Anonymity and In(cid:133)delity: Ethnic Identity, Strategic Cross-Ethnic Sexual Network Formation, and HIV/AIDS in Africa1 Roland Pongou2 This Version: December 2009 Abstract: We develop a theory of how community-level ethnic heterogeneity determines the formation of sexual networks, and how this, in turn, a⁄ects the spread of HIV/AIDS. Speci(cid:133)cally, the model assumes that agentsderiveutilityfromsexualrelationships,butsexualin(cid:133)delityisprohibitedandpunishedifdetectedbythe cheatedpartner. Weshowthatwheninformationcirculatesmoreeasilywithinethnicgroupsthanacross,agents tend to choose their partners from di⁄erent groups to hide their in(cid:133)delity, due to cross-group anonymity. This optimizing behavior in turn implies a mechanism wherein ethnic heterogeneity encourages sexual in(cid:133)delity and directly a⁄ects the spread of HIV/AIDS. We use micro-level data from several sub-Saharan African countries to test these predictions of the model. In doing so, we (cid:133)nd: (1) evidence for strategic cross-ethnic sexual interactions; and (2) a direct e⁄ect of ethnic heterogeneity on both the number of sexual partners and HIV infection. Robustness checks show that this e⁄ect is not driven by a lack of public goods in more ethnically diverse communities. Interestingly, ethnic heterogeneity is shown to have no e⁄ect on anemia, which unlike HIV/AIDS, does not involve socially prohibited human interactions. This (cid:133)nding supports our theory that diseasesdrivenbysociallyprohibitedhumaninteractionsspreadthroughhiddenandanonymousnetworks. Our study o⁄ers a new explanation for the high concentration of the AIDS epidemic in sub-Saharan Africa. JEL Codes: A14, D71, I10, J10, O1. Keywords: Ethnicidentity;ethnicheterogeneity;anonymity;in(cid:133)delity;strategicsexualnetworkformation; HIV/AIDS; Africa. 1I am grateful to Andrew Foster for advice and guidance. My gratitude also goes to Charles Becker, Gervais Beninguisse, Francis Bloch, Ken Chay, Jakina Debnam, Esther Du(cid:135)o, Parfait Eloundou-Enyegue, Jane Fortson, Je⁄rey Greenbaum, Rachel Heath,LeighJohnson,NavinKartik,BlessingUchennaMberu,IsaacMbiti,RogerMyerson,SrinikethNagavarapu,MariePatience Pongou,SamPreston,MichellePoulin,LouisPutterman,PaulSchultz,RobertoSerrano,AdamStoreygard,OlumideTaiwo,David Weil, and several participants at the Brown University Department of Economics Micro Lunch, the 2008 Northeast Universities Development Consortium at Boston University, the 64th European Meeting of the Econometric Society in Barcelona, and the 26thIUSSP InternationalPopulationConferenceinMarrakechforusefuldiscussionsandfeedback. AnneBuvØandLindaMorison provided unique data from the Multicentre Study Group. The Department of Economics, the Graduate School, the Population Studies and Training Centerand the Hewlett Packard Foundation provided generous (cid:133)nancialsupport. 2DepartmentofEconomics,BrownUniversity,64WatermanStreet,Providence,RI02912,U.S.A.;[email protected]. 1 1 Introduction Weproposeatheoryofhowcommunity-levelethnicheterogeneitydeterminestheformationofsexualnetworks, and how these networks a⁄ect the spread of HIV/AIDS. Ethnic groups in a community constitute pre-de(cid:133)ned communication networks wherein information circulates more easily within groups than across. Limited com- munication across groups thus fosters cross-group anonymity, providing incentives for optimizing individuals to choose their sexual partners across groups rather than within, in order to minimize the probability of being caught. We show that this optimizing behavior in turn implies a mechanism wherein ethnic heterogeneity en- courages in(cid:133)delity and positively a⁄ects the spread of HIV/AIDS. We empirically test these predictions of the model. ThereisagrowingbodyofresearchontheeconomiccausesandconsequencesoftheAIDSepidemic, andon howtoreverseitsprogression(seee.g,Canning2006,Fortson2009a,2009b,BloomandMahal1997,Hoganand Salomon 2005). Most of this literature focuses on sub-Saharan Africa, where HIV/AIDS has slowed a century of substantial progress in the (cid:133)ght against infectious diseases, and has caused a sharp decline in life