WPS7699 Policy Research Working Paper 7699 Women’s Empowerment, Sibling Rivalry, and Competitiveness Evidence from a Lab Experiment and a Randomized Control Trial in Uganda Niklas Buehren Markus Goldstein Kenneth Leonard Joao Montalvao Kathryn Vasilaky Africa Region Office of the Chief Economist & Gender Cross Cutting Solution Area June 2016 Policy Research Working Paper 7699 Abstract This study looks at how a community event—adolescent no gender differences in competitiveness among adoles- women’s economic and social empowerment—and a family cents, on average. It also finds no evidence of differences factor—sibling sex composition—interact in shaping in competitiveness between girls in treatment and control gender differences in preferences for competition. To do communities, on average. However, in line with the lit- so, a lab-in-the-field experiment is conducted using com- erature, in control communities the study finds that boys petitive games layered over the randomized rollout of a surrounded by sisters are less competitive. Strikingly, this community program that empowered adolescent girls in pattern is reversed in treatment communities, where boys Uganda. In contrast with the literature, the study finds surrounded by (empowered) sisters are more competitive. This paper is a product of the Office of the Chief Economist, Africa Region and the Gender Cross Cutting Solution Area. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank. org. The authors may be contacted at jmontalvao@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team    Women’s Empowerment, Sibling Rivalry, and Competitiveness:   Evidence from a Lab Experiment and a Randomized Control  Trial in Uganda*    Niklas Buehren     Markus Goldstein     Kenneth Leonard  Joao Montalvao     Kathryn Vasilaky†                                      JEL Classification: C9, I25, J13, J24, O12.  Keywords: Gender and competitiveness, adolescent girls’ empowerment, lab‐in‐the‐field  experiments.      *  We  thank  all  BRAC  Uganda  staff  for  collaborative  efforts  in  this  project.  We  have  benefited  from  discussions  with  Imran  Rasul  and  numerous  seminar  and  conference  participants.  We  are  grateful  to  the  Africa  Gender  Innovation  Lab,  Bank‐Netherlands  Partnership  Programme,  the  World  Bank  Group’s  Umbrella  Facility  for  Gender  Equality  and  Adolescent  Girls  Initiative  Multi‐ Donor Trust Fund for financial support.  †  Buehren:  World  Bank  [nbuehren@worldbank.org];  Goldstein:  World  Bank  [mgoldstein@worldbank.org];  Leonnard:  University  of  Maryland  [kleonard@arec.umd.edu];  Montalvao:  World  Bank  [jmontalvao@worldbank.org];  Vasilaky:  Columbia  University  [katyav@iri.columbia.edu]  1. Introduction  Gender differences in preferences for competition have emerged as a possible explanation for why men  and  women  experience  different  labor  market  outcomes  [Bertrand  2011].  Contributing  to  this  view,  a  large  experimental  literature  documents  that  women  are  on  average  less  competitive  than  men  [Niederle  and  Vedsterlund  2007,  2011],  and  a  few  recent  studies  indicate  that  gender  differences  in  preferences  for  competition  measured  in  the  lab  predict  gender  differences  in  important  education  and  labor market outcomes outside the lab [Buser et al. 2014, Reuben et al. 2015a, 2015b].   A  growing  body  of  evidence  provides  a  richer  picture.  These  studies  highlight  the  importance  of  environmental  factors  such  as  cultural  norms,  family  experiences,  age,  and  the  gender  composition  of  the  environment  in  shaping  gender  differences  for  competition  [e.g.  Gneezy  et  al.  2009,  Booth  and  Nolen  2012,  Cardenas  et  al.  2012,  Andersen  et  al.  2013,  Cameron  et  al.  2013,  Zhang  2015,  Okudaira  et  al.  2014,  Khachatryan  et  al.  2015,  Almås  et  al.  2015].  For  example,  Gneezy  et  al.  [2009]  document  that  gender  differences  in  competitiveness  depend  on  cultural  and  social  norms.  While  the  authors  find  that  men  are  more  competitive  than  women  among  the  patriarchal  Masai  in  Tanzania,  they  also  find  that  women compete more than men among the matrilineal Khasi in India.    These  findings  point  towards  the  importance  of  nurture  in  shaping  gender  differences  in  competitiveness,  and  thus  to  the  potential  role  of  carefully  designed  human  capital  interventions  in  reducing  these  differences.  Yet  rigorous  evidence  on  which  specific  interventions  can  achieve  this  is  nonexistent. We examine this issue with a lab‐in‐the‐field experiment layered over a randomized control  trial  of  a  community  program  that  empowers  adolescent  girls  in  Uganda.  The  program  successfully  empowered  girls  along  economic  and  social  dimensions,  and  shifted  deep  rooted  social  and  gender  norms in communities randomly assigned to the program. In this paper we look at whether the program  also made girls more competitive.   We  also  study  how  a  key  family  environmental  factor  –  sibling  sex  composition  –  interacts  with  the  program in shaping preferences for competition. Gender is about dynamic relations. Thus when girls are  empowered  we  might  expect  boys  who  interact  with  them  to  respond  differently.  These  interactions  take  place  in  different  institutional  settings,  starting  in  the  family.  Previous  research  has  documented  the  importance  of  sibling  composition  in  shaping  preferences  for  competition  [Cameron  et  al.  2013,  Okudaira  et  al.  2015].  In  particular,  Okudaira  et  al.  [2015]  show  that  having  sisters  is  associated  with  reduced competitiveness for men in urban Japan.1 In this paper we examine how these sibling dynamics  vary with a community‐level exogenous shift in adolescent girls’ empowerment.  1 The  psychology  literature  has  long  highlighted  the  importance  of  sibship  sex  composition  in  personality  development.  