Third, committing to deliver basic incomes for along period of time could have effects that differ from the effects of shorter-term transfers them-selves. This is true for (at least) two reasons. First, the basic logic of intertemporal smoothingsuggests that recipients may be less likely to save a marginal dollar today if they expect to con-tinue receiving transfers far into the future. Second, the expectation of income support in thefuture could increase recipients’ tolerance for risk taking today in settings where they are poorlyinsured and close to subsistence

For example, Duflo (2000) finds that pension payments to grandmothersbut not grandfathers led to anthropometric gains for granddaughters in the same households inSouth Africa, while Akresh et al. (2016) find in Burkina Faso that, if anything, transfers to fathershad larger impacts on children’s health, while also yielding greater investment

Second, delivering UBI to each adult in a household (whichis what many basic income proposals contemplate) could have different effects from deliveringthe same amount of money to a single adult representative of the household (which is how mostexisting cash transfer programs function). If the unitary model of the household were correct, then,of course, this distinction would not matter, but there is ample evidence that many households arenonunitary in consequential ways (Chiappori & Donni 2009). Individualizing transfers could thushave important effects, both by changing who decides how those particular dollars are spent andby changing the outside options that each family member holds and thus their influence on variousother decisions.

Mechanically, broader coverage could change the average impact of cash transfers by changingthe identity of the average recipient. Generally speaking, existing programs have tended to targetthe poor or vulnerable—widows, the elderly, low-income parents, and so on. The average recipientof UBI would thus tend to be less poor and less vulnerable than the recipients of existing programs.To forecast how exactly this would affect program impacts, we could potentially adjust existinganalyses by reweighting each observation by an estimate of each household’s (inverse) propensityto be treated, in an effort to give equal weight to impacts on all sorts of people.

Second, evaluations have found a great diversity of positive impacts. For example, a partial listof outcomes affected in a positive way in one study or another, according to Bastagli et al. (2016),includes income, assets, savings, borrowing, total expenditure, food expenditure, dietary diversity,school attendance, test scores, cognitive development, use of health facilities, labor force partici-pation, child labor migration, domestic violence, women’s empowerment, marriage, fertility, anduse of contraception

First, evaluations generally have not found the negative impacts that many feared. Reviewingevidence on temptation goods, Evans & Popova (2017) find that transfers had on average reducedexpenditure on temptation goods by 0.18 standard deviations. In other words, far from blowingtheir transfers on alcohol and tobacco, recipients appear to drink and smoke less

552 million people living in the developing world who receive some form of cashtransfer from their government

UBI in some other potentiallyconsequential way. We emphasize two dimensions in particular. First, existing transfers have notbeen universal but rather targeted, both to subsets of households (through means testing, ordeals,conditions, etc.) and to specific adults within those households (often the female head). Second,existing transfers typically last for relatively short time periods, as opposed to the long-termcommitment envisioned by UBI advocates. Both of these differences could lead to importantdifferences in impacts.

one fortwo years in nine villages in the Indian state of Madhya Pradesh, one in two villages in Namibia,and Iran’s nation-wide cash transfer introduced in 2011 to offset the withdrawal of food andfuel subsidies.

echnocratsmight need to be convinced to hand over money to the poor that they otherwise could have used todesign and implement projects in their own areas of expertise and motivation—projects designedto achieve narrower goals for health, education, nutrition, and so on. Taxpaying voters would needto be convinced to support handouts despite commonly held concerns that they would reducerecipients’ own initiative or trigger self-harmful behaviors such as drinking or drug use.

. Clearly, funding anything close to UBI would require a substantialshare of available resources. Some countries undoubtedly have scope to raise additional revenue,but in general, LICs raise substantially less revenue as a share of GDP than wealthier ones, likelybecause their capacity to do so is limited (Gordon & Li 2009)

In the developing world, large-scale basic income schemes could plausibly be financed from(some combination of ) repurposed foreign aid, repurposed domestic tax revenue, or new taxrevenue—although they would place serious demands on any of these sources.

The net impact of the policy is then the sum of effects on net recipients andnet payers.

. UBImight make sense because it will usually help a bit and will never entirely miss the mark. Moreover,of course, the opportunity cost of basic income as a growth intervention is low, since the bulk ofthe fiscal outlay is simply a progressive transfer

For any particular constraint, there is a plausible argument, and typically some evidence, that UBIcould help relax it.

This review, then, examines what we know and what we do not about these three issues

s a result,while we know that, in some countries, the private sector actually provides quality comparableto the public sector at a lower cost, this may not be true elsewhere.

Even if basic income is the best solution for some people, it is not obvious that it is universallybest. A targeted basic income could be a better use of scarce resources. This in turn depends onthe effectiveness of targeting

The reality, however, is that much of the spending on development is on issues like nutrition,health, and education, which may or may not be the priorities of the people that it is intended tohelp.

e could argue that we do not. One of the central goals of development economics has been tounderstand how to raise the incomes of people who are poor. A sustainable program for universalbasic income (UBI) does that by definition.

wever, it is worth noting that theeffects estimated in this paper may be larger than what wouldbe expected if cable were introduced more widely. Although wehave argued that preexisting trends in attitudes do not drive theresults, we cannot rule out the possibility that television is intro-duced first into areas that have the biggest potential for change;those that are receptive to television may also be receptive tochanging their gender attit

lower. Nevertheless, given the magnitudeof the effects estimated here, even much smaller effects could havesignificant impacts in

n. However, one plausible mechanism is that television eposes rural households to urban lifestyles, values, and behaviothat are radically different from their own and that householdsbegin to adopt or emulate some of these, as suggested by maanthropological and ethnographic studies of te

improves the status of women: women report lower acceptabilitof spousal abuse, lower son preference, more autonomy, and lowfertili

Taken together, the DISE and SARI data suggest that cableleads to increased school enrollment for younger children. Giventhe large literature showing that increases in women's status anddecision-making authority are associated with gains in children'soutcomes, it is certainly plausible that the schooling results arerelated to the improvements in women's status, such as partici-pation in household decision making, documen

ollment. For the fixedcohort ages 6-7, the effect is large, with additional years increas-ing enrollment by 5%. For children ages 6-10, the effect is smaller,between 1% and 2%, but still statistically significant. Finally, con-sistent with the SARI data, we see no effect for the o

e, regressing log enrollment on whether the village has cableand years of cable access. To account for possible serial correla-tion, we use the Prais-Winston estimator. The controls includeelectrification (inferred from whether the school has electricity),population,23 and distance to the neare