expectancy and Disability-Adjusted Life Years (UNAIDS 2008, Lopez et al 2006). This region accounts for 67 percent of the 33 million people living with HIV/AIDS today, and for 72 percent of AIDS deaths, but the epidemic is spreading rapidly in other parts of the world such as South and East Asia (UNAIDS 2008). Beyond its immediate impact on life quality and longevity, the AIDS epidemic has carried devastating social and economic impacts. By primarily striking prime-age adults, it has caused or exacerbated poverty in most households, and has increased the burden of orphanhood and widowhood (Matshalaga 2002, Monash and Boerma 2004, UNAIDS 2008). There is also evidence that the epidemic has negatively a⁄ected human capital investment and economic growth (Fortson 2009b, Bonnel 2000, Kalemli-Ozcan 2006).3 These devastating e⁄ects have made the study of the social and economic determinants of HIV/AIDS crucial for the design of policies to help curtail the growth of the epidemic. Documented determinants include socioeconomic status, spatial mobility, other untreated sexually channeled diseases, and limited behavioral change (Mishra et al. 2007, Fortson 2008, Oster 2005, 2007, Lurie et al 2003, Orubuloye and Oguntimehin 1999, Arna(cid:133)1993, Brokerho⁄and Biddlecom 1999, Canning 2006). Little is known about the role of community attributes in fostering multiple and concurrent partnerships, the most important factor in the rapid spread of HIV/AIDS (Epstein 2007, Halperin and Epstein 2004, Royce 3The claim that the AIDS epidemic has threatened the economic stability and growth of the a⁄ected countries is disputed. A study by Bloom and Mahal (1997) (cid:133)nds no impact of HIV/AIDS on growth. This study is consistent with Werker, Ahuja and Wendell (2006) who (cid:133)nd little e⁄ect of the AIDS epidemic on savings, fertility and growth. In the case of South-Africa, Young (2005)arguesthatAIDSwillreducefertilityandthusresourcedilution,andwillleadtoanimprovementintheeconomicconditions offuturegenerations. ButKalemli-Ozcan(2006)documentsapositivee⁄ectoftheAIDSepidemiconfertility,andFortson(2009a) (cid:133)nds no impact. 2 etal. 1997,MorrisandKretzschmar1997). Bydevelopingandtestinganewmodelthatrelatesethnicdiversity to sexual behavior, the current study increases our knowledge of the rationale and mechanism that underlie the formationofsexualnetworksinmulti-culturalenvironments. First,itshowsthatwithincountries,communities that feature greater ethnic diversity are more able to foster and sustain concurrent partnerships across di⁄erent ethnicgroups,andarethereforepronetogeneralizedepidemics.4 Second,thisstudyenhancesourunderstanding of both the dynamics and high concentration of HIV/AIDS in sub-Saharan Africa, the world(cid:146)s most ethnically diverse region.5 In so doing, it also informs public policies aimed at curbing the growth of this devastating epidemic in the region. In particular, our (cid:133)ndings speak loudly to the role of complete social and ethnic integration in minimizing the spread and e⁄ects of HIV/AIDS. We now summarize the theoretical model of the paper developed in section 3. In a community, reside an equalnumberofmenandwomenpartitionedintodi⁄erentethnicgroups,eachofequalsize. Agentsderiveutility fromsexualintercoursewithagentsoftheoppositesex,withutilityincreasinginthenumberofsexualpartners. However, an agent with multiple partners is committing an act of in(cid:133)delity and is punished if caught. Because information (cid:135)ows more easily within groups than across, the probability of in(cid:133)delity detection is greater if the cheated partners belong to the same group. We show that this underlying communication network structure causes optimizing agents to choose their sexual partners across ethnic groups rather than within, in order to decreasetheprobabilityofin(cid:133)delitydetectionandthusitsassociatedcost(Proposition1). Asacorollaryofthis optimizing behavior, it is shown that increasing the number of ethnic groups in the community results in each agent adding more sexual links from across these groups, which in turn increases the prevalence of HIV/AIDS (Proposition 2).6 