social security in a moral economy java

Upload: educationisagift

Post on 14-Apr-2018

216 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/27/2019 Social Security in a Moral Economy Java

    1/10

  • 7/27/2019 Social Security in a Moral Economy Java

    2/10

    SOCIALSECURITYIN A "MORAL ECONOMY":AN EMPIRICALANALYSIS FORJAVA

    Martin Ravallion nd LorraineDearden*Abstract-Private ransfer ayments re modelled s outcomes

    of a constrained ocial choice problem facing donors. Theapproach s applied to a large household evel data set for Javaand hypotheses re tested oncerning heperformance f the"moral economy" as a social security ystem. Transfer e-haviour s found to be very ifferent etween ural nd urbanareas. While transfer eceipts nd outlays re ncome nequalityreducing n rural areas, this s not the case in urban areas.There is also evidence of 'transfers eing targeted o disad-vantaged households uch as the sick, elderly nd (for urbanareas) the unemployed.

    I. Introduction

    A THOUGHfew DCs have ublicly undedsocial security ystems, imilar istributional

    outcomes may be achievedby a private moraleconomy" ME) based on voluntary nterpersonaltransfers f money r goods.' This begs the entralquestion of this paper: Does a ME perform hefunctions f a social security ystem? n particu-lar: Is there vidence hat a ME targets ransferstowards otherwise isadvantaged roups, uch asthe poor, sick, lderly r unemployed?

    The paper examines hat uestion or he slandof Java n Indonesia. Java has been prominent npast discussions f the ME.2 But, despite large

    literature n this subject, we know surprisinglylittle bout the way the Javanese ME distributesincomes. The present tudy models the distribu-tions of transfer eceipts mongst ecipients ndof donations mongst onors n a large householdlevel data set for Yogyakarta rovince n centralJava. The distributions re interpreted s the out-comes of a constrained ocialchoiceproblem ac-ing potential donors. A theoretical escription fthat problem s offered hich permits dentifica-

    tion of some key parameters f donors' social

    preferences from econometric models of theirtransfer eceipts nd outlays.3A further uestion of interest n this setting

    concerns possible differences n socialpreferencesbetween rural nd urban reas. t has been arguedthat, as developing economies become moreurbanized, here s a decline n the distributionalsignificance f the ME. Casualobservations f thedifferences etween hese nvironments akesuchan argument lausible.For example, neighbourli-ness and reciprocity re widely hought obe moreimportant n villages than cities. Also, institu-tionalized social security ystems re sometimesmore developed n urban areas. With these argu-ments n mind, we shall test for the existence fsignificant ifferences etween ransfer ehaviourin rural nd urban areas of our sample.

    The following ection gives n overview f oureconometric model of transfer eceipts nd out-lays, while section II gives more rigorous reat-ment of its choice-theoretic asis. Section V de-scribes the data and methods of the empiricalinvestigation. ection V discussesthe results for

    rural households, while section VI discusses theurban results. ectionVII offers omeconclusions.

    II. An Overview f the Model

    The following conometricmodel of the distri-bution of transfer eceipts nd outlays s assumed:

    log Tir a, + fl1log 1 Xiyl + Mli f i E RTir 0 otherwise(1)

    logTi' = a2 + /21og i + Xi2Y2 M2iif E DTid 0 otherwise (2)eceived for publication August 6, 1986.Revision ccepted

    for publication July 4, 1987.*The Australian National University.For their helpful omments n this paper, our thanks o to

    Jere Behrman, David Bevan, Anne Booth, David Collard,Mubyarto, R. M. Sundrum, Dominique van de Walle, theReview's referees, nd seminar participants t Gadja MadaUniversity, ogyakarta, nd the Research School of PacificStudies,A.N.U.

    1The term moral conomy" ppears obe due to Thompson(1971)who describes t as "... a consistent raditional iewofsocialnorms nd obligations, f the proper conomic unctionsof severalparties within he community" p. 79).

    2Important contributions ncludeGeertz 1963),Scott 1976),Collier 1981),and Hart 1986).

