American Economic Review: Papers & Proceedings 2013, 103(3): 263–268 92930 http://dx.doi.org/10.1257/aer.103.3.263 The Economics of Gender ‡ Is Ignorance Bliss? The Effect of Asymmetric Information between Spouses on Intra-Household Allocations  † By Carolina Castilla and Thomas Walker* Fundamental questions about how resources I.  Noncooperative Intra-Household are allocated within the family remain unresolved. Contracts in Ghana Because family decision making is a repeated game and there is caring, it is generally posited Southern Ghana is an ideal setting for test- that the marriage relationship exploits the high ing noncooperative intra-household resource degree of information sharing between spouses. allocation because, in the regionally predominant ­ But how valid is this assumption in practice? Akan culture, men and women maintain separate If spouses choose to exploit their information economies, such that no spouse has control over advantages by concealing resources from each all of the household’s resources, while the marital other, the result is a noncooperative equilibrium contract stipulates that husbands should provide and a Pareto suboptimal allocation of resources. a housekeeping allowance (chop money) to their This question has implications for the design of wives (see Castilla and Walker 2012, for refer- cash transfer programs. Even if official cash trans- ences). Lundberg and Pollak (1993) show that fers themselves are common knowledge between intra-household allocations can default to a nonco- household members, the transfer may loosen operative equilibrium when the transaction costs constraints over other income sources less easily associated with cooperation are high. In Castilla monitored. Empirical research has documented and Walker (2012), we show how in a model inefficient allocations between spouses (Udry of intra-household allocation adapted for the 1996) and noncooperative behavior as a result of Ghanaian context, a noncooperative equilibrium asymmetric information in migrant households can exist in which chop money transfers continue (Chen 2006). However, the implications of asym- to be made. Under these conditions, spouses have metric information between cohabiting spouses incentives to hide unobservable resources, and we are still relatively unexplored (for an exception show how these incentives differ depending on the see Ashraf 2009). role each spouse plays within the marital contract. To illustrate how income concealment can be a dominant strategy within Ghanaian households ‡Discussants: Kevin Lang, Boston University; Catalina with gender-differentiated preferences, consider a Amuedo-Dorantes, San Diego State University; Susan household with two decision makers, a husband Averett, Lafayette College. * Castilla: Colgate University, 13 Oak Drive, Hamilton, and a wife. Each spouse allocates their dispos- NY 13346 (e-mail: ccastilla@colgate.edu); Walker: The able income (for husbands, private income net of World Bank, 1818 H Street, NW, Washington, DC 20433 chop money; for wives, private income inclusive (e-mail: twalker@worldbank.org). The authors would like of chop money) among a set of public goods that to thank Chris Barrett, Andrew Agyei-Holmes, Robert Osei, are easily observable, and several private goods Sarah Pearlman, Joyce Chen, seminar participants at Lafayette College, Binghamton University, and College University for which have varying degrees of observability.1 their helpful comments. The survey and experiments were We justify the latter assumption on the basis that supported by grants from the International Growth Centre household public goods, such as assets, food, and [RA-2009-06-010], the National Science Foundation [SES- 0851586] and USAID’s AMA CRSP Program [P686140]. † To view additional materials, and author disclosure statement(s),visit the article page at 1 For full details on the setup of the model, refer to http://dx.doi.org/10.1257/aer.103.3.263. Castilla and Walker (2012). 