In  particular,  research  starting  with  Kock  [1955]  and  Brim  [1958]  has  documented  that  boys  surrounded  by  sisters  tend  to  substitute ‘masculine’ with ‘feminine’ traits.  2 We  present  four  main  findings.  First,  in  contrast  with  the  literature,  we  find  that  adolescent  girls  and  boys  in  Uganda  compete  at  similar  levels.  Second,  we  also  do  not  find  evidence  that  the  empowerment  effects  of  the  program  on  girls  were  accompanied  by  increased  competitiveness.  Third,  in  line  with  previous  evidence  we  document  that  having  sisters,  relative  to  brothers,  is  associated  with  a  lower  willingness  to  compete  for  boys  in  control  communities.  Fourth,  strikingly,  this  pattern  is  reversed  in  treatment  communities.  There,  we  find  that  boys  surrounded  by  (empowered)  sisters  are  more  willing  to engage in competitive behavior.   The  paper  is  organized  as  follows.  Section  2  outlines  the  program  and  its  impacts  in  order  to  sketch  some of the theoretical background motivating the analysis. Section 3 describes the data and the lab‐in‐ the field experimental design. Section 4 presents the results. Section 5 concludes.  2. Background and Theoretical Context  The  Empowerment  and  Livelihood  for  Adolescents  (ELA)  program  in  Uganda,  implemented  by  the  NGO  BRAC,  provides  adolescent  girls  with  training  on  ‘hard’  vocational  skills  to  help  them  successfully  transition  into  the  labor  market,  and  on  ‘soft’  life  skills  to  enable  girls  to  make  informed  choices  about  sex,  reproduction  and  marriage.  The  program  is  delivered  from  development  adolescent  clubs,  which  are  open  five  afternoons  per  week  after  school,  and  are  led  by  a  local  female  mentor.  Participation  is  voluntary for girls aged between 14 and 20. To date BRAC has started 1,505 clubs in Uganda which have  reached 70,000 girls.   Bandiera  et  al.  [2015]  evaluate  the  ELA  program  in  Uganda  using  a  randomized  control  trial  methodology. The study took place in the urban or semi‐urban regions of Kampala and Mukono, and the  mostly rural region around Iganga and Jinja. The research design delivered 100 treatment and 50 control  communities.  Each  treatment  community  hosted  an  ELA  club.  A  baseline  survey  was  administered  just  before  the  establishment  of  the  clubs  to  about  6,000  adolescent  girls,  4,000  (2,000)  of  whom  resided  in  treatment  (control)  communities.  A  first  follow‐up  survey  took  place  in  2010.  Bandiera  et  al.  [2015]  show  that  within  two  years  the  ELA  program  successfully  empowered  girls  along  several  economic  and  social  dimensions.2 We  list  these  impacts  below  in  more  detail  in  order  to  understand  how  they  could  translate  into  increased  competitiveness  for  girls.  We  then  discuss  how  these  impacts  could  interact  with sibling sex composition in shaping boys’ competitiveness.  2.1. Women’s Empowerment and Competitiveness  There  are  at  least  two  possible  channels  through  which  the  program  could  foster  competitiveness  for  girls.  The  first  channel  operates  through  the  program’s  impact  on  economic  empowerment.  Bandiera  et  2 As club participation is voluntary, both Bandiera et al. [2015] and our analysis focus on intent‐to‐treat impacts.  3 al.  [2015]  show  that  the  program  accelerated  girls’  transition  into  the  labor  market.  Specifically  adolescent  girls  in  treatment  communities  were  significantly  more  likely  to  be  self‐employed  relative  to  girls  in  control  communities.  This  effect  was  accompanied  by  significant  increases  in  girls’  entrepreneurial ability as well as reductions in self‐reported anxieties related to transition into the labor  market.  In  light  of  the  evidence  that  preferences  for  competition  are  an  important  determinant  of  success  in  the  labor  market  [Buser  et  al.  2014,  Zhang  2014,  Flory  et  al.  2015,  Berge  et  al.  2015,  Buser  et  al.  2015,  Reuben  et  al.  2015a,  2015b],  it  is  thus  plausible  that  increased  competitiveness  could  have  acted as an additional mechanism through which the program improved girls’ economic empowerment.   The  second  channel  operates  through  the  program’s  impact  on  social  empowerment.  Bandiera  et  al.  [2015]  show  that  the  program  significantly  improved  the  quality  of  girls’  control  over  their  bodies  and  views  on  gender  norms.  Specifically  the  program  reduced  early  childbearing,  early  marriage  and  having  had  sex  unwillingly  for  adolescent  girls  in  treatment  communities  relative  to  girls  in  control  communities.  These  results  were  accompanied  by  a  significant  shift  in  girls’  adherence  to  prevailing  gender  norms  as  measured  by  their  views  on  gender  roles,  early  childbearing,  early  marriage  and  desired  fertility.  To  the  extent  that  girls  in  control  communities  might  shy  away  from  competitive  situations  because  this  is  what  is  expected  from  them  under  traditional  gender  norms  [Akerlof  and  Kranton  2000,  Bertrand  2011],  it  is  thus  possible  that  by  relaxing  the  psychological  pressure  of  these  norms the program increases competitiveness for girls in treatment communities.   2.2. Sibling Sex Composition and Competitiveness  Why  could  sisters  affect  their  brothers’  preferences  for  competition,  and  why  could  the  program  affect  that  relationship?  We  provide  possible  theoretical  underpinnings  drawing  on  literatures  from  both  economics  and  psychology,  and  again  organize  the  discussion  along  the  economic  and  social  empowerment impacts of the program.   First,  the  program  could  affect  the  influence  of  sisters  on  their  brothers’  competitiveness  through  its  economic  empowerment  impacts.  