But if the level of ethnic integration is su¢ ciently high that barriers between groups are weak, then the marginal e⁄ect of ethnic heterogeneity becomes very small. These (cid:133)ndings also introduce ethnic diversity as an underlying factor in the spatial variation of HIV/AIDS prevalence within countries, a topic that has received little attention in the literature.7 Weempiricallytestthetwomainpredictionsofthemodelinsections4and5. Theanalysesarebasedondata from eight sub-Saharan African countries: Benin, Burkina Faso, Cameroon, Ethiopia, Ghana, Kenya, Malawi 4The fact that people choose their partners from di⁄erent ethnic groups gives rise to networks that interconnect these groups, andthatwouldasaresulteasilyleadtoageneralizedepidemicifaconnectedindividualbecameinfectedfromanexogenoussource. 5Note that we do not examine the relationship between ethnic heterogeneity and HIV infection at the country level, but we use micro-level data, therefore controlling for country level factors such as HIV prevention policies, income, and a host of other macro-levelvariables. 6Note thatin general,although information circulatesmore easily within groupsthan across,ourenvironmentisfree ofethnic- based exclusions as regarding intimate relationships. Such exclusions, when they exist, are generally due to strong social norms, anti-miscegenation laws,orethnically driven civilwars,etc. A good analogy tothetypeofenvironmentconsidered in ouranalysis is an academic institution with severaldepartments. While con(cid:135)icts do not generally exist between departments,shared academic interests lead students and faculty to interact more frequently within their department than across. In the concluding section of our paper, we extend the (cid:133)delity model to a setting in which inter-ethnic con(cid:135)icts may prevent cross-group relationships, such as in a situation of an ethnically rooted civil war. In such a setting, HIV prevalence is expected to be low, which may partly explain the slow spread of the AIDS epidemic in countries like Rwanda or Congo. However, there are only such a few countries in Africa, which explains ourfocus on ordinary countries,and ourchoice in the modeling. 7Cross-country di⁄erences have been severally documented and explained by such factors as HIV prevention policies or male circumcision(see,e.g.,Asamoahetal. 2004,Potts2000). Potts(2000)arguesthatHIVprevalenceislowerinWestAfricacompared to East and Southern Africa because of widespread male circumcision practices in the former region (see also Potts et al. 2008). But as mentioned earlier, explaining cross-country variation in HIV prevalence is not our goal in this study. Our analysis controls forcountry-levelvariables as wellas culturalsources ofvariation in factors like male circumcision. 3 andZambia.8 Thesecountriesrepresenttoalargeextentthegeographicalandculturaldiversityofsub-Saharan Africa, and provide a representative mapping of the AIDS situation in the region. They are characterized by markedethnicandlinguisticdiversity(Alesinaetal. 2003),whichisconsistentwithanthropological(cid:133)ndingsthat most African societies are organized around distinct ethnic groups and enclaves. The commonality of language and culture facilitates communication and social interactions among people of the same group. Cultural and linguistic boundaries between ethnic groups however cause information transmission to be more costly and less e¢ cientacrossgroups. However,despiteanotedpersistenceofethnicidentitiesinmostsocieties,Anthropologist FredrikBarth(1969)arguesthattheseidentitiesarenotboundedandcannotbeviewedasculturalisolates. To this e⁄ect, he writes: (cid:147)[...] categorical ethnic distinctions do not depend on an absence of mobility, contact and infor- mation, butdoentailsocialprocessesofexclusionandincorporationwherebydiscretecategoriesare maintaineddespitechangingparticipationandmembershipinthecourseofindividuallifehistories.