i have this paper and i just want a professional reader just look at the paper and make 3 to 4 pages write about your opinion about the paper .

Contraceptive Supply and Fertility Outcomes: Evidence from Ghana

kelly m. jones International Food Policy Research Institute

I. Introduction Sub-Saharan Africa has the highest fertility rates in the world, lagging behind every other region in terms of demographic transition. This region also has a significant unmet need for family planning, with fertility rates outpacing wanted fertility by 24%. Unwanted fertility is highest among the poorest and least educated women, suggesting that achieving target fertility has significant implications for the welfare and productivity of the next generation of Afri- cans.1 Such data suggest that family-planning programs may play a key role in African development. Yet the degree to which contraceptive access actually affects fertility out-

comes is a matter of considerable debate. Some have posited that fertility is largely driven by preferences and that access to family planning has only very small impacts ðPritchett 1994; Miller 2010Þ. Others have shown large pro- gram impacts or have argued that a seemingly small change in fertility can rep- resent a significant change in unwanted fertility ðPhillips et al. 1982; Bongaarts 1994; Debpuur et al. 2002Þ. In sub-Saharan Africa, 80% of contraceptive supplies are provided by in-

ternational population assistance ðRoss, Weissman, and Stover 2009Þ. Such funds are not always reliable, as they are subject to the whims of donors, as in the case presented here. Further, the debate continues regarding whether in-

The author gratefully acknowledges funding from the National Science Foundation’s Doctoral Dissertation Research Improvement Grant, the William and Flora Hewlett Foundation’s Fellowship in Population, Reproductive Health and Economic Development, the UC Berkeley Population Center ðNICHD R21 HD056581Þ, and the Center of Evaluation for Global Action. Thanks are owed to Elisabeth Sadoulet, Alain de Janvry, Jeremy Magruder, Ted Miguel, Pascaline Dupas, Craig McIntosh, Erick Gong, Gil Shapira, Melissa Hidrobo, and participants of the PopPov research conference. Research assistance was provided by Aviva Lipkowitz. All remaining errors are my own. Contact the author at k.jones@cgiar.org. 1 On the basis of the most recent Demographic and Health Surveys from 41 countries in this region, average wanted fertility is 4.1 children per woman, whereas average realized fertility is 5.1 children per woman. Excess fertility, defined as the difference between wanted and realized fertility, is 1.3 children for the poorest ðand for those with no educationÞ and 0.6 children for the least poor ðand those with secondary or higher education; MEASURE-DHS 2013Þ. Electronically published August 4, 2015 © 2015 by The University of Chicago. All rights reserved. 0013-0079/2015/6401-0006$10.00

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

ternational aid on the whole should be dramatically increased or should be scaled back, due to detrimental impacts and aid dependence ðMoyo 2009; Easterly and Williamson 2011Þ. For these reasons, and because poor countries struggle with how to allocate their own funds, it is useful to understand the potential impacts ðor lack thereofÞ of changes in funding for contraceptive provision. If we assume that a woman has a desired fertility target, use of modern

contraceptives is only one way of achieving that target. Traditional methods may also prevent pregnancy.2 In some contexts, induced abortion can also be used to prevent unwanted births. It is thus unknown the degree to which supply-side changes in contraceptive availability will directly translate into changes in fertility. A women may adjust her use of these other options in order to compensate for insufficient supply, leaving fertility unchanged. Or, if women are unable or unwilling to do so, fertility outcomes may respond to supply. Such behavioral questions are interesting theoretically but also have significant implications for designing policies to achieve fertility transition and long-term development in Africa. Equally interesting are questions regarding how such behavioral responses

will differ depending on different demographic characteristics. What factors affect the degree to which a woman will engage in compensating behavior and thereby achieve her target regardless of supply? Are educated women better at this? Are rural women less willing or able to do so? Answers to these questions indicate which groups are most affected by supply-side factors of contracep- tion provision. For example, Pop-Eleches ð2010Þ finds that a drastic increase in the availability of modern contraceptives and abortion in Romania decreased fertility most for the least educated women. To the degree that demographic factors matter for behavioral response, a link may be drawn between contra- ceptive supply and the characteristics of resulting fertility. This is the first study, to my knowledge, that examines the individual-level behavioral response to broad-based, supply-side changes in contraceptives in the context of sub- Saharan Africa.3

2 Modern contraception includes hormonal methods ðe.g., pill, implant, injection, IUDÞ, barrier methods ðe.g., condom, diaphragm, sponge, spermicideÞ, and permanent methods ði.e., steriliza- tionÞ. Traditional methods include abstinence, withdrawal, rhythm, lactational amenorrhea, and other folkloric methods. 3 It is noted that several US-based studies have examined the impacts of access to contraception on a variety of outcomes ðBailey 2006, 2010; Bailey, Hershbein, and Miller 2012; Myers 2012Þ. Other studies based in the United States or Romania have examined the impact of access to abortion ðCur- rie, Nixon, and Cole 1996; Kane and Staiger 1996; Cook 1999; Gruber, Levine, and Staiger 1999; Pop-Eleches 2006; Ananat, Gruber, and Levine 2007; Mitrut and Wolff 2011Þ or combined access to contraception and abortion ðGuldi 2008; Pop-Eleches 2010Þ.

32 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

This article exploits exogenous reductions in contraceptive availability to ex- amine women’s behavioral responses in terms of fertility decisions and outcomes in the context of sub-Saharan Africa. The reductions resulted from significant cuts in US funding to foreign nongovernmental organizations ðNGOsÞ that provide contraceptives. Consistently, the United States is the largest provider of such international population assistance ðAshford 2010Þ. Much of this assis- tance flows to NGOs who are primary providers of rural outreach for family planning in many poor countries ðTurnbull and Bogecho 2003Þ. The funding cuts were a result of domestic politics in the United States, discussed in detail in Section IV, and were in place during 1984–92 and 2001–8. The context of the analysis is Ghana, on the basis of the availability of

individual-level data on induced abortion, which are necessary for examining women’s decisions in the face of supply changes. I first provide evidence that contraceptive availability was reduced during policy periods and that con- traceptive use was affected as well, particularly in rural areas. I then create a woman-by-month panel to examine the impact of these changes on individual fertility outcomes. Estimating within woman, I find that conception signifi- cantly increased during policy periods for rural but not urban women. I also find an increase in the use of induced abortion in rural areas, but only for women in the top three wealth quintiles. The poorest women were either unable or unwilling to offset the increase in pregnancies with induced abor- tion. As a result these women experience increases in fertility that did not occur among other groups. The next section reviews the debate on the magnitude of the impact of

contraceptive access on fertility. Section III sets up a conceptual framework that predicts differential impacts across demographic groups. Section IV gives background on the US policy that provides exogenous variation for this study, with a focus on the policy’s repercussions in Ghana. Evidence of reductions in contraceptive availability and use are presented in Section V. Section VI ex- amines impacts on conception and abortion, and Section VII examines changes in resulting fertility. Section VIII discusses the results and their implications and concludes.

II. Existing Evidence Becker ð1993Þ argues that the major changes in fertility have been caused by changes in demand for children, rather than birth control methods. In agreement, Pritchett ð1994Þ claims that a precedent exists for fertility tran- sition without modern contraception. He cites crude birthrates in Europe around 1800 that were lower than the average rate in low-income countries in

Jones 33

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

1990. However, the 1800 rates cited are far from replacement fertility and are in fact comparable to rates in low-income countries today.4

In Romania and Ireland, fertility declined 30% after the legalization of contraceptives. However, abortion legalization in Romania and changing so- cial norms in Ireland were contemporaneous with contraceptive legalization, making it difficult to assign causality ðBloom and Canning 2003; Pop-Eleches 2010Þ. More experimental evidence in the United States suggests that con- traceptive subsidies reduce fertility by a significant, but small, 2% ðKearney and Levine 2009Þ. Experimental evidence on the impact of contraceptive access in low-income

countries begins with the 1978 Family Planning and Health Services program in Matlab, Bangladesh. Sinha ð2005Þ estimates that women exposed to the program had lifetime fertility reduced by 14%. However, it is unclear how much of that impact is attributable to the family-planning aspects of the program, rather than the introduction of other health services ðPhillips et al. 1982Þ. Similarly, in 1993 the quasi-experimental Navrongo project in north- ern Ghana provided a package intervention for community health, result- ing in lower fertility rates by one full birth, comparable to the 14% effect in Matlab ðDebpuur et al. 2002Þ. Smaller impacts are found in Colombia by Miller ð2010Þ, who estimates that the expansion of a large government family- planning program reduced fertility by 6%–7%. Miller deems this a negligible change in total fertility. His view is consistent with Pritchett ð1994Þ, who claims that contraceptive prevalence has a significant but negligibly small ef- fect on realized fertility, as preferences account for 90% of differences across countries’ fertility rates. In response, Bongaarts ð1994Þ argues that reductions in fertility of 5%–10% are not small, but rather, they represent a meaningfully large share of unwanted fertility. Other works based on experiments or natural experiments have investigated

potential barriers to changing fertility with contraceptive use, such as price, access, and cultural factors. McKelvey, Thomas, and Frankenberg ð2012Þ find that, in Indonesia, exogenous variation in prices of contraceptives hardly af- fects use. In the same setting, Molyneaux and Gertler ð2000Þ instrument for family-planning program placement and find that while subsidies lower fer- tility by 3%–6%, expanding the reach of the supply network by 1 standard deviation decreases fertility by 12%, suggesting that access may be a greater limiting factor than price. An experiment by Ashraf, Field, and Lee ð2014Þ provides evidence that a woman is more likely to use contraception and reduce

4 Europe in 1800 averaged 30 births per 1,000 population ðPritchett 1994Þ. Low-income countries in 1990 averaged 41. By 2011, low-income countries’ average rate was 32; OECD countries’ av- erage rate in was 12.2 ðWorld Bank 2013Þ.

34 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

unwanted births when the contraception is concealable from her husband. This suggests that not all forms of contraception may be perfect substitutes, so that fertility may be affected even if changes in access are limited to only a few of many methods.

III. Conceptual Framework Consider a production function for fertility

F 5 f T; M; A; εð Þ; where a woman’s realized fertility, F, is a function of her effort to prevent pregnancy by traditional methods, T ðincluding periodic abstinence, with- drawal, lactational amenorrhea, and folkloric methodsÞ, her use of modern contraception, M, her use of induced abortion, A, and an idiosyncratic term that includes both natural fecundity and random error, ε.5 Variables T, M, and A have negative and unique relationships with F. There are monetary costs associated with M and A, as well as time costs for acquiring them. Further, there are unique utility costs associated with using T, M, and A ðwhich may include partner negotiationÞ. In optimization, a woman will seek to minimize the difference between her realized fertility and her exogenously given target fertility, F*, while minimizing the utility, monetary, and time costs associated with preventing F from exceeding F*.6

This framework suggests that an increase in the monetary or time costs of M or A will spur reoptimization. I consider first the impact of an increase in the costs of M as might occur as a result of reduced funding for some provid- ers of contraceptives. In a neoclassical framework, this reduction in supply would increase prices in the market generally, at least in the short run until other providers entered the market. I examine here the predicted impacts of increases in both time and monetary costs and allow the results to shed light on which, if either, occurred as a result of the policy in question. In the case of an increase in the cost of M, pM, a woman can choose to pay

the increased costs or reduce her use of M, depending on her elasticity of demand for M. If she reduces her use of M, she may choose to increase T or A or both, in order to keep F as close to F* as possible, depending on her utility costs of T, A, and ðF 2 F*Þ and the functional form of f. This offers several predictions in terms of fertility outcomes.