    3There has been some recent work on the dentification fimplicit ocialpreferences rom hedistribution f governmentexpenditure nd taxation Ahmad and Stern, 984; Behrmanand Sah, 1984) and from parental allocations of schoolingexpenditures mongst children Behrman et al., 1982). Theinfluence f donor's ncome nd taxation n charitable ontri-butions has been studied by Martin Feldstein, avid Collardand others for the United States and United Kingdom; fordiscussion and references ee Collard (1978, ch. 10). Thepresent paper appears to be the first ttempt o model bothtransfer eceipts nd outlays n a choice-theoretic ramework.

    [ 36 ]Copyright 1988

  • 7/27/2019 Social Security in a Moral Economy Java

    3/10

  • 7/27/2019 Social Security in a Moral Economy Java

    4/10

    38 THE REVIEW OFECONOMICS ANDSTATISTICS

    ii. Some transfers re passed on; for example,person A givesRpl,000 to B who givesRp500 toC. In effect, has acted as an intermediary n thetransfer f Rp500 from A to C. Should one think

    of this situation s comprising irect ransfers fRpl,000 (A to B) and Rp500 (B to C) or as adirect ransfer f Rp500 (A to B) and an indirecttransfer f Rp500 (A to C)?

    The choice clearly depends on the assumptionone makes about the nitial donor's expectationsabout the recipient's onations. f one assumesthat donors hold myopic expectations f theirrecipient's behaviour then only the gross directtransfers hould be considered. At the other ex-treme, f one assumes that donors hold rationalexpectations f recipients' ehaviour hen nly nettransfers re relevant nd indirect ransfers viathird arties) houldbe treated n the ame way asdirect nes.

    In our view, it is difficult o choose betweenthese alternative ssumptions n a priori rounds.The informational equirements f the rationalexpectations ssumption re considerable but itmay still be plausible n some settings. or exam-ple, in a traditional illage, t may not be implau-sible that donors have the nformation ecessaryto correctly nticipate on average) recipients' e-haviour. Casual observation uggests hat donorsoften ake account of reciprocity. or example, norder to save face, poor recipient may prefer omake a small gift o a rich donor who expects histo happen. Such "reciprocity" s believed to be acommon characteristic f village economics inSoutheast Asia (Scott, 1976,chapter ).

    Rather hancommit urselves ow on this ssue,the theoretical model of the donor's choice prob-lem discussed in the- ollowing ection can bestatistically mplemented n either of two ways,corresponding o the lternative ssumptions bout

    donors' expectations f recipients' ehaviour. hetransfer made from donor d to recipient r isdenoted tdr. Under myopic xpectations his s thegross direct transfer; nder rational xpectationsit is the net transfer ith ny ntermediate rans-fers ubsumed.

    III. The Donors' Social ChoiceProblemThis section offers choice-theoretic nterpreta-

    tion of the econometric modelsof transfer eceiptsand outlays givenby equations 1) and (2),

    It is assumed that the values of tdr are chosenby d to maximize:

    u(Y15 X15,...,5 Yd5 Xd5 .., Yn Xn) (3)

    where Y1s the ncome fter ransfers f householdi and Xi is a vector f other ttributes f i. Themaximization s constrained y

    Yr= f (tdr 5 Xr) (4)n

    Yd =Yd E tdr (5r=1

    where Yd is the donor's (fixed) pre-transfer n-come. The simplest xample of the function is

    Yr = Yr+ tdr* More generally, ne may supposethat transfers nfluence heir ecipients' re-trans-fer ncome, ither ositively through, or xample,any effects on the recipient's productivity) rnegatively through, for example, disincentiveeffects). Thus aYr/dtdr may be greater r lessthan unity, ut it is assumed o be non-negative.

    Under these conditions, he donor's optimaltransfer must have the property hat

    a u a u aYr

    aYd aYr atdr

    and tdr ? 0 with omplementary lackness. (6)

    In other words, t an optimum, hedonor's margi-nal utility from giving n extra Rupiah cannotexceed the donor's marginal utility of own-income. Otherwise, he xtra Rupiah willbe givenand so the previous tate could not have been anoptimum.) f the two marginal utilities an beequalizedat a positive ransfer evelthen t willbe(at least locally) optimal to make that transfer.5Otherwise no transfer will be made since theequalizationof marginal tilities ould require hepower to tax. Note that no transfers illbe madeif aYryatdr = 0. Non-altruistic ehaviour ntailsau/a Yr 0 and, of course, no transfer ill bemade in this ase either.

