The transition matrix maps changes in

household fortunes for 1997–99 in relation tothe poverty line. For example, we see that ofthe 88 households that were below the povertyline in 1997, two have consumption more thantwice the poverty line in 1999 and 20 havemoved up to consumption levels above thepoverty line but below two times the povertyline. On the other hand, of the 115 householdsbelow the poverty line in 1999, 49 of them wereabove the poverty line in 1997. The 207households on the diagonal have not movedbetween categories. The 51 households abovethe diagonal have seen an improvement inconsumption, while the 89 households belowthe diagonal have seen a worsening.Table 5 further categorizes the households

into three distinct groups: (a) households that

Table 2. Real per capita consumption in 1997 and 1999for the subset of householdsa

Sample Real per capita consumption

1997 1999

Rural 207.0 193.5N 186 186

Urban 278.6 234.8N 161 161

Total 240.2 212.7N 347 347

aConsumption on 1999 was deflated over time using afood price Laspayres index constructed from clusteraverage food prices reported by the 347 households andweighted by their consumption shares. No nonfoodprices were available, and no community-level price datawere available. The use of self-reported prices is unsat-isfactory due to the quality and bulk purchase effectsembodied in them. We were however unsatisfied withusing government reported deflators, because they couldnot be disaggregated at all. Details of the price indexused are provided in Appendix Table 13.

Table 3. Describing the distribution of 1997 and 1999per capita consumption

Year Percentile of per capita consumption

10 25 50 75 90

1999 72.6 109.8 173.1 266.3 388.11997 94.8 128.2 194.0 285.9 403.4

Table 4. Transition matrix for households during1997–99 in relation to the poverty linea

Householdswith per capitaconsumptionin 1997

Households with per capitaconsumption in 1999

Below z Betweenz and 2z

Above2z

Total

Below z 66 20 2 88Between z and 2z 43 78 29 150Above 2z 6 40 63 109Total 115 138 94 347

a z is per capita consumption corresponding to the pov-erty line of 129.19 as described in Datt et al. (1998). Themethodology to estimate the poverty line is the costs ofbasic needs approach with region-specific poverty lines(Ravallion, 1994).

WORLD DEVELOPMENT74

have a per capita consumption below the pov-erty line in both periods (always poor), (b)households for which per capita consumptionfalls below the poverty line in one of the twoyears (sometimes poor), and (c) householdswith a per capita consumption that is neverbelow the poverty line in any year (never poor).From Table 5 it is clear that 66 out of 347 or

19% of all households are always poor. In ruralareas, this percentage is 23% and in urban ar-eas, it is 14%. Overall, 20% of households weresometimes poor––22% in rural areas and 18%in urban areas. Of all the households that werepoor at one time, 48% were always poor (44%in urban areas; 51% in rural areas).How do these results compare to others from

Africa? The ratio of always poor households tosometimes poor households (66/71 or 0.93) ishigher than 11 out of 13 such ratios reportedfor a number of developing countries in Baulchand Hoddinott (2000). Poverty that is persis-tent appears to be relatively high in Egypt.

Table 6 explicitly places the Egyptian results incontext by reporting the Egyptian resultsalongside the African studies reported byBaulch and Hoddinott (2000). There are in-herent difficulties in comparing these studies.They include factors such as different countries,different levels of representativeness of samples,different methods for calculating the povertylines, different duration of spells, and differentnumbers of repeated observations. Neverthe-less, the Egyptian results seem in line with otherAfrican estimates, being at the high end interms of the proportion of poverty that is per-sistent.The analysis so far has relied on the catego-

ries of ‘‘sometimes poor’’ and ‘‘always poor.’’How do we convert these categories intochronic and transitory poverty? One couldsimply classify the ‘‘always poor’’ as thechronic poor and the ‘‘sometimes poor’’ as thetransitory poor. One problem with this ap-proach is that the ‘‘sometimes poor’’ categorybecomes very heterogeneous when there aremore than two observations per household. Forexample, if we had six observations perhousehold, is it equally valid to label a house-hold as transitorily poor if it falls below thepoverty line one time out of six as compared toif it is below the poverty line five times out ofsix? This problem is addressed empirically byJalan and Ravallion (2000), who employ amethod for constructing chronic and transitory

Table 5. Number of households, by poverty status in thesubsample

Poverty status Rural Urban Total

Always poor 43 23 66Sometimes poor 42 29 71Never poor 101 109 210

Total 186 161 347

Table 6. A comparison of the Egypt results with others from Africa

Source Studylocation

Number ofobservationsin panel

Study dates Percent of households Percentagealways pooras a percent-age of everpoor