5 In the sub-Saharan Africa context, methods for increasing fecundity, such as in vitro fertilization, are not generally in a woman’s set of options and are thus excluded from this model. 6 This characterization assumes that the woman has sufficient fecundity to meet or exceed her desired fertility.

Jones 35

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Specifically, given an increase in pM, we would expect to see increases in pregnancy for women who meet both of the following criteria: ð1Þ her demand for M is not perfectly inelastic, and she reduces her use of M because either ðaÞ the monetary cost of M increased and she is unwilling/unable to forgo consumption to pay the increased price, due to high marginal utility of consumption ði.e., poor womenÞ, or ðbÞ the time cost of accessing M in- creased due to stock outs or closed locations, and she is unwilling to commit additional time to acquiring M due to high opportunity costs ði.e., employed womenÞ, and ð2Þ she is unable to fully compensate by increasing T, because either ðaÞ her use of T is not effective for preventing pregnancy, due to low knowledge ði.e., uneducated womenÞ, or ðbÞ her utility cost of using T is high ðshe or her partner may oppose abstinence/rhythm, withdrawal, etc.Þ. As a result, we would expect to see increases in realized fertility for women who both ð1Þ experience increased pregnancy and ð2Þ do not fully offset additional pregnancies with A because either ðaÞ she is unwilling/unable to reduce other consumption in order to purchase an increased amount of A ði.e., poorer womenÞ or ðbÞ she has a high utility cost of A, due to traditional values ði.e., uneducated womenÞ. I next consider the implications of a simultaneous increase in the cost of A,

pA. The resulting impact on abortion use will depend on the relative sizes of the own-price and cross-price elasticities of demand for abortion ðhA and hA,M, respectivelyÞ. First, note that hA,M is a function of the own-price elasticity of demand for contraception, hM. That is, the price of contraception only affects abortion use through the increased pregnancy arising from hM > 0. Further, hM, while likely less than 1, is certain to be far from perfectly inelastic: hM determines the decision a nonpregnant woman makes regarding whether to use contraception that will lower her probability of conception in a given month from low ð∼7%Þ to very low. In contrast, hA determines the decision a pregnant woman makes regarding whether to abort a pregnancy rather than have with near certainty a birth she does not want. Within the limits of her resources, this demand is likely to be near to perfectly inelastic. Given the significant differ- ence in the decision environment, it seems hA ≪ hM. Depending on the effi- ciency with which reductions in M translate into pregnancies, this may imply that the response of A to a cross-price change would be greater than the re- sponse to an own-price change. This suggests that an increase in pA may dampen the impact of pM on A, but it is unlikely to offset it completely. The second potential impact of an increase in pA is a cross-price effect on the

use of M. If nonpregnant women are aware of the increase in pA, one might expect an analogous cross-price effect, whereby the increased price of abortion induces women to substitute with effective contraception. If this is the case,

36 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

than the impacts estimated in this work are underestimates of the effect of pM on pregnancy and fertility. The predicted impacts on conception, abortion, and realized fertility by

poverty and education group are summarized in table 8. The data employed here do not include information on employment, and so the those with high opportunity costs due to employment are proxied by nonpoor women with higher education. For these women, predicted impacts on conception are am- biguous, as increased time costs would reduce their use of M due to high op- portunity costs, while effective substitution of T may offset this. However, we can expect the change in fertility to be less than proportional to any change in conception, as there would be at least partial offsetting with abortion. For nonpoor women with low education, we expect no change in conception as they would be likely to continue using M despite increased monetary or time costs. For poor women with low education, increased monetary costs predict an increase in conception and realized fertility, with no offsetting by induced abor- tion. Finally, for the small group of women who are poor but educated, the predictions for conception are ambiguous as potential price effects may be off- set by effective use of T, although we can say that realized fertility will move as conception does, as these women are also unlikely to offset with abortion.

IV. Background on the Mexico City Policy The United States is consistently one of the largest donors of international population assistance worldwide ðUNFPA 2004Þ.7 In 1984, President Reagan issued an executive order that restricted such funding in the following way: “The United States does not consider abortion an acceptable element of fam- ily planning programs and will no longer contribute to those of which it is a part. . . . Moreover, the United States will no longer contribute to separate nongovernmental organizations which perform or actively promote abortion as a method of family planning in other nations” ðWhite House Office of Pol- icy Development 1984Þ. This executive order is known as the Mexico City Policy, on the basis of its introduction at the International Conference on Pop- ulation held in Mexico City in 1984. It requires foreign NGOs to sign official affidavits stating that they will not perform, lobby for, or educate clients about safe abortion. If they refuse, they forfeit any and all population assistance pro- vided by the United States Agency for International Development ðUSAIDÞ.8

7 Population assistance is defined as funding to support the provision of contraception and family planning in foreign nations. 8 At the time of the policy’s creation, and still today, abortion on request is not legal in many countries that receive US population assistance. Further, the 1973 Helms Amendment already forbade the use of

Jones 37

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

A. Global Repercussions Organizations unwilling to sign affidavits were those for which reproductive health and family planning were the foremost objective. Often, these orga- nizations were not performing abortions ðas it is illegal in many poor coun- triesÞ but were providing a considerable share of contraceptive supply. Such groups lost all funding from USAID, amounting to 10%–60% of organizational budgets. This included large international organizations such as International Planned Parenthood Federation ðIPPFÞ and Marie Stopes International, as well as small local NGOs such as the Family Guidance Association of Ethiopia and the Family Planning Association of Kenya ðTurnbull and Bogecho 2003Þ. Funding shortfalls resulting from lost USAID funds took effect in early

1985. The policy remained in effect until it was repealed by President Clinton in January 1993. It was reinstated by President Bush in January 2001.9 Despite many congressional votes on the matter, the policy remained in effect until it was rescinded by President Obama in January 2009.10 In the interim period 1993–2000, when the Mexico City Policy was not in effect, the United States provided nearly 40% of population assistance worldwide ðUNFPA 2004Þ. On average, about half of that funding flowed to NGOs ðPAI 1999Þ. Advocacy groups consistently report that the primary impact of the policy

was to close clinic locations and force cutbacks in outreach services, reducing access to contraceptives primarily in rural areas. A recent evaluation of the policy by Bendavid, Avila, and Miller ð2011Þ examined its impact on abortion use. They employ cross-country data and an algorithm to infer abortions and conclude that women in high-exposure countries increased their use of abortion after the 2001 reimplementation of the policy.11

9 The policy was extended to apply to State Department funds as well in August 2003. 10 For Presidents Clinton, Bush, and Obama, their change to the policy’s effectiveness was issued on the first or second day after inauguration. It has been the concern of several major court battles, one of which ended in the Supreme Court, and at least 20 congressional debates or votes have been taken on the matter ðsee table A2, available online onlyÞ. Its potential reinstatement in 2011 was one of the “policy riders” that created a roadblock in the congressional budget negotiations, nearly shutting down the federal government. 11 My findings are consistent with theirs, although my methodology differs in a number of ways: ðiÞ I focus on one country, rather than many, which allows me to employ data on actual induced abortions, rather than estimating abortions via algorithm; ðiiÞ I estimate within woman rather than across countries; ðiiiÞ I estimate a more comprehensive effect of the policy, including the first three changes in the policy, rather than just the 2001 reimplementation; ðivÞ I show evidence that the pathway of the effect on abortion is indeed increased conception; and ðvÞ I investigate the impact on fertility outcomes more broadly, including conception rates and fertility rates.

US monies for that purpose. Therefore, it was the forbidding of organizations to use their own funds to educate women about safe abortion options or lobby the government for legalization that earned the policy the derisive nickname “the global gag rule.”

38 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

B. Repercussions of the Policy in Ghana Information regarding NGO funding before the 1984 implementation of the Mexico City Policy is not available. However, the situation surrounding the reimposition of the policy in 2001 provides some insight regarding the policy’s effect. USAID documents from late 1999 list funds slated to specific NGOs, by country, for the 2001 fiscal year. The total per country slated to foreign- based reproductive health NGOs represents the funds at risk for loss after the 2001 reimposition of the policy ðsee table A1, available onlineÞ. In Ghana, such organizations were slated to receive nearly $800,000 in FY2001. That is approximately the median for countries receiving such funding, suggesting that Ghana is a fairly representative case ðUSAID 1999Þ. Planned Parenthood Association of Ghana ðPPAGÞ was ðand isÞ the leading

NGO provider of reproductive and sexual health services in Ghana. As of late 1999, PPAG was slated to receive $472,952 from USAID in 2001 ðUSAID 1999Þ.12 Upon the executive order in January 2001, these funds would only be disbursed if the organization agreed to the Mexico City Policy. Under normal circumstances, the majority of funding for PPAG comes

from the IPPF. However, at this time, USAID was funding a large community- based services ðCBSÞ project through PPAG. As such, USAID was slated to provide 1/4 of PPAG’s budget for FY2001. The CBS project was scheduled to run through 2003, and in order to preserve this project, PPAG agreed to the Mexico City Policy to keep its USAID funding ðIPPF 2002; Turnbull and Bogecho 2003Þ. However, from 2001 to 2003 PPAG did experience significant budget

losses, as its funding from IPPF was reduced by 54% ðreducing the total budget by 40%; IPPF 2002Þ. As IPPF had refused to sign the policy, it had experienced budget cuts. Out of necessity, these were passed on to its member organizations.13 In 2003, at the conclusion of the CBS project, PPAG rejected the policy and lost USAID funding ðand in-kind donations of contraceptivesÞ in addition to previous budget cuts from IPPF. Funding from IPPF did not fully recover until after the repeal in 2009 ðsee table A3, available online onlyÞ. Many PPAG clinics were closed or consolidated during this period, especially those in rural areas, as shown in the before-and-after location maps in figure 1.

12 The remainder of at-risk USAID funds slated for NGOs in Ghana were for the Ghana Social Marketing Foundation, which also primarily provides contraceptives. 13 Before the 2001 reimposition of the Mexico City Policy, USAID was providing 7.3% of income for IPPF ð2002Þ. It is not clear why cuts to PPAG were so large relative to IPPF losses. Perhaps IPPF’s funding cuts to member organizations were inversely proportional to the unilateral budget cuts suffered by the member.