    In deriving quations 1) and (2) as an econo-metric model of the donors' optimal ransfers, he

    5Sufficient onditions or his obe a globaloptimum re thatu is strictly uasi-concave n all incomes nd f is concave ntransfer eceipts.

  • 7/27/2019 Social Security in a Moral Economy Java

    5/10

  • 7/27/2019 Social Security in a Moral Economy Java

    6/10

    40 THE REVIEWOF ECONOMICS ANDSTATISTICS

    IV. Data andEstimation ethods

    Detailed data on household ncomes and ex-penditures or a large sample are available fromresults of the SUSENAS surveysdone by theIndonesian Central Bureauof Statistics BPS).Weuse the data for a sample of 2,261 households nYogyakarta Province n central Java, s containedin the 1981 SUSENAStapes.8These ncludedataon transfer eceipts nd outlays, omprising llgifts of money and goods, the latter valued atcurrent market rices y BPS.

    Our discussions with BPS staff n Jakarta on-vinced us that reasonably erious ffort ad beenmade to accurately uantify ll transfers, nclud-ing (for example) food consumed at selametans(communal feasts commonly rovided by richerhouseholds in Javanese villages). The data arebased on interviewees' ecalland so measurementerrors re to be expected, lthoughwe see no goodreason why those errors wouldnot be whitenoise.However, t is plausible that there re forms ftransfer behaviour which would not be easilypicked up in a conventional urvey, uch as "worksharing" practices mongst nderemployed ork-ers.9

    The use of the household s the unit of observa-tion s forced n us by the data. This means that

    we cannot identify ransfers etween ndividualswithin ne household, lthough ransfers etweenmembers of the same family iving n differenthouseholds re included.

    Yogyakarta Province s a whole turns out tohave been a net recipient f transfers. et receiptsrepresent n average f 2.7% of total onsumptionexpenditure n rural reas. But the figure s muchhigher n urban areas, at about 17%of expendi-ture. This probably reflects wo factors: remit-tances from family members who leave the Pro-vince,particularly hosewho find work n Jakarta,and transfers ade to support tudentswho comefrom utside the Province o attend Yogyakarta's

    numerous educational institutions, articularlyGadja Mada University.

    A similar model to equation 1) has been ap-plied to the nternational istribution f develop-

    ment assistance by Behrman and Sah (1984).Ordinary east squares estimation was used byBehrman nd Sah on the sample of aid recipients.While this is a convenient stimator, t can beexpected to produce asymptotically) iased esti-mates on two counts: therewillbe sample electiv-ity bias associated with the exclusion of non-recipients nd simultaneity ias due to the factthat the post-transfer ncome variable s itselffunction f transfer eceipts nd so will not beindependent f the model's error erm.'0 Our pre-ferred stimator or his tudy as been a two-stage

    Tobit estimator with non-recipients nd non-donors included) in which the post-transfer n-come variable was not used as its own nstrument.

    The income variable n equations 1) and (2)should be a suitable metric f each household'swelfare fter ransfers. he variable we have usedis total household expenditure n all goods andservices, xcluding gifts. We interpret his as aproxy for ost-transfer ermanent ncome. o helpdeal with the simultaneity roblem we have re-placed log Y in equations 1) and (2) by predictedvalues from a regression f log consumptionagainst the following et of household haracteris-tics: income excluding ransfer eceipts, ears ofschooling, dwelling characteristics, ccupationaldummy variables and the X vector n equations(1) and (2). (Full details are available from heauthors.) The use of predicted onsumption s aninstrument or actual consumption ad the ex-pected effects n estimates f the 's, whichwerereduced n the receipts quation and increased nthe outlays equation when compared to the esti-mates obtained using consumption s its own in-

    strument. o the extent hat transfers lso affectpre-transfer ncomes, he nclusion f the atter nthe predictor quation for post-transfer onsump-

    8The choice of Yogyakarta Province was attractive or anumber f reasons. t gave us a sample of manageable ropor-tions. The first uthor was also a visitor at Gadja MadaUniversity n Yogyakarta nd so the area was not unfamiliar.The area also appears to provide reasonable ross-section fconditions n central Java as a whole.