Always poor Sometimespoor

Never poor

Carter (1999) South Africa 2 1993 and1998

22.7 31.5 45.8 42

Dercon andKrishnan(2000)

Ethiopia 2 1994 and1995

24.8 30.1 45.1 45

Grootaert andKanbur (1995)

Coote d�Ivoire 2 1985 and1986

14.5 20.2 65.3 42



13.0 22.9 64.1 36



25.0 22.0 53.0 53

Haddad andAhmed(present paper)

Egypt 2 1997 and1999

19.0 20.4 60.6 48

Source: Adapted from Baulch and Hoddinott (2000). Note that the welfare measure is expenditure per capita in every

case.


poverty based on Ravallion (1988). Specifically,nontransient welfare is proxied by the inter-temporal mean of per capita consumption.Chronic poverty occurs when mean consump-tion falls below the poverty line. Transitorypoverty is the difference between overall pov-erty (the squared poverty gap measure) andchronic poverty. 4 Note that this definition ofchronic poverty goes beyond the classificationof ‘‘always poor.’’ Individuals or householdsthat are sometime poor can have a mean con-sumption below the poverty line and hence bechronically poor.Table 7 presents the estimates of chronic and

transitory poverty by region. The contributionof chronic poverty to the overall total is 66.7%,to rural poverty it is 67.9%, and to urban pov-erty it is 63.9%. In terms of broad regions, theMetropolitan region has the lowest percent ofchronic poverty, followed by the lower regions,with the upper regions having the highest pro-portion of chronic poverty. In terms of the 1997poverty profile of Egypt, the poorer regions––Upper Urban and Upper Rural––have thehighest percentage of chronic poverty. In com-parison to other studies from the developingworld that have used this method, the overallcontribution of chronic poverty to the total(a ¼ 2) is high. For four provinces in SouthernChina, the figure is 50.6% (Jalan & Ravallion,2000) and in a rural Pakistan sample, the figureis 18% (McCulloch & Baulch, 2000). 5

4. DETERMINANTS OF POVERTYCOMPONENTS

The previous section described the changes inwelfare or consumption for the 347 householdsduring 1997–99. This section explores the fac-

tors behind these changes. Before describingthe econometric methods used and the resultsobtained, the explanatory variables used in theregressions are described.

(a) Explanatory variables

The explanatory variables used in the re-gressions are those developed by Datt andJolliffe (1999) for their analysis of the determi-nants of poverty in Egypt in 1997. We use thevalues of the variables in the initial time period.The variables are summarized in Table 8.Household variables cover demographic com-position (household size, the number of indi-viduals below the age of 15, the number above60, the age of the head of household, whetherthe household head is declared to be a female);human capital (the average number of years ofschooling of household members); physicalcapital (the area of cultivated land that isowned, and the value of livestock); and occu-pation (whether the primary occupation of theprimary income earner––typically the house-hold head––is in manufacturing, construction,trade and services, community and recreation,agriculture, and other nonfarm sectors).The main difference between the variable set

used here and those used by Datt and Jolliffe(1999) is that we restrict ourselves to householdcharacteristics and exclude community (that is,primary sampling unit) characteristics. We dothis because the sample of households at ourdisposal (n ¼ 347) proves too small to incor-porate community characteristics.

(b) Determinants of total, chronic, andtransitory poverty

Using this information we estimate the de-terminants of total, chronic and transitory

Table 7. Total, chronic, and transitory poverty, by region

Region Component of poverty (P2)

Total Chronic Transitory

Absolute value Percentageof total



Metro 0.02715 100 0.01533 56.5 0.01182 43.5Lower Urban 0.01975 100 0.00991 50.2 0.00984 49.8Lower Rural 0.02699 100 0.01756 65.1 0.00943 34.9Upper Urban 0.04073 100 0.02982 73.2 0.01091 26.8Upper Rural 0.07449 100 0.05114 68.7 0.02335 31.3All rural 0.05358 100 0.03636 67.9 0.01722 32.1All urban 0.02999 100 0.01917 63.9 0.01082 36.1