(cid:148) (P. 9) In post-colonial Africa, interethnic relationships within countries are further facilitated by the heritage of the colonial language, but ethnic identi(cid:133)cation and cleavages have persisted in most countries (Posner 2003, Eifert, Miguel and Posner 2007), resulting in moderately fragmented societies. As implied by our theoretical model, we expect this sort of cultural diversity to enable people to engage with multiple sexual partners across groups, which sets propitious grounds for the spread of HIV/AIDS. In section 4, we test the proposition that in ethnically diverse communities, optimizing individuals choose their sexual partners from di⁄erent ethnic groups, using standardized cross-sectional population surveys con- ductedinfourcities: Cotonou,Benin;Kisumu,Kenya;Ndola,Zambia;andYaounde,Cameroun. Thesesurveys were conducted in 1997 and 1998 by the Multicentre Study Group led by Anne BuvØ as part of an e⁄ort to understand factors determining the di⁄erential spread of HIV/AIDS in African countries.9 Samples were repre- sentative at the city level, and information was collected on sexual behavior, HIV status, socioeconomic status, ethnica¢ liationandmanyothervariables. Intervieweesalsoprovideddetailedinformationontheirnon-spousal sexual partners, including their ethnic a¢ liation.10 Our analysis of these data reveals that in about half of all relationships, the two partners come from di⁄erent ethnic groups. Also, of those individuals who had at least two sexual partners within the 12 months preceding the survey, more than 56% had partners who did not all belong to the same group. More importantly, we (cid:133)nd that to the extent that it is possible, people tend to choose their partners from di⁄erent groups; and as the number of an individual(cid:146)s partners increases, the degree to which he/she spread them out across di⁄erent groups increases as well. We further examine cross-ethnic 8Thetestofthe(cid:133)rstpredictionisbasedondatafrom fourcities(Cotonou,Benin;Yaounde,Cameroon;Kisumu,Kenya;Ndola, Zambia), and the test of the second prediction is based on nationally representative surveys carried out in six countries (Burkina Faso,Cameroon,Ethiopia,Ghana,Kenya,and Malawi). We therefore use data from eight countries. 9See BuvØ et al. (2001) fora detailed description ofthe study. 10Thisiscertainlyauniqueandinterestingfeatureofthesedata,asmostdataonsexualbehaviordonothavemorethanordinary information on individuals(cid:146)partners. 4 sexual interactions among members of majority and minority groups separately, and demonstrate that such in- teractions are not a result of a random matching process, but are actually strategic. All these (cid:133)ndings validate the (cid:133)rst prediction of our model. The data collected by the Multicentre Study Group provide a unique insight into how ethnic heterogeneity determinesthechoiceofsexualpartners. However,duetothefacttheyareonlyavailableinfourcitiesindi⁄erent countries, we do not use them to test the proposition that community-level ethnic heterogeneity encourages multiple partnerships and facilitates the spread of HIV/AIDS. To test this proposition requires that ethnic heterogeneity be measured at the community level, which would only yield a few values in the Multicentre Study Group data. For this reason, we resort to nationally representative Demographic and Health Surveys (DHS), supplemented by data on population density from censuses. This analysis in conducted in section 5 The DHS constitute a unique source of information on individuals(cid:146)sexual behavior, HIV status, and ethnic a¢ liation, and have been used in recent studies to examine the economic causes and consequences of the AIDS epidemic in sub-Saharan Africa (see, e.g., Oster 2005, Fortson 2008, 2009a, 2009b, etc.). Information on ethnic a¢ liationenablesustocomputeanewlyde(cid:133)nedindexofethnicheterogeneityattheclusterorcensuszonelevel and at the region/urban/rural level.11 We (cid:133)nd a positive e⁄ect of ethnic heterogeneity on the number of sexual partners and on HIV infection. This e⁄ect is robust to the inclusion of a range of demographic, socioeconomic and culturalvariables. Despitethefactthatthecluster-leveland theregion/urban/rural-levelindexes ofethnic heterogeneity are only moderately correlated, the e⁄ect of the (cid:133)rst index remains unchanged while that