    9Such practices are thought o be traditional n Javaneseagriculture, lthough heir mportance ppears to have declinedin recent times. For further iscussion see Collier (1981),Ishikawa 1981), Booth and Sundrum 1984) and Hart 1986).

    10Behrman nd Sah (1984) note existence of the secondproblem and consider the likely bias in the OLS incomecoefficient. owever, his s less than atisfactory n the presentsetting ince all other oefficients n equations 1) and (2) will,in general, lso be biased when even one) regressor s corre-lated with the error term. Behrman nd Craig (1987) haveprovided a more convincing nstrumental ariables stimatorof their model of the spatial allocation of police patrols nBaltimore.

  • 7/27/2019 Social Security in a Moral Economy Java

    7/10

    SOCIALSECURITYIN A "MORAL ECONOMY" 41

    TABLE 1.-ESTIMATES OF EQUATIONS (1) AND (2) FOR RURAL HOUSEHOLDS

    (1) (2)Receipts Outlays

    Gross Net Gross Net

    1 22 38 -25 -39(2.1) (3.1) (6.2) (6.6)

    Income log) - 3.1 - 5.0 2.6 3.7(3.0) (4.0) (6.4) (6.3)

    Household size log) 0.023- 0.86 -0.30 -0.40(0.023) (0.72) (0.76) (0.71)

    Ill-health week) 2.4 2.6 -0.31 - 1.1(2.7) (2.4) (0.82) (2.2)

    Ill-health quarter) 1.9 2.1 -0.48 -1.2(2.4) (2.2) (1.5) (2.6)

    Births 2.4 3.7 0.20 -0.72(2.3) (2.9) (0.46) (1.2)

    Deaths 1.6 - 1.0 - 2.1 -0.34(0.72) (0.38) (2.2) (0.25)

    Age 0.048 0.034 0.13 -0.20Age (0.36) (0.22) (2.3) (2.4)

    Age2 0.0008 0.0018 -0.0010 -0.0022(0.64) (1.2) (2.0) (2.7)

    Unemployment 1.1 2.0 - 2.0 -1.7(0.81) (1.3) (3.5) (2.0)

    SEE 10 11 4.7 6.4Prob T > GIX)

    Predicted 0.34 0.25 0.85 0.63Actual 0.31 0.25 0.72 0.54

    Note: Absolute t-ratios n normalized coefficients n parentheses.

    tion will eave some simultaneity ias in the esti-

    mates. Assuming hat transfers re more ikely ohave, f anything, productive ffect n pre-t-'ans-fer incomes, P, will be over-estimated nd f2under-estimated. he possibility f such biasesshould be borne in mind when interpreting hefollowing esults.11

    V. The Rural Results

    Table 1 givesTobit estimates f the coefficientsof equations 1) and (2) for the 1,539householdsin the SUSENAS sample who were classified s

    living n rural areas of Yogyakarta Province.Re-sults are given for both gross and net transfers,corresponding o the two assumptions oncerningdonors' expectations f recipients' ehaviour dis-cussed in section I. The appendix givesfull defi-

    nitions f the variables nd summary ata on their

    means and standard eviations.The following oints emerge learly from heresults f table 1:

    i. Donors' social preferences xhibit ignificantaversion to inequality nd elastic marginal tilityof own-income."2 he inequality version arame-ter is found to be 4.1 for the model of grossreceipts myopic expectations) nd 6.0 for netreceipts rational expectations). he correspond-ing elasticities f the marginal utility f incomeare 2.6 and 3.7.

    ii. No significant ffects f household size onreceipts or outlays are revealed for either model.Of course, there may still be indirect ffects per-ating through other variables, most particularlyhousehold consumption. Household size washighly ignificant n the instrument quation forconsumption.) he insignificance f size as a com-

    "The direction f any bias in other oefficientss difficult odetermine n a priori grounds. We also tried deleting pre-transfer ncome from hepredictor quations but this resultedin unacceptably ow correlations etween redicted nd actualconsumptions. ssets,notably and, would probably ave beena better choice than ncomes, ut this s not included n theSUSENAS data.

