Development Economics and Public Policy WORKING PAPER SERIES

Development Economcs and Publc Polcy WORKING PAPER SERIES Paper No. 1 What s the mpact of noncontrbutory pensons on poverty? Estmates from Brazl and South Afrca Armando Barrentos IDPM, Unversty of Manchester August 2003 ISBN: 1 904143 42 3 Further detals: Publshed by: Insttute for Development Polcy and Management Unversty of Manchester, Harold Hankns Buldng, Precnct Centre, Oxford Road, Manchester M13 9QH, UK Tel: +44-161 275 2814 Fax: +44-161 273 8829 Emal: dpm@manchester.ac.uk Web: www.manchester.ac.uk/dpm

What s the mpact of non-contrbutory pensons on poverty? Estmates from Brazl and South Afrca Armando Barrentos * IDPM, Unversty of Manchester Abstract Ths paper consders the mpact of cash transfer programmes for the old n Brazl and South Afrca on poverty among households wth older people. Usng datasets collected specfcally for the purpose, the paper constructs condtonal and uncondtonal estmates of the poverty reducton capacty of these programmes. The paper fnds that non-contrbutory pensons have a measurable and sgnfcant mpact upon poverty reducton and poverty preventon n the two countres studed. August 2003 JEL: H55, I32, I38, J14 Keywords: poverty, pensons, South Afrca, Brazl Correspondence to: Armando Barrentos, Insttute for Development Polcy and Management, Unversty of Manchester, Harold Hankns Buldng, Oxford Road,Manchester M13 9QH Phone: +44 (0)161 275 2811 Fax: +44 (0)161 273 8829 E-mal: armando.barrentos@man.ac.uk * The author draws on materals from a research project on Non-contrbutory pensons and poverty preventon n developng countres? A comparatve study of South Afrca and Brazl, funded by the Department for Internatonal Development of the UK Government, and wth the partcpaton of Peter Lloyd-Sherlock and Helena Legdo-Qugley from the UK, Monca Ferrera and Valere Møller from South Afrca, João Saboa and Mara Luca Werneck Vanna from Brazl, and Mark Gorman and Amanda Heslop from HAI. Further nformaton s avalable at http://dpm.man.ac.uk/ncpps. 2

In developng countres, the ncdence of old age poverty s hgh (Delgado and Cardoso 2000c; Barrentos 2002; Barrentos, Gorman et al. 2003), but cash transfer programmes for the old and ther households are scarce (Barrentos and Lloyd- Sherlock 2003). Ths neglect renforces tself. As Kanbur notes n the context of AIDS, poverty measurements focus on the currently lvng wth the mplcaton that poverty measures decrease f the poorest ded as a result of poverty (Kanbur 2002). Gven the lnk between ncome and lfe expectancy, ths non-random attrton works to make old age poverty less vsble and old age support less urgent. There s an mplct resstance to establshng non-contrbutory penson programmes 1 n developng countres. 2 Ths s based on three wdely held generalsatons. Frstly, there s a wdespread, although weakenng, belef that households n low ncome countres provde effectve old age support, and that the ntroducton of formal old age protecton may smply crowd out exstng support (World Bank 1994). Secondly, old age poverty s usually gven a low prorty n development polcy, perhaps because of the assessment that there are other, more lkely, contrbutors to development and other, more pressng, needs (Treas and Logue 1976). Thrdly, the fnancal costs attached to the ntroducton of publc penson programmes appear to be beyond the capacty of poorer developng countres, especally as such programmes look ncreasngly dffcult to fnance n developed countres (Weaver 2003). The evdence emergng from the handful of developng countres wth substantal non-contrbutory publc penson programme suggests that, to an mportant extent, ths resstance to non-contrbutory penson programmes s msplaced (Wllmore 2001; Bertranou, Soloro et al. 2002; Barrentos and Lloyd-Sherlock 2003; Wllmore 2003). The two countres wth the largest non-contrbutory penson programmes are Brazl and South Afrca, and the fndngs from a number of studes ndcate that these programmes have large postve mpacts upon a wde range of varables. Noncontrbutory penson programmes reduce poverty among the elderly and ther households, enable nvestment n human and physcal captal wthn benefcary 1 In the paper non-contrbutory penson programmes refer to cash transfer programmes targeted prmarly on older people. They are non-contrbutory n so far as, n practce, payroll contrbutons to socal nsurance schemes do not consttute a pre-requste for enttlement. These nclude assstental penson programmes. 2 James artculates ths resstance: Probably for most developng countres, a unversal flat old age beneft s a luxury they cannot afford, nor s the best use of ther lmted publc resources (James 2001). 3

households, strengthen ntergeneratonal soldarty and transfers, nsure poorer rural communtes aganst the adverse effects of agrcultural reform, and encourage local economy actvty. 3 A small number of studes have focused on the mpact of non-contrbutory penson programmes n Brazl and South Afrca on poverty. Lund dentfed the poverty reducton and promoton effects of the socal penson n South Afrca, and has traced the expandng lterature (Lund 1993; Ardngton and Lund 1995; Lund 1999). Deaton and Case looked at ths ssue n the context of a 1993 natonwde household dataset and confrmed that the socal penson has sgnfcant effects on poverty. Ther analyss showed that around 35% of Afrcans survved on less than US$1 a day, and that ths fgure would be 40% f the penson ncomes were removed and there was no off-settng change n pre-penson ncomes (Case and Deaton 1998, p.132). Studes usng more recent data have confrmed the poverty reducton effects of the socal penson, albet less drectly through estmatng the correlaton exstng between the presence of pensoners n a households and measures of ncome and poverty (Lebbrandt 2001). In Brazl, researchers at IPEA have nvestgated the mpact of the rural old age penson and have concluded that the programme has sgnfcant effects on poverty (Delgado and Cardoso 2000b; Delgado and Cardoso 2000c; Schwarzer 2000; Schwarzer and Querno 2002). Delgado and Cardoso compared households wth a penson benefcary aganst households wthout one, and found that the ncdence of poor households was hgher among the latter. The proporton of benefcary households who were poor was 38.1 percent n the Northeast regon and 14.3 percent n the South, whereas among non-benefcary households poverty ncdence was 51.5 percent and 18.9 percent respectvely (Delgado and Cardoso 2000a). Ths paper ams to make a contrbuton to ths lterature by constructng relable estmates of the mpact of non-contrbutory penson programmes on poverty. The 3 See nter ala (Ardngton and Lund 1995; Lund 1995; Møller and Sotshangaye 1996; Case and Deaton 1998; Ferrera 1999; Fultz and Pers 1999; Lund 1999; Saad 1999; Sagner and Mtat 1999; Carvalho 2000b, a; Case and Wlson 2000; Delgado and Cardoso 2000c, a; Duflo 2000; Schwarzer 2000; Case 2001; Devereux 2001a, b; van der Berg 2001; Bertranou, Soloro et al. 2002; Brumer 2002; Camarano and Pasnato 2002; Commtte of Inqury nto a Comprehensve System of Socal Securty for South Afrca 2002; Jensen 2002; Lund 2002; Schwarzer and Querno 2002; Barrentos 2003b, a; Barrentos and Lloyd-Sherlock 2003; Deden 2003; Werneck Vanna 2003). 4