To  see  why  it  is  useful  to  extend  the  standard  intra‐household  framework  of  sibling  competition  over  limited  parental  resources  [Becker  and  Lewis  1973].  Parents  allocate  resources  across  children  based  on  returns  to  investment  and  these  returns  are  affected  by  economic  opportunities  outside  the  household.  In  Uganda,  sons  are  strongly  valued,  as  families  traditionally  adhere  to  patrilineal  inheritance  and  patrilocal  residence  customs.  Boys  surrounded  by  (non‐empowered)  sisters  in  control  communities  might  thus  be  less  used  to  competition  over  scarce  resources  [Garg  and  Morduch  1998,  Morduch  2000],  and  choose  to  compete  less.3 But  the  program  raises  the  marginal  return  from  investing  in  daughters  in  treatment  communities.  As  a  result,  boys  3 In  Uganda,  a  girl  cannot  succeed  her  father,  and  when  she  marries  she  leaves  her  village  and  produces  children  for  another  clan. In more traditional communities, girls thus are thought to have little value as children [Beyeza‐Kashesya et. al 2010].    4 surrounded  by  (empowered)  sisters  in  treatment  communities  might  raise  their  competitiveness  in  response to increased sibling rivalry over parental resources.4   Second, the program might also affect the influence that sisters have on their brothers’ competitiveness  through  its  social  empowerment  impacts.  By  challenging  prevailing  gender  norms,  the  program  might  have  threatened  boys’  sense  of  traditional  male  identity  [Akerlof  and  Kranton  2000].  This  is  particularly  true  for  boys  with  relatively  more  sisters,  as  they  are  more  likely  to  be  indirectly  exposed  to  the  program.  This  gender  identity  threat  can  trigger  an  opposing  response  aimed  at  restoring  the  damaged  self‐image  [Bénabou  and  Tirole  2011].  If  being  competitive  is  part  of  the  male  identity  under  traditional  gender  norms,  boys  surrounded  by  sisters  in  treatment  communities  can  subsequently  become  more  competitive as a form of cognitive dissonance reduction.   3. Sampling and Lab Experimental Design  3.1. Sampling    The  participants  in  our  experiment  were  randomly  drawn  from  the  sampling  frame  of  households  surveyed by Bandiera et al. [2015], stratified to include both control and treatment communities for the  randomized  control  trial  impact  evaluation  of  the  ELA  program.  A  total  of  700  adolescents  participated  in the experiment, 40 percent of whom came from control communities and 60 percent from treatment  communities. About 55 percent of the participants were girls, both in treatment and control groups.   At  the  end  of  each  experimental  session,  participants  completed  a  short  exit  survey  asking  basic  background  socio‐demographic  related  questions  such  as  their  age,  education,  marital  status,  and  whether  they  have  any  children.  The  survey  also  contains  sibling  sex  composition  information  on  participants,  namely  the  number  of  brothers  and  sisters  they  have.  Table  1  presents  summary  statistics  on  these  characteristics  for  the  full  sample  of  participants  and  separately  for  girls  and  boys  in  control  and treatment communities.   Four  points  are  of  note.  First,  participants  in  our  experiment  are  in  their  late  adolescence  and  early  adulthood:  the  average  participant  is  17  years  old,  and  just  under  80  percent  are  aged  14‐21  years  (not  shown).  The  bulk  of  the  participants  are  thus  past  the  critical  puberty  age  around  which  gender  differences  in  competitiveness  have  been  documented  to  arise  elsewhere  [Andersen  et  al.  2014].  Second,  the  average  participant  has  2.6  sisters  and  2.7  brothers.  This  large  family  size  is  in  line  with  the  fact  that  Uganda  has  one  of  the  highest  fertility  rates  in  the  world,  and  it  provides  us  with  enough  4 The  notion  that  sibling  rivalry  can  help  shape  personality  traits  has  been  extensively  hypothesized  in  the  psychology  literature  [Sulloway  1966].  While  our  lab  experiment  takes  place  only  two‐years  post‐intervention,  the  psychology  literature  also  documents  that  new  behaviors  in  one  setting  quickly  shape  preferences  over  behaviors  in  other  settings  [Breer  and  Locke  1965].  5 variation  in  the  data  to  estimate  the  effect  of  sibling  sex  composition  on  participants’  competitiveness.5  Third,  there  are  no  systematic  treatment‐control  differences  along  observable  dimensions,  both  among  girls  and  boys.  This  is  consistent  with  the  evidence  in  Bandiera  et  al.  [2015]  documenting  that  the  random  assignment  of  the  program  successfully  balanced  treatment  and  control  groups.  Fourth,  girls  differ  from  boys  along  some  observable  background  characteristics.  Namely,  girls  are  more  likely  than  boys  to  be  out  of  school,  to  be  married,  and  to  have  children.  Girls  also  have  on  average  more  siblings  than  boys.  In  the  main  empirical  analysis  it  will  thus  be  important  to  account  for  these  observable  differences between girls and boys when measuring gender differences in competitiveness.  3.2. Design  We  implemented  the  experimental  protocol  of  Niederle  and  Vesterlund  [2007].  The  experimental  task  consists  of  successfully  ordering  six  eight‐sided  building  blocks  with  various  shapes  (square,  circle,  star,  etc.)  on  each  side  from  smallest  to  largest  in  3  minute  rounds.  The  relative  location  of  the  shapes  on  each  of  the  six  blocks  is  different,  and  the  order  of  the  blocks  for  one  shape  (e.g.  star)  differs  for  the  order  of  the  blocks  for  all  other  shapes  (e.g.  circle,  square,  triangle).  