    "2Any imultaneity ias due to a productive ffect f transferson " pre-transfer" ncomewillresult n 81 being overestimatedand f2 being underestimated. hus our conclusion will alsohold if suchbiases could be eliminated. owever, ny disincen-tive effects f transfers n pre-transfer ncomeswouldmitigatethe conclusion.

  • 7/27/2019 Social Security in a Moral Economy Java

    8/10

    42 THE REVIEWOF ECONOMICSAND STATISTICS

    ponent of the X vector n equations 1) and (2)implies that the relation between transfers ndincomes is not homogeneous n household size;for a given ineome per person, an increase n

    household size will ead to lower transfer eceiptsper person nd higher utlays er person."3iii. A number f other household haracteristics

    affect ransfer ehaviour n rural reas. Ill-healthattracts eceipts nd discourages utlays, lthoughonly significantly o for utlays nder he rationalexpectations ssumption. A recent birth ttractsreceipts but has no significant ffect n outlays,although except for gross utlays) here s little osuggest that deaths have a similar ffect. nem-ployment as no significant ffect n receipts, utoutlays are reduced and by a similar amountunder both assumptions.iv. Receipts appear to be invariant o age ofhousehold head, while outlays xhibit plausibleinverted u relationship, ith urning oints t 61and 45 years for gross and net outlays, respec-tively. However, when age2 is dropped from hereceipts equations, significant ositive relation-ships are found, with regression oefficients f0.13 (t = 5.6) and 0.15 (5.5) for gross and netreceipts, espectively.14Thus transfer eceipts ug-gest targeting oward he lderly.

    v. The magnitudes f a number f these ffectsare far from negligible. Consider household Awhose expenditure s Rp4O,000/month beingabout the mean for the sample) and has meanreceipts f about Rp4,000,but makes no outlays.Compare this ase with hat f householdB whoseconsumption s RplO,000 ower. Then the resultsfor the gross transfers odel suggest hat transferreceipts for B will be Rpl,000 greater han Awhile for the model of net transfers, 's receiptswill exceed A's by Rpl,250: roughly 0% to 12%of the consumption nequalitywill be made up by

    transfer eceipts.15 Comparing households under

    otherwise verage conditions, n extra llness nthe last week increasesnet transfer eceipts y afactor f two, representing bout an extra Rp4,000at the mean, or about 10%of mean expenditure."6

    A recent irth would ncrease eceipts y a factorof 2.5 or about Rp6,000. For an extra year atmean age, net receipts ncrease y about 4.0% andnet outlays all by 1.5%.

    VI. An Urban Moral Economy"?We now look at results or he SUSENAS sam-

    ple of 722 households n urban areas (mainlyYogyakarta City) of Yogyakarta Province. Themethodology s the same, so we turn traight othe results n table 2.

    The contrast with he rural stimates s striking.The following oints eem most notable:i. Neither receipts nor outlays re significantly

    influenced y income for either pecification.value of P, = 0 implies unitary nequality ver-sion. However, P, is not significantly ifferentfrom unity by a likelihood ratio test; so onecannot reject henull hypothesis f zero nequalityaversion. The value of f2 = 0 implies constantmarginal tility f own-income.

    ii. Receipts decreasewith household ize. Thisresult may reflect ransfers eceived y single tu-dents living way from home. As already noted,Yogyakarta City is a major centre for tertiaryeducation.

    iii. While a recent irth has a similar ffect nreceipts, number f other household haracteris-tics have rather different ffects n transfer e-haviour n urban areas when compared to ruralareas. In contrast with the rural results, here slittle suggestion hat ll-health ffects eceipts routlays.

    iv. Receipts exhibit strong shaped relation-ship with ge, while he relationship s inverted oroutlays. The turning oints are well within hebounds of the data, being 52 years and 56 yearsfor receipts nd outlays, espectivelyneither e-pends on whether rossor net transfers re used).

    13Noting that, on defining X = log N to simplify he alge-bra, equations (1) and (2) imply that log(Tj/N) =Pkllog(Y/N) + (k 4- Yk - 1)logN for = r k = 1)and -d (k = 2). The restriction k + Yk = 1 is firmly ejected or llspecifications. Note that this also implies rejection of the(un-tested) estricted orm sedby Behrman nd Sah (1984) inwhich the log of foreign id as a proportion f recipient'snational ncome s regressed gainst per-capita ncome.