analyss makes use of comparable datasets from Brazl and South Afrca collected n October-December 2002 from a sample of households wth older people. The datasets provde detaled nformaton on non-contrbutory pensoner households n the two countres. They also enable a cross-country comparson as an addtonal check on the country specfc results. The paper s organsed as follows: Secton One provdes basc nformaton on the noncontrbutory penson programmes n Brazl and South Afrca. Secton Two descrbes the poverty ndcators and data. Secton Three computes and dscusses poverty measures wth and wthout penson ncome. Secton Four presents estmates of the mpact of the penson programme on the probablty of beng poor n a multvarate settng. A fnal secton dscusses the man conclusons. 1. NON-CONTRIBUTORY PENSION PROGRAMMES IN BRAZIL AND SOUTH AFRICA Ths secton provdes a bref descrpton of the non-contrbutory penson programmes. South Afrca A penson beneft of 640 Rand s pad to men aged 65 and over and women aged 60 and over. Beneft enttlements are means tested on the ncome of the ndvdual benefcary, and hs/her partner f marred, but not on the ncome of other household members. The programme began n the early 1900s as a means of provdng a basc ncome n retrement for whtes who lacked an occupatonal penson (van der Berg 2001). Subsequently, the programme was extended to Coloureds (1928) and Afrcans (1944), but wth dfferent condtons for enttlement and beneft levels. In the 1980s and 1990s, there was a gradual move towards party n benefts, whch was completed n 1996 wth the ntroducton of non-dscrmnatory regulatons. Afrcans are now the man benefcares. In 1993, there were just above 1.5 mllon old age penson beng pad, wth 1.2 mllon beng pad to Afrcans (van der Berg 2001). The most recent estmate s that there are 1.9 mllon benefcares of the state old age penson (Commtte of Inqury nto a Comprehensve System of Socal Securty for South Afrca 2002). The programme s reasonably well admnstered, and reaches the poorer rural areas. The programme s funded through general taxaton, absorbng 60 percent 5

of socal securty expendture, and 1.4 percent of GDP. It s wdely acknowledged that the old age penson produces a sgnfcant redstrbuton of ncome n the country (Commtte of Inqury nto a Comprehensve System of Socal Securty for South Afrca 2002). Brazl Lmted provson of non-contrbutory pensons for workers n the rural sector dates back to 1963, but enttlements were restrcted to the very old. The scheme was gradually upgraded durng the 1970s, n response to moblsatons of rural workers and pressures for land reform (Brumer 2002). The 1988 Consttuton recognsed the rght to socal protecton for workers n the rural sector, and especally for those n nformal employment. Ths led to a range of reforms beng mplemented from 1991 to establsh a new rural old age penson, referred to as Prêvdenca Rural (PR) below. As part of the reforms, the age of penson elgblty was reduced from 65 years of age to 60 for men and 55 for women. Enttlement to old age, dsablty and survvor pensons was extended to workers n subsstence actvtes n agrculture, fshng and mnng, and to those n nformal employment. Whereas pror to 1991 only heads of household were enttled to a penson, the reforms extended enttlement to all qualfyng workers, thus expandng coverage to female rural workers who were not heads of household. The value of the penson benefts was rased from 0.5 to 1 mnmum wage (200 Reas n December 2002). A key aspect of the programme s that access to penson enttlements does not requre earnngs or nactvty tests. In urban areas, provson of old age assstance pensons s much less developed. A socal assstance penson Renda Mensual Vtalíca (RMV) was ntroduced n 1974 payng a flat rate beneft of one half the mnmum wage to older or dsabled people who could not provde for themselves. To be enttled to the RMV, ndvduals needed to be 70 years of age or over and have at least 12 months of contrbutons to socal nsurance. After the 1988 Consttuton, a new socal assstance penson, the Benefco de Prestação Contnuada (BPC) was ntroduced n 1993, payng one mnmum wage to dsabled or elderly people aged 67 and over lvng n urban or rural areas wth per capta household ncome below a quarter of the mnmum wage. Enttlement, ncludng the means test, s revewed every two years. The condtons for enttlement under the BPC are tougher than under the PR. In December 2000, there were 4.6 6