Jones 39

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

F ig u re

1 .P

PA G se rv ic e lo ca ti o n s, b e fo re

ðle ft Þa

n d af te r ðri g h tÞ

re im

p le m e n ta ti o n o ft h e p o lic y in

2 0 0 1 .S

h ad

e d ar e as

ar e u rb an

e xt e n ts ,b

as e d o n d at a fr o m

th e G lo b al R u ra l- U rb an

M ap

p in g P ro je ct . C ro ss e s re p re se n t cl in ic

lo ca ti o n s,

p ro vi d e d to

au th o r b y fo rm

e r P PA

G d ir e ct o r.

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Before the policy, clinics were located within urban extents, shown in shading, and well outside. While both rural and urban clinics were shuttered, the majority of closures were in rural areas. Advocacy groups have claimed that the funding losses resulting from this

policy primarily affected the availability of contraceptives to poor, rural pop- ulations, rather than the provision of abortion services ðCincotta and Crane 2001; Crane and Dusenberry 2004Þ. Inparticular, a report states that in Ghana, “the major cutbacks in PPAG staff and the loss of its community-based dis- tributors have limited its outreach capabilities, particularly in the most remote areas of Ghana” ðTurnbull and Bogecho 2003, 6Þ. Such claims are supported by information received directly from PPAG, which shows that contraceptive provision via community-based distribution in rural areas, as measured in couple-years of protection, dropped by 45% as a result of funding losses.14

While provision of postabortion care increased, no indication was given that availability of abortion services was affected.15

V. Contraceptive Availability and Use The data used here are from the Ghana Demographic and Health Surveys ðDHSÞ, which regularly collect information on reproduction and health for a nationally representative, cross-sectional sample of women age 15–49 ðGSS, GHS, and ICF 2009aÞ. These data are drawn from cross-sections in 1988, 1993, 1998, 2003, and 2008 and provide information on knowledge of modern and traditional contraceptive methods, current method use at the time of the survey, and source for current users for 24,500 women. The data do not provide information on history of contraceptive knowledge or use over time.16

Historically, many African cultures have managed fertility with traditional methods. Many cultures have had taboos requiring postpartum abstinence for up to 3 years after a birth and terminal abstinence once a woman becomes a grandmother ðCaldwell and Caldwell 1987; Brown 2007Þ. Ghanaian women do practice postpartum abstinence, as well as spousal separation and pro-

14 Internal historical documents of PPAG, provided by Joana Nerquaye-Tetteh, executive director of PPAG 1995–2006. 15 Abortion has been legal in Ghana, with significant restrictions, since 1985. Since a change in the legal status of abortion coincided with one of the policy changes of interest, estimates of the policy’s impact on abortion use were also run excluding the 1985 policy change. These results do not differ from the results presented here and are available on request. 16 Unfortunately, there are no data, to my knowledge, that track individuals’ use of traditional or modern contraceptive methods over a time period relevant to these policy changes. The only Ghana DHS with a full calendar of contraceptive use was in 2008; however, the calendar covers the preceding 5 years, during which time there were no changes in the policy.

Jones 41

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

longed breast-feeding ðAdongo et al. 1997; Phillips et al. 2012Þ. Brown ð2007Þ finds that in Ghana, modern contraception serves as a complement to, rather than a substitute for, these traditional forms of postpartum non- susceptibility. Traditional methods for emergency contraception in Ghana include drinking a strong sugar solution, enema, and douching ðBaiden, Awini, and Clerk 2002Þ. Table 1 shows knowledge of method types and mean usage rates for both

modern and traditional methods over 1988–2008, as well as disaggregation of provider type by method. Knowledge of modern methods is nearly double that of traditional methods for both rural and urban women. While most

TABLE 1 USE AND SOURCE BY FAMILY-PLANNING METHOD ð%Þ

Most Recent Source

All Urban Rural Private Public Other

Pregnancies aborted 8.6 15.3 4.8 60 21 19 Modern contraceptive use:

Short acting: Pill 5.5 5.7 5.3 68 27 4 Injection 5.0 5.1 4.9 12 87 1 Condom 4.1 6.7 2.7 75 9 15 Diaphragm/foam/jelly 1.3 1.7 1.1 64 16 19 Female condom .1 .1 .0 75 13 13

Short-acting total 16.0 19.3 14.0 54 39 7 Long acting: IUD .9 1.9 .4 20 80 0 Norplant .7 .7 .7 10 89 1 Sterilization .3 .2 .3 23 76 1

Long-acting total 1.9 2.8 1.4 19 80 1 Long acting as share of modern 10.7 12.7 9.1

Modern contraception use total 17.8 22.1 15.4 48 45 7 Traditional contraceptive use:

Periodic abstinence 8.5 12.0 6.5 Withdrawal 2.1 2.4 1.9 Lactational amenorrhea .2 .3 .2 Other .9 .7 1.0

Traditional contraception use total 11.7 15.4 9.6 Traditional as share of total contraception 39.7 41.3 38.4

Not using contraception 70.5 62.4 75.0 Ever heard of contraceptive method:

Any modern method 76.1 84.2 71.1 Any traditional method 41.6 47.8 37.8

Note. Data are from five cross-sections of Demographic and Health Survey ðDHSÞ data in 1988, 1993, 1998, 2003, and 2008, with the exception that abortion data are from the special DHS in 2007. Abortion data are percentage of pregnancies aborted; other data are total users divided by number of eligible women. Eligible women are defined as sexually active in the past year, not currently pregnant or infecund, not currently practicing postpartum abstinence, and not wanting a birth in the next 2 years ðor not wanting any more birthsÞ. Source data are the share of current users reporting a type as her most recent source; “other” sources are family, friends, church, etc.

42 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

women at risk of an unwanted pregnancy are not using any contraceptive method, the most commonly used methods are periodic abstinence, pills, injections, and condoms.17 Pills and condoms are predominantly sourced from the private sector, while injections are sourced from the public sector. In most data rounds, private sector providers are not disaggregated by NGO versus for-profit status, and so “private” is not disaggregated in these analy- ses.18 PPAG is identified individually as a source in the 1988 data, and at that time PPAG accounted for 42% of all private provision. Surveys of government, private, and NGO providers of family-planning

services in Ghana were undertaken in 1993, 1996, and 2002, and availability of several methods was measured as the share of providers with the commodity in stock. A comprehensive report based on these surveys suggests that con- traceptiveavailability waslowerduring the years the policy was ineffect, asshown in table 2 ðHong et al. 2005Þ.19 The availability of contraceptive methods increased for the two most common methods typically provided privately ðpills and condomsÞ when the policy was rescinded ðin 1996 vs. 1993Þ and decreased for these after reinstatement ðin 2002 vs. 1996Þ. In contrast, the most common publicly provided method, injections, follows an opposite pattern. Availability of other less common methods seems to decrease in the presence of the policy. Considering mean availability of existing facilities, weighted by method preva- lence, weobservean increaseof6 percentagepointswhen the policy turns off and an equal and opposite change when the policy turns back on. Among methods that are predominantly provided by the private sector, this change is 9–10 per- centage points.

17 Those at risk of an unwanted pregnancy are defined as women who are sexually active in the past year, not currently pregnant or infertile, not currently practicing postpartum abstinence and who either want no more children or want to wait 21 years before having aðnotherÞ birth. I exclude those practicing postpartum abstinence and not those practicing amenorrhea, on the basis of their difference in susceptibility, following Brown ð2007Þ. 18 While “type” is given, in terms of clinic vs. pharmacy, it is not clear how a respondent would report supplies received from an outreach worker, and so these are not disaggregated. 19 The sampling frames included government-run or NGO-run hospitals, health centers, and maternity centers; the samples included all PPAG clinics. The providing report states, “Each of the three samples was selected to be nationally representative of facilities providing family planning services, by type of facility, with proportional regional representation. Although the sample sizes for the three surveys varied because of financial considerations, the methodology for selecting the samples was similar” ðHong et al. 2005Þ. Data reported are taken only from the comprehensive report, and no further analysis of changes in availability over time can be undertaken. The data from 1993 and 1996 ðby the Population Council’s Africa Operations Research and Technical Assistance ProjectÞ are no longer available, as they were corrupted while being stored on floppy disks ðe-mail communication with Fred Arnold, senior fellow at ICF InternationalÞ. The 2002 data ðby Macro International as part of the MEASURE DHS1 projectÞ are available but cannot be used alone to see changes over time.

Jones 43

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

A. Changes in Knowledge and Use To test the downstream impact of these supply changes, I also examine wom- en’s knowledge of a source for modern contraception and contraception use as a response to the policy. Because the five cross-sections of data do not coincide well with changes in the policy, I construct a continuous indicator of exposure to the policy for each data wave. Given that data are collected at regular 5-year intervals, I take the share of the preceding 5 years that were exposed to the policy as the independent variable of interest. Nonetheless, the absence of contraceptive access and use histories and the 5-year gaps in in- dicators of current access and use prevent an accurate estimate of how access and use respond to the policy. However, lacking the appropriate data, I present these analyses as attenuated, lower-bound estimates of the policy’s impact on access and use.20

I estimate the linear probability model

Ciy 5 a0 1 a1Policyy 1 X 0 iyf 1 M

0 yv 1 εiy; ð1Þ

20 These estimates are assumed to be lower bounds, as the data gaps prevent observation of nuances in access and use in the immediate vicinity of policy changes.

TABLE 2 SHARE OF FACILITIES WITH CONTRACEPTIVE STOCK

1993 1996 2002 Usage Weight MCP 1 No MCP MCP 2

Mainly private provision: Combined pill 92% 92% 82%** Progesterone pill 62% 86%** 75%** Mean of pills .32 77% 89%** 79%** Condom .27 85% 93%** 87%** Spermicide .06 85% 91%* 74%**

Mainly public provision: Injectable .26 94% 90%* 93% IUD .06 89% 89% 76%** Implant .04 NR 85% 74%**

Weighted mean 85% 91%** 85%** Weighted mean of private 81% 91%** 82%** Weighted mean of public 93% 89%* 88% N 399 313 428

Note. Weights are based on Demographic and Health Survey ðDHSÞ usage data in data rounds 1993, 1998, and 2003. Weight shown for spermicide is weight for diaphragm/foam/jelly, as these cannot be separated in usage data. Availability data are from Hong et al. ð2005Þ. MCP 5 Mexico City Policy; NR 5 not reported and not included in weighted mean. * Significantly different from the measure in the previous survey at the 5% level. ** Significantly different from the measure in the previous survey at the 1% level.