of the second index disappears once both variables are controlled for. This result indicates that the space of strategic sexualnetworkingisnarrowandbestrepresentedbyclusters. Thee⁄ectofethnicheterogeneityisalsorobustto an alternative measure wherein the index is decomposed into two indexes representing respectively the distinct contribution of individuals(cid:146)own and opposing ethnic groups to the overall index.12 In sections 6-8, we conduct several sensitivity and robustness checks that allow to rule out alternative explanations for the e⁄ects of ethnic heterogeneity on sexual behavior and HIV infection. First, we show that these e⁄ects are robust to separate estimation by urban and rural areas, establishing that they do not simply re(cid:135)ect urban-rural di⁄erences in the outcomes. Second, they persist when estimated separately for migrants and non-migrants. Third, they are robust when corrected for sample bias due to HIV non-response. Fourth, they are not driven by a lack of public goods in more ethnically diverse communities.13 Finally, we show that 11Thereare2740censuszonesand100region/urban/ruralstratainthesixcountriesconsideredintheanalysis. Strataareobtained from stratifying each region of a country into urban and rural areas. A stratum is therefore a collection of clusters. Conducting theanalysisatthismoreaggregateleveltakesintoaccountthefactthatthespaceofsexualnetworkingforanindividualmightbe largerthan the clusterin which he/she resides. 12The ethnic heterogeneity index is a sum of e terms, e being the number of ethnic groups, and each term representing the contribution of a speci(cid:133)c group to the index. From an agent(cid:146)s perspective, the decomposition of the index allows to di⁄erentiate betweenhis/herownethnicgroupandtheopposinggroups,butdoesnotdi⁄erentiatebetweentheopposinggroups. Itisreminiscent of the (cid:147)Us/They(cid:148)dichotomization of Duclos, Ray and Esteban (2004), but our index is di⁄erent from theirs. This decomposition also accounts forpotentialgroup and socialpreferences in sexualbehavior. 13Arecentliteratureshowsthatpublicgoodsareunderprovidedinmoreethnicallydiversecommunitiesbecauseofalackoftrust andsocialcapital(EasterlyandLevine1997;Alesina,BaqirandEasterly1999,MiguelandGugerty2005,Habyarimanaetal2008; also seeCosta and Kahn 2003 and thereferencestherein),butthetheoreticalprediction ofthisrelationship isambiguous(Alesina and La Ferrera 2000). 5 ethnic heterogeneity has no e⁄ect on anemia. Anemia is highly prevalent in most African countries, is largely untreated like HIV/AIDS, but unlike HIV/AIDS, it does not involve socially prohibited human interactions. A positive e⁄ect of ethnic heterogeneity on such a condition would largely call into question our interpretation of the positive e⁄ect of ethnic heterogeneity on sexual promiscuity and HIV infection. That we (cid:133)nd no such e⁄ect lends credence to our theory and argument that diseases which involve socially prohibited human interactions spread through hidden and anonymous networks. Insection9,wediscusstwoimplicationsofthe(cid:133)delitymodel. The(cid:133)rsthastodowithhowmarriagenetworks di⁄erfromextra-maritalnetworksintermsoftheethnicbackgroundofpartners. Weprovideempiricalevidence that the former are more segregated than the latter along ethnic lines, due to the fact that most people marry within their group, but cheat outside to hide their in(cid:133)delity. The second implication has to do with how the ability to spread out partners across di⁄erent ethnic groups depends on whether one belongs to a minority group or a majority group. In this respect, we show that members of minority groups are more able to play the optimalstrategy, primarilyduetotheshortsupplyoftheirtypeinthesexualmarket. Empirically, we(cid:133)ndthat the extent to which an average minority individual spreads out his/her partners across groups is similar to the extent to which the overall population is ethnically diverse. This is especially true in Kisumu and Ndola where HIV prevalence happens to be high. These (cid:133)ndings show that cross-ethnic sexual interactions are not random, but are strategic, as a