mllon benefcares of old age pensons under the PR programme, 0.3 mllon old age RMV benefcares, and 0.4 mllon old age BPC benefcares. 4 The fscal cost of the PR programme, ncludng dsablty pensons, has been estmated at 1 percent of GDP (Schwarzer and Querno 2002), whle the cost of the RMV and BPC programmes should be around 0.2 percent of GDP gven the smaller number of benefcares. Overall, a reasonable estmate of the cost of old age pensons under the three noncontrbutory penson programmes for older people n Brazl s 1 percent of GDP. 2. POVERTY INDICATORS AND DATA Ths secton descrbes the poverty measures, poverty lne, and data used n the paper. Poverty measures The poverty headcount and the poverty gap are wdely used poverty measures. These are specal cases of the Foster, Greer, and Thorbecke class of poverty measures (Foster, Greer et al. 1984), P α q 1 z y = n = 1 z α forα 0 (1) where n s the total number of ndvduals, q s the number of poor ndvduals (havng a standard of lvng ndcator no greater than z), z s the poverty lne, y s the standard of lvng ndcator of ndvdual, and α s an averson to poverty parameter. Where α=0, P 0 provdes a poverty headcount measure. Where α=1, P 1 provdes a poverty gap measure, nterpreted as the mprovement n standard of lvng requred to brng the poor to the level of the poverty lne. These two poverty measures wll be used n the emprcal work below. A number of alternatve standards of lvng ndcators are used n the lterature, but the paper focuses on household ncome, because the objectve s to solate the mpact of penson ncome on poverty. 5 More precsely, the standard of lvng ndcator used s adult equvalent per capta household ncome. To account for dfferences n the 4 These fgures exclude benefcares of dsablty pensons under the three programmes. 5 A companon paper consders multdmensonal deprvaton ndcators (Barrentos 2003a). 7

demographc composton of households, and economes of sze wthn the households, total household ncome s transformed usng the followng formula: = J y j 1 j y = (2) θ 1+ [( A 1) + αk] where y s adult equvalent per capta household ncome of ndvdual, y j s household ncome of ndvdual from source j (j = 1,2,,J), A s the number of adults, and K the number of chldren aged 15 and below. The parameter α s a measure of the cost of chldren relatve to adults, whle θ reflects economes of sze n the household. In ths paper, the values chosen for these parameters are α = 0.5 and θ = 0.75, n lne wth the wdely used OECD equvalence scale. 6 Poverty lnes Selectng a poverty lne s a complex ssue at the best of tmes. In the context of ths paper, t s mportant to adopt a poverty lne whch facltates comparatve study. A useful strategy s to fx the poverty lne at the level of the penson beneft n each country, and, as wll be shown below, ths has a number of advantages. In South Afrca, the poverty lne s set at 640 Rand, and n Brazl at 200 Reas, the maxmum penson beneft levels n force at the tme of the survey data collecton n the two countres. There are good reasons for adoptng the level of the penson beneft as the poverty lne, as t represents a socally accepted mnmum lvng standard n both countres. In Brazl, ths s n lne wth the fact that the non-contrbutory penson beneft s equvalent, by law, to one mnmum wage. Ths mnmum lvng standard also apples to other benefts n the socal nsurance scheme, such as dsablty and survvor pensons and mnmum guaranteed pensons for contrbutory programmes. In South Afrca the stuaton s more complex because there s no offcal poverty lne or general mnmum wage, and other socal assstance benefts vary n value. It would be 6 There s a large lterature on ths, but see Klasen (2000) for a general dscusson n the context of South Afrca, and Woolard and Lebbrandt for a senstvty analyss of dfferent equvalence scales (2001). The latter fnd that the senstvty of poverty estmates to dfferent values for the parameters α and θ n the context of South Afrca s low, especally around the values selected n ths paper. Usng 8

useful to compare the poverty and ndgence lnes selected here wth poverty lnes dentfed for South Afrca usng alternatve methodologes. May, Woolard and Lebbrandt compared alternatve poverty lnes n South Afrca, a dstrbutonally based 40 th percentle of populaton mnmum, the commonly used US$2 a day nternatonal poverty lne, and the value of a calorfc ntake of 8,500 KJ per day (May, Woolard et al. 2000; Woolard and Lebbrandt 2001). Updatng ther fgures to November 2002, these three poverty lnes are 586.2 Rand, 408 Rand, and 290 Rand respectvely. The poverty adopted n ths paper s slghtly above the 40 th percentle cut-off pont, 7 and t s one and a half tmes the US$ 2 a day threshold. In Latn Amerca, t s common to defne an ndgence lne, equvalent to the value of a mnmum basket of consumpton needed for physcal survval (Boltvnck n.d.). As t happens, the value of the ndgence level for Brazl was one half of the mnmum wage, or 100 Reas at the tme of the data collecton. Ths ndgence lne wll alsoe b used n the emprcal work below, calculated n South Afrca as one half of the penson beneft level, or 320 Rand. Aganst the alternatve poverty lnes noted n the prevous paragraph, the ndgence lne for South Afrca s very close to the mnmum calorfc ntake poverty lne. What about the comparablty of poverty lnes for Brazl and South Afrca? At the offcal exchange rate for December 2000, the maxmum non-contrbutory penson beneft was US$75.6 for South Afrca and US$ 92.5. At nternatonal PPP US$, the respectve values are 237.2 and 208.3. 8 It s remarkable that the values of the noncontrbutory penson beneft n the two countres are so close to one another, and ths provdes an addtonal justfcaton for usng the value of the penson beneft as the poverty lne and one half of the beneft as the ndgence lne. adult equvalent standard of lvng ndcators n very mportant n the study of old age poverty (Deaton and Paxson 1997; Barrentos, Gorman et al. 2003). 7 The 40 th percentle cut-off pont refers to households ranked by adult equvalent expendture (Woolard and Lebbrandt 2001). 8 The offcal exchange rate and PPP converson rates were taken from the World Development Indcators 2002 (World Bank 2002). Deaton provdes a good summary of the dangers of relyng on cross-country comparsons of poverty based on PPP-set poverty lnes (Deaton 2001). In the context of the paper, t s suffcent to note the poverty lnes are roughly comparable. 9