44 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

where Ciy is an outcome of interest for woman i interviewed in year y; Policyy is the share of the 5 years preceding the survey during which the policy was in place; My is a quadratic time trend; and Xiy contains woman-level controls, including quadratic functions of both age and parity in year y, an indicator for married status, an indicator for poverty ðdefined as the two poorest wealth quintiles specific to the rural or urban sectorÞ, and an indicator for education ðcompletion of primary school or higherÞ. This is estimated for outcomes such as knowledge of any source for contraception, current use of any contracep- tion, and current use of specific categories of contraception. Estimation results are presented in table 3 separately for rural and urban sectors.21

Table 3 columns 1–6 present results in the rural sector. In rural areas, full policy exposure reduces the share of women who know a source of contra- ception by 5% ðcol. 1Þ.22 There is no distinguishable change in total con- traception ðcol. 2Þ, but I estimate a clear shift from modern to traditional contraception as a result of the policy. Columns 3 and 4 indicate that about 12% of all contraceptive users shift away from modern methods toward tra- ditional methods under full policy exposure. This is consistent with recent findings that there was a shift from modern to traditional methods in Ghana during 2003–8 ðAbdul-Rahman, Marrone, and Johansson 2011Þ. Column 5 indicates that the greatest shift is away from short-acting methods, where usage is reduced by 13 percentage points ða 21% effectÞ. To investigate whether this is related to the private sector, column 6 estimates the impact on use of injections, the only short-acting method predominantly supplied by the public sector. I find that any shift away from injection use is not distin- guishable from 0, suggesting that the changes in the short-acting method use are resulting from changes in private sector supply. Table 3 columns 7–12 present the results for the urban sector. In general

the coefficients exhibit the same signs as in rural areas but are less precisely estimated. I cannot reject that there was no change in knowledge of source in urban areas ðcol. 7Þ, and overall use of contraception increases significantly ðcol. 8Þ. However, there is virtually no change in use of modern methods as a share of total contraception ðcol. 10Þ. It is important to note that given the large standard errors on the urban estimates, I cannot reject that effects were the same in rural and urban areas, as their 95% confidence intervals overlap. This reflects the fact that urban areas were also affected by clinic closures, although to a lesser degree as contraceptives are more broadly available in

21 Estimations include only women age 30 or younger, for comparability with analyses in Secs. VI and VII. 22 Full policy exposure would be that all 5 of the years preceding the survey were policy years; e.g., 1993.

Jones 45

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

urban areas, so that changes in NGO provision were less salient. Therefore, the urban areas are not considered quasi-experimental control areas, and thus I focus the analysis solely on rural areas where the impacts are more precisely identified.

B. Overall Reduction in Availability Taking together the findings presented above, I estimate here the full reduc- tion in contraceptive availability nationwide and in rural areas, as detailed in table 4. Among existing facilities, the clinic availability data presented above suggest a nationwide prevalence-weighted reduction in availability of 7% ð6 per- centage points on a mean of 91Þ. However, in addition to these stock outs observed in remaining clinics, a significant impact of the policy was clinic closures, which are not captured in those data. Following Hong et al. ð2005Þ, I can calculate client-weighted shares of facility provision by the categories hos- pital, health center, maternity center, and PPAG clinic. On the basis of this, PPAG represented 21% of facility-based provision of family planning. PPAG

TABLE 3 IMPACT OF POLICY ON CONTRACEPTIVE USE AND KNOWLEDGE OF SOURCE

Knows Source

Uses Contraception

Uses Traditional

Uses Modern

Uses Short Acting

Uses Injection

ð1Þ ð2Þ ð3Þ ð4Þ ð5Þ ð6Þ Rural:

Policy exposure 2.0294* .0165 .1161** 2.1161** 2.1334** 2.0064 ð.0169Þ ð.0116Þ ð.0510Þ ð.0510Þ ð.0523Þ ð.0381Þ

Percentage effect 24.9 . . . 134.8 217.4 220.7 . . . Mean of dependent variable .6018 .1139 .3333 .6667 .6438 .1536

N 8,058 8,058 918 918 918 918 R 2 .142 .044 .118 .118 .103 .161

ð7Þ ð8Þ ð9Þ ð10Þ ð11Þ ð12Þ Urban:

Policy exposure 2.0248 .0473*** .0093 2.0093 2.0167 2.0189 ð.0201Þ ð.0173Þ ð.0515Þ ð.0515Þ ð.0532Þ ð.0337Þ

Percentage effect . . . 129.4 . . . . . . . . . . . . Mean of dependent variable .7297 .1607 .3402 .6598 .6208 .1065

N 5,434 5,434 873 873 873 873 R 2 .114 .044 .152 .152 .138 .145

Note. Data are from five cross-sectional waves of Demographic and Health Surveys in 1988, 1993, 1998, 2003, and 2008. Binary dependent variables are shown as column headings. “Policy exposure” is the share of the 5 years leading up to the survey in which the policy was in effect. Sample is women under age 30, to be comparable with later analysis without making the sample too small; cols. 3–6 are estimated for the subsample of contraceptive users. All estimates include a quadratic time trend, quadratic age, quadratic parity, marital status, poverty status, and completion of primary school. Standard errors in parentheses. * Significant at the 10% level. ** Significant at the 5% level. *** Significant at the 1% level.

46 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

closed 57% of their clinics as shown in figure 1, representing a 12% reduction in total facility access. Combining clinic closures and stock outs in remaining clinics, there was an 18% reduction in availability at family-planning facilities. However, according to the individual data in the DHS, 44% of contraceptive provision was from nonfacility sources, so this represents an overall supply reduction of 10%. As the clinic-level data are not disaggregated by rural and urban sectors, I

rely on differences in impacts on usage across the sectors to roughly estimate the supply reduction in rural areas. In columns 4 and 10 of table 3, the reduction in modern method use in urban areas was 8% of what it was in rural areas. Further, 62% of the population is rural. A 10% national effect, calcu- lated as a population-weighted mean of rural and urban effects, would be composed of a rural drop of 15.4% and an urban drop of 1.2%. However, column 10 is estimated very imprecisely, and we cannot rule out that the effect was as much as 10% in urban areas ðand 10% in rural areasÞ. The upper bound on rural areas is a reduction of 16.1% ðif the effect in urban areas were 0Þ. Note that these are conservative estimates, as reductions in PPAG non- facility outreach services cannot be incorporated.

VI. Impact on Conception and Abortion A. Data The data used in the preceding section are from standard DHS collected every 5 years in many developing countries by Macro International’s MEASURE

TABLE 4 CALCULATION OF CONTRACEPTIVE SUPPLY REDUCTIONS

Share/Reduction ð%Þ Description Client share by facility type:

Hospital 15 Health Center 51 PPAG clinic 21 Maternity center 13

Calculated reductions: Reduction in facility access 12 57% PPAG closures � 21% market share Stock reduction at remaining facilities 7 From weighted means in table 2 Total facility supply reduction 18 12% � 100% 1 88% � 7% Nonfacility market share 44 Based on user reports in Demographic

and Health Survey Total supply reduction 10 ð1 2 .44Þ � .18 Rural supply reduction 15.4 .10 5 .62R 1 ð1 2 .62Þð.08RÞ

Note. Client shares are based on number of facilities by type, and mean number of new clients per month by type, as reported by Hong et al. ð2005Þ. Planned Parenthood Association of Ghana ðPPAGÞ clinic closures are as shown in fig. 1. Nonfacility providers are shops, pharmacies, churches, friends, etc. Data are from 1993.

Jones 47

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

project. In 2007, MEASURE conducted a nonstandard DHS in Ghana com- posed of special modules on maternal mortality and abortion ðGSS, GHS, and ICF 2009bÞ. Unlike most DHS, which collect a woman’s complete birth his- tory, this survey queried each woman’s complete pregnancy history, including pregnancies that ended in miscarriages, stillbirths, and induced abortions. While a handful of other DHS also collect pregnancy ðrather than birthÞ histories, the Ghana 2007 survey is the only one that explicitly records the use of induced abortion.23 These data will be useful for determining whether the supply changes resulting from the policy affected pregnancy and abortion use at the in- dividual level. The data contain information for 10,370 women. For each pregnancy in a

woman’s lifetime, the following information is recorded: the duration of the pregnancy, the month and year it ended ðfrom which one can roughly deduce the month it beganÞ, how it ended, and further information about the child if it was a live birth. Using this, I create a woman-by-month panel, allowing me to estimate within-woman impacts of the policy. In each month, a woman has one of the following seven statuses: conceived,

is pregnant, birthed a live child, had a stillbirth, miscarried, aborted a preg- nancy, or was not pregnant. Moving consecutively through the months, summing the live births, I calculate her existing parity in each month. The survey also collects information regarding the woman’s date of birth and month and year of first marriage ðor cohabiting unionÞ. Using these, each woman- month observation is assigned the woman’s age and parity and whether she has ever been in a union at that time. Months in which the woman is at least age 15 compose the complete data set. There are 1.8 million observations from 1981 to 2007. Each woman has between 1 and 323 observations ðmean is 144; only 5% of women have fewer than 10 observationsÞ. Other information collected about the woman does not vary over time but

is useful for dividing women into demographic subgroups. I classify women according to whether they had any schooling beyond primary school ðgrade 6; “high education,” 39%Þ, an indicator that is likely unchanging by age 15. Also, a wealth index for her household is created on the basis of a principal com- ponents analysis of information about housing quality, drinking water source, toilet facilities, and durable assets ðFilmer and Pritchett 2001Þ. From this, women are sorted into wealth quintiles, specific to rural and urban sectors. The two lowest rural wealth quintiles are labeled “poorest.” It is notable that 70% of

23 Further, other surveys conducted after 2001 that include pregnancy histories are in countries unlikely to be as affected by the Mexico City Policy: Armenia 2005, Azerbaijan 2006, Moldova 2005, Philippines 2008, and Ukraine 2007.

48 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

those in the top three rural quintiles are still poor by international standards ðless than $2/dayÞ, and so this group is labeled “less poor.” The urban population is divided into “poor” ðtwo lowest urban quintilesÞ and “not poor.” While wealth may vary throughout a woman’s life, it seems that a binary indicator of poverty status is likely stable for most individuals.