Participants  were  then  paid  based  on the number of shapes they ordered in the 3 minutes.   There  were  four  different  experimental  rounds,  and  participants  were  paid  for  one  of  the  four  selected  at  random.  Participants  were  first  anonymously  matched  in  groups  of  four  (they  do  not  know  who  else  was  in  their  group,  nor  were  group  members  present  in  the  room).  In  round  1,  participants  were  paid  400  Ugandan  Shillings  (UGX)  for  each  correctly  completed  set  of  shapes  (uncompetitive  piece‐rate).  In  round  2,  the  participant  in  the  group  who  ordered  the  greatest  number  of  shapes  in  the  group  received  four  times  the  piece‐rate  payment,  or  1,600  UGX  per  set,  while  the  other  participants  received  no  payment  (competitive  tournament).6 In  round  3,  participants  first  chose  the  compensation  scheme  (either  piece‐rate  or  tournament),  and  then  performed  the  task.  They  were  told  that  if  they  chose  to  compete  then  their  performance  would  be  compared  against  their  group’s  round  2  performances.  In  round  4,  there  was  no  task,  and  participants  simply  had  to  choose  whether  they  prefer  their  past  round  1  performance  to  be  paid  under  the  piece‐rate  or  the  tournament  compensation  scheme.  If  they  chose  the  latter,  their  performance  was  compared  against  the  round  1  performances  of  the  other  participants  in  the  group.  At  the  end,  all  participants  were  asked  their  beliefs  about  their  relative  performance  in  rounds 1 and 2, and were paid 100 UGX for each correct guess.   The  outcome  of  interest  in  this  experiment  is  the  choice  of  compensation  scheme  in  round  3  (uncompetitive  piece‐rate  versus  competitive  tournament),  from  which  we  identify  participants’  preferences  for  competition.  The  first  two  rounds  serve  to  familiarize  participants  with  the  task  under  5  Uganda  has  the  fifth  highest  fertility  rate  in  the  world  [https://www.cia.gov/library/publications/the‐world‐ factbook/rankorder/2127rank.html].  6 At the time our experiment took place (November, 2011) a 1 USD was worth roughly 2500 UGX.  6 each payment scheme, but most importantly to control for any possible differences in ability. The choice  in  round  4  serves  to  control  for  the  influence  of  factors  that  may  affect  willingness  to  compete  other  than  a  pure  preference  to  perform  under  a  competitive  environment,  such  as  risk‐aversion,  feedback‐ aversion, and self‐confidence [for a detailed discussion see Niederle and Vesterlund 2007].   4. Results  4.1. Descriptive Evidence  Table  2  presents  descriptive  evidence  on  the  experimental  results  to  preview  our  main  findings  on  the  impact  of  the  program  on  gender  differences  in  competitiveness.  The  first  row  focus  on  the  choice  of  compensation  scheme  in  round  3,  the  key  outcome  of  interest.  We  see  that  tournament  entry  does  not  vary  significantly  by  gender  and  treatment  status.  In  control  communities,  36  percent  of  both  girls  and  boys  choose  to  compete  (Columns  2  and  3).  In  treatment  communities,  33  percent  of  girls  and  37  percent  of  boys  make  this  choice  (Columns  4  and  5).  These  small  differences  are  not  statistically  significant  (Columns  6‐9).  The  lack  of  gender  differences  in  tournament  entry  in  our  setting  is  in  sharp  contrast  with  Niederle  and  Versterlund  [2007].  In  their  setting,  conducted  using  university  students  in  the  United  States,  they  found  that  35  percent  of  women  and  73  percent  of  men  choose  to  experience  competition (a statistically significant difference).   The  remainder  rows  of  Panel  B  of  Table  1  focus  on  performance  in  rounds  1  and  2,  the  decision  to  submit  past  performance  in  round  1  to  a  tournament,  and  whether  the  participant  is  confident  about  having  ranked  first  in  round  1  (relative  to  the  other  participants  in  the  same  session).  These  variables  capture  factors  that  could  drive  a  gender  gap  in  tournament  entry  other  than  a  gender  difference  in  competitiveness,  such  as  ability,  risk  and  feedback  aversion,  and  overconfidence  [see  Niederle  and  Versterlund  2007].  Two  points  are  of  note.  First,  overall  these  experimental  controls  are  balanced  between  treatment  and  control  communities,  for  both  boys  and  girls.  The  exception  is  confidence,  where  we  see  that  both  girls  and  boys  in  treatment  communities  are  more  likely  to  believe  to  have  ranked  first  than  in  control  communities.  Second,  these  experimental  controls  are  also  balanced  by  gender, in both control and treatment communities. Taken together with the notion that these different  controls  could  also  influence  the  decision  to  compete,  this  is  consistent  with  the  evidence  that  girls  and  boys are equally likely to compete in both control and treatment communities.  4.2. Women’s Empowerment and Competitiveness  We  now  present  regression  evidence  on  the  impact  of  the  program  on  gender  differences  in  competitiveness.  This  allows  us  to  check  whether  the  above  descriptive  evidence  is  robust  to  the  7 inclusion  of  experimental  controls  and  observable  socio‐demographic  characteristics.  To  do  so  we  estimate the following specification using a linear probability model (LPM):    ,  (1)  where   is  a  dummy  variable  equal  to  1  if  participant   in  community   selects  the  competitive  tournament  in  round  3,  and  0  if  instead  chooses  the  uncompetitive  piece  rate.   is  a  dummy  variable  equal  to  1  for  girls  and  0  for  boys.   is  a  dummy  variable  equal  to  1  for  communities  randomly  assigned  to  the  ELA  program  and  0  for  control  communities.   