One other challenge is that our village survey only reportedthe year the village got cable, not the month. This makes it diffi-cult to match with precision the timing of cable introduction andschool enrollment. For example, villages that received cable in2002 may have added it in August, when it was already too lateto influence the decision on whether to enroll a child in the 2002-2003 academic year, because the school year begins in early Juneand late enrollment is generally not permitted. This suggests thatthe effects of cable on enrollment in many cases may not show upuntil at earliest the academic year starting in the calendar yearafter cable is introdu

, shows less clear results and, in particular, does not appear todemonstrate any "on-introduction" impact of cable: moving from2001 to 2002 in the 2002 adopters or 2002 to 2003 in the 2003adopters does not appear to increase enrollment. However, thereis a large increase between 2002 and 2003 for the 2002 adopters,again perhaps pointing to a delayed effect. Although the resultsfor women's status suggests immediate on-introduction effects, itis possible that any effects on education could take longer, forexample, because plans for education must be made further inadvance and money must be saved for fees and o

here y^ is a full set of individual fixed effects, 8t is a full setof year dummies, and the other controls, Xivt, include householdincome and a quadratic i

ent. In the villages that do not changecable status, enrollment is either flat or slightly decreasing overthe sample period. For those that get cable in 2002, enrollmentincreases between 2001 and 2002, and then further increases be-tween 2002 and 2003, suggesting some increasing impact overtime

n other words, it does not seem that there are changes in the out-comes that anticipate getting cable. Note that not only are thesecoefficients not statistically significant, but the point estimatesare extremely small, and in most cases statistically significantlydifferent from the effect of getting cable this year - so we can rejectthat any changes observed after cable is introduced are simply thecontinuation of preexistin

Cable Access and Female Au

This figure shows son preference (equal to 1 if the respondent reports wantinga son for the next child) from the SARI data, broken down by villages that alwayshave had cable, those that got cable for the first time in 2002, those that got it forthe first time in 2003, and those that never have had cable. The sample is onlyindividuals who report wanting more c

nd 2003. This graph sug-gests a strong connection between cable availability and televisionviewership, with a near doubling in bp

Cable Access and Television

es by these factors, we also include in the regressionsinteractions between a year indicator and state dummies, income,education, age and age-squared, village population density, elec-trification status, and distance to neares

e. Our identifying assumption is thatvillages that added cable would not otherwise have changed dif-ferently than those villages that did not add

ur basic empirical strategy is to compare changes in ourmeasures of women's status for villages that add cable over thecourse of the panel relative to those that do not. We run individual-level fixed-effects regressions of each outcome on cable availability(measured at the village level). Denote the outcome for individuali in village v in year t as stvt and the measure of cable access ascvt. The primary regression estima

Panel B of Table II reports summary statistics for thesemeasures. For the decision-making variables, we condense theresponses to binary indicators for whether the woman partici-pates in the decision (either decides on her own or decides jointlywith others in the household). Overall, slightly more than half ofwomen participate in each of the decisions. There is some overlapin these variables, though not as much as might be suggested bythe similarity of their means; about 20% of women do not par-ticipate in any of the decisions, 25% participate in one, 27% intwo, and 29% in three. About one-half of women report needingpermission to go to the market and two-thirds need permissionto visit family or friends.9 By contrast, nearly three-quarters ofwomen are allowed to keep money set aside to spend as they wish.However, by most measures, women's autonomy overall is quitelo

Summary statistics for these two measures are in Panel B ofTable II. Over 60% of women feel that it is acceptable for a husbandto beat his wife under at least one of the six situations listed.On average, women report 1.6 situations in which it is consideredacceptable. Women are most likely to believe beating is acceptableif a wife neglects her children, goes out without permission, ordoes not show respect toward her husband. Perhaps surprisingly,being unfaithful is reported as valid justification for violence byslightly fewer women. In terms of son preference, 55% of womenwho want another child prefer that child to

The outcome measure we use is the number of situations in whichthe woman reports that beating is accepta

ll women in the sample households ages fifteen and olderwere interviewed (no men were interviewed).

ur primary data set is SARI, a panel survey of 2,700 house-holds, each containing a person age fifty or older, conducted2001, 2002, and 2003 in four states (Bihar, Goa, Haryana, anTamil Nadu) and the capital, Delhi. The sample was selected itwo stages: in the first stage, 180 villages were selected at randomfrom district lists (40 villages in Bihar, Haryana, and Tamil Nadu35 in Delhi; and 25 in Goa), and in the second stage, 15 householdwere chosen within each village through random sampling basedon registration lists. Other than Delhi, the survey was confinedrural areas. Attrition over the panel was low, with just 108 (4%of the original households dropping out by t

. Note, however, that the R2 valuesin the regressions in Table I are small, indicating that much ofthe variation in cable access remains unexplai

These quantitative results are supportive of the qualitativeevidence from the interviews and suggest the importance of con-trolling for these determinants of cable access in our ana

anel B showsmultivariate regressions with these variables and, in column (4),includes state fixed effects to capture the fact that access variessignificantly across states. The broad patterns remain the same.In these data, there is some evidence of a role for income as adeterminant of cable access, suggesting it may be important tocontrol for it even though it was not explicitly mentioned by cableoperator