Balancing the Panel

A primary concern in creation of panel data from cross-sectional data is the loss of representativeness. The survey is nationally representative of women age 15– 49 in 2007 ðmean age is 29Þ. For each woman, her panel begins when she turns 15 ðor in 1981 if she is already 15 or olderÞ and ends when she is interviewed in 2007 ðmax age is 49Þ. However, the resulting panel is not representative of women of these ages for each year. For example, in 1981, the data only contain women age 15–24 ðmean 18Þ; in 1997 they contain women age 15–40 ðmean 25Þ. This “aging” of the sample confounds comparisons of births from different time periods. In order to limit the bias arising from the conversion of cross- sectional to panel data, I limit the age range of women in the sample. This effectively creates a “rolling panel” where women age into and out of the sample over time. But what is the appropriate age range for this analysis? Figure 2 shows the conception rate by age; that is, the share of fecund

woman-months in which a conception occurred.24 The conception rates are highest ðover 2.5%Þ for women age 22–27. A gradual decline begins around age 28, becoming steeper at age 37. For women younger than age 17, or age 401, the chance of conception in a given month is less than 1%. Figure 3 shows the abortion rate by age; that is, the share of pregnancy conclusions that are abortions. The likelihood of aborting a pregnancy is greatest for the youngest women; over 15% for 15-year-olds. However, considering their low number of pregnancies, this represents a small share of total procedures. The likelihood of aborting a pregnancy declines with age, generally remaining below 5% for women over age 25. Figure 4 shows the confluence of likelihood of conception and likelihood of aborting an existing pregnancy, that is, the unconditional probability of having an abortion, by age. The combination of high conception rates and high abortion rates yields the greatest chance of having an abortion for women age 18–20: about 2% per year ð.0018 � 12Þ. Women outside the 17–25 age range have a considerably lower probability: less than 1% per year. Given the focus of the analysis on conception and abortion decisions, I

select the 17–25 age range as the default for the analysis, varying this in

24 Fecund indicates she is not continuing or concluding a pregnancy.

Jones 49

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

robustness checks. In order to keep the sample consistent, I employ data only from months in which the sample is representative of this age range ð1981– 2007Þ.25 These data are divided according to policy periods: PRE from 1981 to 1984, ON1 from 1985 to 1992, OFF from 1993 to 2000, and ON2 from 2001 onward. Table A4 ðavailable online onlyÞ shows how the mean age for the full sample is significantly increasing over the periods. The last two col- umns show that when restricting observations to those for women age 17–25, the mean age is much more similar across periods. Further restricting to the group in which most abortions occur ð18–20-year-oldsÞ produces mean ages nearly identical across periods. There are 8,344 women that are observed while age 17–25 during the study

period 1981–2004. The conception analysis includes all of these women, in each month when she is in this age range. Of the 6,422 women that are ever pregnant in the sample, 91% had a pregnancy while age 17–25, yielding an effective sample size for the abortion analysis of 5,860 women.26 However, it is useful to note that, when employing woman fixed effects, the identifica- tion of policy effects on abortion use arises from women who have at least two

25 While the respondents report pregnancies as far back as 1977, the panel used for this analysis begins in 1981, for reasons of representativeness as discussed. 26 This figure is 93% for the rural sample.

Figure 2. Conception rate, by age. Data are from special Demographic and Health Survey, 2007

50 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

pregnancies in that age range, with variation across the pregnancies in the status of the policy: 66% of the effective sample have at least two pregnancies during the 9 years when age 17–25.27 This potentially introduces some selection bias, so it is important to note that the estimated effects are specific to women having two or more conceptions during that period of life. The source of identification is further reduced by the fact that only 59% of these women have at least one pregnancy during a policy period and at least one during a nonpolicy period. However, while this reduces the size of the sample used for identification, it does not introduce any further bias. For each woman, the timing of this 9-year period in her life is orthogonal to the imposition and removal of the policy. Table A5 ðavailable online onlyÞ shows the effective sample sizes for the analysis of conception and abortion use.

Measurement Error

One concern regarding these data is noisy measurement of pregnancy timing due to recall error. While the selection of respondents that misreport may be nonrandom, we expect that the direction of misreporting ðtoo early or too lateÞ will be random. As such, the direction of misclassification of pregnancies ðregarding whether they were conceived under the policy or notÞ will also be

27 This figure is 72% in the rural sample.

Figure 3. Abortion rate, by age. Data are from special Demographic and Health Survey, 2007

Jones 51

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

random. Further, measurement error of pregnancy timing would only induce troublesome bias if it were the case that women who are more likely to conceive ðor abortÞ were also more likely to misreport nonpolicy pregnancies as policy pregnancies. This would require these women to systematically report incorrectly late dates when conceiving just before the policy turns on and systematically report incorrectly early dates when conceiving just before the policy turns off. Given the low probability of this type of systematic misreporting, this measurement error should not create any troublesome bias. Another artifact of recall error is that less recent pregnancies are more likely

to be unreported due to forgetting, especially if the pregnancy ended in mis- carriage, stillbirth, or abortion. Therefore, I include a fixed effect for the pol- icy change nearest in time to the month of observation. In this way, observa- tions are directly compared only with others within the same 8-year time span, so that misreporting should be more consistent within these groups. However, even estimating within period, this recall bias may still yield an artificially lower rate of conception or abortion in earlier years. For periods sur- rounding a policy change from on to off ðoff to onÞ, this could artificially asso- ciate the policy with lower ðhigherÞ conception and abortion rates. For this reason, it is important that the analysis here includes both an on-to-off period as well as off-to-on periods. As a robustness check, I restrict the analysis to just one of each type of period, and the results are not significantly different.

Figure 4. Probability of abortion, by age. Data are from special Demographic and Health Survey, 2007

52 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Finally, while abortion is legal in Ghana during the period of analysis, there may be social desirability bias resulting in intentional underreporting of in- duced abortion. Some aborted pregnancies may be reported as miscarriages or may be unreported as pregnancies. The probability of exhibiting such re- porting bias is likely correlated with important unobservable characteristics, implying that we are underestimating the use of abortion for certain women ðperhaps the poor, the rural, or the less educatedÞ. However, unless such social desirability bias is also correlated with the timing of the policy, this should not present a threat to the analysis. At worst it will induce an underestimate of the impact on abortion for women who underreport, as the lower incidence will reduce the statistical power to detect changes.

B. Estimation In estimating the impact of changes in contraceptive supply on conception and abortion, one concern is the degree to which conception and abortion are affected by environmental and situational concerns beyond the change in supply. For example, birthrates fluctuate in tandem with business cycles, as couples are more reluctant to have children during recessions ðKirk and Thomas 1960Þ. Fertility decisions are also affected by seasonal changes. For this reason, it is important to control for other unobservable factors changing over time. To deal with seasonality, I include calendar month fixed effects. However, because the imposition ðor removalÞ of the policy always coincided with the change in calendar year, year fixed effects would be perfectly collinear with an indicator for the policy. I employ several alternatives to deal with this concern. First, I include a cubic time trend in all specifications. Second, I focus the estimation on a fixed time window on either side of each policy change to reduce the impact of time-varying unobservables. The default win- dow includes observations within 4 years of a policy change, and this is nar- rowed in robustness checks. Third, I include a fixed effect for the policy change that is nearest in time, effectively estimating within change. For example, an observation occurring in January 1990 is nearer the 1993 policy change than the 1985 policychange. Inthis way, Icompare observations just before a change to those just after it, rather than comparing observations from, say, 1982, to those from 2004. Finally, in order to control for the host of unobservable characteristics about

each woman that certainly affect such decisions, I employ woman fixed effects to compare each woman only with herself. Further, because a woman’s pref- erence for having a child changes throughout her life, I include controls for time-varying characteristics that strongly predict conception and childbirth:

Jones 53

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

quadratic functions of her age and parity, and whether she has ever lived in union with a man.28

The primary estimations are

Cimyc 5 b0 1 b1ONmy 1 X 0 imyf 1 M

0 myv 1 ni 1 nm 1 nc 1 εimyc; ð2Þ

Aimyc 5 g0 1 g1ONmy 1 X 0 imyf 1 M

0 myv 1 ni 1 nm 1 nc 1 εimyc; ð3Þ

where Cimyc indicates that woman i conceived in month m of year y. The index c takes the value 1, 2, or 3, representing which policy change is within the fewest months of my. Specifically, c 5 1 represents the change in late 1984 from PRE to ON1, including observations in the window from 1981 through 1988; c 5 2 represents the change in January 1993 from ON1 to OFF, in- cluding observations from 1989 through 1996; c 5 3 represents the change in January 2001 from OFF to ON2, including observations from 1997 through 2004; Ximy is a vector containing quadratic functions of age and parity specific to woman and month, plus an indicator for whether she has ever been in a cohabiting union; and M represents a cubic time trend. Fixed effects for the individual, the calendar month, and the nearest policy change are included as ni, nm, and nc, respectively; Aimyc 5 1 indicates that the pregnancy of woman i that ended in month m in year y was aborted; and Aimyc 5 0 for all pregnancies ending in live birth, stillbirth, or miscarriage. For the estimation of equation ð3Þ, all woman-months are included when the woman is concluding a preg- nancy while age 17–25. The independent regressor of interest is ONmy, which indicates that the

policy was in effect in month-year my, and b1 and g1 indicate the estimated impacts of the policy on conception and abortion, respectively, of the sub- group for which equation ð2Þ, or ð3Þ, is estimated, conditional on age, existing parity, ever-unioned status, and the secular trend.

C. Results Conception

The top panel of table 5 provides estimates of the policy’s effect on the probability of conception for women age 17–25.29 Overall, the probability of

28 Note that having ever been in union will be equal to 0 for a woman in all months preceding her first union and will be equal to 1 in all months after that. Note also that the results do not depend on the inclusion of these controls, as shown in Sec. VI.C. 29 None of the estimates of the policy’s impact on conception is different when restricting the sample to months when a woman is not already pregnant or concluding a pregnancy. Results available on request.

54 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

conception per month increased by 0.0011, representing a 5.6% increase that is significant at the 5% level. Results for the urban population are not sig- nificantly different from 0, although we cannot reject that the results are the same in rural and urban areas ð95% confidence intervals overlapÞ. This is consistent with the findings on contraceptive use in Section V, where results were more precisely estimated in rural areas, although urban effects were not

TABLE 5 POLICY’S EFFECT ON PROBABILITY OF CONCEPTION AND ABORTION

Rural Subgroups

All Urban Rural Poorest Less Poor

ð1Þ ð2Þ ð3Þ ð4Þ ð5Þ ð6Þ ð7Þ Conception:a

Policy .0011** 2.0003 .0022*** .0019* .0024** .0021** .0020* ð.000Þ ð.001Þ ð.001Þ ð.001Þ ð.001Þ ð.001Þ ð.001Þ

Policy � schooling 2.0032* .0003 ð.002Þ ð.001Þ

Linear combination 2.0007 .0023** ð.0020Þ ð.0012Þ

N 686,449 322,506 363,943 147,832 147,832 216,111 216,111 R2 .007 .007 .008 .010 .010 .008 .008 Mean of dependent variable .0195 .0164 .0217 .0228 .0228 .0210 .0210

ð8Þ ð9Þ ð10Þ ð11Þ ð12Þ ð13Þ ð14Þ Abortion:b

Policy .0144 .0044 .0235** 2.0019 .0059 .0393*** .0443*** ð.009Þ ð.020Þ ð.009Þ ð.013Þ ð.013Þ ð.013Þ ð.013Þ

Policy � schooling 2.0470 2.0135 ð.037Þ ð.019Þ

Linear Combination 2.0411 .0308 .0369 .0195

N 12,439 4,945 7,494 3,155 3,155 4,339 4,339 R2 .132 .239 .074 .067 .070 .090 .091 Mean of dependent variable .0823 .1484 .0457 .0263 .0263 .0585 .0585

Note. Data are from special Demographic and Health Survey, 2007. “Poorest” indicates the lowest two wealth quintiles of the rural sample. “Less poor” is the top three quintiles in the rural sample. “Schooling” indicates having education beyond primary school ðgrade 6Þ. “Linear combination” shows the sum of “policy” and “policy � schooling” coefficients with its standard errors in parentheses. All specifications include woman fixed effects, woman level controls as described in the text, a cubic time trend, calendar month fixed effects, and indicators for which policy change is relevant. Sampling weights are employed. Standard errors in parentheses, clustered at the village level. a Dependent variable is a binary indicator for conceiving in a given month. Samples include all months in which a woman was age 17–25. b Dependent variable is a binary indicator for whether a pregnancy conclusion was an abortion. Samples include all pregnancy conclusions for women age 17–25. * Significant at the 10% level. ** Significant at the 5% level. *** Significant at the 1% level.