263 264 AEA PAPERS AND PROCEEDINGS MAY 2013 and home expenses are typically easy to moni- The field experiment took place over four tor in households where both spouses live under rounds, one week before each survey round the same roof. However, private goods may dif- (except for round one). Windfalls of cash and fer in their levels of observability and the costs goods in kind were generated by distributing lot- of monitoring; for instance, while purchases of tery prizes to randomly selected survey respon- personal care items may be easily observed, inter- dents. Husbands and wives had an equal and household transfers may not. Now assume that independent probability of winning a prize in one spouse receives windfall cash income, such each round. The experiment was a t ­wo-by-two as from a public cash transfer scheme. One design, with half of the prizes raffled in ­public, would expect the couple to bargain over the half in private, ten of each in cash, and ten as allocation of this additional income. However, if livestock (chickens and goats). The private the windfall was received privately, the recipient prizes were drawn and given to the winner in a could allocate the unobserved resources toward closed room, so that the winner’s identity was concealable private expenditure. unknown even to their spouse.4 We treat the Why would a spouse conceal income? In the privately-won livestock prizes as public, since it Ghanaian context, the wife’s incentive to hide is unlikely that they could be concealed. Over arises because the additional private income the four lotteries, 42 percent of individuals and could crowd out her chop money allowance; by 62 percent of households won at least one prize. concealing it from her husband she can increase The prizes were of a substantial size and their her command over total household resources. The values varied between GH¢10 and GH¢70.5 husband is indifferent between hiding and reveal- The lotteries and lucky dips took place one ing the private prize since he indirectly deter- week before the commencement of the survey mines his household good consumption through interviews. Great care was taken to make clear to the chop money allowance. Nonetheless, if he participants that the allocation of prizes was ran- wishes to avoid renegotiating the chop money dom, and that each respondent had an equal chance allowance with his wife, he may be inclined to of winning in each round. A village meeting was hide the windfall as well. Note that these results held in the community, and all respondents were only hold provided that there is gender specializa- invited to attend. The team explained that respon- tion within the marital contract and spouses have dents had a chance to win one of 20 prizes that preferences over different private goods, since day, framing the lottery tickets as a gratuity for otherwise the allocation of income would not participation in the survey. Winners for the ten affect final household consumption outcomes. public prizes were then drawn (without replace- ment) from a bucket containing the names of the II.  Data Collection and Empirical Specification survey respondents. Each winner was announced, and asked to come forward to receive their prize. To test the model, we use data from a field After the lottery prizes were distributed, the experiment conducted in 2009 in conjunction with lucky dip began. Respondents were asked to a year-long panel household survey in four com- identify themselves to an enumerator, who took munities in Akwapim South district of Ghana’s their thumbprint or signature and issued them Eastern Region. The survey included detailed an identification number. The respondents then information on spousal relationships, expendi- entered a closed classroom, one at a time, where ture, and social networks. The sample consisted another enumerator invited them to draw a bottle of 70 married monogamous and polygamous cap without replacement from a bag. There was households from each of the four communities.2 one bottle cap for each of the respondents in the Spouses were interviewed separately every two sample, ten of which were of different colors months between February and November 2009.3 corresponding to the prizes. Those who drew winning tokens were informed of their prize and, for cash prize winners, given the money on 2 We only include the monogamous couples as bargaining in polygamous households is very different. This amounts to 4 The winner had the choice to disclose his/her windfall; dropping 16 households out of the 280 total. this design provided them the opportunity to conceal it. 3 5 For further details on the survey and experiments, refer For comparison, mean monthly per capita consumption to Walker (2011). in the four survey communities during 2009 was GH¢65. VOL. 103 NO. 3 asymmetric information and intra-household allocations 265 Table 1—Balance of Treatment for Key Expenditures data collected before the experiments were con- at Baseline (Round 1) ducted, and follow-up data collected afterwards, we can test the effect of asymmetric information Non Private Public Livestock Expenditure winners cash cash prize on actual household expenditure. We estimate reduced-form demand equations Household public goods for expenditure separately for husbands and Home 24.78 17.65 16.78 19.38  expenses (1.87) (3.51) (3.16) (2.25) wives. We consider household public goods such Assets 14.43 8.11 13.6 6.97* as home expenses (includes household items, (2.22) (2.99) (4.97) (1.57) utilities, and rent) and asset purchases which should be observable to both spouses. For both Husband private expenditure the husband and the wife’s private expenditure, Public 8.93 9.02 8.40* 14.8 personal care and public transportation expenses  transportation (0.78) (2.26) (6.25) (4.14) as well as gifts are considered. Personal care Personal care 4.53 3.89 3.06 2.41 (0.50) (0.65) (1.75) (0.71) expenses and public transportation are eas- ily observed. The gifts to each social network Cash gifts 13.40 24.47 1.875 19.68 (3.25) (14.9) (0.93) (4.02) are much harder to monitor because the money In-kind gifts 7.67 6.71 −39.0 13.9 effectively leaves the household, and the recipi- (2.57) (2.16) (5.62) (3.64) ents have an incentive to keep the gifts private as otherwise, the giver would have to negotiate with Wife private expenditure his/her spouse over how the money is allocated. Public 8.42 6.4 7.07 13.3 In Table 1 we present tests on mean expendi-  transportation (1.56) (1.49) (2.95) (7.65) ture across treatment groups and control (non- Personal care 8.00 (0.40) 7.71 (1.12) 3.55* (1.47) 1.89 (1.35) winners). There are no significant differences between winners and non-winners, except for Cash gifts 2.83 0.43 4.33 4.77 (0.77) (0.25) (1.50) (1.59) husbands’ average expenditure on public trans- In-kind gifts 3.07 1.73 73.5 1.77 portation and wives’ personal care expenses, (0.67) (1.48) (13.0) (0.71) significant at the 10 percent level. Because the possibility exists of zero spend- Note: Standard errors in parentheses. ing at any particular round on a given item, an *** Significant at the 1 percent level.  **  Significant at the 5 percent level. unobserved random effects Tobit model is used.7   *  Significant at the 10 percent level. For spouse i, in household h, village v, and round r, the demand for good x ​ ​  g i,    ​ h,v,r​can be expressed as the spot. They were told that they did not have x​  g ​  =   h,v,r​ i,  ​ to tell anyone else that they had won. The survey ∑ 1 team made clear that they would not divulge the identities of the lucky dip prize winners. 6 ​   ​​[ ​      ​δ​  Pu​ 1j​ C​    i,      v,r−j​ + ​ δ​   C​     Pr​ 2j​     v,r−j​ i,  δ​ + ​   3j​ ​L​  i,v,  ]​ r−j​ ​   Identification of noncooperative behavior j=0 in the form of income hiding requires data on θ​ +  ​ A​  liq 2​  ln ​ h,v    ​   ,1​ + ​ θ​ A​  ill 3​  ln ​ h,v   ​    ,1​ expenditure and income with varying degrees  ∑   ∑  of observability. The field experiment lotteries 4 5 provide an exogenous shock to spouse’s pri- +  ​     ​​α  ​   ​  + ​ v​     ​​σ     ​ ​ + ​ r​ ε​   , i,h,v,r​ vate income and random variation on the ease v=2 r=3 of observability by the other spouse (as well as the entire village). Husbands and wives had where Pu​C​    Pr​C​     v,r​ ​ i,  ,v,r​ i,  ​ ​     and L v,r​  are the val- i,  the same probability of winning a prize, which ues of the public and private cash prizes, and allows us to compare spouses’ responses to livestock prizes, respectively, won by spouse i; ­ ​  ​  h  ∑ ​   ​​ are village fixed-effects;  ​∑ ​  r= ​​α   ​​ are ​   ​​σ 4 5 prize winning by gender. Further, using baseline =2 v 3 r 6 Respondents who were absent at the time of drawing 7 Results are robust if we use an inverse sign hyperbolic [ ​x​   i​  ​  + ​​( ​x​  2 2 ] were called to pick up their prize in person, if possible. _ ​  1   transformation, ln ​x​    i​  ​  =  ln​  + 1 )​ i​  ​ ​  ​ Livestock prizes were delivered two days later to the winner , as well as ln​ ​  ​ in person, or to another household member if absent. x​  i​  ​  =  ln​   [ ​x​  i​  ​  +  1 ]​   , and fit is slightly better. 266 AEA PAPERS AND PROCEEDINGS MAY 2013 Table 2—Treatment Effects on Observable Expenditure Household goods Observable private expenditure Husband Wife Home Public Personal Public Personal expenses Assets transporation care transporation care Husband Public cash prize 0.232** 0.555** 0.060 −0.009 −0.008 0.020 (0.105) (0.237) (0.054) (0.014) (0.124) (0.023) Private cash prize −0.033 −0.545 0.125** 0.012 −0.121 0.027 (0.109) (0.345) (0.054) (0.014) (0.130) (0.023) Livestock 0.035 −0.065 −0.040 −0.018* −0.058 −0.008 (0.067) (0.176) (0.038) (0.010) (0.077) (0.014) Public cash prize (lag) 0.196 0.350 0.026 −0.013 −0.157 −0.027 (0.130) (0.313) (0.072) (0.018) (0.164) (0.030) Private cash prize (lag) 0.172 0.047 0.103 −0.009 0.026 0.013 (0.135) (0.327) (0.067) (0.017) (0.155) (0.029) Livestock (lag) 0.107 0.214 −0.009 −0.010 −0.078 −0.005 (0.080) (0.181) (0.046) (0.012) (0.092) (0.017) Wife Public cash prize 0.199** −0.038 −0.011 0.004 0.069 0.036* (0.092) (0.250) (0.049) (0.012) (0.100) (0.019) Private cash prize 0.098 0.052 0.104* −0.009 −0.017 0.019 (0.108) (0.253) (0.056) (0.015) (0.118) (0.022) Livestock 0.010 −0.180 0.053 −0.011 0.003 −0.010 (0.067) (0.181) (0.036) (0.009) (0.078) (0.014) Public cash prize (lag) −0.013 −0.065 0.001 −0.002 0.071 0.052** (0.094) (0.242) (0.049) (0.013) (0.102) (0.019) Private cash prize (lag) 0.034 0.182 −0.012 −0.004 −0.033 −0.013 (0.119) (0.270) (0.064) (0.016) (0.131) (0.024) Livestock (lag) −0.018 0.277 −0.035 −0.015 −0.045 −0.010 (0.073) (0.169) (0.040) (0.010) (0.085) (0.016) Uncensored observations 826 342 830 839 845 1,009 Observations 1,070 1,070 1,027 1,027 1,070 1,070 Notes: Results include controls for initial assets, village, and round fixed effects. Standard errors in parentheses. *** Significant at the 1 percent level.  **  Significant at the 5 percent level.   *  Significant at the 10 percent level. round fixed-effects; and A​ ​  liq h,v   and A ​ ,1​ ​ ​  ill h,v  are ​  ,1​ under asymmetric information, if spouses were household liquid and illiquid assets in round cooperative. In the absence of hiding, there should one, respectively. We use assets rather than be no differences between spouses in the treatment income (as in the usual Engel specification) effect of public and private cash prizes. Any wind- because the income data proved too noisy to fall, public or private, would increase allocations reliably capture households’ living standards. to items preferred by the beneficiary as a result of her increased b ­argaining power (due to the III.  Empirical Results and Findings change in her resources relative to her spouse). If the prize is private and the recipient wishes to shift The counterfactual to identify noncooperative her bargaining power it would have to be revealed. behavior in the form of hiding is the set of allo- A possible alternative counterfactual is the set of cations that would be chosen by the household, allocations that would result under a gender-based VOL. 103 NO. 3 asymmetric information and intra-household allocations 267 marital contract (as described above) in the Table 3—Treatment Effects on Concealable absence of hiding. Both public and private lottery Expenditure prizes would have no differential effect and would Concealable private expenditure only influence allocations in the winning spouse’s Husband Wife realm of responsibilities. The empirical results on observable and con- Cash Inkind Cash Inkind cealable goods are presented in Tables 2 and 3, gifts gifts gifts gifts respectively. The results indicate that there is no Husband shift in bargaining power as a result of winning a Public cash −0.142 −0.290 −0.109 −0.058 prize. The effect of lottery prizes on assets, utili-  prize (0.301) (0.185) (0.237) (0.083) ties, household items, and rent depends on the Private cash 0.049 0.153** 0.128 −0.029 gender of the recipient. When husbands win a  prize (0.276) (0.067) (0.196) (0.092) public prize, expenditure in home goods and assets Livestock −0.066 0.117** −0.051 −0.192** increases, but there is no effect on public transpor- (0.195) (0.048) (0.132) (0.072) tation or personal care expenses. If the wife wins Public cash −0.400 −0.078 0.170 −0.029 a public prize, her expenditure on personal items  prize (lag) (0.501) (0.136) (0.236) (0.112) increases in following rounds, but other expenses Private cash −0.016 −0.005 −0.361 −0.227  prize (lag) (0.376) (0.104) (0.389) (0.181) are unaffected by lottery prizes won by the wife (public or private). It is possible that the prizes Livestock −0.028 −0.021 0.039 −0.079   (lag) (0.237) (0.069) (0.149) (0.076) were not large enough to shift bargaining power. However the cooperative contract can be rejected Wife as there are differences in the effect of public ver- Public cash −0.127 −0.042 −0.018 0.078 sus private cash prizes on assets, home expenses,  prize (0.219) (0.068) (0.149) (0.060) husbands’ public transportation and wives’ per- Private cash 0.017 0.012 0.232a 0.037 sonal care items. While winning a private prize  prize (0.274) (0.075) (0.152) (0.075) has no effect on public good expenses, it increases Livestock −0.070 0.062 −0.205 0.096** (0.187) (0.048) (0.141) (0.048) husbands’ expenditure on public transportation. Public cash −0.112 −0.017 0.392** 0.118* Lottery winnings by the wife have no effect  prize (lag) (0.239) (0.074) (0.136) (0.064) on husbands’ expenditure, and vice versa for the Private cash 0.046 0.020 0.056 −0.281 husband. These results support the hypothesis  prize (lag) (0.310) (0.090) (0.215) (0.185) that spouses maintain separate