8 (column (2)).tors, villages thatoverall

The cable operator data can be used to examine the deter-minants of cable access more quantitatively by merging villageswith administrative data from an education database (again, de-scribed in more detail in Section V). Doing so allows us to ex-amine the village-level relationship between cable access and thecorrelates suggested above: distance to a town, population, andelectrification

n moving to a more quantitative analysis of the impacts ofcable, we must recognize that variation in access is certainly non-random. Therefore, understanding the determinants of the timingand placement of cable is important for our ability to attributechanges in women's status to the introduction of ca

nt. Another woman reports that because of tele-vision, men and women are able to "open up a lot more" (Scrase2002). Johnson (2001) quotes a number of respondents describingchanges in gender roles as a result of television. One man notes,"Since TV has come to our village, women are doing less workthan before. They only want to watch TV. So we [men] have to domore work. Many times I help my wife clean the h

lso), a show based aroundthe life of a wealthy industrial family in the large city of Mumbai.As can be seen from the title, the main themes and plots of theshow revolve around issues of family and g

rly 1990s. In the five yearsfrom 2001 to 2006, about 30 million households, representing ap-proximately 150 million individuals, added c

ess so. In fact, the government of TamilNadu has recently begun a program to provide free color tele-visions to 7.5 million households with the goal of ensuring thatevery household has one by

rollment. Using both ourhousehold panel data and administrative data for roughly 1,000villages in the state of Tamil Nadu, we provide evidence that theintroduction of cable increases school enrollment for younger chil-dr

tus. We argue that this does not appearto be the case, first showing visually that there are no preexist-ing differential trends in women's status for villages that do anddo not add cable, and that the timing of changes in outcomes isclosely aligned with the introduction of cable; and second, that theoutcomes are not correlated with future cable access.

sing these data, we find that cable television has large ef-fects on women's status. After cable is introduced to a village, thereare significant changes in gender attitudes: women are less likelyto report that it is acceptable for a husband to beat his wife, andless likely to express a preference for sons. Behaviors tradition-ally associated with women's status also change: women reportincreased autonomy (e.g., the ability to go out without permissionand to participate in household decision making) and lower fertil-ity

The primary analysis relies on a three-year panel data setcovering women in five Indian states between 2001 and 2003.These years represent a time of rapid growth in rural cable ac-cess. During the panel, cable television was newly introduced in21 of the 180 sample villa

men. Although issues of gender equality are im-portant throughout the world, they are particularly salient inIndia. Sen (1992) argued that there were 41 million "missingwomen" in India - women and girls who died prematurely dueto mistreatment - resulting in a dramatically male-biased popu-latio

. Most popular cable programming features urban settingswhere lifestyles differ in prominent and salient ways from those inrural areas. For example, many characters on popular soap operashave more education, marry later, and have smaller families, allthings rarely found in rural areas; and many female characterswork outside the home, sometimes as professionals, running busi-nesses or in other positions of auth

ndia has not been left out of the cable and satellite revolution:a recent survey finds that 112 million households in India owna television, with 61% of those homes having cable or satelliteservice (National Readership Studies Council 2006). This figurerepresents a doubling in cable access in just five years from aprevious survey. The survey finds that in some states, the changehas been even more dramatic; in the span of just ten to fifteenyears since it first became available, cable or satellite penetrationhas reached an astonishing 60% in states such as Tamil Nadu,even though the average income is below the World Bank povertyline of two dollars per person per

Several studies have demonstrated that the information andexposure provided by television can influence a wide range ofattitudes and behavior. Gentzkow and Shapiro (2004) find thattelevision viewership in the Muslim world affects attitudes to-ward the West, and DellaVigna and Kaplan (2007) show largeeffects of the Fox News Channel on voting patterns in the UnitedStates. In the developing world, Olken (2006) shows that televi-sion decreases participation in social organizations in Indonesia,and Chong, Duryea, and La Ferrara (2007) find that exposure tosoap operas in Brazil reduces f

We argue that the results are not driven by preexisting diffe

. Using a three-year, individual-level panel data set, we find that the intro-duction of cable television is associated with significant decreases in the reportedacceptability of domestic violence toward women and son preference, as well asincreases in women's autonomy and decrease

nd behaviors. This paperexplores the effect of the introduction of cable television on women's status in ruralIndi

. These media sources expose viewers to new information about the outsideworld and other ways of life, which may affect attitudes a

We document an important consequence of bride price, a paymentmade by the groom to the bride’s family at marriage. Revisiting Indone-sia’s school construction program, we find that among ethnic groups with-out the custom, it had no effect on girls’ schooling. Among ethnic groupswith the custom, it had large positive effects.

From a policy perspective, our estimates suggest that expanding the hiring of female teachers- both at the margin of the current patterns of hiring (assuming that the marginal female teacherhired has the same characteristics as the average female teacher), and also when holding othercharacteristics constant, would improve overall learning outcomes and be especially useful as atool for bridging gender gaps in learning trajectories over time

This result is interesting because the rhetoric ofhiring female teachers is often based on the belief that having female teachers increases thesafety and comfort of girls in school, and that their presence therefore encourages girls to attendschool. Our results suggest however, that the mechanism for the positive impact of a gendermatch on test scores is less likely to be due to effects on the extensive margin of schoolparticipation, but more due to the increased effectiveness of classroom transactions betweenteachers and students.

Since all test scores are normalized by grade and subject, theestimated coefficients can be directly interpreted as the correlation between the covariate andannual gains in normalized test scores

presents summary statistics for the teachers in our analysis. Femaleteachers comprise 46% of the total teacher body, but are less experienced, less likely to havecompleted high school or a masters degree, and less likely to hold a head-teacher position. Notsurprisingly, their mean salaries are also lower. They also comprise a much greater share of thecontract teacher work-force than that of regular civil-service teachers. Since teachercharacteristics vary systematically by gender, we will report our key results on the impact ofmatching teacher and student gender, both with and without controls for these additional teachercharacteristics

However, this attrition over time is not a first-order concern forthis paper because it is highly unlikely that the additional 3% of female students who appear forthe test each year (relative to boys) would have test scores that are differentially affected byhaving a female teacher

Together, these results suggest that the overall gains from hiring female teachers comemainly from improving mean math test scores relative to male teachers (positive for girls, noeffect for boys) than from language (positive for girls, negative for boys, and no overall effect)

The result above is a difference-in-difference estimate that compares the relative advantageof female teachers in teaching girls rather than boys with the relative disadvantage of maleteachers in teaching girls rather than boys, and is symmetric by construction

Girls have significantly higher test scores inlanguage and equal test scores in math relative to boys at the end of grade one, but score almoston par with boys in language and significantly worse in math by the end of grade five. Theseresults are consistent with evidence of gender gaps in test scores (particularly in math)documented in both high and low income countries (Fryer and Levitt 2010, Bharadwaj et al.2012), and suggest that the growing gender gaps documented at later ages in both these papersprobably reflect a cumulative effect of a trend that starts as early as primary schoo

he dataset features annual longitudinal data on student learning measuredthrough independent assessments of learning conducted over five years across a representativesample of 500 rural schools and over 90,000 students in the Indian state of Andhra Pradesh

(6) Finally, the increasing probability of having a male teacher inhigher grades can account for around 10-20% of the negative trend we find in girls' test scores as theymove to higher grades.