Jones 55

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

statistically different.30 For the rural sample, however, the probability of con- ception per month increased by 0.0022 when the policy was in place. This rep- resents a 10% increase in pregnancies and is statistically different from 0 with 99% confidence. To gauge the magnitude of this effect relative to a change in contraceptive

use, I simulate the share of women who would need to switch from modern to traditional methods or quit using contraception altogether to achieve this change, as detailed in appendix B ðapps. A–C available onlineÞ. I find that about 1%–4% of women switching, with an additional 1.5%–2.5% quitting entirely, would predict the estimated change in the probability of conception of 0.002. These changes would represent ðin totalÞ about 20%–30% of users of modern contraception, which is comparable with the lower-bound effects estimated in Section V, as this would represent 12%–18% of all contraceptive users. I find that these changes would be reasonable in the face of a 10%– 16% reduction in availability of modern contraception. Further, this suggests an elasticity of pregnancy with respect to contraceptive availability between 0.62 and 1. The conceptual framework laid out in Section III suggests that, even within

the rural sector, we may observe differences in these impacts across wealth and education groups. Columns 4–7 of table 5 explore these differences. Columns 4 and 5 estimate using only the poorest two quintiles, where column 5 adds an interaction term to examine the differential impact of the policy according to education. Columns 6 and 7 are parallel estimations using the “less poor” subsample. The linear combination effects for the educated groups are re- ported below the estimated coefficients. I find that conception increased by 10% for women with low education,

regardless of wealth ð95% confidence intervals overlap for policy coefficients in table 5 cols. 5 and 7Þ. This was expected for the uneducated poor, con- firming that they are both unable to pay the increased cost of contraception and unable to effectively prevent pregnancy with traditional methods. However, this is surprising for the uneducated less poor; given that these women are un- likely to have unusually high opportunity costs, this effect must arise from an increase in monetary costs. I conclude that there was an increase in monetary costs significant enough to affect the quantity demanded for both poor and less poor. This is supported by the fact that effects are also not statistically distin- guishable across wealth groups among educated women ð95% confidence in- tervals overlap for the linear combinations in cols. 5 and 7Þ.

30 Four urban subgroups are explored, and none shows significant changes in conception as a result of the policy ðresults not shownÞ.

56 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

The model also predicted that higher education levels may improve off- setting with traditional methods, reducing the impact on conception. Among the poorest, I do find that the more educated present a lower impact of the policy on conception ðinteraction term in table 5 col. 5, significant at 10%Þ, consistent with a higher ability to substitute with traditional methods. Among the less poor, I do not find significant differences by education level ðinteraction term in col. 7Þ. Given that the educated less poor are the most likely to be employed and exhibit higher opportunity costs, this finding is consistent with the more effective use of traditional substitutes being offset by the higher opportunity costs that reduce access to modern contraception. While these demographic analyses are intriguing and consistent with both

increased time costs and significantly increased monetary costs, it is worth noting that I cannot clearly identify the magnitude ðor existenceÞ of these increased costs solely on the basis of these subgroup estimations. The findings from the demographic disaggregation are suggestive at best.

Induced Abortion

The lower panel of table 5 shows the estimates of the policy’s impact on the share of pregnancies ending in abortion for the full sample and the same subgroups.31 For the full sample, the coefficient is positive, although not statistically different from 0 at a standard level. The point estimate for the urban population is also positive and imprecise; we cannot reject that the effect in urban areas is 0, nor can we reject that it is the same as the rural effect. In rural areas, the estimation suggests that the policy increased the use of

abortion by 2.35 percentage points, an estimate significant at the 5% level ðcol. 10Þ. Given that only 4.7% of pregnancies are aborted in rural areas, this change reflects a 50% increase in the use of abortion. However, this additional 2.35% of pregnancies aborted only partially offsets the estimated 10% in- crease in pregnancies estimated by equation ð2Þ. That is, of the additional unwanted pregnancies resulting from the policy, one in five was aborted. The last four columns of table 5 present results for rural subgroups. For the

poorest two quintiles of the rural population, shown in columns 11 and 12, the effect is not distinguishable from 0, regardless of education status. This is consistent with the prediction that abortion is prohibitively costly for the

31 What may also be of interest is the unconditional change in the occurrence of abortion, that is, the probability of having an abortion in any month, regardless of pregnancy status. For the rural population, this probability is only .0009, which makes changes in this low rate very difficult to detect. Nonetheless, I find that the policy increased the unconditional probability of abortion for rural women by .0003, a 33% increase with a p-value of .117. These results are shown by rural subgroup in table A6, available online.

Jones 57

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

poorest of the poor. For the less poor population, the policy increases the share of pregnancies aborted by 3–4 percentage points ðcols. 13 and 14Þ. For the uneducated, the estimate is significant at the 1% level; for the educated, the point estimate is marginally significant ðp-value 5 .115Þ, and we cannot reject that the effects are the same across education groups. For the less poor, the increase in abortion offsets 30%–40% of conceptions that result from the supply reduction. The fact that education status has no discernible effect on the policy’s impact on abortion use implies that the assumption that uned- ucated women would hold traditional values preventing them from increasing use is false. Although education is generally a stronger predictor of abortion use than is wealth, using abortion to compensate for reductions in contra- ception is not driven by educational attainment. The results on conception and abortion by demographic subgroup are summarized in table 8.

Specification and Placebo Tests

Varying Estimation Window

In the specifications presented above, I include all years within 4 years of a policy change ðwhich includes 1981–2004Þ. However, one might expect that the effect of a policy ðor its removalÞ would be most salient within a narrower time frame. Table 6 shows the estimation of policy impacts on conception and abortion for rural women using successively tighter windows of estima- tion. The estimates for abortion use are focused on women in the upper three quintiles of wealth, as the poorest of the poor showed no significant effect. I gradually narrow the window by 6 months at a time ð42, 36, 30, 24 monthsÞ. The smallest feasible window that allows enough women to have at least two pregnancies, and thereby allows the use of woman fixed effects, employs dates within 24 months of a policy change. The estimates show that as the window is narrowed, the effect becomes gradually larger for estimates of abortion use. For conception, estimates remain in the neighborhood of the original estimate of .0022, although coefficients do increase as the window is narrowed for conception estimations that employ only months when a woman is not al- ready pregnant or concluding a pregnancy ðresults available on requestÞ. Stronger estimates within a narrower time frame may suggest that the

impacts of the policy wane over time. A variety of factors could contribute to this, including market adjustment. If prices rise in response to the supply shortage, this should induce new suppliers into the market, eventually bring- ing the market back to equilibrium. Additionally, NGOs that experienced funding cuts may have been able to secure compensatory funding in the longer term ðas suggested by the data in table A3, available onlineÞ. Finally, women who lose access to their method of choice may take time to learn about or

58 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

become comfortable with substitute methods, such as injections, for example. The increase of estimates in a narrower window suggests that the reduction in contraceptive access was concentrated in the short term after the policy turned on both in 1985 and 2001. The response of pregnancy and abortion rates to even these short-term supply shocks suggests that fertility outcomes respond acutely to contraceptive access.

TABLE 6 ROBUSTNESS AND PLACEBO CHECKS

Policy Coefficient SE N R 2

Conception:a

Window around policy change:b

42 months .002** .001 322,152 .008 36 months .002** .001 277,406 .008 30 months .003*** .001 232,915 .008 24 months .002** .001 187,891 .008

Robustness checks: Controlling for NHIS .002*** .001 363,943 .008 Excluding 2001 change .003*** .001 217,817 .008 Age 16–26 .001** .001 441,720 .008 Age 15–27 .002** .001 285,348 .009 No weights .002*** .001 363,943 .008 No controls .002** .001 363,943 .000

Placebo electionc .003 .002 111,031 .008 Abortion:d

Window around policy change:b

42 months .040*** .014 3,826 .087 36 months .048*** .019 3,334 .095 30 months .063*** .024 2,820 .094 24 months .073** .036 2,252 .114

Robustness checks: Controlling for NHIS .040*** .014 4,339 .090 Excluding 2001 change .053*** .016 2,643 .092 Age 16–26 .029** .013 5,021 .083 Age 15–27 .034** .015 3,515 .107 No weights .038*** .012 4,339 .085 No controls .041*** .013 4,339 .014

Placebo electionc .028 .035 1,413 .128 Stillbirths placebo .009 .007 4,339 .038

Note. Data are from special Demographic and Health Survey, 2007. Each row represents a separate estimation. All specifications include controls as described in note to table 5. Sampling weights are employed, except as noted. Standard errors clustered at the village level. NHIS 5 national health insurance scheme. a Dependent variable is conception. Samples include rural women in months when age 17–25. b Includes only observationswithinthe specified numberofmonthsof a policy change. c Includes observations within 4 years before or after a placebo policy change ðto OFFÞ in 1988, coinciding with a US presidential election. d Dependent variable is abortion, except in last row ðwhich employs stillbirth as the outcome, rather than abortionÞ. Samples include pregnancy conclusions for rural women in the upper three rural wealth quintiles, age 17–25. ** Significant at the 5% level. *** Significant at the 1% level.

Jones 59

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Robustness Checks

During the period of this analysis, there were other policies changing in Ghana that may have affected fertility decisions. In particular, in 2003 the government introduced a national health insurance scheme ðNHISÞ that made all prenatal and maternity care available free of charge. To check whether this policy is affecting the results presented here, I include an indicator for the presence of the NHIS ðsee table 6Þ. Its inclusion does not affect the point estimate in mag- nitude or precision for either conception or abortion. Another potential concern is that women are more likely to conceive and less

likely to abort as they age through the 17–25-years-of-age period. Therefore, a policy change from off to on, viewed in isolation, will mechanically indicate that the policy increased conception and vice versa for abortion. Therefore, I restrict the estimation sample to the time windows surrounding just the first two policy changes, so that the sample is balanced in terms of direction of policy change. Table 6 shows the estimation including only the years surrounding the changes in 1985 and 1993 ði.e., excluding the 2001 changeÞ, and the estimated effects are not significantly different from the original estimates. In Section VI.A, I discuss the need to restrict the age range of women in

included observations. The default age range is 17–25, on the basis of the natural breaks in abortion use on either side of this range. Table 6 also presents results under the larger age ranges of 16–26 and 15–27. None of these differs signifi- cantly from the primary estimations. Finally, table 6 presents estimations with- out employing the sampling weights and estimates without the woman-level controls for age, parity, and marital status; the estimates are not significantly different from the original point estimates and remain statistically significant.