Data The data used n ths paper comes from household surveys mplemented n 2002 n Brazl and South Afrca. The surveys targeted households wth older people and ncluded a household component, mplemented on the most knowledgeable person n the household, and an addtonal supplement for household members 55 years of age and older, mplemented on the older persons themselves. In South Afrca, data was collected on 1111 households n Cape Town and the Eastern Cape. The sample of urban households from Cape Town ncludes 324 Black households and 413 Coloured households. The rural sample ncludes 374 Black households. In Brazl, data was collected on 1006 households, 250 from the Muncpalty of Ro de Janero, 255 from the Metropoltan Regon n Ro de Janero, 269 from the urban areas under the Muncpalty of Ilhéus, and 232 from rural areas n the same Muncpalty. The sample was constructed by selectng census cells (Enumerator Areas n South Afrca and Setores Censtáros n Brazl, on the bass of soco-economc - and n South Afrca ethnc composton - ndcators. Wthn the sampled cells, households were vsted randomly, and ntervewed f they had an old age member. A common survey nstrument was used n the two countres wth the am of enablng full comparablty. The questonnare covers household locaton, housng, and demographc nformaton; employment, ncome and expendture, health care, and self-reported well beng. The older person supplement covers pensons and other state benefts, health status, socal partcpaton and transfers. 9 2. POVERTY IN THE ABSENCE OF NON-CONTRIBUTORY PENSIONS As a frst step n dentfyng the mpact of penson programmes on poverty, we could compare poverty measures computed usng full household ncome aganst the same poverty measures computed excludng non-contrbutory penson ncome. 10 There are number of drawbacks assocated wth ths strategy, n as much as the household response to the absence of non-contrbutory penson s unobserved. It may well be the case that other ncome sources could be used to make up ths shortfall, for example 9 The survey nstruments and feld reports are avalable at http://dpm.man.ac.uk/ncpps. 10 Ths nvolves comparng P y α 1 = n q = 1 z y z α wth P y ncy α 1 = n q = 1 z y ncy z α 10

through ncreased labour supply or prvate transfers, but we are not able to account for ths. Nonetheless, ths rough and ready comparson provdes some ndcaton of the sgnfcance of non-contrbutory penson ncome for the sampled households. Table 1 below summarses the poverty ncdence and gap wth and wthout noncontrbutory penson ncome for the two samples. Full ncome nvolves ncomes receved by the household from all sources, ncludng ncome from prvate transfers and the value of own produce consumed wthn the household. Table 1. Poverty headcount and gap measures wth and wthout penson ncome (usng adult equvalent household ncome per capta) wth n-c penson Brazl (n=3523) wthout n-c penson South Afrca (n=5560) wth n-c wthout n-c penson penson poverty headcount 61.0 65.3 41.3 43.1 poverty gap as % of poverty lne 21.9 29.7 18.6 33.7 ndgence headcount 21.0 30.6 17.0 19.3 ndgence gap as % of ndgence lne 4.3 12.0 7.3 9.0 Own calculatons usng household datasets for Brazl and South Afrca The fgures n the Table show that non-contrbutory penson ncome reduces, other thngs beng equal, both the poverty headcount and gap n the two countres nvolved. Non-contrbutory penson ncome reduces the poverty headcount among members of households wth older people 4.2 percentage ponts n Brazl and 2.8 percentage ponts n South Afrca. It also reduces the ndgence headcount by 9.6 percentage ponts n Brazl, and 2.3 percentage ponts n South Afrca. The mpact on headcount poverty s markedly lower for South Afrca than for Brazl. Ths can be explaned to a large extent by the poorer dversfcaton of ncome sources and the ncdence of larger households n South Afrca, wth the mplcaton that non-contrbutory penson ncome has to be spread more wdely. 11

Non-contrbutory penson ncome has a stronger mpact upon the poverty gap n both countres. In Brazl, the absence of non-contrbutory penson ncome would make the poverty gap among older households, other thngs beng equal, 40 percent larger, and the ndgence gap almost 3 tmes larger. For South Afrca, the absence of noncontrbutory penson ncome would ncrease the poverty gap, other thngs beng equal, by 81 percent, but would have a much smaller effect upon the ndgence gap, ncreasng t by just over one ffth. These are estmates of the mpact of non-contrbutory penson ncome at the mean of the sample, but t s also mportant to dentfy the mpact of penson ncome at other ponts n the dstrbuton. Table 2 below shows the poverty and ndgence headcounts dsaggregated by quntle of adult equvalent household ncome per capta. Table 2. Poverty headcount wth and wthout penson ncome, by quntles of adult equvalent household ncome per capta Brazl (n=3523) wth n-c wthout n-c penson penson South Afrca (n=5560) wth n-c wthout n-c penson penson per capta household poverty headcount ncome quntles 1 100.0 100.0 100.0 100.0 2 100.0 100.0 93.5 94.2 3 86.9 88.8 13.0 20.7 4 18.4 34.8 0.0 0.5 5 0.0 3.1 0.0 0.0 ndgence headcount 1 87.8 87.8 83.3 84.6 2 17.1 37.2 2.0 9.1 3 0.0 17.5 0.0 3.4 4 0.0 10.1 0.0 0.0 5 0.0 0.3 0.0 0.0 Own calculatons usng household datasets for Brazl and South Afrca The fgures n Table 2 show that the mpact of non-contrbutory penson ncome on poverty headcount s restrcted to some quntles. For Brazl, the mpact of gnorng non-contrbutory penson ncome s stronger for the fourth quntle, where the probablty of poverty would almost double. In South Afrca, the mpact s stronger for the thrd quntle. Elsewhere, the mpact s margnal n both countres. In terms of 12

the ndgence headcount, the mpact of non-contrbutory penson ncome s stronger, and dstrbuted more wdely. Table 3 shows poverty and ndgence gap measures wth and wthout takng account of non-contrbutory penson ncome. The mpact of the penson programmes on the extent of poverty s stronger. In the case of Brazl, and focusng on the poverty gap as a proporton of the poverty lne, non-contrbutory penson ncome has effects on the poverty gap across the full dstrbuton, although these are stronger n the mddle quntles. The same apples to the ndgence gap calculated as a proporton of the ndgence lne. Ths rses for the four lower quntles n Brazl. For South Afrca, the mpact of non-contrbutory penson ncome s observable for the lower three quntles only. Table 3. Poverty and Indgence gaps wth and wthout penson ncome, by quntles of adult equvalent household ncome per capta Brazl (n=3523) wth n-c wthout n-c penson penson South Afrca (n=5560) wth n-c wthout n-c penson penson per capta household poverty gap as % of poverty lne ncome quntles 1 60.0 61.1 67.3 69.3 2 38.2 49.4 21.5 29.0 3 10.6 24.5 0.9 4.5 4 0 9.6 0 0 5 0 0.6 0 0 ndgence gap as % of ndgence lne 1 21.6 31.6 36.7 40.6 2 0.2 14.8 0 3.3 3 0 9.9 0 1.4 4 0 3.5 0 0 5 0 0 0 0 Own calculatons usng household datasets for Brazl and South To recap the man conclusons from ths secton, a comparson of poverty measures calculated on full household ncome, and household ncome excludng noncontrbutory penson ncome ndcate that these programmes have an mportant mpact on poverty among households wth older people n Brazl and South Afrca. 13