random matching process would imply the same probability of matching across groups for members of minority and majority groups. In our concluding section (section 10), we discuss the implications of our main theoretical results for more integrated societies, including Muslim societies where religion can be viewed as a unifying factor possibly eclipsing the divisive role of ethnicity. We also show how our model can be extended to "separated" societies, such as those marked by ethnically driven civil wars. In both cases, HIV prevalence is expected to be low, but for di⁄erent reasons. In the (cid:133)rst case, there is no incentive for strategic cross-ethnic sexual interactions, and in the second case, sexual interactions across groups are very costly and are thus precluded. For most African countries however, ethnic integration is incomplete. This is because policies of integration undertaken after independence were only partially successful in these countries, providing an institutional platform on which ethnicnationalitiescould interactwithoutcompletelyuniting. Inthispaper, we showthatethnicheterogeneity is a driving factor in the spread of HIV/AIDS in such countries. In the next section (section 2), we discuss the literature. 2 Related literature To the best of our knowledge, no study has examined the role of ethnic diversity in sexual network formation and the spread of HIV/AIDS in the way we do in this paper. We however view our work as relating to several literatures, aside from contributing to the literature on the social and economic determinants of HIV/AIDS. 6 First, our theoretical model is an extension to a multi-ethnic environment of the basic framework of the (cid:133)delity model developed in Pongou (2009a). This paper studies network formation in a mating economy with two types of agents (e.g., men and women), where each enjoys having relationships with the opposite type, while having multiple partners is viewed as in(cid:133)delity and is punished if detected by the cheated partner. It provides a characterization of pairwise stable networks, examines their welfare properties, and derives conditions under which female discrimination leads to higher HIV prevalence among women.14 The current study di⁄ers in the modeling, as agents derive utility not only from the number of partners they have, but also from their ethnic composition. It also has a di⁄erent scope, as we are interested in how ethnic diversity a⁄ects sexual behavior and the spread of HIV/AIDS, with a focus on empirically testing the predictions of our generalized model. In assuming that agents trade o⁄the cost and bene(cid:133)t of forming relationships with each other, our study also shares some features of the emerging literature on endogenous network formation (see, e.g., Aumann and Myerson 1988, Jackson and Wolinsky 1996, Dutta and Mutuswami 1997, Jackson and Nouweland 2000, Bala and Goyal 2000, Page, Wooders and Kamat 2001, Jackson and Watts 2002).15 Most of this literature assumes the observability of network con(cid:133)gurations by agents. In our study however, the mechanism driving network formation,whichis"(cid:133)delitytopartners",impliesthatapriori,agentsdonotknowtheirpartners(cid:146)otherpartners, and thus do not observe the structure of the network to which they belong.16 Our networks end up having very di⁄erent incentive properties. Our study also relates to the literature on how identity and socio-cultural distance a⁄ect social interactions as well as decisions such as schooling and childbearing (see, e.g., Lazersfeld and Merton 1954, Akerlof 1997, Akerlof and Kranton 2000, Mcpherson, Smith-Lovin and Cook 2001, Currarini, Jackson and Pin 2009). Most of these studies document a tendency in people to interact with those who are similar to them. We however elucidateamechanismwhereinculturaldissimilaritygivesrisetoanothersortofinteractions-sexualpromiscuity - which by their nature are socially "prohibited", and which occur among culturally distant individuals. Our analysisthussuggeststhattheextenttowhichmoreculturallysimilarindividualsinteractwitheachothermay