Total 0.04355 100 0.02905 66.7 0.01450 33.3

WORLD DEVELOPMENT76

poverty. We follow the approach of Jalan andRavallion (2000). Our dependent variables are(i) the squared poverty gap measure for eachhousehold, (ii) the chronic component, and (iii)the transitory component. Our dependentvariable will have zero values for householdsabove the poverty line. The presence of zeros inthe dependent variable is indicative of censor-ing of an underlying variable and requiresTobit estimators, but as Jalan and Ravillionindicate, these estimators are sensitive to themisspecification of the error term (i.e., if it isnot distributed normally and its variance is notconstant). As in Jalan and Ravallion, we cir-cumvent the censoring problem by doubling thepoverty line and recalculating total, chronic,and transitory poverty. We thus increase thepool of chronic and transitory poor householdsand decrease the number of never poor house-holds. By doing this, 70% of the householdshave some chronic poverty and 73% have sometransitory poverty. These numbers are similarto Jalan and Ravallion (2000). This meanshowever, that our goal should be to estimatethe dependent variable around the 70th quan-tile, not around the mean, as ordinary leastsquares (OLS) would do. The model is there-fore estimated by quantile regression (70thquantile). 6

The results are presented in Table 9. 7

Larger households are more likely to experi-ence poverty and chronic poverty, but nottransitory poverty. Similarly, households con-taining more members under the age of 15 aremore likely to be poor and chronically poor.Older household heads are less likely to be poorand chronically poor. Households containingmore members over 60 are more vulnerable to

transitory poverty (but not to chronic poverty).The value of livestock assets helps householdsto reduce total and chronic poverty. Theamount of owned land that is suitable for cul-tivation is negatively associated with total andchronic poverty, but is positively associatedwith transitory poverty.The mean years of schooling of the household

members has a significant and negative associ-ation with all forms of poverty (although onlyat the 9% level for transitory poverty), with theestimated coefficient for chronic poverty being30 times the estimated coefficient for transitorypoverty. Location in an urban area has no as-sociation with chronic poverty, but has a neg-ative association with transitory poverty (at the6% level). Finally, the sector containing theprimary occupation of the primary incomeearner is important. Being in the manufacturingsector, the community and recreation sectorand in the other nonfarm sector means thathouseholds are less likely to experience totalpoverty and chronic poverty compared tohouseholds employed in the agriculture sector.But the sector of employment of the primaryincome earner demonstrates no significant as-sociation with transitory poverty.Several things are worth noting about these

results. First, the results for total poverty andchronic poverty are similar. This echoes theresults of Jalan and Ravallion (2000) forsouthern China. Second, note that fewer vari-ables have estimated coefficients that are sig-nificantly different from zero in the transitorypoverty regression. We are better able to pre-dict chronic poverty than transitory poverty.This is in line with the studies cited in Baulchand Hoddinott (2000).

Table 8. Explanatory variables used in the regression equationsa

Variable description Mean Minimum Maximum

Household size 6.00 1 15Number of household below age 15 2.25 0 9Number of household above age 60 0.369 0 7Age of household head in years 48.24 18.78 109.58Household members, mean years of schooling 6.08 0 17Log: value of livestock owned 2.313 0 9.695Log: area cultivated land that is owned by the household 0.517 0 7.783Industry: manufacturing 0.156 0 1Industry: construction 0.055 0 1Industry: trade and service 0.144 0 1Industry: community and recreation 0.248 0 1Industry: other nonfarm employment 0.179 0 1Industry: agriculture 0.219 0 1Urban ¼ 1, rural ¼ 0 0.466 0 1

a n ¼ 347 for all variables.


Third, note that the determinants of totalpoverty are similar to those derived from thecross-section estimates of Datt and Jolliffe(1999). But two of the variables that are foundto be positively associated with transitorypoverty here––the number of household mem-bers above 60 and the amount of cultivatedland owned––were either not significant in thecross-section regression (number of individualsabove 60) or had the opposite sign (cultivatedland owned).Fourth, the significance of the negative as-

sociation of years of schooling with bothcomponents of poverty is not sensitive tospecification. 8 We are unable to decomposethis result into the returns from high qualityschooling, from credentialism, and from theself-selection of innately intelligent individualsinto longer periods in school. Nevertheless,given the strength of the relationships estimatedbetween schooling and poverty reduction, anin-depth investigation of the access and quality

of public schooling in Egypt would seem to beworthwhile. Despite the strong result for yearsof schooling on poverty reduction, the rela-tionship between education and employment inEgypt is a complex one. For example, Assaad,El-Hamidi, and Ahmed (1999) use the EHIS toconclude that for men, the employment returnsare highest for basic education in rural areasand for university education in urban areas.For women, the returns to more education inurban areas are especially poor as public-sectorjobs have dried up. Furthermore, the private-sector jobs that are currently being created areculturally inappropriate for women and in anycase do not require high levels of education.But, no matter how complex the relationship,improving access to education, and raising thequality of educational services would seem tobe crucial components of an Egyptian povertyreducing strategy, particularly a long-term one.Fifth, the other significant correlates of

transitory poverty––living in an urban area

Table 9. Determinants of total, chronic, and transitory poverty: quantile regressions (70th quantile)a

Explanatory variables Total poverty Chronic poverty Transitorypoverty

Household size 0.012424(3.55)�

0.012942(3.08)�

0.000293(0.37)

Number of household members below age 15 0.026589(5.08)�

0.027103(4.39)�

)0.001171(1.07)

Number of household members above age 60 0.011927(0.90)

0.009799(0.61)

0.004924(2.13)��

Age in years of household head )0.002498(4.29)�

)0.002599(3.81)�

)0.000092(0.74)

Log: area of cultivated land that is owned by thehousehold

)0.015992(3.42)�

)0.015450(2.79)�

0.002755(2.59)�

Log: value of livestock owned )0.006972(2.73)�

)0.006590(2.19)��

)0.000308(0.54)

Household members, mean years of schooling )0.020087(12.78)�

)0.018982(10.19)�

)0.000623(1.71)

Industry: manufacturing ¼ 1 )0.043844(2.04)��

)0.050566(1.95)��

0.000618(0.12)

Industry: construction ¼ 1 )0.036204(1.26)

)0.029720(0.86)

0.001134(0.16)

Industry: trade and service ¼ 1 )0.033786(1.63)

)0.026622(1.07)

)0.004985(0.99)

Industry: community and recreation ¼ 1 )0.060617(3.34)�

)0.062766(2.83)�

)0.001576(0.34)

Industry: other nonfarm ¼ 1 )0.085343(4.10)�

)0.092771(3.74)�

)0.003599(0.73)

Urban ¼ 1, rural ¼ 0 0.011178(0.77)

0.005947(0.35)

)0.006005(1.88)

Constant 0.393111(11.50)�

0.380087(9.56)�

0.025799(3.16)�

Observations 347 347 347

aAgriculture is the omitted industry employment sector. Absolute value of z-statistics in parentheses.* Significant at 1% level.** Significant at 5% level.

WORLD DEVELOPMENT78

(less likely to experience it, but significant onlyat the 6% level), more household members over60 and more owned cultivated land (more likelyto experience it)––offer some clues as to thenature of risk faced by these households. Recallthat these are households that have a meanconsumption level above the poverty line, buthave fallen below the line in one of the years.On average, urban livelihoods are probably lessrisky––being based more on wage work (al-though being in an urban area may be a riskybusiness for the very poor should they losewage employment), while the holding of culti-vatable land seems to open up the holder toincome risk. The fact that households con-taining more members over the age of 60 arevulnerable to transitory but not chronic pov-erty suggests that the social security and welfareaid program in Egypt may not be performingwell. Assaad and Rouchdy (1998) state that‘‘Although the social security and welfare aidsystem is the largest social assistance programin Egypt, very little is known about its deliverysystem and its success in reaching the mostneedy’’ (p. 57). Our results are consistent withthe need to review the effectiveness of this sys-tem. 9 The result on the ownership of landcan be viewed in a number of ways. 10 Sea-sonality is not an issue in Egyptian agriculture,given the almost exclusive reliance on irrigationfrom the Nile, but the result may reflect land-owners� inability to insure completely againstweather or pest shocks. Another land-relatedshock during this period was a major land re-form that came into effect in 1997 and gavelandlords increased power to terminate theusufruct rights of tenants. Unfortunately, wehave no information on shocks in the data setwith which to further explore this result, orwith which to add to the explanatory power ofthe transitory poverty model.Finally, the total and chronic poverty-

reducing sectors of the economy appear to be inthe manufacturing, community, and recreationactivities and in residual nonfarm activities notcovered by the other categories. The ‘‘commu-nity and recreation’’ category includes public-sector jobs in administration and defense,social services, and jobs in recreation/culturalamenities and services. The residual nonfarmcategory consists largely of personal/householdservices. These sectors likely provide moresteady income flows and help families avoidchronic poverty, although they have no asso-ciation––positive or negative––with transi-tory poverty. 11 The findings on nonfarm

income are consistent with the observations ofAdams (1999) on the importance of this type ofincome for decreasing income inequality inEgypt. The findings are also consistent withSingerman�s (1995) conclusions as to the im-portance of informal networks to livelihoodgeneration in Cairo. Also note that despite itscontraction, employment in the public sectorstill reduces the likelihood that a household willexperience chronic poverty.

5. CONCLUSIONS AND IMPLICATIONS

This paper has used data from a 1999 surveyof 347 households that were first surveyed as apart of the EIHS in 1997. The householdsrepresent 19 PSUs and are drawn from sevengovernorates in all regions of Egypt. Ouranalysis indicated that the subsample was nottoo dissimilar from the full sample of 2,450households and that our failure to interview all380 households in the 19 PSUs was not likely todo violence to subsequent analyses of the 347households.As a sample, per capita consumption has

decreased for these 347 households. The de-crease has not been dramatic––from a mean240 LE per capita per month to 213 LE, but ithas occurred at all points along the distribu-tion. The reductions in consumption are rela-tively large for the Upper Rural region and forthe Metro region. The fortunes of somehouseholds have improved over the two-yearperiod––22 have climbed out of poverty, butmore have worsened (49 fell into poverty). Themost striking aspect of poverty for thesehouseholds has been the low mobility betweenconsumption groups. In fact, 60% of thehouseholds did not move out of their threeconsumption groups (below the poverty line,between the poverty line and twice the povertyline, and above twice the poverty line). This isreflected in the high proportion––67%––ofoverall poverty that is chronic (average con-sumption over time is below the poverty line).Moreover, almost half of all poverty is persis-tent (that is, persistently below the poverty lineor ‘‘always poor’’).We attempted to identify the determinants of

total, chronic, and transitory poverty usingquantile regression. Our efforts were hamperedby the small sample size relative to the measuredchange in consumption that occurred over thetwo periods. Nevertheless, we were able toidentify some factors that influence total,


chronic, and transitory poverty. These factorsinclude the average years of schooling of adulthousehold members (which reduces both formsof poverty, but chronic poverty most strongly),the value of land and livestock (but not fortransitory poverty), the number of childrenunder 15 and household size (which both in-crease total and chronic poverty), the locationof residence (with urban households less likelyto experience transitory poverty), and employ-ment activity in manufacturing, community,and recreational and other nonfarm sectors(which reduces total and chronic poverty).The predominantly chronic nature of poverty

in our sample, and our ability to identify as-sociated characteristics, highlights the need forthe Egyptian government to improve the assetaccumulation process for the poor. Unfortu-nately we cannot distinguish between the dif-ferent processes that deter accumulation (doshocks undermine the accumulation process oris no accumulation taking place?). The chronicnature of poverty also has implications for thetargeting of anti-poverty interventions such asoil and sugar subsidies. If most poverty weretransitory, it would be more difficult to justifysuch a targeted approach, because the targetwould keep shifting. The nature of poverty asrevealed in our sample of households thereforeis consistent with an administrative approachto the targeting of anti-poverty interventions.Our analysis is able to document the extent

of transitory poverty, but does relatively poorly

in explaining the determinants of this type ofpoverty. This is in line with other similarquantitative work. Our estimates would cer-tainly have benefited from a larger samplesize and more rounds of data, but we suspectthat more qualitative work would have thelargest payoff in terms of identifying the trig-ger events that plunge people into or elevatethem out of poverty. Nevertheless, the find-ings that urban households are less likely toexperience transitory poverty and that house-holds that have older members or own culti-vated land are more likely to experience it havesome implications for the design of insurancemechanisms in rural areas. As Morduch (1999)notes in a more general context, the appro-priate role for governments is to facilitatethe provision of insurance from the privatesector.Finally, the analysis here has demonstrated

the possibilities opened up by a longitudinaldata set. On the one hand, the paper hasdemonstrated the limitations of a small data setwhen the change in welfare has been small––nottoo many of our explanatory variables do anyexplaining. On the other hand, even from thissmall data set, we can get an insight as to thestructure of poverty and understand why somehouseholds have improved their welfare whileothers have not. The returns to introducing alongitudinal component to poverty data col-lection in Egypt and elsewhere in Africa wouldseem to be quite large.

NOTES

1. As described in Datt, Jolliffe, and Sharma (1998).

2. The studies use essentially the same data sources,

but they use different methods and assumptions in

constructing a poverty line and in constructing real total

expenditure or income. This results in some studies that

find poverty increasing between 1990–91 and 1995–96

(see El-Laithy, 1996; Cardiff, 1997, and for the 1996

Egyptian Human Development Report (EHDR) in El-

Laithy & Osman, 1996 in urban areas at the lower

poverty line) and some that find a decline (EHDR, 1996,

upper poverty lines and EHDR, 1996, lower poverty line

in rural areas). Note that some may be tempted to use

panel data sets to track poverty rates over time, but

because the original cohort becomes less representative

of the total population over time (for example, the

households age), one cannot make statements about

change in poverty. The shorter the period between

surveys, the less of a problem this is, but to be totally

confident, additional households need to be sampled in

subsequent surveys to restore the sample to a represen-

tative subset of the population.