Ther mpact on poverty and ndgence gaps s stronger than that on poverty and ndgence headcount measures. In South Afrca, n partcular, the presence of extended households together wth poorer ncome dversfcaton means that the old age penson s not suffcent, by tself to lft households our of poverty or ndgence. In Brazl, the mpact of non-contrbutory penson programmes s more wdely felt across the dstrbuton of households, and ther mpact upon the extent of poverty s strong. 3. NON-CONTRIBUTORY PENSIONS AND THE PROBABILITY OF BEING POOR As a second step n dentfyng the mpact of non-contrbutory penson ncome on poverty, ths secton reports on probt regressons of the determnants of the probablty that a household member s poor. A multvarate settng enables the dentfcaton of the mpact of havng a non-contrbutory penson benefcary on the probablty that household members are poor, havng controlled for the nfluence of household and ndvdual characterstcs, as well as other ncome sources. The model to be estmated s of the poverty profle type. In poverty profles, t s postulated that the rato of ndvdual or household ncome to the poverty lne s a functon of a vector X of ndvdual and household characterstcs, more specfcally y /z = β X + e, where e s ~N[0,σ 2 ]. Ths can be estmated by regressng the ndvdual or household poverty measure y /z on a range of ndvdual, household and envronmental characterstcs. As for each ndvdual the poverty headcount measure P 0 s a bnary ndcator wth a value of 1 f y /z 1 and 0 otherwse, the probablty that an ndvdual wll be found to be poor s gven by: Pr ob = Pr[ P0 = 1 X ] = Pr[ ε < 1 β ' X ] = Φ[1 β ' X ] (3) where Φ [.] s the cumulatve densty functon of the standard normal. Separatng out the varables of nterest, the model to be estmated s: Pr[ P 0 = 1 X ] = α + β 'X + λ' NCP + θ ' IS + ε (4) 14

where X s a vector of household and ndvdual characterstcs, NCP s a dummy varable ndcatng whether someone n the household receves a non-contrbutory penson beneft, IS s a vector of dummy varables ndcatng whether the household receves a range of ncome sources other than from non-contrbutory penson. The parameter of nterest s λ, provdng an estmate of the mpact of non-contrbutory penson recept on the probablty of poverty. A number of ssues assocated wth ths specfcaton have been examned n the relevant lterature (Glewwe 1991; Damond, Smon et al. 1999; Deden 2003). In a sense, ncome sources cannot be taken as exogenous to the ndvdual and household. Some studes for South Afrca have consdered whether recept of the penson encourages other relatves to co-resde wth pensoners, as a means of avodng poverty or desttuton, or as an nsurance aganst frequent unemployment spells or varable ncome (Edmonds, Mammen et al. 2001). There are other complcatons that arse from the juxtaposton of ncome source varables and varables ndcatng characterstcs whch enable or dsable access to ncome sources, such as labour earnngs and schoolng. And, to an mportant extent, ncome levels may help determne the ncome sources of households, as t s the case where a means test s appled for the determnaton of non-contrbutory penson enttlement. It can be argued that these ssues are less mportant n the context of older people and ther households. Especally as regards older people themselves, t can be argued that whatever choces may have led to the presence of current ncome sources, these can be taken to be predetermned and rreversble. The labour supply effects of penson recepts, for example, are lkely to be smaller for older people than for mddle age adults by an order of magntude. Nonetheless, ssues of endogenety reman mportant for other members of the household. Regardng the feedback effects of ncome levels on ncome sources, the model estmated below uses dummy varables ndcatng the presence of ncome sources rather than the amounts receved. Whle exercsng consderable care, t s possble to nterpret the parameters estmated by the model as reflectng a statstcal assocaton between the dfferent sources of ncome on the one hand and poverty on the other, controllng for a number of household and ndvdual characterstcs. 15

The varables ncluded n the model reflect ndvdual characterstcs, such as age, martal and work status, whch have been found to determne poverty n smlar studes (May 2000; May, Woolard et al. 2000; Bhorat, Lebbrandt et al. 2001; Lebbrandt 2001; Woolard and Lebbrandt 2001; Woolard and Klasen 2003). The same goes for household characterstcs, such as the number of household members, whether the household s located n a rural area, the number of rooms and durables, and the presence of extraordnary expenses and recent death of household members. In addton, ncome sources have been grouped n meanngful categores, and dummy varables ndcate ther presence wthn the household, ncludng non-contrbutory penson recept. The full lst of varable defntons can be found n Appendx One. Tables 4 and 5 report on the estmates for the Brazl and South Afrca samples respectvely. The reported parameters are margnal effects computed at the mean of the regressors. Broadly, the probt regresson results are n lne wth expectatons. Members of larger households n rural areas have a hgher probablty of beng poor. Household members n work who belong to households wth more durables are less lkely to be poor. Households wth labour earnngs are emphatcally less lkely to be poor. 11 Prvate transfers have a sgnfcant negatve mpact upon poverty probabltes n the Brazl and South Afrca samples. Some parameter estmates have dfferent sgns n the two countres. Members of households recevng state transfers other than pensons are more lkely to be poor n the Brazl sample, but less lkely to be poor n the South Afrca sample. Ths can be explaned n the context of the dfferent socal protecton systems n these countres. As non-contrbutory pensons are almost unversal n South Afrca, other state transfers complement penson ncome, but n Brazl, and especally for urban households unable to access non-contrbutory pensons, other state transfers can the only lfelne and could be an ndcator of acute poverty. 11 See (Woolard and Klasen 2003) 16