dependonthesocialacceptabilityofsuchinteractions. Whereasthefocusintheexistingliteraturedocumenting same-type bias has been on socially accepted relationships like friendships, business partnerships and marriage relationships, we consider sexual promiscuity, which is of a di⁄erent nature. The prohibition of sexual in(cid:133)delity implies that optimizing agents choose their partners from di⁄erent ethnic groups, taking advantage of cross- groupanonymitytohidetheirbehavior. Inthissense,ourresultsareconsistentwithexperimentalstudiesonthe e⁄ect of anonymity on the reporting of sensitive behaviors.17 In general, our analysis implies that ethnic-based exclusions are limited in the sphere of sexual relationships, unlike in marriage relationships, something that we 14The analysis in this paper is extended to a dynamic setting in Pongou and Serrano (2009). 15Also see Jackson (2003) and the references therein. 16They form what Anthrologist Robert Thornton (2008) has coined an "unimagined community". This phrase re(cid:135)ects the fact that individuals involved in a sexual network do not generally have full information on their indirect links and do not know the othermembers ofthe network. 17Ithasbeenshownforinstancethataudiocomputer-assistedself-interviewingproducesmoreaccuratereportingofsexualactivity and drug injections than face-to-face interviews (Mensch et al. 2003,Jaya et al. 2008,Des Jarlais et al. 1999). 7 also show empirically. Finally,bydemonstratingapositivee⁄ectofethnicheterogeneityonthespreadofHIV/AIDS,ourstudycan alsobeseenascontributingtotheliteratureonhowethnicheterogeneitya⁄ectseconomicdevelopment(see,e.g., EasterlyandLevine1997;Alesina,BaqirandEasterly1999,CinyabugumaandPutterman2006,Horowitz1985, CostaandKahn2003, etc.). Thisisbecausehealthisanimportantdimensionofhumancapitalandasignicant input into the production of wealth. We however demonstrate a di⁄erent mechanism or pathway of impact. The existing literature hypothesizes that ethnic diversity leads to ethnic-based exclusions, which in turn lead to growth-retarding policies and sub-optimal contributions to public goods. Our study shows that the positive e⁄ect of ethnic heterogeneity on the spread of HIV/AIDS is driven by strategic cross-ethnic sexual network formation. Infact,inthepresenceofstrongethnic-basedexclusionsinthesphereofintimaterelationships,only segregatedsexualnetworkswouldform, andthespreadofHIV/AIDSwouldbeslowasaresult. Wedonot(cid:133)nd any evidence for this in the data. In addition, we show that our results are not driven by a lack of public goods in more ethnically diverse communities. Our study thus suggests that ethnic diversity plays out di⁄erently in the sphere of public policies and that of intimate relationships. In the latter sphere, it brings people together, withtheiractingstrategicallytominimizein(cid:133)delitydetection, givingrisetonetworksthatarepropitioustothe spread of HIV/AIDS. As such, we o⁄er a new and fresh explanation for the high concentration of the AIDS epidemic in sub-Saharan Africa. 3 The (cid:133)delity model 3.1 Individual decision-making Wemodeltheindividualdecision-makingofsexualpartnerschoiceinacommunitycomposedofeethnicgroups G , ..., G , each of equal size. Each group has a distinct language and culture, which makes social interactions 1 e easier and more intense within groups than across. We assume that each individual derives utility from sexual relationships with the opposite sex. However, sexual in(cid:133)delity, which consists of having many partners, is punished if detected by the cheated partner. We now de(cid:133)ne agents preferences more formally. Assume that an agent i has s sexual partners. His/her utility function is: u(s)=v(s) c(s) (cid:0) where v(s) is a concave bene(cid:133)t function increasing in s, and c(s) the cost of in(cid:133)delity. We would like to answer the three following questions: What is the optimal number of partners s for agent i? (cid:3) (cid:15) 8 How should he/she allocate these partners across ethnic groups? (cid:15) How does the number of ethnic groups e a⁄ect s ? (cid:3) (cid:15) To answer these questions requires that we de(cid:133)ne the in(cid:133)delity cost function c(s) more