(5) We find no effect of having a same-genderteacher on student attendance, suggesting that the mechanism for the impact on learning outcomesis not on the extensive margin of increased school participation, but on the intensive margin of moreeffective classroom interactions;

(4) The overall gainsfrom having a female teacher are mainly attributable to their greater effectiveness at improving mathtest scores than male teachers (especially for girls);

(3) However, female teachers are more effective overall,resulting in girls' test scores improving by an additional 0.036 in years when they are taught by a femaleteacher, with no adverse effects on boys when they are taught by female teachers;

We finda small but significant negative trend in girls' test scores in both math (0.02/year) and language (0.01/year)as they progress through the public primary school system; (2) Using five years of panel data, school-gradeand student gender by grade fixed effects, we find that both male and female teachers are more effectiveat teaching students of their own gender;

Recruiting female teachers is frequently suggested as a policy option for improving girls' educationoutcomes in developing countries, but there is surprisingly little evidence on the effectiveness of sucha policy. We study gender gaps in learning outcomes, and the effectiveness of female teachers in reducingthese gaps using a large, representative, annual panel data set on learning outcomes in rural publicschools in the Indian state of Andhra Pradesh.

But there is considerable slack to im-prove institutions and policy at the margin

The focus on thebroad INSTITUTIONS as a necessary and sufficientcondition for anything good to happen is somewhatmisplaced.

This result suggests that a credible message canconvince the voters to vote in favor of general-in-terest policies. Once the trust is there, the individualpolitician’s incentives also change. He can start tofeel that if he does something good he will be appre-ciated and reelected.

Invillages that had never had reserved seats for wo-men, and therefore had no experience of a womanleader, men who heard the “male” speech gave high-er approval ratings than those who heard the “fe-male” speech. On the other hand, in villages thathad been reserved for women before, men tended tolike the “female” speech better.

The nurses’ workload was based on an ideologythat wants to see nurses as dedicated social workers,designed in ignorance of the conditions on theground, that lives on, mostly just on paper, becaus

This is whatwe call, for short, the “three Is” problem: ideology,ignorance, inertia. This problem plagues many ef-forts to supposedly help the poor

In our visits to the village, we couldhardly find anyone to acknowledge that the nurseswere absen

To see whether this new policy made a difference,we sent independent survey researchers to recordabsence both in the centers that Seva Mandir wasmonitoring and in the other centers (where thesame rules applied in principle but where there wasno monitoring). 34 Initially, everything went accord-ing to plan. Nurse attendance, which was around 30percent before the launch of the program, jumpedto 60 percent by August 2006 in centers where SevaMandir was monitoring, but it remained unchangedelsewhere.

The risk of corruption and neglect is thus endem-ic in any government, but it is likely to be moresevere in three circumstances:

Governments exist toa large extent to solve problems that markets cannotsolve—we have already seen that in many instancesgovernment intervention is necessary preciselywhen, for some reason, the free market cannot dothe job.

hepolicy, as we have reported, led to an increase ineducation and, for the generations that benefitedfrom this schooling, an increase in wages. The edu-cation expansion was accompanied by a massiveprogram promoting better nutritional practices forchildren, in part by training 1 million village volun-teers who were supposed to bring the message totheir villages.

The import-ant lesson is to take advantage of whatever slackthere is

Political economy is the view (embraced, as we haveseen, by a number of development scholars) thatpolitics has primacy over economics: Institutionsdefine and limit the scope of economic policy

In the “clientelist” villages, thespeech stressed the ethnic origin of the candidateand promised to bring schools and hospitals to theregion and government jobs to his people. In the“national unity” villages, the same candidate prom-ised to work for a national reform of the health andeducation sector and to work for peace among allthe ethnic groups of Benin. The villages were ran-domly chosen to get different speeches, but all ofthem were in the candidate’s political stronghold.The clientelist speech was a clear winner: On aver-age, the clientelist candidate got 80 percent of thevotes, as opposed to 70 percent in the national unityvillages.

Further studies elsewhere in India have made itclear that women leaders almost always make a dif-ference. Furthermore, over time, women also ap-pear to be doing more than men with the same lim-ited budget and are reported to be less inclined totake bribes.

India’s system of village government, or grampanchayat (the GP, or village council), has such re-strictions.

The so-called Community Driven Development pro-jects, in which the communities choose and managecollective projects, are quite the rage in post-conflictenvironments like those in Sierra Leone, Rwanda,Liberia, and Indonesia

In particular, there was an increase inpublic health expenditures and a reduction in thenumber of low-birth-weight babies among less edu-cated mothers. A seemingly minor technical fix, in-volving no major political battle, changed the way inwhich the voice of the poor was taken into accountin Brazil’s political process

of a large num-ber of voters. In the average election, almost 25 per-cent of the votes were invalid and not counted. Inthe late 1990s, electronic voting was introduced, atfirst in the largest municipalities, and then in all ofthem

no guarantee that good INSTITUTIONSnecessarily work well in practice.

Once thisis done, it will be possible, for example, to requirepeople to scan their fingerprints to take delivery ofany subsidized grains from the government fair-price shops. This will make it much harder for theshop owners to sell off the grains at market pricesand claim that they have sold to the poor.