Table 4. Probt regresson results for the Brazl sample Dependent varable s poverty ndcator (poor = 1; non-poor = 0) Varable Margnal s.e. t-stat Mean of X Constant 1.300 *.07 18.1 Wdow -.020.03 -.63.12 Age -.002.004 -.58 43.19 Age 2.00002.00009.28 2540.02 Age 3 -.0000002.0000006 -.37 168044.85 Number n household.075 *.005 14.80 4.76 Rural.221 *.02 10.75.26 Number of rooms -.056 *.008-6.66 4.28 Durables -.118 *.007-15.15 5.58 Work -.148 *.02-5.11.24 Expendture shock -.050.03-1.55.11 Death n the household -.081 ***.04-1.77.05 N-c Penson -.182 *.03-5.72.23 Publc transfers.058 **.01 1.95.14 Prvate transfers -.068 *.02-2.60.19 Labour earnngs -.322 *.02-15.63.65 Prvate Penson.096.09 1.05.006 Publc penson -.311 *.02-14.80.66 Rent -.359 *.05-6.70.03 NGO transfers.031.11.27.004 *sgnfcant at 1%; ** sgnfcant at 5%; *** sgnfcant at 10% N=3253; P1 = 0.389; P0 = 0.610; LogL = -1424.02 ; LogL(0) = -2174.68 Estrella = 0.432; Efron = 0.400 ; McFadden = 0.345; Ben/Lerman = 0.712; Cramer = 0.396; Veall/Zm = 0.551 Table 5. Probt regresson results for the South Afrca sample Dependent varable s poverty ndcator (poor = 1; non-poor = 0) Varable Margnal s.e. t-stat Mean of X Constant -.186 *.036-5.08 Wdow.022.064.35.01 Age.004.0025 1.58 31.96 Age 2 -.00009.00007-1.29 1555.51 Age 3.0000006.0000006 1.06 90331.24 Number n household.044 *.002 17.93 6.69 Rural -.0007.021 -.035.36 Number of rooms.013 *.004 3.12 3.88 Durables -.017 *.004-4.33 3.86 Work -.042 ***.023-1.82.13 Expendture shock.095 *.019 4.97.19 Death n the household.035 **.018 1.96.23 N-c Penson -.125 *.016-7.53.63 Publc transfers -.067 *.016-4.15.31 Prvate transfers -.074 *.019-3.81.13 Labour earnngs -.249 *.016-14.87.48 Prvate Penson -.160 *.024-6.54.07 Rent -.142.091-1.57.004 *sgnfcant at 1%; ** sgnfcant at 5%; *** sgnfcant at 10% N=5560; P1 = 0.561; P0 = 0.413 ; LogL = -3292.6 ; LogL(0) = -3770.6 Estrella = 0.167; Efron = 0.164 ; McFadden = 0.126; Ben/Lerman = 0.539; Cramer = 0.162; Veall/Zm = 0.254 17

As regards the man parameter of nterest, lvng n a household wth a noncontrbutory penson recpent reduces the probablty of a household member fndng herself n poverty. Ths s true for both countres. For the Brazl sample, the margnal effect of non-contrbutory penson recept wthn the household s to reduce the probablty of poverty by 18 percent for ts members. For the South Afrca sample, non-contrbutory penson recept wthn the household reduces the probablty of a household member beng n poverty by 12.5 percent. In sum, the non-contrbutory penson programmes n these two countres have a strong and sgnfcant mpact upon the probablty of poverty, havng controlled for other ncome sources and ndvdual and household characterstcs. These results confrm the uncondtonal mpact estmates reported n the prevous secton. CONCLUSIONS It s mportant to keep n mnd the potental bases n the data used n the paper, as the samplng s regonally bound and targets households n poorer areas. Takng these on board, the man concluson from ths paper s that, n Brazl and South Afrca, noncontrbutory penson programmes have a sgnfcant mpact on ncome poverty among households wth older people. For the sample of households wth older people n Brazl and South Afrca, estmates of the uncondtonal mpact of non-contrbutory penson programmes ndcate that: Poverty headcount would be 4.2 percent hgher for the Brazl sample and 2.8 percent hgher for the South Afrca sample f penson ncome s removed and there are no off-settng changes. Indgence headcount would rse by around 9.6 percent n the Brazl sample, and 2.3 percent n the South Afrca sample n smlar crcumstances. The mpact of non-contrbutory penson programmes s stronger on poverty and ndgence gap measures. The poverty gap would be 40 percent larger for the Brazl sample and 81 percent larger for the South Afrca sample f penson ncome s removed and there are no off-settng changes. The ndgence gap would be 18

almost three tmes larger n the Brazl sample, and over ffth larger n the South Afrca sample n smlar crcumstances. Dsaggregatng ths uncondtonal estmate by quntles of ncome shows that the mpact of non-contrbutory pensons on the poverty gap can be observed across the sampled populaton, but that t s stronger on the lower ncome quntles. When consderng the mpact of non-contrbutory penson ncome on the ndgence gap, ths s focused on the lowest two quntles can be observed. Ths confrms the vtal contrbuton of these programmes to the standards of lvng of the poorest, and suggests these have a key poverty protecton role among households wth older persons. A multvarate analyss of the mpact of non-contrbutory penson programmes on the probablty of a members of a household beng poor fnds that belongng to a household wth a penson recpent reduces ths probablty by 18 percent for the Brazl sample and 12.5 percent for the South Afrcan sample. It s hoped that the evdence presented n ths paper adds to the wder case for the desrablty and feasblty of non-contrbutory old age penson programmes n developng countres. Non-contrbutory penson programmes could make a sgnfcant contrbuton to poverty reducton and preventon n these countres, and are an ndspensable nstrument n achevng the MDGs. 19