Placebo Tests

Perhaps there is something about US presidential elections that affects fertility outcomes in Ghana by some mechanism outside the Mexico City Policy. Perhaps elections affect business cycles in the United States, which affect economies elsewhere. Or perhaps there are expectations in the months approaching an election regarding what might change about the policy. In order to test for these possibilities, I take advantage of the 1988 US presidential changeover ðfrom Reagan to George H. W. BushÞ that did not change the policy in any way. I estimate the effect of the 1988 election on conception and the use of abortion and find effects that are not statistically distinguishable from 0 ðtable 6Þ. Finally, I check whether a presumably unrelated birth outcome, stillbirths,

could also be predicted by the policy. I find that there is no statistically sig- nificant effect of the policy on stillbirths.

60 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

VII. Net Impact on Fertility The total fertility rate ðTFRÞ of a population in a given year is the average number of live births a woman would have in her lifetime if she were to experience the given year’s age-specific fertility rates ðASFRÞ for each age throughout her lifetime.32 This is the summation of the single-age ASFR in the given year. In order to calculate the TFR of a population, one needs a representative sample of reproductive-aged women in the year of interest. I rely again on the five standard DHS employed in Section V. These data are

repeated cross-sections, but collection of retrospective birth histories allows for the creation of a woman-by-year panel, similar to that described in Section VI. This allows for a reasonable calculation of the TFR for women age 15–45 in Ghana for the years 1984–2007.33 Note that, while the fertility histories of respondents stretch back to 1951, the first year in which the sample is rep- resentative of women age 15–45 is 1984.34

A longer panel of fertility rate history can be achieved by focusing on ASFR. Given the focus in Section VI on women age 17–25, I also calculate the ASFR for this age group. The 17–25 ASFR for a given year is the number of children a woman would have while she was age 17–25 if she experienced the relevant single-age ASFR for the given year at each of those nine ages in her life. This can be calculated from 1965 to 2007, as women in this age range in 1965 would be fully represented in the 1988 survey ðas women age 41–49Þ. I calculate the 17–25 ASFR separately for each of the eight demographic subgroups created by pairwise combinations of poverty and education status for both rural and urban sectors.35

For all women, there is a clear declining secular trend in fertility from 1965 to 2007 ðsee table A7, available onlineÞ. Therefore, an estimation of the impact of the policy on fertility rates must carefully account for this, including as many years as possible. Focusing on the 17–25 ASFR for demographic subgroups, I estimate

ASFRyg 5 d0 1 d1ONy 1 Y 0 y v 1 εyg; ð4Þ

32 This assumes no mortality of reproductive-age females. 33 Only survey years are representative of women up to age 49; the interim years are only con- sistently representative up to age 45. For example, any woman age 46 or older in 1994 would not be enumerated in 1998, as she would be over 49 at that time. As the births for 1994 are drawn from surveys in 1998 or later, these women’s births in 1994 would be excluded. 34 This is because women age 45 in 1983, for example, or at younger ages in earlier years, were not included in any surveys as they were past the target age by 1988. 35 Table A7 summarizes the fertility trends in Ghana by rural and urban sectors ðsince 1989Þ and by education level within the rural sector for the 17–25 ASFR ðsince 1970Þ.

Jones 61

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

where ASFRyg is the year- and group-specific ASFR for the 17–25-year-old age range. The estimation includes ASFR for eight groups across 43 years, for a total of 344 observations; estimations are weighted according to the size of each subgroup. Variable ONy is interacted with indicators for urban, nonpoor, and high-education statuses in successive estimations; d1 estimates the impact of the presence of the policy on the 17–25 ASFR ðsummations of d1 with coefficients from successive interaction terms provide estimates of this impact for specific subgroupsÞ; Y represents a quadratic time trend; and ε is a mean 0 error. Table 7 shows the results of estimations of variations of equation ð4Þ. Col-

umn 1 shows a positive effect for the full population that is not distinguishable from 0. In column 2, we see that the impact on rural fertility is positive and significant. Column 3 suggests that the only precisely estimated effect among rural women is for the poorest, although we cannot reject that the impacts are the same for the less poor. The estimated effect for each of the subgroups is calculated via summation of the relevant coefficients from column 4, and the effects for rural subgroups are presented in table 8. The full list of subgroups is presented in table A8, available online. No significant changes in fertility were found among urban women, and we

can reject that the effects are the same as rural effects with 5% significance. For the less poor rural women, estimated increases are 3%–6%, which are fairly consistent with a 10% increase in conception, where 3–4 percentage points are offset with induced abortion. However, given the standard errors, we cannot reject that these effects are 0.36 For the poorest rural women, fertility increased by 7.4% for the uneducated and by 10.7% for the educated, net of the secular trend. For the uneducated women, this is broadly consistent with the 10% observed increase in conception, with little to no offsetting with induced abor- tion. The small group of educated poor women present a puzzle: point estimates for both conception and abortion are negative and insignificant, but estimates show an increase in fertility. However, for 92% of the rural population, the findings are both consistent with the conceptual framework proposed and con- sistent across estimation of different outcomes, even using different data sources.

VIII. Conclusion This study has examined Ghanaian women’s response to a reduction in the availability of modern contraceptives in terms of contraceptive access and use,

36 Large standard errors likely result from noise, as some annual ASFRs are estimated on the basis of as few as 166 women per year of age. In order to calculate a fertility rate, one needs an accurate estimate of the percentage of women giving birth in each 1-year age group.

62 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

resulting pregnancies, use of induced abortion, and resulting births. The exogenous change in availability results from a US policy, driven entirely by domestic politics, that cut funding to NGOs providing reproductive health services in poor countries. These organizations closed clinics in both rural and urban areas but report that the primary change in service provision resulting from the funding cuts was a reduction in contraceptive supplies and outreach in rural areas. These claims are supported by method availability data from clinic surveys and lower-bound estimations of changes in contraceptive use during policy periods. On the basis of clinic availability data, stocks of modern birth control methods dropped by about 10% nationwide, driven by reduc- tions in the private sector. On the basis of rural and urban changes in use, it appears that the nationwide change is composed of a 10%–16% reduction in rural areas and a 0%–10% reduction in urban areas. Using reproductive histories that are unusually detailed for a poor-country

setting, I find that many young rural women were either unwilling or unable to fully offset the reduction in modern contraception with traditional methods for preventing pregnancy, resulting in increases in pregnancies of 10%. I also examine an additional potentially mediating behavior, that is, whether induced abortion is used to offset increases in unplanned pregnancies. As predicted by the conceptual framework, the poorest women did not increase abortion use.

TABLE 7 POLICY IMPACT ON FERTILITY RATES

ð1Þ ð2Þ ð3Þ ð4Þ Policy .0487 .1202* .1715* .1644*

ð.043Þ ð.061Þ ð.089Þ ð.094Þ Policy � urban 2.1921** 2.1916** 2.2038**

ð.084Þ ð.084Þ ð.082Þ Policy � nonpoor 2.0937 2.1024

ð.091Þ ð.091Þ Policy � primary education .0397

ð.080Þ Linear combination:

Urban 2.0719 2.0201 See table A8 ð.0588Þ ð.0765Þ

Nonpoor .0778 See table A8 ð.0660Þ

R 2 .607 .614 .615 .616

Note. Dependent variable is the age-specific fertility rate for 17–25-year-olds in 1965– 2007, calculated separately for eight demographic groups ðby rural/urban, poor/nonpoor, and low/high educationÞ. Data are from standard Demographic and Health Surveys in 1988, 1993, 1998, 2003, and 2008. For descriptions of subgroups by schooling and wealth, see note to table 5. Estimations are weighted by the size of the demographic group; sizes are given in table A8, as are all linear combinations from col. 4. Standard errors in parentheses. N 5 344. * Significant at the 10% level. ** Significant at the 5% level.

Jones 63

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

However, women in the upper three wealth quintiles aborted four out of every 10 additional pregnancies that resulted from the supply reduction. This com- pensating behavior did not differ according to educational attainment. Given that the increase in pregnancy was only partially offset by abortion

for some women, and not at all for others, we would expect to see fertility rates increase in response to the supply reductions as well. Using separate data to calculate ASFR, I find increases in fertility that are remarkably consistent with the estimates of impact on conception and abortion. The poorest women realize fertility that is 7%–10% higher than predicted by the secular trend. The less poor women experience increases of 3%–6%, although these esti- mates are not as precise. While changes in birth timing may contribute to these impacts, secondary analyses suggest that timing cannot fully explain the results, and reductions in total fertility are contributing as well ðsee app. CÞ. Taken together, these changes in fertility are consistent with previous esti-

mates of the impact of contraceptive availability on fertility outcomes. Pritchett

TABLE 8 SUMMARY OF PREDICTED AND ESTIMATED IMPACTS

Low Education High Education

Predicted impacts: Poor: Conception Increase Ambiguous Abortion No change No change Fertility Changes with conception Changes with conception

Nonpoor: Conception No change Ambiguous Abortion No change Increase if conception

increases Fertility Changes with conception Changes less than conception

Estimated impacts—rural: Poorest: Population share ð%Þ 33 8 Conception 10.5% increase Cannot reject no change Abortion Cannot reject no change Cannot reject no change Fertility 7.4% increase 10.7% increase Change in ASFR .1644* ð2.23Þ .2041** ð1.9Þ

Less poor: Population share ð%Þ 29 30 Conception 9.5% increase 11% increase Abortion Increased; offset 4 of 10

additional pregnancies Increased; offset 3 of 10 additional pregnancies

Fertility 2.8% increase 6.0% increase Change in ASFR .0620 ð2.19Þ .1018 ð1.71Þ

Note. For descriptions of subgroups by schooling and wealth, see note to table 5. “Change in ASFR” rows show the subsample’s ASFR (age-specific fertility rate) in parentheses for comparison to the effect size. * Significant at the 10% level. ** Significant at the 5% level.