Indian police stations are evaluated on the basis ofthe number of unsolved cases, that is, the more un-resolved cases, the worse the evaluation. Therefore,an easy way to get better evaluations is to register asfew cases as possible. In our first set of decoy visits,only 40 percent of the cases actually got to the pointwhere the police were willing to register them (atwhich point our decoys were required to reveal thatit was just a test). It is therefore no surprise that thepoor rarely attempt to report any petty crimes to thepolice.The police in India represent a near-perfect ex-ample of a persistent colonial institution. Despitethe fact that they were originally designed to protectthe interests of the colonists, there was no attemptto reform the Indian police after Independence

Yet, Olkenshowed that the threat of audits reduced the theft ofwages and materials by one-third, compared to thevillages where audits were not conducted (the vil-lages where audits were conducted were randomlyselected).

After a villagestarts holding elections, the village chiefs are morelikely to relax unpopular central policies, such asthe one-child policy.

donesiaunder Suharto, Brazil during the military dictator-ship, and Mexico under the Institutional Revolu-tionary Party (PRI)

CHANGES AT THE MARGIN

are very persistent, and there may be no naturalprocess to eliminate them.

Easterly also wantsgovernments to stop pushing education and healthcare on an indifferent populace but rather allowthem the freedom to find ways to get themselveseducated and healthy, through their own collectiveaction

Easterly has one piece of ex-pert advice—freedom.

William Easterly. 12 The centralproblem, he rightly points out, is that it is easier totake over a country than to know how to make it runwell.

Paul Collier, an OxfordUniversity professor and former World Bank eco-nomist, argues that there are sixty “basket case”countries (think Chad, Congo, and so forth) inwhich about 1 billion people live. 11 These countriesare stuck in a vicious circle of bad economic and badpolitical institutions, and it is the duty of theWestern world to get them out, if necessary throughmilitary interventions.

If you can-not run your country, subcontract it to someonewho can. 10 Still, running an entire country may bedifficult. So he proposes starting with cities, smallenough to be manageable but large enough to makea difference. Inspired by the example of Hong Kong,developed with great success by the British and thenhanded back to China, he developed the concept of“charter cities.” Countries would hand over anempty strip of territory to a foreign power, whowould then take the responsibility for developing anew city with good institutions

incumbent on the rich countries of the world to helpthem get better institutions, by force, if need be. Inthe other view, any attempt at manipulating institu-tions or policies from the top down is doomed tofail, and changes can only come from within

The ex-amples they emphasize are the Glorious Revolutionin England and the French Revolution.

were individually responsible for their own taxes:These regions developed more cooperative and ho-rizontal social relationships.

“iron law of oligarchy.”

Acemoglu andRobinson define institutions as follows: “Economicinstitutions shape economic incentives, the incent-ives to become educated, to save and invest, to in-novate and adopt new technologies, and so on.Political institutions determine the ability of citizensto control politicians.

Corruption, or the simple dereliction of duty, cre-ates massive inefficiencies. If teachers or nurses donot come to work, no education or health policy canreally be implemented. If truck drivers can pay asmall bribe to drive massively overloaded trucks,billions of dollars will be wasted in building roadsthat will be destroyed under their wheels.

If rural school headmasters could fightcorruption, perhaps it is not necessary to wait forthe overthrow of the government or the profoundtransformation of society before better policies canbe implemented. Careful thinking and rigorousevaluations can help us design systems to keep cor-ruption and inefficiency in check. We are not“lowering our ambitions”: Incremental progress andthe accumulation of these small changes, we be-lieve, can sometimes end in a quiet revolution

However, there was an interesting coda to Rein-ikka and Svensson’s story. When their results werereleased in Uganda, there was something of an up-roar, with the result that the Ministry of Financestarted giving the main national newspapers (andtheir local-language editions) month-by-month in-formation about how much money had been sent tothe districts for the schools. By 2001, when Rein-ikka and Svensson repeated their school survey,they found the schools were getting, on average, 80percent of the discretionary money that they wereentitled to.

Jeffrey Sachs sees corrup-tion, perhaps not surprisingly, as a poverty trap:Poverty causes corruption, and corruption causespoverty.

There is no point to figuringout the best way to spend a dollar on schools, if 87cents will never reach the school anyway. It follows(or so it is assumed) that “big questions” require“big answers”—social revolutions, such as a trans-ition to effective democracy

It is to sort out the political process. If

William Easterly, for one, criticized ran-domized control trials (RCTs) on his blog in theseterms: “RCTs are infeasible for many of the bigquestions in development, like the economy-wideeffects of good institutions or good macroeconomicpolicies.” Then, he concluded that “embracing RCTshas led development researchers to lower their am-bitions.” 3

The answer they got was noth-ing short of stunning: Only 13 percent of the fundsever reached the schools. More than half the schoolsgot nothing at all. Inquiries suggested that a lot ofthe money most likely ended up in the pockets ofdistrict officials

In 1996, Ritva Reinikka andJakob Svensson

Unfortunately, the gapbetween intention and implementation can be quitewide.

he task, for researchers, is to expand the evi-dence base for exactly which democratic reformsare the most effective in providing the poor witha productive political voice, so that all people—the poor, insiders, and outsiders—can best takeadvantage of opportunities to effect change whenthey appear.

Berliner and Erlichshow that political competition, by creating un-certainty over future political control, gaveMexican political parties incentives to undertaketransparency reforms that “serve as insurancemechanisms enabling ruling groups to protecttheir access to government information, and topreserve means of monitoring future incum-bents, in case they lose power” (38)

uring the19th century, the United Kingdom (like manyWestern countries) extended the vote from wealthylandowners to all males. Lizzeri and Persicoargue that British elites broadened the franchisebecause it better aligned citizen and politicianpolicy preferences, increased the electoral valueof policies with diffuse benefits, and reduced thereturns to politicians from clientelistic policies(32).

(i) the poor and disadvantaged who would ben-efit from it, (ii) the economic and political elitewho control the levers of power from withinthe system, and (iii) actors, such as internationaldevelopment and human rights agencies, whoinfluence reform from the outside.

One way to resolve the dilemma, inspired bythe modernization hypothesis, is to focus onpolicies that affect growth. The main argu-ment of the modernization hypothesis is thatthe social transformations which accompanyeconomic development create more politicallyaware citizens and, thereby, the conditions fordemocratic development (29). Yet, at the levelof cross-country analysis, evidence supportingthis hypothesis remains weak (30)

Fig. 2. What citizens consider to be an essential characteristic of democracy.