3. For Egypt, the cross-section study of the correlates

of poverty is by Datt and Jolliffe (1999), who base their

analysis on the 1997 EIHS. For rural areas, they find

that years of schooling, the completion of primary

school, the ownership of cultivated land, the value of

livestock, and work in the recreation and community

sector to have positive effects on real per capita

consumption, while female headedness, household size,

the number of children, and unemployment (male and

female) had negative effects. For urban areas, they found

that years of schooling had an even larger positive effect

on real per capita consumption than in the rural areas,

as did the value of livestock. Household size and female

unemployment had a negative impact on real per capita

consumption. One could undertake such an analysis for

WORLD DEVELOPMENT80

the Household Income and Expenditure Surveys (HIES)

collected by Central Agency for Public Mobilization and

Statistics (CAPMAS), but with a much more restricted

set of household and community determinants, given

that the HIES does not collect much information outside

of incomes and expenditures. If the HIES were more

broad-based in the information it collects, one could get

an invaluable moving snapshot of the determinants of

poverty over time.

4. This method is not valid for the headcount and

poverty gap measures but is valid for the squared

poverty gap measure as discussed by Jalan and Raval-

lion (2000).

5. Although these studies are not strictly comparable

because they use more than two rounds of data. In a

rural Ethiopia sample, chronic poverty ranges between

75% and 92% of the total, but it is reported only for the

poverty gap measure (a ¼ 1) and so it is not comparable(Dercon & Krishnan, 2000).

6. One alternative to this investigation of the determi-

nants of chronic and transitory poverty is multinomial

logit with never, chronic and transitory as the categor-

ical outcomes. Results from this estimation procedure

are very similar to the results in Table 9 in magnitude,

but are weaker in terms of statistical significance. This is

perhaps not surprising given that the multinomial logit

procedure does not capture all the information available

on the dependent variable (i.e., given the availability of

the underlying continuous variable).

7. The qreg command in STATA (v6) is used. The

equality of the estimated coefficient on average years of

schooling (which has the most precisely estimated

coefficient) estimated at the 50th, 70th, and 90th

quantiles could not be rejected, indicating that the

estimates are not too sensitive to the quantile selected.

When bootstrapped standard errors are generated

instead of the estimates derived from the estimated

variance–covariance matrix under the qreg procedure,

the standard errors increase somewhat, but the main

results hold. This indicates that heteroskedasticity is not

a large problem (STATA, 1999).

8. Several additional econometric models were tried

(including OLS and multinomial logit). In addition,

several different quantiles were used in the quantile

regressions, with full interaction terms for the rural/

urban breakdown.

9. The introduction of a squared term for the head of

household to test whether the elderly variable is picking

up lifecycle effects does not eliminate the significant

positive result on the elderly variable in the same

regression. The estimated coefficient on the square of

age is positive, but not close to significance.

10. Seventeen percent of rural households own some

land that is cultivated as do 7% of urban households, so

the urban households are not treated any differently

from a regression viewpoint. When the land owned

variable is dropped, the urban result on transitory

poverty still holds.

11. An alternative specification was tried by including

variables describing the type of employment household

members were engaged in during the previous 12

months. Variables were constructed that summed up

the number of people above the age of 10 in the

household by type of primary occupation in the last 12

months; casual labor, salaried, own business, student,

and looking but unable to find work. These were added

to the regression analysis. None of these variables were

significant at the 5% level, with or without the industry

variables.

REFERENCES

Adams, R. (1999). Nonfarm income, inequality and landin rural Egypt. PRMPO/MNSED. Washington, DC:World Bank.

Alderman, H., & Lindert, K. (1998). The potential andlimitations of self-targeted food subsidies. Worldbank Research Observer, 13(2), 213–229.

Assaad, R., & Rouchdy, M. (1998). Poverty and povertyalleviation strategies in Egypt. Report to the FordFoundation. Humphrey Institute of Public Affairs,University of Minnesota, USA.

Assaad, R., El-Hamidi, F., & Ahmed, A. U. (1999). Thedeterminants of employment status in Egypt. Reviseddraft. Washington, DC: International Food PolicyResearch Institute.

Baulch, R., & Hoddinott, J. (2000). Economic mobilityand poverty dynamics in developing countries. Jour-nal of Development Studies, 36(6), 1–24.

