Poverty in Tunisia: A Non-Monetary Approach *

Transcription

1 Poverty in Tunisia: A Non-Monetary Approach * Mohamed Ayadi, AbdelRahmen El Lahga Abstract In this study we construct a composite index from a set of non-monetary household living conditions indicators (WCI) in order to analyze the evolution of poverty and inequality in Tunisia between 1988 and 2001 from a multidimensional perspective. We show that poverty significantly decreased during the period, although regional and rural/urban disparities remained unchanged. Poverty is essentially a rural phenomenon. The Northwest and Central West (NW, CW) regions remained the poorest areas of the country. Improvement in housing conditions and in the access to communication tools may be important in the effort to fight poverty.. JEL classification: D31, D63 Key words: poverty, Tunisia * This work was carried out with the grant from the Poverty and Economic Policy (PEP) Research Network, financed by the International Development Research Centre (IDRC). We thank Abdelkrim Araar, Louis-Marie Asselin, Sami Bibi, John Cockburn, Jean-Yves Duclos, Jean-Bosko Ki and Mohamed Kriaa. Our thanks also go to all the administrative PEP and CIRPÉE personnel. In particular, we are very grateful to Aissatou Diop, Évelyne Joyal, Gaétane Marcoux, Sonia Moreau and Johanne Perron. Financial assistance from Tunisian Ministry of higher education are also acknowledged. Tunis Business School and UAQUAP, University Of Tunis, Tunisia. med.ayadi@gnet.tn Tunis Business School and UAQUAP, University Of Tunis, Tunisia. rahmen.lahga@gmail.com 2

2 1. Introduction Over the past four decades Tunisia has implemented several programs to fight poverty. Since the independence in 1956, compulsory education has been instated. As a consequence average educational attainment nearly doubled between 1960 and 1980, and continues to increase. The ratio of illiterate people has been about 72% for men in 1970 and 85% for women. In 1998, this ratio dropped to 31% and 42% for men and women, respectively. In the same time an indirect subsidy of basic commodities and medical care has been generalized, that improved the nutritional and healthcare status of the poor. Life expectancy in Tunisia is now over 70 years. These policies have been accompanied by direct transfer programs such as the program of assistance to needy families programme d aide aux familles nécessiteuses. The development of export and labour- intensive industries employing female labour in particular, and the improvement of infrastructure in disadvantaged areas have contributed to the amelioration of employment levels. Macroeconomic performances in Tunisia were noticeable between 1990 and Economic growth was particularly impressive between 1995 and 2000, a period characterized by an annual increase of GDP by 5.6%. These investments and economic strategies have improved the human capital quality. The consequence of this economic performances are reflected by the HDI (Human Development Index) improvement which rises from.516 to.745 between 1975 and Nevertheless, an objective assessment of the impact of these different policies on poverty alleviation requires a rigorous study based on microeconomic data. A long literature has grown up around the poverty phenomena in Tunisia. For instance (Ayadi et al., 1995, 2001, 2004; Bibi, 2001;INS, 1985,1990,1995, 2000; World Bank, 1990, 2003) show that a significant reduction in the level of poverty. Western regions and especially rural areas highly contribute to overall poverty. All these studies are based on a monetary approach using income as the sole indicator of well-being. However using income or expenditure as an indicator of household wealth, is no longer unanimously accepted among economists as the only poverty analysis framework for many conceptual and technical problems. At the technical level, while many of household income and expenditures surveys are available for Tunisia, using these surveys to make inter-temporal comparisons of poverty is problematic. Indeed, as documented by Sahn and Stifel (2003) in more 1