explicitly. Let j and k be two sexual partners that agent i has, and D(j : (i;k)) the event that j detects the liaison between i and k. We assume that D(j :(i;k)) and D(k :(i;j)) occur with the same probability, which is de(cid:133)ned as follows: p if h 1;::e :j G and k G h h 9 2f g 2 2 Prob(D(j :(i;k)))=Prob(D(k :(i;j)))=8 q if not >>>< with q <p: That is, D(j :(i;k)) and D(k :(i;j)) occurs with pr>>>:obability p if j and k belong to the same ethnic group, andwithprobabilityqifthetwoindividualsbelongtodi⁄erentethnicgroups. Theassumptionthatq <pmeans that the cheater, who is i, is more likely to be caught by his/her cheated upon partners j and k if the latter belong to the same group than if they belong to two di⁄erent groups. This assumption comes from the fact that linguistic and cultural barriers between groups make the circulation of information easier, more e¢ cient, and less costly within groups than across. Because i may choose his/her sexual partners from di⁄erent groups, his/her decision function can be ex- pressed as: f :R+ !Re+ s (s ;:::;s ) with s =s 1 e h ! P where f is a mapping from the set of positive real numbers R+ to the e dimensional cartesian product of (cid:0) R+ Re+.18 f associateswithanynumberofsexualpartnerssofagentianallocation(s1;:::;se)ofthesepartners across the e ethnic groups that make up the community; s is the number of sexual partners chosen by agent i h from ethnic group G , with 1 h e. h (cid:20) (cid:20) Assume that i has s partners allocated across groups according to the vector (s ;:::;s ). A partner j 1 e of agent i belonging to ethnic group G is expected to detect p(s 1)+q(s s ) liaisons of i with other h h h (cid:0) (cid:0) partners. The s sexual partners of agent i that belong to ethnic group G are thus expected to detect h h s [p(s 1)+q(s s )] liaisons of i with other partners. Summing across all partners and ethnic groups yields h h h (cid:0) (cid:0) e s [p(s 1)+q(s s )], which is the total expected number of liaisons that the s sexual partners that i h=1 h h(cid:0) (cid:0) h Phas will detect. Assuming that i incurs a punishment cost of c per detection, he/she will incur a total expected cost of: c(s;s ;:::;s )= e s [p(s 1)+q(s s )]c 1 e h=1 h h(cid:0) (cid:0) h 18TheanalysisshouldnormallybeconductedwithintPhediscretesetofnaturalnumbersN. Wechoosetodoitinthecontinuous space R+, which allows fractional number of partners, to simplify the exposition of the results, but all these results qualitatively hold in the discrete space Nas well. 9 Agent i(cid:146)s expected utility function is therefore: Eu(s;s ;:::;s )=v(s) e s [p(s 1)+q(s s )]c: 1 e (cid:0) h=1 h h(cid:0) (cid:0) h P We will derive the answer to our (cid:133)rst question from that of the second. Assume that i is an expected utility maximizer. His/her decision-making problem is thus: maxEu(s;s ;:::;s )=v(s) e s [p(s 1)+q(s s )]c 1 e (cid:0) h=1 h h(cid:0) (cid:0) h subject to s =s h (s;s1;:::;se) P P Lets betheoptimalnumberofsexualpartnersfori,and(s ;:::;s )theoptimalallocationofthesepartners (cid:3) (cid:3)1 (cid:3)e across the e ethnic groups. If we assume p, q and c to all be strictly positive, the following proposition states that it is optimal for i to equally distribute his/her sexual partners across groups. Proposition 1 : s =s =:::=s = s(cid:3). (cid:3)1 (cid:3)2 (cid:3)e e Proof : The Lagrangian is: =v(s) e s [p(s 1)+q(s s )]c (cid:21)( e s s): L (cid:0) h=1 h h(cid:0) (cid:0) h (cid:0) h=1 h(cid:0) P P For any h 1;:::;e , taking the derivative of with respect to s yields: h 2f g L [p(s 1)+q(s s )]c+s (p q)c (cid:21)=0 h h h (cid:0) (cid:0) (cid:0) (cid:0) which implies that for any h; l 1;:::;e : 2 f g [p(s 1)+q(s s )]c+s (p q)c=[p(s 1)+q(s s )]c+s (p q)c h h h l l l (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) which in turn implies that for any h; l 1;:::;e : 2f g 2s (p q)c+( p+qs)c=2s (p q)c+( p+qs)c: h l (cid:0) (cid:0) (cid:0) (cid:0) Solving for s yields: h s =s h l which implies: s =s =:::=s (*). 1 2 e Taking the derivative of with respect to (cid:21) yields: L e s s=0 (**). h=1 h(cid:0) P Given (*), substituting for each s with s in (**) yields: h 1 10
Description: