J Popul Econ (2010) 23:1177–1187 DOI 10.1007/s00148-008-0226-3 ORIGINAL PAPER

Far above rubies: Bride price and extramarital sexual relations in Uganda David Bishai · Shoshana Grossbard

Received: 2 November 2006 / Accepted: 1 October 2008 / Published online: 11 November 2008 © Springer-Verlag 2008

Abstract The custom of bride price involves the payment of goods or cash from the groom’s family to the bride’s family at the time of marriage. Data from a household survey in Uganda were used to estimate the relationship between payment of bride price and non-marital sexual relationships. A robust correlation between bride price payment and lower rates of non-marital sexual relationships is found for women but not for men. One interpretation we offer for these findings is that bride price reflects the price of women’s sexual fidelity to men. This interpretation makes sense in light of the refundable nature of bride price in Uganda. Keywords Marriage · Extramarital relations · Bride price · Uganda JEL Classification D13 · I12 · J13 · 015

Responsible editor: Junsen Zhang “Who can find a virtuous woman? For her price is far above rubies.” –Proverbs 31:10 D. Bishai (B) Department of Population Family and Reproductive Health, Johns Hopkins University, Bloomberg School of Public Health, 615 N. Wolfe St., Baltimore, MD 21030, USA e-mail: [email protected] S. Grossbard IZA Fellow and Department of Economics, San Diego State University, San Diego, CA 92182, USA

1178

D. Bishai, S. Grossbard

1 Introduction In this paper we use Ugandan data to test for linkage between payment of a bride price and the extramarital behavior of husbands and wives. Like many other Africans, many Ugandan men make monetary bride price payments to a girl’s family prior to marriage. Ethnographical perspectives from Uganda report that husbands are considered as owners of sexual rights over their wives, that unauthorized sexual contact between a married woman and another man is considered to be theft committed against her husband, and that bride price transfers rights over a woman’s sexuality (Parikh 2007). This study gives an economic interpretation to bride price payments to assess whether they incentivize faithful female sexual behavior. Existing studies of extramarital relationships in Africa highlight the effects of male wealth and a wife’s pregnancy as factors affecting male promiscuity, (Kimuna and Djamba 2005; Onah et al. 2002) but there has been little attention given to women’s incentives or disincentives to be sexually faithful. Our empirical analysis uses data from a household survey conducted in the capital city Kampala and 12 rural districts of Uganda in 2001. We find that if a bride price was paid, women are less likely to have extramarital relations, but the same is not true for men. We discuss possible reasons for this asymmetry. It could be that men pay for their wife’s fidelity when they pay a bride price to her relatives. The refundable nature of the bride price gives women incentives to comply. This interpretation is consistent with previous economic analyses of marriage.

2 Theoretical discussion Commitment devices can make verbal agreements binding. In the Ugandan context bride price refundability could reinforce verbal promises to be sexually faithful. Even though it is officially illegal to request the refund of bride price in Uganda, (Government of Uganda 2001) if a wife is not faithful, a husband can claim his money back and return his wife to her male relatives. Conversely, paying for their wives’ fidelity in the form of bride price may reduce men’s need to exchange vows of sexual fidelity with their brides in order to obtain their fidelity. As a result, we do not expect a significant positive relationship between male payment of bride price and male fidelity in marriage. The logic of our conceptual model is as follows: (1) men deposit bride prices with their wives’ families, knowing that this money is available to them in case their wives commit adultery; (2) this payment, which men can recover conditional on their wives’ behavior, raises married men’s bargaining power and well-being in marriage; and (3) an expression of men’s higher well-being is more wives’ fidelity without corresponding increases in husbands’ fidelity. The next section presents some tests of whether women for whom a refundable bride price was paid at marriage are more likely to be faithful to their

Far above rubies

1179

husbands, whether bride price payments affect the likelihood of male infidelity, and of the determinants of bride price.