3 detail, the continual changes in surveys designs (e.g., the recall period, the number and choice of item codes listed...), the absence of valid regional price indices, rental market, self-employment and seasonal variability in earnings can have a non negligible effect on the measurement of household expenditures or income. At the conceptual level, and at least since Sen's (1985) paper, it is common to argue that income covers one aspect of living standard. Others aspects should be included in the evaluation of well-being, like living conditions and access to basic facilities. Even if you have enough income, but you live in an unsanitary house, you will obtain a little benefit from your expenditures. Given these problems one can resort to a non monetary approach based on the construction of a composite asset index as a proxy of household wealth. The underlying idea of this approach is that some non-monetary attributes such as the nutritional status, housing conditions, or the health environment, may be considered as personal welfare indicators. Constructing a welfare composite index based on these indicators provides a reliable 4 alternative to the welfare monetary indicator and falls within the general framework of the multidimensional poverty analysis. The aim of this paper is to establish a poverty profile in Tunisia for the periods, using a non-monetary approach. More specifically, a welfare composite index is constructed based on household non-monetary attributes in order to deal with the multidimensional nature of poverty. Thus our approach offer a new framework toward a better understanding of poverty. The paper is organized as follow. Section 2 presents our Welfare composite index. Section 3 analyzes the poverty profile at the global level. We decompose overall poverty by region and by source in section 4. Section 5 concludes. 2. The Welfare Composite Index (WCI) A prerequisite to our empirical analysis is a clear definition of what we mean by household welfare indicator. As discussed above, unlike the widely used procedure that proxies household wealth by current income or expenditure, we generate an index based on asset ownership (e.g, owning Tv, radio) and housing characteristics (e.g, type of toilet facilities, floor material), referred to as Welfare Composite Index 4 See Sahn and Stifel (2003) for an evaluation of the performances of a composite index constructed on the basis of African household non-monetary attributes, in attempting to predict their well-being as compared to their income. 2

4 (WCI), as an alternative proxy of household wealth. For the sake of brevity, we will sketch here the general methodology followed for the construction of the WCI. A more detailed technical presentation can be found in Ayadi et al. (2006). 2.1 Methodology Used for the Construction of the WCI Consider K primary indicators which reflect household living conditions such as the ownership of some durable goods and housing conditions. The basic idea is to summarize the information provided by these qualitative indicators into single composite index A which can be written for a household i as: K i = γ j, (1) j= 1 A I ij with I ij the primary indicator j( j = 1.. K) for household ii ( = 1... n) and γ j is the weight of the indicator I ij, which we must estimate. Previous literature uses different methods, which are almost based on multivariate statistical analysis, to estimate weightsγ j. For instance, Sahn and Stifel (2000) use factor analysis while Filmer and Pritchett (1998) use Principal component analysis (PCA), in order to determine the weights. In this paper we use another factor analysis variant, namely the multiple correspondence analysis (MCA), as suggested by Asselin (2002). This method is particularly suitable for the available data which include a set of binary variables representing the different modalities that primary indicators can take. Each primary indicator I ij can take J modalities, thus A i the composite index for household i can be rewritten as: A i with K Jk k k W 1 k 1 j I k= j = k ijk =, (2) K K : Number of primary indicators; J k : Number of indicator k jk k modalities; W : The weight attributed to modality j k ; I k ijk : A binary variable equal 1 when household i has modality j k, 0 otherwise. The Welfare Compose Index, for a householdi, is simply the average of the A i 3

5 k weights of binary variables I. ij k The weight to attribute to each component of index A i is the normalized score W k j k score = of the modality λ α eigen value for axisα I obtained from MCA. k ijk 2.2 The Data For our purpose we use data drawn from three surveys. The first one is the well known Demographic and Health Survey (DHS), conducted in The second and third surveys, entitled National Survey on Family Health, were conducted respectively in 1994 and 2001 by the Office national de la population et de la Famille (ONPF). The two latter surveys are considered as the continuation of the DHS program in Tunisia. Therefore, the three surveys are comparable in their methodologies and sampling designs. The sizes of the samples obtained from these surveys after eliminating the missing data are: 4184, 6080 and 6059 for the years 1988, 1994 and 2001 respectively. Although they were designed to assess the Tunisian women health status, these surveys contain rather precise information on household living conditions and can be readily used to construct our WCI. We select eleven primary indicators that can be classified into three categories: ownership of durable goods, housing conditions and education. Table 1 presents a detailed description of these indicators. 5 Table 1 : Primary indicators Variable Categories Scores Ownership of durables indicators Radio Yes 0.30 no TV Yes 0.45 no Refrigerator Yes 0.82 no Gas cooker Yes 0.30 no Telephone Yes 1.21 no Housing conditions Water Tap water Private well It would have been advisable to include other attributes such as the ownership of a car, household head s educational attainment or children s nutritional status. Unfortunately, such information is not available in all three surveys. 4