is  a  vector  of  experimental  controls  including  performance  in  round  1;  difference  in  performance  between  rounds  1  and  2;  a  dummy  variable  equal  to  1  if  the  participant  submits  past  piece‐rate  performance  in  round  1  to  a  tournament,  and  0  otherwise;  and  a  dummy  variable  equal  to  1  if  the  participant  believes  to  have  ranked  first  during  round  1’s  performance,  and  0  otherwise.   is  a  vector  of  socio‐demographic  characteristics,  including  total  number  of  siblings  and  number  of  sisters,  age,  years  of  schooling,  and  indicators for whether the participant is out of school, is married and has children.   The  parameters   and   in  equation  (1)  indicate  the  gender  gap  in  competitiveness  in  control  and  treatment  communities,  respectively.  Given  the  random  assignment  of  the  ELA  program  across  communities,  the  parameters   and   indicate  the  causal  impact  of  the  program  on  boys’  and  7 girls’  competitiveness,  respectively.  Additionally,   indicates  the  program’s  impact  on  the  gender  gap  in  competitiveness.  Throughout,  we  cluster  the  standard  errors  at  the  experimental  session  level.8 We  also report estimates based on a non‐linear probit specification analogous to equation (1) to account for  the binary nature of the competitiveness outcome.  Table 3 presents the results. We report estimates for the coefficients of interest. Column 1 only controls  for the girl and treatment dummy variables, and their interaction. Consistent with the above descriptive  evidence,  we  see  that:  (i)  there  are  no  significant  gender  differences  in  tournament  entry  in  control  communities,  (ii)  the  program  had  no  impact  on  tournament  entry  for  girls  or  boys,  and  (iii)  as  a  result  there  are  also  no  gender  differences  in  tournament  entry  in  treatment  communities.  Column  2  and  3  sequentially  condition  on  experimental  controls  and  socio‐demographic  characteristics.  Throughout  we  see  that  the  results  on  gender  and  treatment‐control  differences  in  competitiveness  are  robust  to  the  inclusion  of  these  controls.  Column  4  estimates  equation  (1)  using  a  probit  model.  We  also  see  that  the  results remain unchanged when using this alternative model.  7 Since  the  sampling  of  girls  from  treatment  communities  was  independent  of  their  actual  participation  status  in  the  ELA  program  and  boys  were  denied  participation  in  the  ELA  clubs,  ours  is  an  intent‐to‐treat  (ITT)  analysis  for  girls  and  a  pure  spillover effect analysis for boys.  8 There  were  50 experimental  sessions.  Each  session  was  run  in  a lab  setting  with  dividers,  and  facilitators  using  stopwatches  to  queue participants on when to start and stop.  8 4.3. Sibling Sex Composition and Competitiveness  We  now  examine  the  effect  of  sibling  sex  composition  of  a  girl  or  boy’s  siblings  on  that  adolescent’s  competitiveness  in  both  control  and  treatment  communities.  To  do  so  we  estimate  the  following  LPM  specification for participant  , sibship size  , and community  , separately for boys and girls:    ,  (2)  where  ,  ,  and   are  defined  as  before.   denotes  participant  ’s  total  number  of  sisters  .   is a set of fixed effects for the total number of siblings, which we allow to vary by treatment  9 status.    includes  other  socio‐demographic  characteristics:  age,  years  of  schooling,  and  indicators  for  whether the participant is out of school, is married, and has children.  The  parameters  of  interest  in  equation  (2)  are   and  ,  which  indicate  the  impact  on  competitiveness  of  having  an  additional  sister  (as  compared  to  an  additional  brother)  in  control  and  treatment  communities, respectively. The identification of   exploits the  notion  that  conditional on the  total  number  of  siblings  a  participant  has,  the  gender  composition  of  the  siblings  is  close  to  being  exogenous.10 Furthermore,  since  the  ELA  program  was  randomly  assigned  across  communities,    identifies  the  causal  impact  of  the  program  on  the  impact  that  sisters  have  on  their  siblings’  competitiveness.   Table  4  presents  the  results  separately  for  girls  (Panel  A)  and  boys  (Panel  B),  reporting  only  the  coefficients of interest (  and  ). Column 1 only controls for the number of sisters, the treatment  dummy,  their  interaction,  and  sibship  size  fixed  effects  (in  isolation  and  interaction  with  the  treatment  dummy). Columns 2 and 3 sequentially introduce experimental and socio‐demographic controls. Column  4 estimates equation  (2) using a probit  model. At the foot of the  table we report the p‐value on the test  of  hypothesis  0,  i.e.  that  the  program  had  no  impact  on  the  effect  of  sisters  on  their  siblings’  competitiveness.  In  all  specifications,  boys’  competitiveness  is  systematically  affected  by  the  sex  composition  of  his  siblings,  and  girls’  competitiveness  is  not.  While  in  control  communities  boys  surrounded  by  sisters  are  less  competitive,  in  treatment  communities  boys  surrounded  by  sisters  are  more  competitive.  Our  preferred specification is in Column 3 with experimental and socio‐demographic controls. In Panel B this  9 In practice,   is a vector of dummy variables for each sibship size, in isolation and in interaction with  .  10 Strictly  speaking  this  assumption  is  violated  in  the  presence  of  son‐preferring  differential  fertility  stopping  behavior,  whereby  parents  continue  having  children  until  a  certain  number  of  boys  are  born.  In  such  case,  a  key  theoretical  prediction  is  that  girls  will  have  more  siblings  than  boys,  on  average  [Yamaguchi  1989,  Jensen  2005].  