3 Empirical Study Data Data for the study come from a household survey conducted in Kampala and 12 rural districts of Uganda in 2001 (Bishai et al. 2004). The survey included a module to assess respondent’s risk of HIV/AIDS via sexual behavior, including numbers of extramarital sexual partners. In addition to sexual behavior, one of the final sections of the survey asked all currently married women whether money or valuable items were transferred to their family by way of bride price. They were also asked the date of completion of the payment and the total value of the payment. A total of 1,758 individual interviews were conducted with respondents usually residing in the sampled households, aged 18–60, and not absent for more than 6 months in the previous 12 months, of these 839 were interviews with currently married women, and 430 with the husbands of these women. Husbands were only interviewed by male interviewers, wives by female interviewers. We eliminated 247 women with missing bride price data from the analysis, which left 552 married women with valid observations of both women’s self-reported infidelity and self-reported bride price and 340 valid observations about these women’s husbands. Since only 157 respondents provided nominal estimates of the size of their bride prices and the year payment was completed we only use a dichotomous variable: whether a bride price was paid or not. Although the absence of data on bride price was not associated with women’s schooling, it was associated with age. The women with missing bride price had husbands who were on average 5 years older and they themselves were on average 3 years older than the sample who reported bride price status. The possibility that unobservable individual factors were also associated with failure to report bride price will limit the generalizability of the analysis. Table 1 lists the characteristics of the women and men in the analytical sample. Table 1 shows that 5% of wives and 19% of husbands report nonspousal sex during the prior 12 months. The sample of the 340 wives whose husbands were also interviewed report the same rates of non-spousal sex (not shown in table). The overall prevalence of bride price was 68% in the 552 women studied. There were age/period effects: the prevalence of bride price declined with age from 91% for women ages 50–60; to 72% for women ages 30–49, and 63% for women ages 18–29 reporting bride price (not shown in table). Our data also indicate that there was wide variation in the timing of bride price payments: in the case of six respondents bride price was reportedly paid prior to a girl’s birth, and for an additional 13 women bride price was paid prior to a girl’s 10th birthday. The median age of the wife at the time that bride price was paid in full was 18. We settled on a simple dichotomous indicator equal to 1 for households that owned more than three assets from the asset list. Schooling was coded as a quadratic for the husband and the wife. Our results

Based on only 340 observations

0 (0.00) 19 (0.67) 69 (0.93) 31.27 (9.62) 4.16 (3.54) 24 (0.80) 38.38 (12.27) 5.52 (3.11) 0.5 (1.63) 57 (1.07) 32 (0.95) 7 (0.280) 17 (0.62) 28 (0.88) 18 (0.64) 19 (0.67) 18 (0.64)

19 (0.66) 68 (0.93)

31.1 (9.6) 4.19 (3.52)

23 (0.75)

38.15 (12.25) 5.52 (3.13)

0.52 (1.63) 56 (1.05)

32 (0.93) 7 (0.28) 17 (0.60) 28 (0.86) 18 (0.63) 20 (0.68) 17 (0.60)

526

5 (0.20)

552

Wives with no infidelity

27 (0.27) 8 (1.44) 12 (2.07) 42 (4.78) 23 (3.47) 23 (3.47) 0 (0.0)

0.73 (1.59) 45 (4.65)

33.4 (11.12) 5.46 (3.67)

8 (1.44)

27.81 (8.36) 4.81 (3.00)

38 (4.62) 35 (0.45)

100 (0.00)

26

Wives with infidelity

31 (1.10) 8 (0.38) 22 (0.88) 30 (1.08) 11 (0.50) 22 (0.88) 15 (0.66)

0.40 (1.42) 63 (1.20)

39.38 (12.72) 5.52 (3.02)

27 (1.02)

32.13 (9.91) 3.88 (3.41)

21 (0.86) 100 (0.0)

2 (0.10)

376

Wives with bride price

34 (1.68) 5 (0.36) 5 (0.36) 25 (1.41) 34 (1.68) 16 (1.01) 20 (1.20)

0.75 (1.98) 42 (1.83)

35.56 (10.79) 5.52 (3.35)

15 (0.96)

28.96 (8.49) 4.84 (3.65)

16 (1.01) 0 (0.0)

10 (0.67)

178

Wives without bride price

Schooling was constructed by coding all individuals with 1–4 years of schooling as 2.5, all with 5–7 as 6, all with 8–11 as 9.5, all with 12–13 as 12.5, and all with over 13 as 13

b

a

Number of observations Variables % infidelity reported by wives in last 12 months % infidelity reported by husbands in last 12 monthsa % paid bride price Wife’s characteristics Age Schooling in yearsb % perceives HIV not possible Husband’s characteristics Age Schooling in yearsb Number of months away in last 12 months % farmer Household characteristics % owns 3 or more items % polygynous % North % East % Kampala % Central % West

Whole sample

Table 1 Means, percent, and standard deviations (in parentheses)

1180 D. Bishai, S. Grossbard

Far above rubies

1181

are robust to different treatments of the asset variables and different ways of coding schooling. A dummy variable controls for region in our regressions. We also include an interaction term between bride price and the dummy for residence in western Uganda. A statistical appendix with alternative ways of coding schooling and assets is available from the authors. Statistical specification In order to preserve maximum sample size while including husband’s characteristics in models predicting wife’s behavior we chose to do all of the following: (a) run models without any husband’s characteristics, (b) run models including husband’s characteristics despite the reduced sample size, and (c) run models in which husband’s characteristics were imputed by regressing each variable on all other right-hand-side covariates. After determining that the coefficient on bride price did not vary by more than 14% regardless of the strategy and that its statistical significance was improved from a p-value of 0.07 to 0.01 in models where husband’s characteristics were imputed, we opted for models including imputed values. Our major goal was to estimate regressions of marital infidelity. We define P∗ as the probability of participating in any extramarital relations in the past 12 months, since we have data on participation but not on number of infidelities. We model P∗gh , the infidelity probability of the “g-th” wife married to the “h-th” husband, as a function of another dichotomous variable, B∗ , whether or not bride price was paid at the time of marriage, characteristics of the wife Xg , and characteristics of the h-th” husband Xh : P∗gh = c + γ1 B∗gh + γ2 Xg + γ3 Xh + ηgh

(1)

∗ Likewise, we model Phg , the probability that a particular husband h participated in an extramarital affair, as a linear function ∗ = c + β1 B∗hg + β2 Xh + β3 Xg + μhg . Phg

(2)

Whether a bride price was paid, B∗gh , is also modeled as a linear equation: B∗gh = c + δ1 Xh + δ2 Xg + εgh

(3)

Because male and female behavior could be correlated, we explored a bivariate probit model proposing a non-zero covariance, i.e. Cov(ηhg μgh ) > 0. We obtained an insignificant ρ statistic, thus leading us to reject the hypothesis that Cov(ηhg μgh ) >0. We then proceeded to estimate models applied to one spouse at a time. Probit models would have been natural given the dichotomous outcomes studied, and they were estimated, but they forced the exclusion of the 141 women, primarily from the western regions, who reported no extramarital relations. If a model contains a dummy variable for a group which shares common values of the dependent variable, this group of dummy coefficient cannot be estimated by probit or logit models. Ordinary least squares (OLS) models that assume linear probability are not sensitive to this problem (Caudill 1988). We therefore estimated OLS models of marital infidelity that enabled us to preserve sample size, even though some predicted probabilities from an OLS model would be outside the 0–1 interval. Both the

1182

D. Bishai, S. Grossbard

probit and OLS model give significant coefficients on the relationship between bride price and wife’s extramarital affairs, but the advantage of the OLS results is that they include the western region. This allowed us to estimate models that include an interaction between ‘bride price’ and ‘west’ and attain higher statistical significance, as presented in Table 2 and discussed in the next section. We included demographic characteristics like age and husband’s farmer status as well as factors that have previously been associated with marital fidelity like wealth and education (Kressel 1977; Cameron 2002). Other plausible variables that could affect a woman’s infidelity were included such as husband’s absence and whether the wife perceives a high risk of AIDS. In a fruitless search for plausible instruments for bride price we tested whether within our Ugandan sample regional polygyny rates and prevalence of ethnic groups were related to whether a bride price was paid. We tried both two-stage least squares and bivariate probit specifications that included markers of ethnic variation and district level polygyny as potential instruments for bride price. These specifications failed tests for weak instruments, in addition to their questionable exclusion restrictions. We thus abandoned attempts to use instrumental variables for bride price and do not present these estimates. Results One can appreciate the relationship between bride price and infidelity from the descriptive statistics presented in Table 1. A comparison of the last two columns in Table 1 indicates that 2% of women with bride price reported infidelity, compared to 10% of women without bride pride reporting infidelity. This ratio of 1 to 5 is statistically significant. The same two columns in Table 1 also indicate that male infidelity rates are higher (21%) in unions with bride price, compared to (16%) in unions without bride price, but this difference is not statistically significant. Of the 88 men who cheated on their wives, six report sex with a short-term partner, 53 report sex with a long-term partner, and 29 report sex with both types. Only one of the husbands reported frequenting what was defined as a commercial sex worker. The tendency of Ugandan men to prefer to have extramarital sex in the context of a relationship instead of as a brief commercial transaction was also reported elsewhere (Parikh 2007). Of the women who cheated on their husbands, five report sex with a short-term partner, 22 report sex with a long-term partner, and two report sex with both types. For reasons explained above, none of the models presented in this paper are based on an instrumental variables approach so the regressions presented in Table 2 must be interpreted as only indicative of causation. Models A1 and A2 are linear probability models of an individual woman’s probability of reporting non-spousal sexual contact in the last 12 months. Model A1 does not include an interaction with western—imposing a common slope on bride price for the entire sample. Model A2 relaxes that assumption with an interaction term. Models A1 and A2 show that having bride price is correlated with a woman’s probability of infidelity that is 0.06 to 0.09 points lower, controlling for both spouse’s age and schooling and a number of other characteristics. Model B