6 Well or public tap Surface water, other source Toilet facilities private toilets connected to sewage toilets not connected to sewage No toilet facilities Type of housing Hoovel Yes 0.21 No Quality of the floor Cemented and tiled floor 0.20 Earth or clay floor Number of people per bedroom Less than 2 people per bedroom 0.64 More than 2 people per bedroom Education Illiterate wife Wife can read easily Wife cannot read easily The MCA results Before presenting the MCA results, an important remark is in order. To implement coherent inter-temporal poverty comparisons, it is necessary to hold the weight given to each indicator in our WCI constant over time. Thus, we use the initial period (1988) survey to estimate the WCI weights. The resulting weights are then used to calculate the subsequent WCI for 1994 and 2001 surveys. 6 Weights for each primary indicator are presented in third column of Table Consistency of the WCI At this stage, a natural question that arises is whether the WCI reflect a welfare situation? Examination of indicators weight presented in table 1 ensures that the welfare ordinal structure of each variable is respected by the ordinal structure of scores of its modalities. For example, the MCA associates a positive weight to the private tap water modality and negative weights for unsafe and less accessible water sources. The second sensitivity analysis which indicates whether the estimated WCI describes welfare is informed by the percentage of households, not affording a good or a service, declines with movement from a lower to a higher 6 An alternative strategy consists in pooling all available data (three surveys) into a single database, and then applying an MCA analysis to estimate WCI weights. As the choice of either method remains an empirical issue, the two strategies have been applied to calculate two welfare composite indices. However, preliminary experiments, not reported here, show a strong correlation of between the two indices. Furthermore, the two indices classify households in the same manner. Thus for the sake of brevity we present only the results based on the WCI calculated following the first method. 5

7 quintile of the WCI. The detailed distribution of the primary indicator presented in table 2 shows that the WCI satisfies such statement. Table 2: Distribution of indicators by the Welfare Composite Index Quintile in 1988 Attributes in % Q1 Q2 Q3 Q4 Q5 Illiterate wife Hovel Earth floor No radio No TV No refrigerator No telephone No gas cooker More than 2 people in bedroom Have a reliable water source Have a toilet connected to sewage Note: A reliable water source means water supply facilities. Finally, table 3 demonstrates that area and regional classification according to WCI and per capita expenditure (extracted from the 1990 INS Household Budget survey) are similar. This finding gives an additional argument in favor of WCI as a welfare indicator. Table 3: Area and regional classification of WCI and per capita expenditure Dimension 1988 WCI Rank 1990 per capita expend. Rank Area Urban Rural WCI Rank 1990 per capita expend. Rank Region Greater Tunis Northeast Northwest Central West Central East South An Informal look to the WCI Descriptive statistics on WCI are presented in table 4. There is a net improvement of the index value between 1988 and 2001, which reflects the trend in household living standards as approximated by the WCI. This observed trend is in line with other aggregated indicators such as GDP per capita and Human Development Index (HDI) published by the UNDP. Table 4: WCI Descriptive Statistics Variable Mean Standard - Min. Max. N 6

8 deviation WCI WCI WCI Figure 1 presents the WCI cumulative distributions for each period. It can be shown that 2001 WCI distribution first order dominates the two previous distributions. Figure 1: WCI Cumulative Distribution in Percentile Welfare index dis1988 dis2001 dis1994 Table 5 reports the number of the unique values by WCI quintile. Some diversity in the household profiles is observed, especially in the first two quintiles. However, the most well-off households present profiles which are rather similar. It can be inferred that household deprivation is a multifaceted phenomenon. Table 5: WCI Unique Values by Quintile Q Q Q Q Q By construction the MCA may produce negative WCI for some households, as in the initial period the WCI has a zero mean and a variance equal to the eigen value of the 7

9 first axis. Consequently, traditional analysis using poverty measures normalized by a poverty line is not useful for the WCI. Two solutions are conceivable to overcome such a problem. The first consists to making a translation of the WCI initial values to obtain a distribution of a welfare indicator defined on a positive support. In other word, we can add the absolute value of the lowest score in the distribution to each household s scores. However, such translation modifies the distribution mean and affects poverty measurements and may influence our results. A more direct solution consists to use absolute poverty indices (not normalized by a poverty line). In what follow we opt for the latter solution. 7 3 Poverty Analysis 3.1 Poverty line Poverty analysis requires definition of a poverty line below which every individual is considered as poor. Studies based on monetary welfare indicators (income, expenditures) are often characterized by different points of view concerning the poverty line choice. An absolute poverty line, representing the minimum income required by an individual to meet his/her basic needs, or a relative poverty line defined as a proportion of the median income, representing the minimum income necessary to achieve a given level of well-being accepted by the society. The debate remains open among analysts, and the choice of either strategy depends on the objectives of each study and the ethical assumptions implicitly adopted by researchers 8. In our non-monetary framework the choice of the poverty line is somewhat less debatable for two reasons. Firstly, the definition of an absolute poverty line is not obvious since the welfare composite indicator used here does not include the nutritional dimension which is helpful in determining a minimum subsistence threshold. Secondly, there is need to determine a set of welfare indicators deemed essential for every individual to achieve a minimum level of well-being. The choice of such indicators will be arbitrary, regardless of our knowledge of Tunisians lifestyles. To give some robustness to our findings we define two relative poverty lines. A lower 7 It should, however, be noted that in previous versions, poverty analyses were carried out based on the first solution and that the findings were qualitatively comparable. 8 See Duclos and Araar (2005: chapter. 7) for a discussion on this issue. 8

10 line L= -.425, which corresponds to the 25 th percentile of the 1988 distribution, and an upper line U=-.049, which corresponds to the 40 th percentile of the same distribution. 9 Obviously, the arbitrary nature of this choice is unavoidable. However, as we will see, we conduct a robustness analysis to asses the validity of our conclusion for larger poverty line interval. 3.2 FGT Poverty Measures Several poverty yardsticks are proposed in the literature of poverty analysis. In this paper we use the family of poverty measures proposed by Foster, Greer and Thorbecke---FGT--- (1984), which satisfy several desirable properties especially decomposability by sub-groups suitable for assessing the contribution of geographical regions to over-all poverty. FGT measures are defined by: FGT α 1 = n n i= 1 I ( ) α z y y i, with I y an indicator function equal to 1 if y i z, 0 otherwise. y i the individual s i welfare indicator ( WCI), z the poverty line, n the size of the population, α a non negative parameter. Forα = 0, FGT 0 simply represents the proportion of the poor referred to as headcount (HC) or poverty incidence (PI). Forα = 1, FGT 1 represents the average poverty gap and expresses the average WCI necessary for an individual to be able to reach the poverty line. This measurement reflects the intensity of poverty. 3.3 Poverty Trends According to Filmer and Pritchett (2001) monetary based poverty analysis are sensitive to economic conjuncture; however the non monetary one may only be affected by welfare conditions. Poverty indices based on the WCI, reported in table 6, reveal a considerable decline in poverty over time for both the incidence and intensity of poverty. Interestingly our results reveal a clear decline of poverty between 1988 and 1994 which is in contradiction with those of the World bank, based on monetary approach (see table 6b in appendices) due to poor economic situations and years of 9 This strategy was adopted by Sahn and Stifel (2000). 9