Appendx One. Varable defntons Varable AEHY WIDOW AGE AGE 2 AGE 3 NUMBER OF HOUSEHOLD MEMBERS RURAL NUMBER OF ROOMS DURABLES WORK EXPENDITURE SHOCKS DEATH IN THE HOUSEHOLD N-C PENSION PUBLIC PENSION PRIVATE PENSION PUBLIC TRANSFERS PRIVATE TRANSFERS NGO TRANSFERS LABOUR EARNINGS RENT Varable defnton Adult equvalent per capta household ncome, constructed by addng up household ncome from all sources, and ncludng prvate transfers, and value of own produce Dummy ndcatng wdowhood (yes=1; no=0) Age n completed years Age x Age Aged x Age x Age Number of household members Dummy ndcatng whether household lves n rural area (yes=1; no=0) Number of rooms ncludng ktchen but excludng bathroom Number of durable goods n workng order (max=10; ndvdual tems are: telephone, gas or electrc stove, electrcty, televson, rado or stereo, refrgerator/deep freeze, sewng machne, car, bcycle, motorcycle) Whether person s workng or lookng for work Whether the household has had extraordnary expenses n the last 12 months (yes=1; no=0) Whether a member of the household ded n the last 2 years (yes=1; no=0) Dummy ndcatng whether a member of the household receves a non-contrbutory penson (yes=1; no=0) Dummy ndcatng whether the household receves ncome from publc pensons (yes=1; no=0) Dummy ndcatng whether the household receves a prvate penson (yes=1; no=0) Dummy ndcatng whether the household receves state benefts other than non-contrbutory pensons (yes=1; no=0) Dummy ndcatng whether the household receves ncome from a person outsde the household, ncludng remttances and goods, valued by respondent (yes=1; no=0) Dummy ndcatng whether the household receves transfers from Church or NGOs (yes=1; no=0) Dummy ndcatng whether the household receves labour ncome (yes=1; no=0) Dummy ndcatng whether the household receves ncome from savngs, property rental, or lodgers (yes=1; no=0) 20

REFERENCES Ardngton, E. and F. Lund (1995), Pensons and development: socal securty as complementary to programmes of reconstructon and development, Development Southern Afrca, 12, (4), 557-577. Barrentos, A. (2002), Old age, poverty, and socal nvestment, Journal of Internatonal Development, 14, (8), 1133-1142. Barrentos, A. (2003a), Non-contrbutory pensons and the well-beng of older people: Evdence on Multdmensonal Deprvaton from Brazl and South Afrca, mmeo, Manchester: IDPM, Unversty of Manchester. Barrentos, A. (2003b), Pensons and development n the South, Geneva Papers on Rsk and Insurance, 28, (4). Barrentos, A.; M. Gorman and A. Heslop (2003), Old age poverty n developng countres: contrbutons and dependence n later lfe, World Development, 3, (3), 555-570. Barrentos, A. and P. Lloyd-Sherlock (2003), Non-contrbutory pensons and socal protecton, mmeo, Geneva: ILO. Bertranou, F.; C. Soloro and W. van Gnneken (2002), Pensones no-contrbutvas y asstencales. Argentna, Brazl, Chle, Costa Rca y Uruguay, Santago: ILO. Bhorat, H.; M. Lebbrandt; M. Mazya; S. van der Berg and I. Woolard (2001), Fghtng poverty. Labour markets and nequalty n South Afrca, Cape Town: UCT Press. Boltvnck, J. (n.d.), Poverty n Latn Amerca: A crtcal analyss of three studes, mmeo, Santago: ECLAC. Brumer, A. (2002), Gender Relatons and Rural Socal Securty n Southern Brazl, n C. Abel and C. Lews (eds.), Excluson and Engagement: Socal Polcy n Latn Amerca, London: Insttute of Latn Amercan Studes. Camarano, A. A. and M. T. Pasnato (2002), Envelhecmento, condcoes de vda e poltca prevdencara. Como fcam as mulheres?, Dscusson Paper 833, Ro de Janero: IPEA. Carvalho, I. (2000a), Elderly Women and ther Lvng Arrangements, mmeo, Cambrdge MA: MIT. Carvalho, I. (2000b), Old-Age Benefts and the Labour Supply of Rural Elderly n Brazl, mmeo: MIT. Case, A. (2001), Does Money Protect Health Status? Evdence from South Afrcan Pensons, mmeo, Prnceton: Prnceton Unversty. 21

Case, A. and A. Deaton (1998), Large Scale Transfers to the Elderly n South Afrca, Economc Journal, 108, (450), 1330-1261. Case, A. and F. Wlson (2000), Health and well-beng n South Afrca: evdence from the Langeberg Survey, mmeo, Prnceton: Prnceton Unversty. Commtte of Inqury nto a Comprehensve System of Socal Securty for South Afrca (2002), Transformng the present- Protectng the future, Draft Consoldated Report, Pretora: Commtte of Inqury nto a Comprehensve System of Socal Securty for South Afrca. Deaton, A. (2001), Countng the World's poor: Problems and possble solutons, World Bank Research Observer, 16, (2), 125-147. Deaton, A. and C. Paxson (1997), Poverty among chldren and the elderly n developng countres, mmeo, Prnceton: Research Program n Development Studes, Prnceton Unversty. Delgado, G. C. and J. C. Cardoso (2000a), Condcões de reproducão econômca e combate à pobreza, n G. C. Delgado and J. C. Cardoso (eds.), A Unversalzação de Dretos Socas no Brazl: a Prêvdenca Rural nos anos 90, Brasla: IPEA. Delgado, G. C. and J. C. Cardoso (2000b), Prncpas Resultados da Pesqusa Domclar sobre a Provdênca Rural na Regão Sul do Brasl, Texto para Dscussao 734, Ro de Janero: Insttuto de Pesqusa Econômca Aplcada. Delgado, G. C. and J. C. Cardoso, eds.) (2000c), A Unversalzação de Dretos Socas no Brazl: a Prêvdenca Rural nos anos 90. Brasla: IPEA. Devereux, S. (2001a), Can socal safety nets reduce chronc poverty?, mmeo, Falmer: IDS, Unversty of Sussex. Devereux, S. (2001b), Socal Pensons n Namba and South Afrca, IDS Dscusson Paper 379, Falmer: IDS. Damond, C. A.; C. J. Smon and J. T. Warner (1999), A Multnomal Probablty Model of Sze Income Dstrbuton, Journal of Econometrcs, 43, 43-61. Deden, S. (2003), Levels of poverty and ncome sources among South Afrcan households - a multnomal approach, mmeo. Duflo, E. (2000), Grandmothers and granddaughters: old age penson and ntrahousehold allocaton n South Afrca, Workng Paper 8061, Cambrdge MA: Natonal Bureau of Economc Research. Edmonds, E.; K. Mammen and D. Mller (2001), Rearrangng the famly? Household composton responses to large penson recepts, mmeo, Hanover NH: Darmouth College. 22