Kananga’s governmentrandomized property tax collection across its431 neighborhoods. In taxed neighborhoods within-person collection, collectors went door-to-doorregistering households and collecting the ap-proximately $2 property tax. In control neigh-borhoods, citizens were left to voluntarily payat the tax ministry. All citizens were encouragedto attend government-hosted town hall meetings,where officials and citizens discussed tax andpublic spending in Kananga, and to submit anon-ymous evaluations of the provincial government.The campaign increased the probability ofvisits from tax collectors by 81.5 percentagepoints (from 0.05% in control) and increasedproperty tax compliance by 11.5 percentagepoints (from 0.001% in control)

Along these lines, Olken reports 20 timeshigher citizen participation and greater citi-zen satisfaction when local village projects inIndonesia were chosen by direct plebisciterather than representative village meetings (28).

Banerjee et al. investigate this question di-rectly and at scale, in the context of the citygovernment of Delhi, where roughly a third ofthe population of 20 million live in slums (25).Delhi’s elected councilors legislate on how toredistribute state resources, and they also haveaccess to discretionary funds that can be spenton infrastructure. Survey data show a drasticmismatch between councilor spending andcitizen preferences: Although sanitation wasa priority for most slum-dwellers, most moneywent toward road construction. Against thisbackdrop, a random sample of councilors wereinformed 2 years prior to city elections that aleading newspaper would publish report cardson their performance just before the city elec-tion. The informed councilors subsequentlymoved their spending in a pro-poor direction,a move that was rewarded by political parties.

idwell et al. conducteda large-scale experiment during the 2012 par-liamentary elections in Sierra Leone, where theyrandomized citizen exposure to pre-electioncandidate debates hosted and screened by athird party (24). Watching debates increasedpolitical knowledge, improved voter-candidatealignment, and increased both the number ofvotes cast and vote shares of the best-performingcandidates.

Using data from Brazilian municipalities,Ferraz and Finan showed that the public re-lease of audit reports lowered reelection ratesof mayors from more corrupt municipalities,with more pronounced effects observed inmunicipalities with a local radio station (23)

This,in turn, was associated with a 34% increasein public health care spending over an 8-yearperiod.

Universal suffrage has been effectively ubiq-uitous in democracies and near-democraciessince 1980, with a handful of exceptions (1). Butcomplex de facto voting procedures often con-strain the ability of the poor to exercise their vote.Using technology to make voting procedures moreaccessible to the less educated can strengthentheir political voice. For example, Brazil historical-ly used a paper ballot system under which it wascommon for more than a quarter of the votes tobe deemed invalid, that is, either blank or error-ridden. In 1998, electronic voting devices wereintroduced to municipalities above a certain pop-ulation threshold, and the electronic interfacewas universally adopted in 2002. By providingvoters step-by-step guidance and introducingvisual aids, this technology facilitated voting byless literate citizens.

A growing body of empirical political economypapers use naturally random events or field ex-periments to examine which reforms to demo-cratic institutions can achieve these goals

A direct implication is that, for the poor, ef-fective democracy requires more than regularlyoccurring elections. It also requires democraticinstitutions that successfully disengage an indi-vidual’s ability to freely exercise her electoralrights from her economic power.

In economics, each of these situa-tions embodies a principal-agent problem, inwhich one group (principals) delegates policyimplementation to another group (agents) insettings typified by incomplete information andvarying incentives (19). When the principal hasless information than the agent and the twoparties face differing incentives, the agent candeviate from prescribed actions and, instead,make personally beneficial choices

lientel-istic policies

The political voice afforded the extreme poor indemocracies is typically constrained by eco-nomic and social disadvantage. In many settings,the economic elite can exploit their social con-nections and economic power to provide poo

In practice, however, both the rules and theirimplementation depend on how well the stateaccommodates interests of groups with vary-ing economic and political power.

But fixing the state’s plumbing need not im-ply progressive redistribution. In reality, a stateconsists of groups of actors with often mis-aligned interests—and the state machinery thathas been built for service delivery may not, inpractice, deliver for the poor.

Duflo has argued for a problem-by-problemapproach: Economists should embrace the roleof “plumbers” and identify ways of improvingthe so-called “plumbing” of the state machinery(8).

nd yet the opposite is true. Sowhy are low- and middle-income democraticstates less responsive to the preferences of theirpoorer citizens? One possible reason is a state’slimited capacity to deliver antipoverty programs.The state may want to redistribute resourcesyet finds itself hobbled by low revenues, lim-ited manpower, and, potentially, a lack of tech-nical know-how (7

If the poor lack the means...to use theirde jure power effectively, then they will failto negotiate a fair share of...growth.”

Equally poor citizens in middle-incomecountries do not benefit from the same support.This is in part because citizens of rich countriesare reluctant to donate to countries with obviousconcentrations of wealth (a political constraintthat results in, for example, the country incomecriteria that underlie grant and loan program-ming by most bilateral or multilateral aid agen-cies), and so international aid is far less likely totarget the extreme poor in these countries (2).

How can the poor securegreater democratic influence when the existing democratic playing field is tilted against them

However, citizens in those countries have not succeeded in using the vote toensure adequate progressive redistribution. Interventions aiming to provide the economically vulnerable withneeded resources must go beyond assisting them directly, they must also improve democratic institutions so thatvulnerable populations themselves can push their representatives to implement redistributive policies

Millions of the world’s poorest people now live in middle-income democracies that, in theory, could use theirresources to end extreme poverty.

clear indicators of regime legitimacy as opposed to government legitimacy:citizens’ express support for the system as a whole; the degree to which theysupport key institutions of liberal democracy, such as civil rights; their will-ingness to advance their political causes within the existing political sys-tem; and their openness to authoritarian alternatives such as military rule

The Census Bureau raises the poverty line every year by the same percentage as the increase in the Consumer Price Index for All Urban Consumers (CPI–U). The CPI–U does not try to measure changes in what people need, or why they think they need it. Nor does it measure changes in what Americans mean when they talk about poverty. The CPI–U just measures changes in prices. But if we want to know how close America has come to eliminating the kind of poverty that existed in 1964, the official measure is supposed to provide an answer. However, the world has also changed in other ways that the official poverty count either ignores or mismeasures. As a result, Figure 1 does not actually tell us much about changes in the kind of poverty that Lyndon Johnson promised to eliminate. Four changes are especially important when we try to measure changes in the poverty rate since 1964.

The poverty rate could havedeclined despite the War on Poverty, not because of it.

We observe that the T2 × gender interaction is signifi-cant at p < 0005 in both models. This provides evidencethat women are not more but are less likely than mento sort out of workplaces that explicitly offer nego-tiable wages

Second, interestingly, the gender gap in applicationprobabilities is more pronounced in T1 than in T2.In T1, women’s application probability is 31.9% andmen’s application probability is 46.6%. In contrast,in T2, women’s application probability is slightlyhigher (33.0%; p = 00630, Fisher’s exact test), whereasmen’s application probability is lower (41.2%; p = 00133,Fisher’s exact test). Thus, the gender gap almost halves(from 14.7% to 8.2%) when moving from T1 to T2

Figure 1 illustrates the application probabilities depend-ing on treatment and gender. First, we observe thatwomen are overall less likely to apply than men (32.5%versus 43.8%; Fisher’s exact test, p < 00001), a generalpattern that is also reported in Flory et al. (2015).

Baby Name Guesser,available at http://www.gpeters.com/names/baby-names.php, accessed December 2011), which calcu-lates gender ratios by first name using the Inter-net to analyze patterns of name usage for morethan 100,000 first names.

hat we advertised, women earned 84.3% ofthe median weekly earnings of their male counterparts(for high-skilled occupations, women earned 80.1% of2 See Online Appendix B for more information on posted wages.(Online appendices available at https://sites.google.com/site/leibbrandt/research.)the median weekly earnings of their male counterparts).In addition, Maxwell (2006) finds that 90% of low-skilled jobs have promotional opportunities. Second,we were able to offer real administrative assistantpositions (and afford to pay salaries). Third, the tasksfor these jobs could be performed at the employees’homes, which rendered it possible to advertise thepositions in different cities and hire individuals fromdifferent cities. One drawback of using administrativeassistant positions is that they are mainly occupied bywomen (79%) and thus may make extrapolations toother jobs difficult where the gender distribution issignificantly different.

To carry out the 2 × 2 design, we employ a two-stageexperimental method. In the first stage, we adver-tise the position without reference to the contractenvironment. In the second stage, after job seekersexpress interest in the position, we inform them ofthe contract environment and record whether theyultimately choose to apply for the job. These two stepsare important because they allow us to randomizecontract environment across subjects without affectingthe normalcy of the field setting

Our second result is based on sorting into jobs: thegender gap in applications is much more pronouncedfor jobs that leave the negotiation of wage ambiguous.Interestingly, this result is driven by men preferringjobs where negotiation of initial wage is ambiguousrather than when negotiations are expected.

A first result is that we findthat when there is no explicit statement that wages arenegotiable, women are more likely to signal their will-ingness to work for a lower wage than advertised andslightly less likely to negotiate a higher wage than men.However, when we explicitly mention the possibilitythat wages are negotiable, these gender differencesvanish completely, suggesting that women react morestrongly to simple changes in the wage description

This investi-gation permits us to compare two quite different, butrelevant, situations: one that is ambiguously structuredso that people arrive at their own interpretations of thesituation and one where the negotiations are muchmore concretely structured. We proceed to investigatethe importance of triggering gender by placing twodistinct ads: one for a “masculine” job task (involvingsports) and the other for a gender-neutral job.

is its capacity to disentangle twoimportant levels where gender differences in negotia-tion might play a role: (i) sorting into negotiable-salaryworkplaces and (ii) initiation of salary negotiations oncethe person applies.

Finally, Babcock and Laschever (2003) report surveyevidence that shows men are four times more likely tonegotiate on first salaries and that individuals who donot negotiate first salaries lose more than $500,000 byage 60.

There is important laboratory and survey evidencesuggesting a quite different determinant of genderdifferences in labor markets: men are significantly morelikely to engage in salary negotiations and are better atthem

However, when we explicitly mention the possibility that wages are negotiable, these differences disappearcompletely. In terms of sorting, we find that men, in contrast to women, prefer job environments where the “rulesof wage determination” are ambiguous.

tural field experiment that randomizes nearly 2,500 job seekers into jobs thatvary important details of the labor contract, we are able to observe both the extent of salary negotiations and thenature of sorting. We find that when there is no explicit statement that wages are negotiable, men are more likelyto negotiate for a higher wage, whereas women are more likely to signal their willingness to work for a lowerwage.

We should, however, caution the reader that eventhough we find suggestive results, care should be taken when making inferencefrom the data patterns observed herein because several important factors varyacross the two societies. And, we have sampled a limited number of villages.We suspect that our results will not be a universal truth amongst all matrilinealvillages, rather other important factors will interact with matriliny to producethe data patterns observed herein. More research is certainly warranted.

Maasai men com-pete at roughly twice the rate as Maasai women, evidence that is consistentwith data from Western societies that use different tasks and smaller relativestake levels. Yet, this data pattern is reversed among the Khasi, where womenchoose the competitive environment more often than Khasi men

The importance of these results should not be under-stated: in both a positive and normative sense, these insights have the potentialto explain important puzzles in economics and in social science more generally.In this study we use an experimental task to explore whether there are genderdifferences in selecting into competitive environments across two distinct soci-eties: the Maasai in Tanzania and the Khasi in India. The societies are uniquein that the Maasai represent an example of a patriarchal society, whereas theKhasi are matrilineal

n other words, the cultural choice to displace men or women fromtheir maternal homes, by itself, alters the process of genetic inheritance.

Girlswho imitate the behavior of successful competitive women are more likely tosurvive childhood and will inherit greater wealth if those women are also theirmothers or grandmothers. In turn, their wealth and success make them morepreferable as a model for younger girls (likely their daughters and nieces) andmore likely to have surviving children

The study of gene-culture co-evolution in mathematical modeling suggests thatwhen a particular genetic characteristic favors the transmission of a particularcultural feature, and that cultural feature also increases the fecundity of thegenetic characteristic, evolution can occur at a much faster pac

Our data show that Khasi women are more likely to choose to compete thanKhasi men. Furthermore, the Khasi women compete more often than Maasaiwomen or any group of women in the various settings in which preferencesfor this type of behavior have been elicited. In the very least, these findingsrepresent existence results: it is not universally true that the average femalein every society avoids competition more often than the average male in thatsociety because we have discovered at least one setting in which this is not true.