economies and, Livestock −0.192 0.052 0.062 −0.046 at least with respect to public and observable   (lag) (0.216) (0.054) (0.129) (0.063) private expenditure, provide further evidence in support of the noncooperative contract. Uncensored 335 270 182 292 Livestock and private cash winnings by hus-  observations bands significantly increase their in-kind gifts Observations 1,048 1,048 1,100 1,100 and there are no lagged effects. Also, when Notes: Results include controls for initial assets, village, and the husband wins a livestock prize, wives’ in- round fixed effects. Standard errors in parentheses. kind gifts decrease by a similar amount as the *** Significant at the 1 percent level. increase observed in husbands’ gifts. Livestock  ** Significant at the 5 percent level. prizes were awarded in front of the entire vil-   * Significant at the 10 percent level.     a p-value < 0.13 lage, thus the resulting gifts may be driven by social pressure. However, cash prizes are only known by others if the recipient chooses to share, suggesting that inter-household transfers s ­ urveys were conducted a week after the lotter- are driven by alternative motives. ies, suggesting the wife commits the prize toward Wives’ private prizes increase cash gifts in an allocation that is both difficult to monitor and the immediate round, though only significantly cannot be easily recovered if the husband were at the 13 percent level.8 Recall that follow-up to find out she won. Public prizes won by wives significantly increase both cash and in-kind gifts in following rounds. 8 This result is significant at the 5 percent level when The differences in the timing of commit- using logarithmic specifications. ting the private prizes toward cash gifts, and 268 AEA PAPERS AND PROCEEDINGS MAY 2013 lagged public prizes toward both kinds of gifts spouse (consistent with Ashraf 2009). As hid- consistent with differing motives. The wife is ­ ing in this context occurs in the form of gifts, commits the private prize to cash gifts immedi- it is not clear whether hiding is inefficient with- ately. On the other hand, she shares (probably out observing reciprocity effects of risk sharing what is left from) public prizes from previous among these households. In Castilla and Walker rounds. The effect on cash gifts is larger than (2012) we examine this issue in more detail. the effect on in-kind gifts, further strengthening the argument of an intention to conceal from her husband, as in-kind gifts are easier to monitor. In REFERENCES the Ghanaian context, inter-household transfers are usually reciprocated at a later date based on Ashraf, Nava. 2009. “Spousal Control and Intra- need, thus this result may indicate that women household Decision Making: An Experimental use gifts as a means of safeguarding their win- Study in the Philippines.” American Economic nings for unexpected emergencies in the future. Review 99 (4): 1245–77. The use of “money guards” by women in this Castilla, Carolina, and Thomas Walker. 2012. way has been noted in other country contexts as “Gender Roles and Intra-Household Alloca- well (Collins et al. 2009). tion.” Unpublished. Chen, Joyce J. 2006. “Migration and Imperfect IV. Conclusions Monitoring: Implications for Intra-household Allocation.” American Economic Review 96 In this paper we used data from a field experi- (2): 227–31. ment in Southern Ghana to test the effect of Collins, Daryl, Jonathan Morduch, Stuart Ruth- windfalls on intra-household allocation. The erford, and Orlanda Ruthven. 2009. Portfolios results from the experiment indicate that spouses of the Poor: How the World’s Poor Live on $2 behave according to a noncooperative marital a Day. Princeton: Princeton University Press. allocation contract with differences in spend- Lundberg, Shelly, and Robert A. Pollak. 1993. ing patterns by gender. Furthermore, asymmet- “Separate Spheres Bargaining and the Mar- ric information over windfalls has a differential riage Market.” Journal of Political Economy effect on observable and concealable expenses, 101 (6): 988–1010. consistent with hiding. Public windfalls received Udry, Christopher. 1996. “Gender, Agricultural by husbands lead to increased investment in Production, and the Theory of the Household.” assets and social capital, while there is no such Journal of Political Economy 104 (5): 1010– effect when wives win. Private windfalls of both 46. spouses are committed to cash (wives) or in- Walker, Thomas. 2011. “Risk Coping, Social kind gifts (husband) which are either difficult to Networks and Investment in Rural Ghana.” monitor or to reverse if discovered by the other Unpublished PhD diss., Cornell University.