    14Full details on the restricted egressions re available fromthe authors.

    15Note that the elasticities sed here and in later discussionare the total ffects f changes n the regressor n the xpected

    value of the regressand. or the Tobit model, hese ffects regivenby the product f the Tobit coefficientsas given n table1) and the proportion f the sample above the imit see, forexample,McDonald and Moffitt, 980).

    16Note that the figure f 2.0 is the exponential alue of theproduct of the Tobit coefficient nd the fraction f the ampleabove the imit see footnote 5).

  • 7/27/2019 Social Security in a Moral Economy Java

    9/10

    SOCIAL SECURITY IN A"MORAL ECONOMY" 43

    TABLE 2.-ESTIMATES OFEQUATIONS1) AND(2) FORURBANHOUSEHOLDS

    (1) (2)Receipts Outlays

    Gross Net Gross Net

    1 -3.8 -2.2 -17 -28(0.39) (0.21) (2.0) (2.5)Income log) 1.9 1.7 0.50 0.80

    (2.1) (1.8) (0.65) (0.79)Householdsize log) - 2.7 -2.6 0.90 1.3

    (3.0) (2.8) (1.1) (1.2)Ill-health week) 1.2 1.7 -0.56 -0.70

    (1.2) (1.7) (0.61) (0.56)Ill-health quarter) -0.19 -0.41 0.81 0.70

    (0.22) (0.46) (1.1) (0.70)Births 2.8 2.2 -0.27 -1.9

    (2.2) (1.6) (0.24) (1.3)Deaths 3.4 4.5 2.1 -0.82

    (1.0) (1.3) (0.67) (0.18)Age -0.63 -0.68 0.45 0.59

    (4.3)(4.4) (3.0) (2.7)Age2 0.0060 0.0065 -0.0040 -0.0052

    (3.9) (4.0) (2.5) (2.3)Unemployment 4.7 5.6 - 3.3 - 7.0

    (4.6) (5.3) (3.3) (4.7)-------------------------EE--------------------------------------__-------SEE 8.3 8.5 7.6 9.2Prob T > OIX)

    Predicted 0.50 0.45 0.50 0.31Actual 0.44 0.40 0.45 0.31

    Thus, n contrast o the rural esults, rban trans-

    fers reveal significant argeting oward both theyoung and the elderly. his may reflect differ-ence in age distributions etween ural nd urbanareas.17

    v. In marked ontrast o the rural esults, nem-ployment merges s a significant eterminant furban transfers; nemployment n the previousweek increases receipts nd there s also a fall noutlays. This difference etween the samples isconsistent with the widely held view that greaterwage flexibility nd/or the existence f work har-ing arrangements n the rural abour markets f

    LDCs, render "unemployment" as convention-ally defined) s less serious welfare isadvantagethan n the cities.'8

    VII. Conclusions

    The modelled distributions f transfer eceiptsin rural Java suggest hatdonorshold a preferencefor less inequality. Receipts ncrease more thanproportionally o a fall n recipient's ncome.Weinterpret his s indicating hat hedonor's margi-nal utility f recipient's ncome s not only posi-tive, but also elastic n that ncome.Donors alsoassist the sick, lderly, nd householdswith ecentbirths. Donors' marginal utility f own-incomeappears to be elastic, o their overall generosityincreases more than proportionally ith their n-come. And most of those

    attributes f recipientswhich make donors more generous, lso tend tomake donors with hose ttributes essgenerous.

    Transfer ehaviour smarkedly ifferent mongsturban households.The distribution f receipts e-veals little aversion to inequality nd little vari-ation in the marginal tility f own-income. hustransfers ave negligible ffect n the distributionof ncome.There are other ifferences: e find oevidenceof significant ompensation or ll-healthin the pattern f urban transfers. hile transfers

    17Mean age of householdhead is lower for heurban ample(see appendix), nd the difference s significant t = 8.6). Againthis s likely o be due to students iving way from home nYogyakarta City. t is also thought o be common or migrantsin the city o return o their illage f origin t old age.