Overall, if subjects were making inferenceon potential competitors based on the mix they observed, then the effects re-ported in Table IV are consonant with the notion that women are more likelyto compete against other women, especially in matrilineal societies.

More importantly for our purposes, roughly half ofthe Khasi subjects opted to compete, whereas only 39 percent of the Maasaichose to compete. When broken down by gender, the underlying force behindthe competitiveness differences across the two societies becomes clear.In the Maasai data, the gender result that we oftentimes observe in the lit-erature is evident: whereas 50 percent of men choose to compete, only 26 per-cent of women select to compete. Alternatively, as Figure 1 highlights, Khasiwomen choose to compete more often than Khasi men—whereas 54 percent ofKhasi women choose to compete, only 39 percent of Khasi men select the com-petitive incentive scheme.

The experimental task was to toss a tennis ball into a bucket that wasplaced 3 meters away. Participants were informed that they had 10 chances.A successful shot meant that the tennis ball entered the bucket and stayedthere. The task was chosen because it was simple to explain and implement,and no gender differences in ability were expected (as was found in a pilotexperiment and reinforced in the results discussed below). Furthermore, weare aware of no other popular task in these societies that is similar to the ballgames that we implemented.

This attenuated selec-tion problems since everyone was interested in participating in the experimentafter they were made aware of the pecuniary incentives involved. The experi-ment with the Maasai was conducted in two villages in the Arumeru district inthe Arusha region of Tanzania. The experiment with the Khasi was conductedin the Meghalaya region of India. Upon arrival at each experimental site, par-ticipants were directed into one of two groups randomly. These groups wereseparated for the entire experiment

Khasi men are more important in their sisters’ households thanin their wives’ households, and Maasai women can enjoy prestige and powerin their roles as widows (if they have sons).

Age and cattle dominate the Maasai social structure. The most importantdistinctions between men are age-based, and almost all wealth is in cattle. Theage structure prevents men from marrying until they are roughly 30 years oldand polygamy is the most common form of marriage.

Though Khasi women do not generally assume the roles held by men in pa-triarchal societies (they do not become warriors or hunters, for example), theyalways live in households in which they or their mothers have authority overmost household decisions. On the other hand, men frequently hold roles thatseem to mirror those of women in patriarchal societies. The Khasi husbanddwells in a household in which he has no authority or property, is expected towork for the gain of his wife’s family, and has no social roles deemed impor-tant.

In addition, even ordinalclassification of societies on any dimension is dangerous, as culture and societyare not static fixtures handed down from prehistory. Certain reports of extremefemale domination in the Khasi or strong male domination among the Maasaiare somewhat exaggerated and subject to charges of ethnocentrism.

Our experimental results reveal interesting differences in competitiveness:in the patriarchal society, women are less competitive than men, a result con-sistent with student data drawn from Western cultures. Yet, this result reversesin the matrilineal society, where we find that women are more competitive thanmen.

Nevertheless, the role of nurture, or the fact that culture might be criticallylinked to competitive inclinations, is also an important consideration. Morethan a handful of our male readership can likely recall vividly their grammarschool physical education teacher scorning them with the proverbial “you’replaying like a girl” rant to induce greater levels of competitive spirit. Clearly,however, the explanations might not be competing; rather the nature–nurtureinteraction might be of utmost importance, either because nurture enables theexpression of nature

We use a controlled experiment to explore whether there are gender differencesin selecting into competitive environments across two distinct societies: the Maasaiin Tanzania and the Khasi in India.

However, in thispaper we have examined an environment where women and menperform equally well and where issues of discrimination or timespent on the job do not have any explanatory power. Nonethelesswe find large gender differences in the propensity to choose com-petitive environments. It appears that these differences aredriven by gender differences in confidence and preferences forentering and performing in a competition. These differences seemsufficiently strong to call for greater attention of standard eco-nomics to explanations of gender differences that so far havemostly been left in the hands of psychologists and sociologists.

Overall, we find that about 57 percent of the original gendereffect can be accounted for by general differences in overconfi-dence and risk and feedback aversion while the residual “compet-itive” component is 43 percent. This makes clear that the gendergap in choice of compensation scheme is exacerbated when indi-viduals subsequently have to perform under the selected compen-sation scheme.

An ordered probit of the guessed rank as a function of afemale dummy and performance shows that, conditional on per-formance, women are significantly less confident about their rel-ative ranking than men and that participants with a highertournament performance think they have higher relative perfor-mance (see Table V).1

A probit regression reveals that while the participant’s per-formance under the two compensation schemes does not signifi-cantly affect the decision to enter the tournament, the partici-pant’s gender does.

both genders perform significantlybetter under the tournament than the piece rate (one-sided p .01 for each gender separately).

As expected we find no gender difference in performanceunder the piece rate or under the tournament. In the piecerate, the average number of problems solved is 10.15 forwomen and 10.68 for men. Using a two-sided t-test, this dif-ference is not significant (p
.459). The gender difference inperformance is also not significant in the tournament where,on average, women correctly solve 11.8 problems, and men 12.1(p
.643). Throughout the paper the reported test statisticsrefer to a two-sided t-test, unless otherwise noted. The conclu-sions of the reported t-test do not differ from those of a Mann–Whitney test.

e selected this five-minute addition task because itrequires both skill and effort and because research suggests thatthere are no gender differences in ability on easy math tests

First, there is evidence that women tend to incorporate negativefeedback more than men (see, e.g., Roberts and Nolen-Hoeksema[1989]). Second, women, more than men, may view a negativesignal as indicative of their self-worth rather than simply theirone-time performance on a task. Women may therefore fall into“confidence traps” from which they do not recover easily (see, e.g.,Dweck [2000] and references therein).

Explanation 4: Men enter the tournament more than womenbecause they are less averse to feedback.