Bouis, H. E., Ahmed, A. U., & Hamza, A. S. (1999).Patterns of food consumption and nutrition in Egypt.Washington, DC: International Food Policy Re-search Institute.

Cardiff, P. (1997). The 1995/96 household income,expenditure and consumption survey: Final analysisreport. Paper prepared for the US Agency forInternational Development/Egypt and US Bureauof Census, Washington, DC.

Carter, M. (1999). Getting ahead or falling behind? Thedynamics of poverty in post-apartheid South Africa.


Department of Agricultural and Resource Econom-ics, University of Wisconsin, Madison.

Datt, G., & Jolliffe, D. (1999). Determinants of poverty:1997. International Food Policy Research Institute,Washington, DC.

Datt, G., Jolliffe, D., & Sharma, M. (1998). A profile ofpoverty in Egypt: 1997. Food Consumption andNutrition Division Discussion Paper No. 49. Interna-tional Food Policy Research Institute, Washington,DC.

Dercon, S., & Krishnan, P. (2000). Vulnerability,seasonality and poverty in Ethiopia. Journal ofDevelopment Studies, 36(6), 25–53.

El-Laithy, H. (1996). Structural adjustment and poverty.Paper presented at the International Conference onStructural Adjustment, University of California, LosAngeles, CA.

El-Laithy, H., & Mukherjee, S. (1999). Review ofpoverty in Egypt: Estimates, methods, and compa-rability. Egypt Food Security Project Paper No. 6.International Food Policy Research Institute, Wash-ington, DC.

El-Laithy, H., & Osman, M. (1996). Profile and trend ofpoverty and economic growth in Egypt. ResearchPapers Series. Egypt Human Development Report.United Nations Development Programme/Instituteof National Planning.

Grootaert, C., & Kanbur, R. (1995). The lucky fewamidst economic decline. Distributional changes inCoote d�Ivoire as seen through panel data sets, 1985–1988. Journal of Development Studies, 31(4), 603–619.

Jalan, J., & Ravallion, M. (2000). Is transient povertydifferent? Journal of Development Studies, 36(6), 82–99.

May, J., Carter, M., Haddad, L., & Maluccio, J. (2000).KwaZulu-Natal income dynamics study (KIDS)1993–1998: A longitudinal household data set forSouth Africa policy analysis. Development SouthernAfrica.

McCulloch, N., & Baulch, B. (2000). Simulating theimpact of policy upon chronic and transitory povertyin rural Pakistan. Journal of Development Studies,36(6), 100–130.

Morduch, J. (1999). Between market and state: Caninformal insurance patch the safety net?World BankResearch Observer, 14(2), 187–208.

Ravallion, M. (1988). Expected poverty under risk-induced welfare variability. Economic Journal, 98,1171–1182.

Ravallion, M. (1994). Poverty comparisons. Chur, Swit-zerland: Harwood Academic Publishers.

Singerman, D. (1995). Avenues of participation: Family,politics and networks in urban quarters of Cairo.Princeton, NJ: Princeton University Press.

STATA (1999). Reference manual: P-St. Release 6.College Station, Texas, Stata Corp.

Stats SA (2000). Statistics South Africa. Government ofSouth Africa, Pretoria. Available: http://www.stats-sa.gov.za.

Tuck, L., & Lindert, K. (1996). From universal foodsubsidies to a self-targeted program. World BankDiscussion Paper No. 351. World Bank, Washington,DC.

World Bank (1999). World development indicators. TheWorld Bank, Washington, DC.

APPENDIX A

Table 10. Sample frame for 1999 subsample resurvey

Region Region code Governorate Governoratecode

PSU PSU code Urban ¼ 1,rural ¼ 0

Metro 1 Cairo 1 El Sabteya 104 1Shagaret Mariam 106 1Masaken ElAmeria

107 1

Lower Urban 2 El Sharkia 7 Mashtool ElSouk

134 1

El Gharbia 10 Nasser 156 1El Beheira 12 El Mahmoudia 167 1

Lower Rural 3 El Sharkia 7 Sheibat El Nika-ria

137 0

Mashtool ElKady

139 0

El Gharbia 10 Kafr Dima 159 0El Beheira 12 Badr 169 0

KombaniatAbou Keir

172 0

WORLD DEVELOPMENT82

http://www.statssa.gov.za

http://www.statssa.gov.za

Table 10—(continued)