    "Note that he rate of unemployment n the urban ample sa good deal higher han the rural ample; 25%in the formerversus 6% in the atter appendix).

  • 7/27/2019 Social Security in a Moral Economy Java

    10/10

    44 THE REVIEWOF ECONOMICSAND STATISTICS

    tend to be targeted oward the elderly n ruralareas (other things being equal), urban transfersalso go to the young. And, unlike rural areas,unemployment merges s an important nfluence

    on transfer ehaviour.

    REFERENCESAhmad, Ehtisham, nd Nicholas Stem, "Theory of Reform

    and Indian Indirect axes," Journal f PublicEconomics25 (1984), 259-298.

    Atkinson, Anthony B., "On the Measurement f Inequality,"Journal f Economic heory (June 1970), 244-263.

    Behrman, Jere R., and Steven G. Craig, The Distribution fPublic Services:An Exploration f Local GovernmentalPreferences," he American conomic Review77 (Mar.1987), 37-49.

    Behrman, Jere R., Robert A. Pollak, and Paul Taubman,

    "Parental Preferences nd Provision for Progeny,"Journal f Political Economy 0 (Feb. 1982), 52-73.Behrman, Jere R., and Raaj Kumar Sah, "What Role Does

    Equity Play in the nternational istribution f Devel-opment Aid?" in M. Syrquin, L. Taylor and L. E.

    Westphal eds.), Economic tructure nd Performance(Orlando: Academic Press, 984).

    Booth, Anne, and R. M. Sundrum, Labour Absorption nAgriculture Oxford: Oxford University ress, 984).

    Collard, David, Altruism nd Economy: A Study n Non-SelfishEconomics Oxford: Martin Robertson, 978).

    Collier, William L., "Agricultural volution n Java," n G. E.Hansen (ed.), Agricultural nd Rural Development nIndonesia Boulder, Colorado:Westview ress, 981).

    Geertz, Clifford, gricultural nvolution Berkeley:Universityof California ress, 963).

    Hart, Gillian, Power, Labour and Livelihood. Processes ofChange in Rural Java (Berkeley: University f Cali-fornia ress, 1986).

    Ishikawa, Shigeru, Essays on Technology, mployment ndInstitutions n EconomicDevelopment Tokyo: Kino-kuniya Company, 981).

    McDonald, John F., and Robert A. Moffitt, The Uses ofTobit Analysis," hisREVIEW 62 (Aug. 1980), 318-321.

    Scott, James C., The Moral Economy f the Peasant (NewHaven: Yale University ress, 976).

    Sugden, Robert, The Economics f Rights, Cooperation ndWelfare Oxford: BasilBlackwell, 986).Thompson, E. P., "The Moral Economy f the EnglishCrowd

    in the Eighteenth entury," ast and Present 0 (Feb.1971),76-136.

    APPENDIX

    TABLEA.1.-DEFIMTIONOF VARIABLESND UMMARYATA

    Rural UrbanVariable Definition Mean St. Dev. Mean St. Dev

    Transfer eceipts Value of all gifts (gross)3998 14252 16242 38373received n past month (net) 3293 13370 15728 38069

    Transfer utlays Value of all gifts (gross)3030 5768 2558 7567made n past month (net) 2324 5572 2045 7504

    Income Total food ndnon-food xpenditure,excluding ifts made 39652 36599 80488 80231

    Household Size Total number fhouseholdmembers 4.5 2.0 4.7 2.7

    Ill-health week) Dummy ariable:1 ifillness eported orlast week, otherwise 0.15 0.35 0.15 0.35

    Ill-health quarter) As above,but forlast 3 months 0.20 0.40 0.24 0.43

    Births Dummy ariable:1 ifbirth n ast year,0 otherwise. 0.10 0.31 0.090 0.29Deaths As above,for eath 0.019 0.14 0.0097 0.098

    Age Ageof head ofhousehold n years 50 15 44 16

    Unemployment Dummy ariable:1ifhead s in workforce utout of work n previousweek, otherwise. 0.064 0.25 0.25 0.43

    Note: All monetory variables are in Rupiah per month, 1981 prices.