c

b

Significant at 1% Significant at 10%

Robust t-statistics in parentheses a Significant at 5%

Wife had bride price Wife’s characteristics Wife’s schooling Square of wife’s schooling Wife’s age Wife perceives no risk of AIDS Husband’s characteristics Husband’s schooling Square of husband’s schooling Husband is a farmer Husband’s age Number of months husband was away in last 12 months Household characteristics Household owns 3 or more items North East Kampala Polygynous household West Interaction of West and had bride price Constant Number of observations Adjusted R-squared Controls for ethnicity

−0.087 (−2.73)b 0.015 (1.98)a −0.002 (−1.90)c 0 (−0.01) −0.022 (−1.47) −0.02 (−1.58) 0.002 (1.47) −0.04 (−1.30) −0.001 (−0.76) 0.004 (0.75) −0.021 (−1.13) 0.008 (0.34) 0.016 (0.53) −0.03 (−0.83) −0.006 (−0.15) −0.144 (−3.97)b 0.113 (3.23)b 0.222 (3.33)b 552 0.04 No

−0.066 (2.51)a 0.013 (1.76)c −0.001 (−1.64) 0 (0.16) −0.02 (−1.36) −0.017 (−1.43) 0.001 (1.34) −0.039 (−1.24) −0.001 (−0.89) 0.003 (0.59) −0.02 (−1.11) 0.004 (0.17) 0.016 (0.54) −0.025 (−0.71) −0.005 (−0.13) −0.07 (−3.63)b 0.199 (3.03)b 552 0.03 No

A2

Wife’s infidelity A1

Table 2 OLS regressions of wife’s infidelity, husband’s infidelity, and bride price

−0.042 (−0.73) −0.122 (−1.84)c −0.05 (−0.82) −0.162 (−1.58) 0.047 (0.34) −0.201 (−2.78)b 0.01 (0.11) 0.531 (3.99)b 344 0.04 No

−0.039 (−1.80)c 0.003 (1.72)c −0.067 (−1.43) −0.003 (−0.86) 0.005 (0.26)

0.028 (1.52) −0.002 (−1.04) −0.003 (−0.82) −0.069 (−1.41)

0.078 (1.33)

Husband’s infidelity B

0.293 (2.43)a 590 0.12 No

0.044 (1.02) 0.071 (1.18) −0.054 (−1.01) −0.33 (−4.61)b 0.054 (0.66) −0.091 (−1.29)

0.035 (1.60) −0.002 (−0.91) 0.142 (2.53)a 0.002 (1.13)

0.007 (0.41) 0 (−0.32) 0.004 (1.41) 0.103 (2.19)a

Wife had bride price C

0.201 (1.52) 578 0.13 Yes

0.047 (1.06) 0.214 (2.60)a −0.04 (−0.72) −0.257 (−3.63)b 0.018 (0.21) 0.183 (1.86)c

0.036 (1.60) −0.002 (−0.90) 0.144 (2.54)a 0.002 (1.01)