11 drought that affected the country during this period. Table 6 : Evolution of Non-monetary Poverty Lower poverty line FGT 0 (percentile 25 of the 1988 distribution) FGT 1 Upper poverty line FGT 0 (percentile 40 of the 1988 distribution) FGT (.007) (.005) (.003) (.004) (.002) (.001) (.008) (.007) (.005) (.007) (.004) (.002) Note: Number in parentheses represent the standard deviations of poverty estimates. Annual growth rate of poverty index are presented (in %) in Table 7. First a considerable decreases of the headcount is observed for the whole period ( ) and for the two sub-periods ( ) and ( ). Table 7: Annual Variation Rate of Poverty Levels WB Lower line WB Upper line INS Lower line WCI Lower line WCI Upper line Year Global 0,51 1,11-1,49-8, ,61 1,96-0,55 1,35 1,21-2,81 Year Global -9,88-8,42-6,45-12, ,38-7,72-6,2-9,49-8,54-8,16 Year Global -4,81-3,89-3,73-10, ,85-3,26-3,29-4,39-3,92-4,91 Source: Our calculations. It is interesting to note that the World Bank s analysis reveal that the poverty index registered a 0.5% annual increase in the early 1990s, after the three years of drought (1993, 1994 and especially 1995) (see Table 6 b (in appendices) and table 7). However, our conclusions are in line with those of the National Institute of Statistics INS -- and show that poverty continued to decline between 1990 and 1995, but at a higher rate (4.29 %), and (1.49%) according to the INS, although the basic assumptions in these two studies are not similar, the fact that the WCI is less 10

12 sensitive to the cyclical variations in incomes might explain this situation. 3.4 Robustness Analysis Our inter-temporal comparisons of poverty are based on the chosen family of poverty measures. Thus, our conclusions could also be mitigated by an alternate choice of poverty line. In order to assess their robustness we resort to stochastic dominance analysis to establish the conditions under which poverty comparisons are robust within a plausible range of poverty lines and across a pre-defined family of poverty measures. 10 For the sake of simplicity, the distribution of the WCI will be standardized in such a way to get poverty line z equal to 100. Figures 2,3,4 in appendices, depict the differences between poverty incidence for the different sub-periods and for various poverty lines. It can easily be observed that the 1994 distribution dominates that of 1988, unless we consider an extremely high poverty line (over 175% when compared to the initial line). However the 2001 distribution dominates both the 1988 and 1994 distributions. Hence our first conclusion is a marked reduction of poverty in Tunisia between 1988 and 2001, irrespective of the poverty line selected. In order to ensure that the conclusions are not affected by sampling variability, a t statistical test has been computed to test the assumption of a difference between two distributions, for different values of z. Test resultss presented in Table 8 are a confirmation of the previous conclusions. Table 8: Difference between the Poverty indices Poverty line P ns 0.29 ns 0 P0 P P0 P ns P0 Note: (ns) means non-significant difference at a 5% threshold. 4. Poverty Decomposition The FGT α indices satisfy the propriety of decomposability by sub-group. In other words, the overall poverty index can be expressed as a weighted sum of poverty level within each sub-group. 10 On this see Atkinson (1987) Davidson and Duclos (2000). 11

13 Consider a partition of the whole population in K exclusive sub-groups, φ ( k) the relative size sub-group k. The FGT α index can be expressed as: FGT 1 n n α = φ ( k) FGTα ( z, k ) i= 1 (4) where FGTα ( z, k) denotes the poverty index of the sub -group k. Ceteris paribus, the improvement of the well-being of a given sub-group implies the improvement of the well-being of the entire population. Such decomposition have the advantage to permits the implementation of a targeting program in a decentralized manner, for each sub-groups. In what follows, we present the decomposition of the FGT index for α= 0 or 1, by area of residence and geographical regions. 4.1 Urban rural decomposition Table 9 presents FGT0 and FGT1 indices in urban and rural areas as well as the relative contribution of each area to the overall poverty for the entire period. Table 9: Decomposition of Poverty by Area Lower poverty line FGT0 FGT1 FGT0 FGT1 FGT0 FGT1 Area Urban (.118) (.068) (.077) (.035) (.05) (.03) Rural (.882) (.932) (.923) (.965) (.95) (.97) Higher poverty line Urban (.21) (.11) (.2) (.1) (.05) (.1) Rural (.79) (.89) (.8) (.9) (.85) (.9) Note: The figures in parentheses represent the relative contribution of each group to the overall poverty measurement. Table 9 show that there was a significant drop in poverty, for both the urban and the rural areas. However, the high poverty incidence in the rural areas can also be observed for various years. Such trends are the same as those of the World Bank conclusion based on monetary welfare indicator, and contrary to those of INS who claims that poverty is an urban phenomenon. (See Table 7 and Table 6b in 12

14 appendix). These findings can be predicted since the WCI used here is based on indicators whose distributions are more concentrated in the urban areas (such as sanitation, type of housing, ownership of certain durable goods). However, although the incidence of poverty dropped between 1988 and 2001, the relative contribution of the rural areas to the overall poverty remained constant, close to 75% during the three years. A huge differential can also be observed with regard to the average poverty gap ( FGT 1 ) between the urban and the rural areas: in the urban areas, as against in the rural areas in Such findings confirm the dual nature which usually characterizes the distribution of wealth in developing countries. 4.2 Regional Decomposition Table 10 presents FGT0 and FGT1 indices for the different geographical regions, as well as their relative contributions to overall poverty. Table 10: Decomposition of Poverty by Region Lower poverty line FGT0 FGT1 FGT0 FGT1 FGT0 FGT1 Region Gr. Tunis (.027) (.025) (.05) (.035) (.03) (.03) Northeast (.114) (.104) (.11) (.11) (.08) (.08) Northwest (.3) (.405) (.35) (.41) (.24) (.23) Central West (.255) (.233) (.32) (.32) (.45) (.48) Central East (.147) (.085) (.1) (.075) (.09) (.08) South (.157) (.148) (.07) (.05) (.11) (.1) Higher poverty line Gr. Tunis (.05) (.03) (.08) (.05) (.06) (.05) Northeast (.14) (.11) (.15) (.13) (.12) (.1) Northwest (.24) (.34) (.29) (.35) (.21) (.22) Central West (.23) (.24) (.25) (.3) (.33) (.4) Central East (.16) (.12) (.15) (.1) (.15) (.11) South (.17) (.16) (.08) (.07) (.12) (.12) Note:Numbers in patentheses represent the relative contribution of each group to the overall poverty. 13

15 Table 10 show that North-West and Central-West are the most deprived regions. They contribute by more that 50% to overall poverty, regardless of poverty lines used. In other hand the poverty gap registers its lowest level in the Greater Tunis (FGT1=0.043) against for the Central East the second most advantaged region. Finally, we note that regional ranking according to our results are in line with the conclusions of monetary based studies, whereby the Northwest and the Central West regions are the most affected by this phenomenon. Such a conclusion is very important insofar as whatever approach is adopted to study poverty, the result is that the North West and the Central West should be considered as priority regions to fight against poverty. 4.3 Decomposing poverty by source An advantage of the non-monetary approach is that it can easily identify the contribution of the various attributes to household welfare. Poverty alleviation policies will thus be easier to implement as they will be based on more easily observable indicators. This paragraph deals with a marginal analysis which is a simulation of the potential effect of each welfare attributes on the reduction of poverty. Recall that WCI, for a household i, can be written as a weighted sum of the primary indicators of the household standard of living. Equation 2 can be re-written as: K WCI = s V k k 1 k =, (5) where V is the indicator k of household living condition and s its weight. k For the sake of simplicity, we assume that all the are binary variable equals to 1 when the household enjoy the indicator Vk and 0 otherwise. The variable sv k k = S k may be interpreted as the contribution of the k-th source to household welfare. This interpretation is similar to the one made within the monetary framework, where income, considered as a welfare measurement, is composed by several sources: V k k 14