64 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

ð1994Þ, Molyneaux and Gertler ð2000Þ, and Miller ð2010Þ estimate that access to family planning can explain up to 4%–8%, 6%, and 5% of fertility in In- donesia, Colombia, and across nations, respectively. My estimates from rural Ghana on average are comparable with these, although higher for the poorest women, as they do not offset with abortion. This suggests that estimating the impact of contraceptive access only on realized fertility understates the full impact, which includes effects on both pregnancy and abortion use. My esti- mates are notably smaller than the 15% reductions in fertility found in the Matlab and Navrongo experiments, perhaps suggesting that the difference may be resulting from the complementary community health aspects of those programs. Both comparisons also imply that access to contraception has a greater scope for impact among populations with higher excess fertility. The pattern of differential impacts by demographic group is consistent with

increases in both time and monetary costs for accessing modern contraception. Women with low opportunity costs experienced increases in conception re- gardless of wealth status, suggesting that the monetary cost increase was signif- icant enough to affect demand across income groups. Among the poor, we find that the small group of highly educated women experienced a lesser increase in pregnancy, consistent with more effective use of substitute traditional methods by educated women. Among the less poor, education differentials are not sig- nificant, perhaps indicating a countervailing impact of further reduced access to modern methods for educated women due to higher opportunity costs of employed women. However, while this pattern is consistent with increased monetary and time costs, it is impossible to draw definitive conclusions on the basis of these rough demographic subgroup analyses. Finally, the assump- tion that uneducated women hold traditional values, preventing them from in- creasing abortion in response to the supply change does not hold; among non- poor women, the uneducated were no less likely to compensate for the supply reduction by increasing abortion use. The supply of contraceptives in many African countries is heavily depen-

dent on donor funding. As shown here, such funds are subject to the whims of donors and may change unpredictably. While some researchers have argued that contraceptive availability plays only a small role in fertility outcomes, the results of this study suggest otherwise. In the face of temporary supply re- ductions, all the major subgroups of rural women experienced increased pregnancy. Among those who could afford to respond by increasing their use of induced abortion, even those with ex ante low use rates chose to do so. Increases in realized fertility, relative to the secular trend, arguably represent increases in unwanted or unplanned births. While a 5%–10% increase in total

Jones 65

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

fertility may seem negligibly small as some have claimed, in fact, this repre- sents a significant 20%–40% increase in unwanted fertility. The burden of these additional unplanned births fell disproportionately on

the poorest women, where adding additional unwanted births strains their ability to raise a healthy and productive next generation. For others, increased use of induced abortion put women at greater risk of unsafe procedures and long-lasting complications, in a setting that already has very high maternal mortality.37 While contraceptive access may have deceivingly small impacts on TFR, I find that it has significant impacts on excess fertility and maternal risk, as well as distributional implications. In Africa, total fertility is higher than anywhere else in the world, and 25%

of births are unwanted. International funding for contraceptive supplies clearly has a role in changing these statistics and may have distributional impacts as well. However, the reproductive health sectors in poor countries must even- tually develop donor independence to ensure reliability of provision in the longer run. The results presented here suggest that interruptions or reductions in supply can hinder the demographic transition of Africa and result in a poorer, less healthy, and less productive population in the future.

References Abdul-Rahman, L., G. Marrone, and A. Johansson. 2011. “Trends in Contraceptive Use among Female Adolescents in Ghana.” African Journal of Reproductive Health 15:45–55.

Adongo, P. B., J. F. Phillips, B. Kajihara, C. Fayorsey, C. Debpuur, and F. N. Binka. 1997. “Cultural Factors Constraining the Introduction of Family Planning among the Kassena-Nankana of Northern Ghana.” Social Science and Medicine 45:1789– 1804.

Ananat, E. O., J. Gruber, and P. Levine. 2007. “Abortion Legalization and Life-Cycle Fertility.” Journal of Human Resources 42:375–97.

Ashford, L. 2010. “Resource Flows for International Population Assistance and UNFPA.” Background paper prepared for the Center for Global Development. http://www.cgdev.org/doc/Resources-at-UNFPA.pdf.

Ashraf, N., E. Field, and J. Lee. 2014. “Household Bargaining and Excess Fertility: An Experimental Study in Zambia.” American Economic Review 104, no. 7:2210–37.

Baiden, F., E. Awini, and C. Clerk. 2002. “Perception of University Students in Ghana about Emergency Contraception.” Contraception 66:23–26.

37 In 2013, maternal mortality in Ghana was 18 times the rate in OECD countries; in the top 20% in the world ðWorld Bank 2013Þ.

66 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Bailey, M. J. 2006. “More Power to the Pill: The Impact of Contraceptive Freedom on Women’s Lifecycle Labor Supply.” Quarterly Journal of Economics 121:289–320.

———. 2010. “‘Momma’s Got the Pill’: How Anthony Comstock and Griswold v. Connecticut Shaped US Childbearing.” American Economic Review 100:98–129.

Bailey, M. J., B. Hershbein, and A. R. Miller. 2012. “The Opt-In Revolution? Contraception and the Gender Gap in Wages.” American Economic Journal: Applied Economics 4:225–54.

Becker, G. 1993. A Treatise on the Family. Rev. ed. Cambridge, MA: Harvard Uni- versity Press.

Bendavid, E., P. Avila, and G. Miller. 2011. “United States Aid Policy and Induced Abortion in Sub-Saharan Africa.” Bulletin of the World Health Organization 89: 873–80.

Bloom, D. E., and D. Canning. 2003. “Contraception and the Celtic Tiger.” Eco- nomic and Social Review 34:229–47.

Bongaarts, J. 1994. “The Impact of Population Policies: Comment.” Population and Development Review 20:616–20.

Brown, M. 2007. “When Ancient Meets Modern: The Relationship between Post- partum Non-susceptibility and Contraception in Sub-Saharan Africa.” Journal of Biosocial Science 39:493–515.

Caldwell, J. C., and P. Caldwell. 1987. “The Cultural Context of High Fertility in Sub-Saharan Africa.” Population and Development Review 13:409–37.

Cincotta, R. P., and B. B. Crane. 2001. “Public Health: The Mexico City Policy and U.S. Family Planning Assistance.” Science 294:525–26.

Cook, P. J. 1999. “The Effects of Short-Term Variation in Abortion Funding on Pregnancy Outcomes.” Journal of Health Economics 18:241–57.

Crane, B. B., and J. Dusenberry. 2004. “Power and Politics in International Funding for Reproductive Health: The US Global Gag Rule.” Reproductive Health Matters 12: 128–37.

Currie, J., L. Nixon, and N. Cole. 1996. “Restrictions on Medicaid Funding of Abortion: Effects on Birth Weight and Pregnancy Resolutions.” Journal of Human Resources 31:159–88.

Debpuur, C., J. F. Phillips, E. F. Jackson, A. Nazzar, P. Ngom, and F. N. Binka. 2002. “The Impact of the Navrongo Project on Contraceptive Knowledge and Use, Reproductive Preferences, and Fertility.” Studies in Family Planning 33:141–64.

Easterly, W., and C. Williamson. 2011. “Rhetoric versus Reality: The Best and Worst of Aid Agency Practices.” World Development 39:1930–49.

Filmer, D., and L. Pritchett. 2001. “Estimating Wealth Effects without Expenditure Data or Tears: An Application to Educational Enrollments in States of India.” Demography 38:115–32.

Gruber, J., P. Levine, and D. Staiger. 1999. “Abortion Legalization and Child Living Circumstances: Who Is the ‘Marginal Child’?” Quarterly Journal of Economics 114: 263–91.

GSS, GHS, and ICF. 2009a. “Ghana Demographic and Health Surveys, 1988– 2008.” Ghana Statistical Service, Ghana Health Service, and ICF Macro Interna- tional, Calverton, MD.

Jones 67

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

———. 2009b. “Ghana Maternal Health Survey 2007.” Data set GHIQ51DT. DTA. Ghana Statistical Service, Ghana Health Service, and ICF Macro Interna- tional, Calverton, MD.

Guldi, M. 2008. “Fertility Effects of Abortion and Birth Control Pill Access for Minors.” Demography 45:817–27.

Hong, R., N. Fronczak, A. Chinbuah, and R. Miller. 2005. “Ghana Trend Analysis for Family Planning Services, 1993, 1996 and 2002.” DHS Trend Report no. 1 ðJanuaryÞ, ORC Macro.

IPPF. 2002. “Financial Statements: Annual Report of the Governing Council for the Year Ended 31 December 2001.” International Planned Parenthood Federation, May.

Kane, T. J., and D. Staiger. 1996. “Teen Motherhood and Abortion Access.” Quarterly Journal of Economics 111:467–506.

Kearney, M. S., and P. B. Levine. 2009. “Subsidized Contraception, Fertility, and Sexual Behavior.” Review of Economics and Statistics 91:137–51.

Kirk, D., and D. S. Thomas. 1960. “The Influence of Business Cycles on Marriage and Birth Rates.” In Demographic and Economic Change in Developed Countries, 241–60. Princeton, NJ: Princeton University Press.

McKelvey, C., D. Thomas, and E. Frankenberg. 2012. “Fertility Regulation in an Economic Crisis.” Economic Development and Cultural Change 61:7–38.

MEASURE-DHS. 2013. “Demographic and Health Surveys STATcompiler.” http:// www.statcompiler.com.

Miller, G. 2010. “Contraception as Development? New Evidence from Family Plan- ning in Colombia.” Economic Journal 120:709–36.

Mitrut, A., and F. C. Wolff. 2011. “The Impact of Legalized Abortion on Child Health Outcomes and Abandonment: Evidence from Romania.” Journal of Health Economics 30:1219–31.

Molyneaux, J. W., and P. J. Gertler. 2000. “The Impact of Targeted Family Plan- ning Programs in Indonesia.” Population and Development Review 26:61–85.

Moyo, D. 2009. Dead Aid: Why Aid Is Not Working, and How There Is a Better Way for Africa. London: Penguin.

Myers, C. K. 2012. “Power of the Pill or Power of Abortion? Re-examining the Effects of Young Women’s Access to Reproductive Control.” IZA Discussion Paper no. 6661, Institute for the Study of Labor, Bonn.

PAI. 1999. “Donor Report Card and Country Profiles.” Population Action Inter- national, Washington, DC.

Phillips, J., W. Stinson, S. Bhatia, M. Rahman, and J. Chakraborty. 1982. “The Demographic Impact of Family Planning–Health Services Project in Matlab, Bangladesh.” Studies in Family Planning 13:131–40.

Phillips, J. F., E. F. Jackson, A. A. Bawah, B. MacLeod, P. Adongo, C. Baynes, and J. Williams. 2012. “The Long-Term Fertility Impact of the Navrongo Project in Northern Ghana.” Studies in Family Planning 43:175–90.

Pop-Eleches, C. 2006. “The Impact of an Abortion Ban on Socioeconomic Out- comes of Children: Evidence from Romania.” Journal of Political Economy 114: 744–73.

68 E C O N O M I C D E V E L O P M E N T A N D C U L T U R A L C H A N G E

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

———. 2010. “The Supply of Birth Control Methods, Education, and Fertility: Evidence from Romania.” Journal of Human Resources 45:971–97.

Pritchett, L. H. 1994. “Desired Fertility and the Impact of Population Policies.” Popu- lation and Development Review 20:1–55.

Ross, J., E. Weissman, and J. Stover. 2009. “Contraceptive Projections and the Donor Gap.” Working paper, Reproductive Health Supplies Coalition.

Sinha, N. 2005. “Fertility, Child Work, and Schooling Consequences of Family Planning Programs: Evidence from an Experiment in Rural Bangladesh.” Economic Development and Cultural Change 54:97–128.

Turnbull, W., and D. Bogecho. 2003. “Access Denied: U.S. Restrictions on Inter- national Family Planning.” Population Action International, Washington, DC.

UNFPA. 2004. “Financial Resource Flows for Population Activities, 2004.” United Nations Population Fund. http://www.resourceflows.org/.

USAID. 1999. “Contracts and Grants and Cooperative Agreements with Universities, Firms and Non-profit Institutions by Country/Region Active as of October 1, 1999: Fiscal Year 2001.” Yellow book, Office of Procurement, U.S. Agency for International Development.

White House Office of Policy Development. 1984. “US Policy Statement for the International Conference on Population.” Population and Development Review 10:574–79.

World Bank. 2013. “World Development Indicators.” http://databank.worldbank.org.

Jones 69

This content downloaded from 135.026.202.153 on November 28, 2016 19:22:18 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).