0.002 (0.10) 0 (−0.03) 0.004 (1.40) 0.101 (2.01)a

D

Far above rubies 1183

1184

D. Bishai, S. Grossbard

shows that bride price had a positive but not statistically significant correlation with the probability of male infidelity. Our results show non-linear correlations between marital fidelity and schooling. According to Models A1 and A2, each additional year of wife’s schooling increases the probability of her infidelity if she has between 0 and 4 years of schooling. However, an additional year of schooling lowers the probability of her infidelity if she has more than 4 years of schooling. Model B shows that for husbands the net “effect” of a year of schooling is to decrease the probability of infidelity for men who have between 0 and 5 years of schooling, and to increase that probability for those with more than 5 years of schooling. The lack of correlation between measures of household wealth and male infidelity was unexpected and may be occurring because lists of shared household assets may be a poor proxy for disposable male income. It is also possible that unfaithful men can shift assets out of the shared house towards the upkeep of a long-term mistress. It was also unexpected that male infidelity would show no correlation with husbands’ time spent living outside the household. Because most of the male infidelity does not occur in formally organized commercial sex markets, it is not possible to use data on the supply of formal sex workers to validate the men’s reports of infidelity. The survey did not include any questions on the location or other attributes of the sex partner. We conducted extensive tests to establish that the results were robust to alternative specifications. The general pattern of results shown in the OLS models is not altered by the use of probit models. These results are available from the authors upon request. If bride price simply marks out more traditional women who adhere to sexual norms, then including a variable for schooling—a better proxy for traditionalism—should lower the effects of bride price on infidelity. In models of wife’s infidelity that excluded schooling (not in tables) the coefficient of bride price was −0.089 (t-statistic −2.84), which is almost identical to the −0.087 obtained in model A2 that includes schooling. Bride price thus does not seem to be correlated with infidelity simply because it signals traditionalism. Our finding of a negative correlation between women’s marital infidelity and bride price is robust to the inclusion of additional variables not shown in Table 2. Prior work with the same data estimated models of infidelity including ethnic composition, particularly heterogeneity (Bishai et al. 2006). Including ethnic dummies in addition to bride price did not change the correlation between infidelity and bride price. Community amenities, such as distance to a market and an urbanicity scale, did not have important effects on the level or significance of the bride price variable either. Rates of wife’s infidelity were higher (13.8%) for households with only one child compared to 4.5% in households with more than one child. This difference in infidelity by number of children was only significant in bivariate analysis ( p = 0.01), but not in multivariate regression models. Including the number of children did not change the correlation between bride price and infidelity. Our basic result is also robust to different definitions of schooling. We obtained similar results when estimating alternative regressions of infidelity omitting quadratic terms

Far above rubies

1185

in schooling, including spousal schooling differences, and coding schooling categories as dummies. When data were stratified to urban vs. rural samples the effects of bride price on wife’s infidelity were stronger and estimated with more precision in urban samples. We assessed a possible non-response bias. In the sample of women who did not respond to the infidelity question, the proportion having bride price was higher than average at 76%. Thus the non-responding women appear to be somewhat more traditional. It is speculative to suggest how including these women would have changed the results, but we nevertheless conducted an experiment to see what happens if the non-responders were recoded as all unfaithful or as all faithful. If all of the 38 women who did not respond to the infidelity question are recoded as being unfaithful, the coefficient on bride price in model A2 is reduced to −0.071 (t-statistic = −1.82). If these 38 women with no response are recoded as being faithful (which is more likely, since a majority of women in our sample were faithful) the coefficient becomes −0.083 (t-statistic −2.84). This exercise suggests that excluding the women who did not report their fidelity (as we did in our estimations) leads to results similar to those obtained if these respondents were all faithful. Models C and D of Table 2 show the correlates of the likelihood that bride price was paid. The difference between Model C and Model D of Table 2 is that Model D includes markers for ethnic group. Bride price is associated with households involved in farming and is significantly less common in the urban setting of Kampala. It is also associated with a higher probability that the wife will state that she perceives no risk of AIDS. The inclusion of the ethnic group markers accounts for a modest increase in R2 of 0.02 points, suggesting that it would make a weak instrument (Hahn and Hausman 2003). That husband’s income or wealth is not associated with bride price could possibly reflect compensating differentials: women and their guardians prefer wealthy husbands. This preference could lead to a large supply of brides willing to marry wealthy husbands and thus less of a need for wealthy men to pay a bride price . Alternatively, the total compensation that wives receive from husbands over a lifetime could include a number of components: bride price payment at marriage, payments during marriage, and post-marriage payments. The higher the expected payments after marriage (due to husbands’ higher wealth), the less husbands pay at marriage (see Grossbard-Shechtman 1993).