As  a  result,  conditional  on  family  size,  families  with  a  higher  share  of  daughters  must  desire  fewer  sons  [see  Barcellos  et  al.  2014].  If  parents  who  want  more  daughters  (plausibly)  invest  more  in  daughters,  than  sisters  might  actually  offer  more  competition  to  their  siblings.  Although  we  cannot  reject  the  hypothesis  that  families  in  our  sample  engage  in  son‐preferring  fertility  stopping  rules  (Table  1  shows  that  girls  have  on  average  0.85  more  siblings  than  boys,  p‐value<.05),  our  results  indicate  that  in  control  communities  sisters  are  associated  with reduced (not increased) competitiveness among boys ( 0).    9 shows  that  in  control  communities  an  additional  sister  (relative  to  a  brother)  corresponds  to  an  8.8  percentage  points  decrease  in  tournament  entry  among  boys.  In  contrast,  in  treatment  communities  an  additional  sister  corresponds  to  9.8  percentage  points  increase  in  the  likelihood  that  boys  enter  tournament. To quantify these magnitudes, we note the mean of the dependent variable is 34.9 percent  (see  Column  1  in  Table  1).  The  hypothesis  that  the  empowerment  of  girls  (through  the  program)  had  no  impact on their brothers’ willingness to compete is rejected at conventional statistical significance levels  (p‐value<.01).   5. Conclusion  We  conduct  a  lab‐in‐the‐field  experiment  in  Uganda  to  measure  the  impact  of  a  randomly  placed  adolescent  girls’  empowerment  program  on  gender  differences  in  preferences  for  competition.  We  also  exploit  variation  in  sibling  sex  composition  in  both  control  and  treatment  communities  to  identify  the  impact of adolescent girls’ empowerment on their brothers’ competitiveness.  We  present  four  main  findings.  First,  in  contrast  with  the  evidence  drawn  mostly  from  developed  countries,  we  find  that  young  men  and  women  in  Uganda  are  equally  competitive.  Second,  we  also  find  that  despite  the  program  empowering  girls  along  economic  and  social  dimensions,  these  changes  were  not  accompanied  by  increased  competitiveness  for  girls.  Third,  in  line  with  previous  research,  we  document  that  in  control  communities  boys  surrounded  by  sisters  are  less  competitive.  Fourth,  the  program  reverses  this  pattern:  boys  surrounded  by  (empowered)  sisters  in  treatment  communities  are  more competitive.   Our  findings  push  forward  an  emerging  literature  highlighting  the  importance  of  studying  different  cultures  and  family  environments  in  order  to  understand  gender  differences  in  competitiveness.  While  much  more  research  is  needed  to  identify  policy  interventions  that  effectively  foster  competitiveness,  ours  is  one  of  the  first  pieces  of  evidence  suggesting  that  large  scale  female  empowerment  programs  and  other  programs  that  shift  gender  dynamics,  also  have  the  potential  to  significantly  alter  gender  differences in competitiveness.  References  [1] AKERLOF.G.A  AND  R.E.KRANTON  (2000).  “Economics  and  Identity.”  Quarterly  journal  of  Economics,  115(3): 715‐753.  [2] ALMAS.I,  A.W.CAPPELEN,  K.G.SALVANES,  K.G.,  E.Ø.SØRENSON  AND  B.TUNDOGGEN  (2015).  “Willingness to compete: Family matters.” Management Science.  10 [3] ANDERON.S,  S.ERTAC,  U.GNEEZY,  J.A.LIST  AND  S.MAXIMIANO  (2013).  “Gender,  Competitiveness  and  Socialization  at  a  Young  Age:  Evidence  from  a  Matrilineal  and  a  Patriarchal  Society.”  The  Review of Economics and Statistics 95: 1438–1443.  [4] BECKER.G.S  AND  H.G.LEWIS  (1973)  "On  the  Interaction  between  the  Quantity  and  Quality  of  Children." Journal of Political Economy 81(2): S279‐S288.  [5] BENABOU.R  AND  J.TIROLE  (2011)  "Identity,  Morals,  and  Taboos:  Beliefs  as  Assets."  The  Quarterly  Journal of Economics 126(2): 805‐855.  [6] BANDIERA.O, N.BUEHREN, R.BURGESS,  M.GOLDSTEIN, I.RASUL, S.GUESCI AND M.SULAIMAN (2015)  “Women's  Empowerment  in  Action:  Evidence  from  a  Randomized  Control  Trial  in  Africa.”  Working  Paper.  [7] CARVALHO.L.S,  S.H.BARCELLOS  AND  A.LLERAS‐MUNEY  (2014).  “Child  Gender  and  Parental  Investments  in  India:  Are  Boys  and  Girls  Treated  Differently?”  American  Economic  Journal:  Applied  Economics, 6(1): 157‐189.  [8] BEYEZA‐KASHESYA.J,  S.NEEMA,  A.M.EKSTROM,  F.KAHARUZA,  F.MIREMBE  AND  A.KULANE  (2010)  “"Not a Boy, Not a Child": A Qualitative Study on Young People's Views on Childbearing in Uganda.”  African Journal of Reproductive Health 14(1): 71‐81.  [9] BERGE.L,  I.O.K.BJORVATN,  A.J.G.PIRES  AND  B.TUNGODDEN  (2015).  “Competitive  in  the  Lab,  Successful in the Field?” Journal of Economic Behavior & Organization, 118: 303–317.  [10] BERTRAND.M  (2010)  “New  Perspectives  on  Gender.”  in  Handbook  of  Labor  Economics,  O.  Ashenfelter, and D. Card, eds. Amsterdam: North‐Holland, 1545–1592.  [11] BOOTH.A.L  AND  P.J.NOLEN  (2012).  “Choosing  to  Compete:  How  Different  are  Girls  and  Boys?”  Journal of Economic Behavior & Organization, 82: 542–555.  [12] BREER.P.E  AND  E.A.LOCKE  (1965).  Task  Experience  as  a  Source  of  Attitudes.  Homewood,  IL:  The  Dorsey Press.  [13] BRIM.O.G  (1958)  "Family  Structure  and  Sex  Role  Learning  By  Children:  A  Further  Analysis  of  Helen  Koch's Data." Sociometry 21: 1‐16.  [14] BUSER.T,  L.GEIJTENBEEK  AND  E.PLUG  (2015)  “Do  Gays  Shy  Away  from  Competition?  Do  Lesbians  Compete Too Much?” Working Paper No. 9382. Institute for the Study of Labor (IZA).  [15] BUSER.T,  M.NIEDERLE  AND  H.OOSTERBEEK  (2014)  “Gender,  Competitiveness  and  Career  Choices.”  Quarterly Journal of Economics, 129 (3): 1409‐1447.  [16] CAMERON.L,  N.ERKAL,  L.GANGADHARAN  AND  X.MENG  (2013).  “Little  Emperors:  Behavioral  Impacts of China's One‐Child Policy”. Science 22 339(6122): 953‐957.  [17] CARDENAS.J.C,  A.DREBER,  E.VON.ESSEN  AND  E.RANEHILL  (2012).  “Gender  Differences  in  Competitiveness  and  Risk  Taking:  Comparing  Children  in  Colombia  and  Sweden.”  Journal  of  Economic Behavior & Organization, 83: 11–23.  11 [18] FLORY.J,  A.LEIBBRANDT  AND  J.LIST  (2015).  “Do  Competitive  Work  Places  Deter  Female  Workers?  A  Large‐Scale  Natural  Field  Experiment  on  Gender  Differences  in  Job‐Entry  Decisions.”  The  Review  of  Economic Studies, 82: 122–155.  [19] GARG.A,  J.MORDUCH  (1998)  “Sibling  Rivalry  and  the  Gender  Gap:  Evidence  from  Child  Health  Outcomes in Ghana.” Journal of Population Economics, 11: 471‐493.  [20] GNEEZY.U, K.L.LEONARD AND J.A.LIST (2009). “Gender Differences in Competition: Evidence from a  Matrilineal and a Patriarchal Society.” Econometrica, 77: 1637–1664.  [21] JENSEN.R  (2012).  “Do  Labor  Market  Opportunities  Affect  Young  Women’s  Work  and  Family  Decisions? Experimental Evidence from India.” Quarterly Journal of Economics, 127(2): 753–92.  [22] KHACHATRYAN.K,  A.DREBER,  E.VON.ESSEN  AND  E.RANEHILL  (2015).  “Gender  and  Preferences  at  a  Young age: Evidence from Armenia.” Journal of Economic Behavior & Organization, 118: 318‐332.  [23] KOCH.H  (1955)  “Some  Personality  Correlates  of  Sex,  Sibling  Position,  and  Sex  of  Sibling  Among  Five  and Six Year Children." Genetic Psychology Monographs, 52: 3‐50  [24] MORDUCH.J (2000) “Sibling rivalry in Africa.” American Economic Review, 90(2): 405‐409.  [25] NIEDERLE.M.  AND  L.VESTERLUND  (2007)  “Do  Women  Shy  Away  from  Competition?  Do  Men  Compete Too Much?” Quarterly Journal of Economics, 122(3): 1067–1101.   [26] NIEDERLE.M.  AND  L.VESTERLUND  (2011)  “Gender  and  Competition.”  Annual  Review  of  Economics,  3 (1): 601‐630.  [27] OKUDAIRA.H,  Y.KINARI,  N.MZUTANI,  F.OHTAKE  AND  A.KAWAGUCHI  (2015).  “Older  Sisters  and  Younger  Brothers:  The  Impact  of  Siblings  on  Preference  for  Competition.”  Personality  and  Individual Differences, 82: 81‐89.  [28] REUBEN.E,  P.SAPIENZA  AND  L.ZINGALES  (2015a)  “Taste  for  Competition  and  the  Gender  Gap  Among Young Business Professionals.” NBER Working Paper No. 21695.  [29] REUBEN.E,  M.WISWALL  AND  B.ZAFAR  (2015b).  “Preferences  and  Biases  in  Educational  Choices  and  Labour  Market  Expectations:  Shrinking  the  Black  Box  of  Gender.”  The  Economic  Journal,  forthcoming.  [30] SULLOWAY.F.J  (1996)  Born  to  Rebel:  Birth  Order,  Family  Dynamics,  and  Creative  Lives.  Pantheon  Books.  [31] YAMAGUCHI.K.  (1989)  “A  Formal  Theory  for  Male‐Preferring  Stopping  Rules  of  Chilbearing:  Sex  Differences in Birth Order and the Number of Siblings.” Demography 26 (3): 451–65.  [32] ZHANG.Y.J (2014). “Can Experimental Economics Explain Competitive Behavior Outside of the Lab?”  Working Paper.  [33] ZHANG.Y.J  (2015).  “Culture,  Institutions,  and  the  Gender  Gap  in  Competitive  Inclination:  Evidence  from the Communist Experiment in China.” Working Paper.  12 Table 1. Descriptive Statistics: Sample Characteristics  Means, standard deviations reported in parentheses, p‐values reported in brackets.                Differences [p‐values]                Girls vs Boys    Treatment vs Control    (1) Full    Control    Treatment    (6) Control   (7) Treatment     (8) Girls  (9) Boys     Sample    (2) Girls  (3) Boys    (4) Girls  (5) Boys    (3) vs (2)  (5) vs (4)    (4) vs (2)  (5) vs (3)  Age  16.9    17.2  16.7    16.9  16.9    [.157]  [.902]    [.535]  [.525]  (3.15)  (3.05)  (2.80)  (3.38)  (3.25)  Married [yes =1]  .052    .072  .023    .076  .023    [.128]  [.012]    [.916]  [.975]  (.222)  (.260)  (.150)  (.265)  (.151)  Has child(ren) [yes =1]  .095    .171  .008    .135  .041    [.000]  [.001]    [.429]  [.039]  (.293)  (.378)  (.008)  (.342)  (.198)  Out of school [yes =1]  .256    .303  .205    .291  .203    [.057]  [.114]    [.846]  [.984]  (.437)  (.461)  (.405)  (.455)  (.404)  Years of schooling  9.95    10.3  9.76    9.92  9.86    [.127]  [.822]    [.324]  [.777]  (3.01)  (2.90)  (3.03)  (2.90)  (3.21)  Number of siblings  5.34    6.15  5.30    6.16  5.31    [.012]  [.013]    [.990]  [.988]  (2.54)  (3.49)  (3.01)  (3.31)  (2.99)  Number of brothers  2.74    3.13  2.95    3.20  2.81    [.446]  [.059]    [.763]  [.601]  (1.55)  (2.16)  (2.57)  (2.24)  (1.94)  Number of sisters  2.60    3.14  2.55    3.05  2.51    [.024]  [.013]    [.741]  [.886]  (1.59)  (2.45)  (2.23)  (2.25)  (1.75)  Notes:  Column 1 refers to all participants (girls and boys) across control and treatment communities. Columns 2 and 3 refers to girls and boys from control communities, respectively. Columns 4 and 5 refers to girls and  boys from treatment communities, respectively. Columns 6‐9 show p‐values on tests of equality obtained from an OLS regressions and allows the error term to be clustered by experimental session.  Table 2. Descriptive Statistics: Experimental Results  Means, standard deviations reported in parentheses, p‐values reported in brackets.                Differences [p‐values]                Girls vs Boys    Treatment vs Control    (1) Full    Control    Treatment    (6) Control   (7) Treatment     (8) Girls  (9) Boys     Sample    (2) Girls  (3) Boys    (4) Girls  (5) Boys    (3) vs (2)  (5) vs (4)    (4) vs (2)  (5) vs (3)  Competitive [yes =1]  .349    .362  .356    .325  .372    [.919]  [.323]    [.460]  [.787]  (.477)  (.482)  (.481)  (.469)  (.485)  Round 1 performance   8.69    8.62  8.64    8.82  8.61    [.942]  [.320]    [.442]  [.923]  (1.88)  (1.90)  (1.97)  (1.81)  (1.90)  Round 2 performance  10.4    10.3  10.2    10.6  10.3    [.759]  [.259]    [.224]  [.693]  (1.99)  (1.98)  (2.03)  (1.94)  (2.03)  Submit round 1 to tournament [yes =1]  .375    .342  .417    .367  .378    [.210]  [.788]    [.606]  [.517]  (.484)  (.476)  (.495)  (.483)  (.486)  Confident [yes =1]  .478    .401  .432    .498  .541    [.618]  [.408]    [.059]  [.052]  (.499)  (.492)  (.497)  (.501)  (.500)  Notes:  Column 1 refers to all participants (girls and boys) across control and treatment communities. Columns 2 and 3 refers to girls and boys from control communities, respectively. Columns 4 and 5 refers to girls and  boys  from  treatment  communities,  respectively.  