Region Region code Governorate Governoratecode

PSU PSU code Urban ¼ 1,rural ¼ 0

Upper Urban 4 El Menia 17 Thaleth 196 1Asyout 18 Al Walidiah 204 1Sohag 19 Sakolta 211 1

Upper Rural 5 El Menia 17 Beni Ali 200 0Talah 202 0

Asyout 18 Shotb 207 0Nagea Sabae 206 0

Sohag 19 Aldayabat 214 0

Table 11. Distribution of PSUs, by region and by governorate, 1997 and 1999

PSUs in 1999 subsample PSUs in 1997 EIHS

Region1 3 192 3 183 5 344 3 205 5 34

Governorate1 3 112 0 74 0 35 0 76 3 77 0 128 0 49 2 610 0 511 3 612 0 213 0 1714 0 315 0 516 2 717 3 718 2 619 1 820 0 2

All Egypt 19 125

Table 12. Logit estimates to assess representativeness of the subsample and the extent of attrition biasa

Explanatory variable Dependent variable (1)(1 ¼ in subsample,0 ¼ not in subsample)

Dependent variable (2)(1 ¼ did not attrit from subsample,0 ¼ did attrit from subsample)

Real per capita expenditure )0.000464 (1.44) )0.000495 (0.63)Household size 0.016112 (0.53) )0.084249 (0.68)Number of household members below15 years

)0.020254 (0.44) 0.243595 (1.34)

Number of household members above60 years

0.011380 (0.10) 0.131767 (0.41)

Age of head of household in years )0.001282 (0.34) )0.010495 (0.80)(continued next page)



Explanatory variable Dependent variable (1)(1 ¼ in subsample,0 ¼ not in subsample)

Dependent variable (2)(1 ¼ did not attrit from subsample,0 ¼ did attrit from subsample)

Dummy: female-headedhousehold ¼ 1

)0.187315 (1.09) 0.036334 (0.06)

Average years of schooling,household members

0.007481 (0.47) 0.019676 (0.35)

Mother or father: primary schoolingor more ¼ 1

)0.137068 (0.89) 0.227050 (0.40)

Log: owned cultivated land )0.074563 (1.72)� )0.240205 (1.72)�Log: livestock value )0.000424 (0.02) 0.172769 (1.69)�

Industry: manufacturing ¼ 1 )0.395656 (2.18)�� 1.020709 (1.22)Industry: construction ¼ 1 )0.516705 (1.98)�� 0.232389 (0.21)Industry: trade and service ¼ 1 )0.196893 (1.08) )0.388722 (0.64)Industry: community andrecreation ¼ 1

)0.077621 (0.48) )0.205251 (0.38)

Urban ¼ 1 0.122262 (0.89) )0.568126 (1.20)Constant )1.446430 (5.72)�� 3.109337 (3.22)��

Observations 2449 374Log likelihood )1039.04 )90.24Pseudo R2 0.0073 0.0694

aAbsolute value of z-statistics in parentheses.* Significant at 10% level.** Significant at 5% level.*** Significant at 1% level.

Table 13. PSU-level food price index, 1999 (base: 1997 ¼ 100Þa

Region code Governorate Governoratecode

PSU PSU code 1999 Price index

Metro 1 Cairo 1 El Sabteya 104 106.6ShagaretMariam

106 107.1

Masaken ElAmeria

107 99.8

Lower Urban 2 El Sharkia 7 Mashtool ElSouk

134 107.3

El Gharbia 10 Nasser 156 102.4El Beheira 12 El Mahmoudia 167 100.5

Lower Rural 3 El Sharkia 7 Sheibat ElNikaria

137 106.8

Mashtool ElKady

139 111.1

El Gharbia 10 Kafr Dima 159 107.8El Beheira 12 Badr 169 106.2

KombaniatAbou Keir

172 106.8

Upper Urban 4 El Menia 17 Thaleth 196 109.0Asyout 18 Al Walidiah 204 112.1Sohag 19 Sakolta 211 114.4

WORLD DEVELOPMENT84


Region code Governorate Governoratecode

PSU PSU code 1999 Price index

Upper Rural 5 El Menia 17 Beni Ali 200 100.0Talah 202 111.7

Asyout 18 Shotb 207 119.4Nagea Sabae 206 110.1

Sohag 19 Aldayabat 214 118.4

a PSU-level food price index for 1999 is a weighted aggregate Laspeyres� price index with 1997 as base year. Regionalfood budget shares used as weights. These weights are derived from Bouis, Ahmed, & Hamza (1999), based on datafrom the 1997 EIHS.


Documents

Chronic and Transitory Poverty: Evidence from Egypt, 1997–99