4 Discussion and conclusions In most of our regressions we find that payment of a bride price is significantly correlated with reduced participation in extramarital relations by women, but not by men. Two ways of interpreting this finding are that (1) bride price includes a price that men pay to acquire their future wife’s fidelity and (2) bride price is a marker for unmeasured characteristics associated with higher female fidelity.

1186

D. Bishai, S. Grossbard

It is difficult to think of such unmeasured characteristics that would explain why women would be more faithful but have no effect on male fidelity. For instance, it could be that men who paid bride price are richer (husband’s income not being fully captured by ‘education’ and ‘farmer status’) and that women are less inclined to cheat on a richer husband. However, if higher income makes men more attractive to women, one expects men who paid bride price to have more infidelities. Alternatively, it is possible that women who were paid bride price are more physically attractive. This would create more opportunities for female infidelity, and we would observe a significantly positive correlation between bride price and female fidelity. Furthermore, husbands of more attractive wives would be less likely to have extramarital relations and we would observe a significantly negative correlation between bride price payment and male infidelity. In light of bride price’s refundable nature in Uganda our findings can be interpreted as follows: when men pay a refundable bride price at marriage, they partially pay for their future wives’ fidelity. Paying for their wives’ fidelity in the form of bride price may reduce men’s need to comply with vows of sexual fidelity to their brides in order to obtain their fidelity, thus we don’t find a positive relationship between paying bride price and men’s fidelity in marriage. In Africa in general, and in Uganda in particular, there has been active policy discussion about abolishing the refundability of bride price. It has been argued that such refundability enables men to control their wives by threatening to divorce and request a refund of the bride price (Okioma 2004; Rogers 2004; Wendo 2004). Our analysis reinforces such arguments. The elimination of a bride price system, or the weakening of such system via the elimination of refundability, may encourage men to find other ways of obtaining their wife’s fidelity, including reciprocation. Reform-minded Ugandans may adapt Solomon’s Proverb as follows: Who can find a virtuous woman? A virtuous man, for rubies no longer buy virtue. Acknowledgements Helpful comments were received from the editor, Junsen Zhang, two anonymous referees, Stephane Mechoulan, Catherine Sofer, and Howard Yourow.

References Bishai D, Pariyo G, Ainsworth M et al. (2004) Determinants of personal demand for an AIDS vaccine in Uganda. Bull WHO 82(9):652–660 Bishai D, Patil P, Pariyo G et al. (2006) The Babel effect: community linguistic diversity and extramarital sex in Uganda. AIDS Behav 10(4):369–376 Cameron S (2002) The economics of partner out trading in sexual markets. J Bioecon 4(3):195–222 Caudill S (1988) An advantage of the linear probability model over probit or logit. Oxf Bull Econ Stat 50(4):425–427 Government of Uganda (2001) A compendium of laws relating to domestic relations in Uganda. LDC, Kampala Grossbard-Shechtman S (1993) On the economics of marriage. Westview Press, Boulder, CO

Far above rubies

1187

Hahn J, Hausman J (2003) Weak instruments: diagnosis and cures in empirical econometrics. Am Econ Rev 93(2):118–125 Kimuna S, Djamba Y (2005) Wealth and extramarital sex among men in Zambia. Int Fam Plan Perspect 31(2):83–89 Kressel GM (1977) Bride-price reconsidered. Curr Anthropol 18(3):441–458 Okioma M (2004) Brideprice—paving the way for a killer. International Conference on Brideprice and Development, 11 April 2004 Onah HE, Iloabachie GC, Obi SN et al. (2002) Nigerian male sexual activity during pregnancy. Int J Gynaecol Obstet 76(2):219–223 Parikh S (2007) The political economy of marriage and HIV: the ABC approach, “safe” infidelity, and managing moral risk in Uganda. Am J Public Health 97(7):1198–1208 Rogers N (2004) The human and economic costs of bride price in Uganda. International Conference on Brideprice and Development, 11 April 2004 Wendo C (2004) African women denounce bride price. Campaigners claim payment for wives damages sexual health and contributes to aids spread. Lancet 363(9410):716