Ferrera, M. (1999), The generosty and unversalty of South Afrca's penson system, The EU Courer, 176. Foster, J. E.; J. Greer and E. Thorbecke (1984), A class of decomposable poverty measures, Econometrca, 52, (3), 761-766. Fultz, E. and B. Pers (1999), Socal Securty Schemes n Southern Afrca, Dscusson Paper 11, Geneva: ILO/SAMAT. Glewwe, P. (1991), Investgatng the determnants of household welfare n Côte d'ivore, Journal of Development Economcs, 35, 307-337. James, E. (2001), Coverage under Old Age Socal Securty Programs and Protecton for the Unnsured: What are the Issues?, n N. C. Lustg (ed.) Sheldng the Poor. Socal Protecton n the Developng World, Washngton DC: Brookngs Insttuton Press and InterAmercan Development Bank, 149-174. Jensen, R. (2002), Do prvate transfers 'dsplace' the benefts of publc tranfers? Evdence from South Afrca, Journal of Publc Economcs, n prnt. Kanbur, R. (2002), Conceptual challenges n poverty and nequalty: One development economst's perspectve, mmeo, Ithaca: Cornell Unversty. Klasen, S. (2000), Measurng poverty and deprvaton n South Afrca, Revew of Income and Wealth, 46, (1), 33-58. Lebbrandt, M. (2001), Household ncomes, poverty and nequalty n a multvarate framework, n H. Bhorat; M. Lebbrandt; M. Mazya; S. van der Berg and I. Woolard (eds.), Fghtng poverty. Labour markets and nequalty n South Afrca, Cape Town: UCT Press, 130-154. Lund, F. (1993), State socal benefts n South Afrca, Internatonal Socal Securty Revew, 46, (1), 5-25. Lund, F. (1995), Changng socal polcy n South Afrca, Socal Polcy Revew, 7. Lund, F. (1999), Understandng South Afrcan socal securty through recent household surveys: new opportuntes and contnung gaps, Development Southern Afrca, 16, (1), 55-67. Lund, F. (2002), 'Crowdng n' care, securty, and macro-enterprse formaton - revstng the role of the state n poverty reducton, and n development, Journal of Internatonal Development, 14, (6), 681-694. May, J. (2000), Poverty and nequalty n South Afrca: Meetng the challenge, London: Zed Books. May, J.; I. Woolard and S. Klasen (2000), The nature and measurement of poverty and nequalty, n J. May (ed.) Poverty and nequalty n South Afrca: Meetng the challenge, London: Zed Press. 23

Møller, V. and Sotshangaye (1996), "My famly eats ths money too": penson sharng and self-respect among Zulu grandmothers, Southern Afrcan Journal of Gerontology, 5, (2), 9-19. Saad, P. M. (1999), Transferêncas de apoo entre geracões do Brasl: um studo para São Paulo e Fortaleza, n A. Caramano (ed.) Muto além dos 60. Os novos odosos brasleros, Ro de Janero: IPEA. Sagner, A. and R. Mtat (1999), Poltcs of penson sharng n urban South Afrca, Ageng and Socety, 19, (4), 393-416. Schwarzer, H. (2000), Impactos socoeconômcos do sstema de aposentadoras ruras no Brazl - Evdêncas empírcas de un estudo de caso no estado de Pará, Dscusson Paper 729, Ro de Janero: IPEA. Schwarzer, H. and A. C. Querno (2002), Non-contrbutory pensons n Brazl: The mpact on poverty reducton, ESS Paper 11, Geneva: Socal Securty Polcy and Development Branch. ILO. Treas, J. and B. Logue (1976), Economc Develoment and the Older Populaton, Populaton and Development Revew, 12, 645-673. van der Berg, S. (2001), Socal polcy to address poverty, n H. Bhorat; M. Lebbrandt; M. Mazya; S. van der Berg and I. Woolard (eds.), Fghtng poverty. Labour markets and nequalty n South Afrca, London: Zed Press, 171-204. Weaver, K. R. (2003), The Poltcs of Publc Penson Reform, CRR Workng Paper CRR WP 2003-06, Chestnut Hll, MA: Center for Retrement Research at Boston College. Werneck Vanna, M. L. T. (2003), Segurdade socal e combate à pobreza no Brazl: O papel dos benefcos não-contrbutvos, mmeo, Ro de Janero: Unversdade Federal do Ro de Janero. Wllmore, L. (2001), Unversal Pensons n Low-Income Countres, mmeo, Washngton DC: Department for Economcs and Socal Affars, Unted Natons. Wllmore, L. (2003), Unversal pensons n Maurtus: Lessons for the rest of us, DESA Dscusson Paper 32, New York: Unted Natons Department of Economc and Socal Affars. Woolard, I. and S. Klasen (2003), Income Moblty and Household Dynamcs n South Afrca, Conference Paper - CPRC Conference on 'Stayng Poor: Chronc Poverty and Development Polcy', Manchester: IDPM, Unversty of Manchester. Woolard, I. and M. Lebbrandt (2001), Measurng poverty n South Afrca, n H. Bhorat; M. Lebbrandt; M. Mazya; S. van der Berg and I. Woolard (eds.), Fghtng poverty. Labour markets and nequalty n South Afrca, Cape Town: UCT Press, 41-73. 24

World Bank (1994), Avertng the Old Age Crss, London: Oxford Unversty Press. World Bank (2002), World Development Indcators 2002, Washngton DC: The World Bank. 25