Land reform, distribution of land and institutions in rural Ethiopia: Analysis of inequality with dirty data Bereket Kebede School of Development Studies, University of East Anglia (UEA) and Centre for the Study of African Economies (CSAE), Oxford University 12-13 October 2006 Addis Ababa Structure of presentation 1. Research questions 2. Data and general patterns in land distribution 3. Methodology 4. Empirical results 5. Link between current distribution and old tenures? 6. Conclusions 1. Research questions • Widespread acceptance that – rural land distribution is highly equitable in Ethiopia (for example compared to other African countries with private land ownership) – no strong link between current land distribution patterns and pre-reform tenures – seldom are current distribution patterns related to pre-reform distributions 1. Research questions (cont’d) “The low of level of [income] inequality is consistent with the overall picture of Ethiopia as a very poor country, with a low per capita income. In addition, the egalitarian land holding system might have contributed to a more equal income distribution in rural Ethiopia.” (FDRE, 2002; italics mine) 1. Research questions (cont’d) “This paper is an attempt to argue, in broad terms, that the social homogeneity in rural Ethiopia today, which is in large measure a consequence of the land system, is an inhibiting factor for agrarian development.” (D. Rahmato, 2005; italics mine) 1. Research questions (cont’d) • This study questions the two widely accepted views 1. Is rural land distribution in Ethiopia highly equitable as generally accepted? 2. Is there a link between current land distribution patterns and pre-land reform tenures? 1. Research questions (cont’d) • Main conclusions from this study: – rural land distribution in Ethiopia is not as equitable as generally accepted; for example, it is as inequitable as, if not more inequitable, as in some other African countries with private ownership and land markets – post-reform land distribution is probably partially explained by pre-reform land tenures 2. Data and general patterns in land distribution • Data from Ethiopian Rural Household Surveys (ERHS) • Covered 15 villages/sites (1500 households) in the main settled agriculture farming systems • Longitudinal survey: 1989, 1994-95 (3 rounds), 1997, 1999, 2004 • This study used data from 1995-1997: 20 years after land reform 2. Data and general patterns in land distribution (cont’d) • Distribution of what? –PA allocated land or cultivated land? –Total household or per capita or per adult holdings of households? • Focuses on PA allocated land and looks at household, per capita and per adult holdings Summary statistics PA allocated land (ha.) Total Per capita Per adult Mean 1.688 0.335 0.586 Median 1.000 0.188 0.313 1% 0.000 0.000 0.000 5% 0.000 0.000 0.000 95% 5.500 1.125 2 .000 99% 9.688 2.125 4 .000 For village mean and median figures see Table 3 0 .02 .04 .06 .08 .1 .12 .14 Histogram of paland for all sites (with bin width 0.1) 0 1 2 3 4 5 6 7 8 9 10 Land through PA allocation/approval 0 .1 .2 .3 .4 Histogram of pcpaland for all sites (with bin width 0.1) 0 1 2 3 4 5 6 7 PA allocated land per capita 8 9 10 0 .1 .2 .3 .4 Histogram of papaland for all sites (with bin width 0.1) 0 1 2 3 4 5 6 7 8 9 10 PA allocated land per active adult (15-54) geblen dinki debre berhan yetmen shumsheha sirbana godeti adele keke korodegaga turufe kechema imdibir aze deboa 0 .2 .4 .6 .8 0 .2 .4 .6 .8 0 .2 .4 .6 .8 haresaw 0 gara godo domaa 0 .2 .4 .6 .8 adado 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 Land through PA allocation/approval Graphs by site number 5 1 2 3 4 5 geblen dinki debre berhan yetmen shumsheha sirbana godeti adele keke korodegaga turufe kechema imdibir aze deboa 0 .2 .4 .6 .8 0 .2 .4 .6 .8 0 .2 .4 .6 .8 haresaw 0 gara godo domaa 0 .2 .4 .6 .8 adado 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 PA allocated land per capita Graphs by site number 4 5 1 2 3 4 5 geblen dinki debre berhan yetmen shumsheha sirbana godeti adele keke korodegaga turufe kechema imdibir aze deboa 0 .2 .4 .6 .8 0 .2 .4 .6 .8 0 .2 .4 .6 .8 haresaw 0 gara godo domaa 0 .2 .4 .6 .8 adado 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 PA allocated land per active adult (15-54) Graphs by site number 1 2 3 4 5 2. Data and general patterns in land distribution (cont’d) • Do PAs compensate with quality for size of land? – If size and quality of land are strongly negatively related, use of land holdings for analysing distribution would be problematic – Quality of land indicators are not significantly correlated to size of land after controlling for village level fixed effects (see Table 1) 3. Methodology • If data are completely unreliable, of course don’t use them – the survey data compares well with others including nationally representative samples • But inequality measures can be sensitive even for minor contamination • Dirty data may arise due to – errors in collection, coding, transcribing, etc., of data – under (over) reporting by respondents – but also from true outliers: influential observations (observations with high leverage) 3. Methodology (cont’d) Does G Lorenz dominates F? ●μ(Fε(z)) G Fε(z) (Cowell _& _ Victoria Feser ,2002) Elementary _ distributi on : H ( z ) ( x) ( x z ) indicator _ function : ( D) {1 _ if _ D _ is _ true,0 _ otherwise) Observed _ distributi on : F (z) ( x) [1 ]F ( x) H ( z ) ( x) Influence _ function : (z) IF ( z; T , F ) : T ( F ) 0 3. Methodology (cont’d) • Sensitivity of inequality measures to data contamination at high and low incomes vary – Generalised entropy index with α>1: very sensitive to high incomes – Generalised entropy index with α<0 and Atkinson index with ε>1:very sensitive to small incomes – Middle level values have small influence Values of IF for inequality measures for 10 highest, middle & smallest observations (Cowell & Flachaire, 2002) 3. Methodology (cont’d) • Re-estimating inequality measures using trimmed distributions to see how robust are results to outliers • In our case trimming from above is the more appropriate procedure because – zeros represent important phenomenon – landlessness (not the same as with zero incomes) – generally influence of lower values smaller than higher values 4. Empirical results • A cursory look at – Box and Whiskers graphs – Cumulative distributions – Non-parametric kernel densities makes one suspect that there could be some influential observations significantly affecting inequality measures 0 5 10 15 20 Box and whiskers graph for total PA land 0 2 4 6 8 Box and whiskers graph for PA land per capita 0 5 10 15 Box and whiskers graph for PA land per adult 0 5 10 15 20 Box and whiskers graphs for total PA land by village haresaw geblendinki debre berhan yetmen shumsheha sirbanaadele godeti korodegaga keke turufe kechema imdibir aze deboa adado gara godo domaa 0 5 10 15 20 Pen's parade of dwarfs and few giants for paland 0 .2 .4 .6 cupaland .8 1 0 2 4 6 8 Pen's parade of dwarfs and few giants for pcpaland 0 .2 .4 .6 cupcpaland .8 1 0 5 10 15 Pen's parade of dwarfs and few giants for papaland 0 .2 .4 .6 cupapaland .8 1 4. Empirical results (cont’d) • Even after treating influential observations inequality measures are high (Table 4) – Gini coefficients range from • 42%-56% for total • 44%-59% for per capita • 44%-60% for per adult • These are within the same range or higher than estimates for other African countries (see Table 5) 4. Empirical results (cont’d) • As expected, inequality within villages more sensitive to treatment (Table 6) – partly due to lower number of observations – but still significant levels of within village inequality • with 5% trimming from top, Gini ranges from 23%62% • Lower but still significant intra-village inequality 4. Empirical results (cont’d) • Looking at whole distributions for villages after treatment – Lorenz (relative) curves of some villages dominate others - first order stochastic dominance – Generalised Lorenz curves of some villages dominate others - second order stochastic dominance • There are significant regional inequalities in size as well as distribution of land 0 .2 .4 .6 .8 1 Lorenz curve for paland5a (5% trimmed from top) 0 .2 .4 .6 Cumulative population proportion paland5a[1] paland5a[3] paland5a[5] paland5a[7] paland5a[9] paland5a[11] paland5a[13] paland5a[15] .8 paland5a[2] paland5a[4] paland5a[6] paland5a[8] paland5a[10] paland5a[12] paland5a[14] 1 0 1 2 3 4 Generalised lorenz curve for paland5a (5% trimmed from top) 0 .2 .4 .6 Cumulative population proportion paland5a[1] paland5a[3] paland5a[5] paland5a[7] paland5a[9] paland5a[11] paland5a[13] paland5a[15] .8 paland5a[2] paland5a[4] paland5a[6] paland5a[8] paland5a[10] paland5a[12] paland5a[14] 1 5. Link between current distribution and old tenures? • Why should we expect a link between pre- and post-reform distributions? 1. Significant amounts of land held by households in 1995 were acquired from pre-reform periods – on the average, 36.6% of household land holdings in 1995 were acquired through pre-land reform inheritance and purchases! – for some villages pre-land reform inheritance and purchases accounted for the majority of land holding in 1995 • • • • Adele Keke & Adado:>90% Imdibir & Aze Deboa:>80% Gara Godo: 70% Haresaw: 41% 5. Link between current distribution and old tenures? (cont’d) 2. No strict national level guideline/rule for allocation of land by PAs • Family size considered as rough guide but can be interpreted in different ways • If PAs were left to their own device, wouldn’t the different land tenures within which farmers previously lived in influence the way they allocate land after reform? 5. Link between current distribution and old tenures? (cont’d) • For a proper analysis an economic history approach is required – detailed information on land tenure in each village before the land reform – how post-reform distribution was done over time in villages – examine if the pattern of distribution after reform systematically varied with pre-reform tenures 5. Link between current distribution and old tenures? (cont’d) • If the current distribution is at least partly influenced by pre-reform tenures, the current distribution may contain some information on previous tenures • Classified the villages into four – villages where the rist tenure (inheritance through ambilineal/cognatic descent) was dominant (group 1) – Non-rist tenures but with significant changes in traditional tenures (commercialisation, settlement, etc.) (group 2) – Non-rist tenures with traditional systems preserved (group 3) – a resettlement village established after the land reform (group 4) 5. Link between current distribution and old tenures? (cont’d) • Gini coefficients for the four groups (Table 7) – as expected the lowest Gini is in re-settlement village – villages located in previous more traditional tenures (both rist and non-rist) have higher inequality • If these results capture features of distributions before the reform → more individualised (privatised) tenures were more egalitarian 5. Link between current distribution and old tenures? (cont’d) • The Lorenz curve for group 4 (resettlement village) dominates all as expected • The Lorenz curve for group 2 (nontraditional) dominates that for group 1 (rist) & 3 (non-rist) – higher equality in those villages that have passed through more individualised tenures 0 .2 .4 .6 .8 1 Lorenz curve for paland5t (by old land tenure) 0 .2 .4 .6 Cumulative population proportion paland5t[1] paland5t[3] .8 paland5t[2] paland5t[4] 1 6. Conclusions • The land reform didn’t seem to have significantly solved the problem of inequality as usually assumed – Distribution of land is part of a wider dynamic distribution of social, economic and political power • PA allocation influenced by livestock ownership, demographics, etc. – Land reform created an institutional barrier to ruralrural migration and hence to regional inequality – The reform couldn’t avoid the development of significant landlessness