16 labour income, family allowances, assets income, etc. We further assume that WCI is based on six sources of welfare: water, sanitary facilities (toilets), education, durable goods (gas cooker and refrigerator), communication (TV, radio and telephone) and housing conditions (nature of the floor, number of people per bedroom and the quality of the housing). Here we deal the questions: What sources of welfare are the most likely to further reduce the observed poverty level? What order of priority should be given to each source? A marginal analysis of a potential effect of each welfare source on the reduction of poverty will enable us to answer the above questions. This will involve measuring the effect of a k th welfare source on the reduction of poverty as being equal to the poverty index differential between the present distribution of attributes and a hypothetical distribution in which all the households are given the k th attribute. For example, in order to evaluate the potential effect of education attribute j on the reduction of poverty, we calculate the difference between the poverty index currently observed and that, under the hypothesis that all individuals are educated, maintaining constant the distribution of the remaining attributes. Technically, this reduces to increasing WCI by scores differential between not deprived and deprived for households not affording an attribute j. For example if we consider education attribute, this means increasing the value of the WCI by 1.16 (score of the educated modality - score of the uneducated modality). This analysis is applied to each source. Results are reported in table 11. Table 11: Marginal Effects of HC attributes during the year 2001 Attribute Head Count Marginal Effect Education Housing Toilets Water Durables Communication NB: The marginal effect is the difference between the HC in 2001 evaluated at.194 and the hypothetical HC of the second column. The first conclusions of this decomposition stress the need to emphasize the improvement of access to communication tools for future efforts to alleviate poverty. 15

17 This bears a considerably high marginal effect almost threefold as compared to all the other attributes. Access to safe water source has the lowest marginal effect. Ownership of durable goods, housing conditions, women s education, and health conditions have moderate marginal effects. 5. Conclusion It is widely accepted that poverty is a multidimensional phenomena. We construct a Welfare composite index that helps us to analyze poverty trends in a multidimensional perspective between 1988 and Using a relative poverty line we show a significant drop of poverty in the 1990s, with some signals of weakness in the middle of the decade, following years of drought (1993, 1994 and 1995). We conclude that poverty is essentially a rural phenomenon: over 70% of poor people live in rural areas. Such conclusions reflect the trends reported by the macroeconomic indicators, which give certain robustness to our approach. This study reveals that poverty have not decreased in the rural areas as much as previous studies indicate. As a result, if the Government hopes to further improve the wellbeing of Tunisian families, its efforts towards the alleviation of poverty must be more intensified in these rural areas. Despite the relative improvement observed during the last decade in the nutritional status of rural population, the housing, health conditions and ownership of some durable goods must be taken into consideration to guarantee a better standard of living to the rural populations and to reduce poverty. Interestingly the more deprived areas remain unchanged. Indeed, despite the reduction in poverty incidence in the Northwest and Southwest regions, they still occupy the last positions in the regional ranking in term of poverty. Finally, our decomposition of poverty by source reveals the necessity to further focus on the improvement of education, health conditions and access to a reliable water source in order to improve household well-being. 16

18 References Asselin, L-M ''Multidimensional Poverty: Composite Indicator of Multidimensional Poverty'', Institut de Mathématique Gauss: Lévis, Québec. Ayadi, M., M. S. Matoussi and M.P.Victoria-Feser ''Putting Robust Statistical Methods into Practice : Poverty Analysis in Tunisia", Swiss Journal of Economics and Statistics Vol.137, (3). Ayadi, M., R. Baccouche, M. Goaied and M. S. Matoussi ''Variation spatiale des prix et analyse de la demande des ménages en Tunisie", 7th World Congress of the Econometric Society, Tokyo, Japan. Ayadi, M. and M. S. Matoussi ''Urban-Rural Poverty Comparisons in Tunisia", 11th Annual Congress of the Eurorean Economic Association August 1996, Istanbul, Turkey. Ayadi, M., G. Boulila, M. H. Lahouel and P. Montigny ''Pro-Poor Growth in Tunisia", Mimeo Université de Tunis. Bibi, S "Les Dépenses publiques et le ciblage de la population pauvre en Tunisie'', Thèse Faculté des Sciences Économiques et de Gestion de Tunis. Bibi, S "Comparing Multidimensional Poverty Between Egypt and Tunisia'', 10th ERF`s Annual Conference, Marrakech, Morocco. Booysen, F. le R., Van der Berg, S., Du Rand, G., Von Maltitz, M. and Burger, R ''Poverty and Inequality Analysis for Seven African Countries, Using Asset Indices Constructed from DHS Data'', Document de travail, PEP-PMMA Davidson, R. and J.Y. Duclos "Statistical Inference for Stochastic Dominance and for the Measurement of Poverty and Inequality", Econometrica, Vol. 52 (3), pp Duclos, J.-Y., A. Araar Poverty and Equity: Measurement, Policy and Estimation with DAD, Université Laval. Filmer, D., Pritchett, L ''Estimating Wealth Effects Without Expenditure Data - or Tears: An Application to Educational Enrollments in States of India'', World Bank Policy Research Working Paper No. 1994, Washington D.C.: World Bank. Foster, J. E., J. Greer and E. Thorbecke "A Class of Decomposable Poverty Measures", Econometrica, Vol. 68 (6), pp Institut National de la Statistiques. 1990, "Enquête sur la consommation et le budget des ménages''. Kanbur, R., et al "L'évolution de notre manière d'envisager la pauvreté: analyse des interactions'', in Aux frontières de l'économie du développement: le futur en perspective eds. G. Meier and J. E. Stiglitz, Banque Mondiale, Washington D.C. Ki, J. B., S. Faye and B. Faye ''Pauvreté multidimensionnelle au Sénégal : une approche non monétaire par les besoins de base'', Working Paper , PEP Network. 17

19 Runciman, W. G ''Relative Deprivation and Social Justice: A Study of Attitudes to Social Inequality in twentieth-century England, Berkeley and Los Angeles'', University of California Press. Sahn, D. E. and D. C. Stifel "Poverty Comparisons Over Time and Across Countries in Africa'', World Development, 28, pp Sahn, D. E. and D. C. Stifel "Exploring Alternative Measures of Welfare in the Absence of Expenditure Data'', Review of Income and Wealth, 49 (4) pp Sen, A ''Commodities and Capabilities'', North-Holland, Amsterdam. Sen, A ''The Standard of Living. Cambridge'', Cambridge University Press. World Bank ''Republic of Tunisia, Poverty Update'', rapport en deux volumes. World Bank ''Republic of Tunisia, Poverty Update'', rapport en deux volumes. 18

20 APPENDICES Table 6b: Evolution of Monetary Poverty Incidence WB Lower line WB Upper line INS Lower line 1990 Global Urban Rural Global Urban Rural Global Urban Rural Source: WB (2003). Note: The lower poverty line =2$/day for urban areas and 1.8$/day for rural areas. The higher poverty line =2.9$/day and 2.3$/day for urban and rural areas, respectively. 19

21 Figure 2 - FOD: Difference Between the 1994 and1988 Headcounts Figure 3 - FOD: Difference Between the 2001 and1988 Headcounts Figure 4 - FOD: Difference Between the 2001 and 1994 Headcounts 20