Columns  6‐9  show  p‐values  on  tests  of  equality  obtained  from  an  OLS  regressions  and  allows  the  error  term  to  be  clustered  by  experimental  session.  Competitive  is  a  dummy variable equal to 1 if the participant chooses the competitive tournament in round 3, and 0 if instead chooses the uncompetitive piece rate. Confident is a dummy variable equal to 1 if the participant believes to  have ranked first during round 1 performance (relative to the other participants in the same session), and 0 otherwise.  Table 3. Women’s Empowerment and Gender Differences in Competitiveness  Dependent variable =1 if participant chooses tournament in Round 3, =0 if instead chooses piece‐rate.  Linear probability model estimates in Columns 1‐3, Probit estimates in Column 4, standard errors reported in parentheses.     LPM Estimates    Probit Estimates  (1) No   (2) Experimental  (3) Other  (4) All     controls  controls  controls    controls  Gender gap in control communities [ ]  .006  .037  .042    .045  (.056)  (.051)  (.051)  (.050)  Program impact on boys [ ]  .016  .034  .034    .036  (.059)  (.050)  (.049)  (.048)  Gender gap in treatment communities [ ]  ‐.047  ‐.046  ‐.047    ‐.036  (.047)  (.043)  (.044)  (.048)  Program impact on girls [ ]  ‐.037  ‐.050  ‐.054    ‐.056  (.050)  (.044)  (.043)  (.042)  Experimental controls  No  Yes  Yes    Yes  Other controls (socio‐demographics)  No  No  Yes    Yes  P‐value:    [.475]  [.206]  [.169]    [.135]  R‐Squared  .002  .178  .183    –  Observations  693  693  693    693  Notes:  ***  (**)  (*)  indicates  significance  at  1%  (5%)  (10%)  level.  Each  column  corresponds  to  a  separate  regression.  In  all  columns  the  dependent  variable  is  a  dummy  equal to 1 if the participant chooses competitive tournament in round 3, and 0 if instead chooses uncompetitive piece‐rate. Linear probability model (OLS) estimates are  reported  in  Columns  1‐3.  Probit  marginal  estimates  are  reported  in  Column  4.  Robust  standard  errors  clustered  by  experimental  session  reported  in  parentheses  (50  clusters).  Column  1  only  controls  for  a  girl  dummy,  a  treatment  dummy,  and  their  interaction.  Column  2  additionally  includes  experimental  controls:  participant’s  performance  in  round  1,  improvement  in  performance  from  round  1  to  round  2,  a  dummy  equal  to  1  if  the  participant  believes  to  have  ranked  first  during  round  1  performance  and 0  otherwise,  and a  dummy  equal  to  1  if  the  participant chooses to  submit  past performance  in  round 1  to  tournament  and 0 to piece‐rate.  Columns 3  and  4  additionally  control  for  participant‐level  socio‐demographic  characteristics:  age,  years  of  schooling,  indicators  for  whether  the  participant  is  out  of  school,  is  married, and has children, total number of siblings, and number of sisters.    Table 4. Sibling Sex Composition and Competitiveness  Dependent variable =1 if participant chooses tournament in Round 3, =0 if instead chooses piece‐rate.  Linear probability model estimates in Columns 1‐3, Probit estimates in Column 4, standard errors reported in parentheses.     LPM Estimates    Probit Estimates  (1) No   (2) Experimental  (3) Other  (4) All     controls  controls  controls    controls  Panel A. Girls            Impact of sisters in control communities [ ]  .028  .037  .043    .037  (.053)  (.045)  (.044)  (.039)  Impact of sisters in treatment communities [ ]  ‐.011  ‐.020  ‐.019    ‐.018  (.036)  (.034)  (.034)  (.032)  Siblings fixed effects  Yes  Yes  Yes    Yes  Experimental controls  No  Yes  Yes    Yes  Other controls (socio‐demographics)  No  No  Yes    Yes  Test (p‐value):   [.523]  [.274]  [.584]    [.248]  R‐Squared  .043  .221  .227    –  Observations  389  389  389    389  Panel B. Boys            Impact of sisters in control communities [ ]  ‐.103**  ‐.089**  ‐.087**    ‐.088**  (.043)  (.043)  (.044)  (.044)  Impact of sisters in treatment communities [ ]  .077**  .094***  .098***    .105***  (.038)  (.033)  (.030)  (.029)  Siblings fixed effects  Yes  Yes  Yes    Yes  Experimental controls  No  Yes  Yes    Yes  Other controls (socio‐demographics)  No  No  Yes    Yes  P‐value:    [.002]  [.002]  [.002]    [.000]  R‐Squared  .090  .259  .268    –  Observations  304  304  304    303  Notes:  *** (**) (*) indicates significance at 1% (5%) (10%) level. Each column corresponds to a separate regression. In all columns the dependent variable is a dummy equal to 1 if the  participant  chooses  competitive  tournament  in  round  3,  and  0  if  instead  chooses  uncompetitive  piece‐rate.  Linear  probability  model  (OLS)  estimates  are  reported  in  Columns  1‐3.  Probit  marginal  estimates  are  reported  in  Column  4.  Robust  standard  errors  clustered  by  experimental  session  reported  in  parentheses  (50  clusters).  Column  1  only  controls  for  the  number  of  sisters,  the  treatment  dummy,  their  interaction,  and  sibship  size  fixed  effects  (in  isolation  and  interacted  with  the  treatment  dummy).  Column  2  additionally  includes  experimental controls: participant’s performance in round 1, improvement in performance from round 1 to round 2, a dummy equal to 1 if the participant believes to have ranked first  during round 1 performance and 0 otherwise, and a dummy equal to 1 if the participant chooses to submit past performance in round 1 to tournament and 0 to piece‐rate. Columns 3  and  4  additionally  control  for  participant‐level  socio‐demographic  characteristics:  age,  years  of  schooling,  indicators  for  whether  the  participant  is  out  of  school,  is  married,  and  has  children.