AEA Papers and Proceedings 2019, 109: 133–137 https://doi.org/10.1257/pandp.20191073 Entertainment, Education, and Attitudes Toward Domestic Violence† By Abhijit Banerjee, Eliana La Ferrara, and Victor Orozco* Over one-third of women around the world tudes toward domestic violence. We find broadly are victims of physical or sexual violence positive effects. Moreover, the effect seems to be (World Health Organization 2013). Many of concentrated among people who recall the show them live in low income countries where indi- and the narrative around the characters well, viduals are often socialized to accept and tol- consistent with the idea of edutainment. erate gender based violence (GBV). GBV has We contribute to the n ­ onexperimental litera- serious consequences for women’s mental and ture on the impact of commercial TV on gen- physical ­ well-being and significant resources der outcomes (e.g., Chong and La Ferrara 2009; are invested in policies to change such attitudes Jensen and Oster 2009; La Ferrara, Chong, and and behaviors. In this paper we evaluate an inno- Duryea 2012; Kearney and Levine 2015) and to vative approach that uses entertainment televi- recent experimental work that uses edutainment sion to reduce GBV. for public policy (e.g., Banerjee, Barnhardt, and Entertainment education (“edutainment”) is Duflo 2015; Ravallion et al. 2015; Berg and Zia a communication strategy that works through 2017). We differ from the latter in focusing on mass entertainment media with the aim of pro- changing norms toward GBV. moting a better context for behavior change than the delivery of information alone. We experi- I.  Experiment and Data mentally evaluate season 3 of the edutainment TV series MTV Shuga, produced by MTV To evaluate the impact of MTV Shuga we Staying Alive Foundation and filmed in Nigeria. conducted a randomized controlled trial. We set Shuga 3 consists of eight episodes of 22 min- up ​80​screening centers in urban and ­ peri-urban utes each. While the main focus of the series is locations of ​7​towns in ­South West Nigeria and HIV, a s­ ubplot involves a married couple with a invited to a movie showing a random sample violent husband.1 In this paper we focus on this of individuals aged 18–25 who lived within theme and assess the impact of Shuga on atti- a 10 minute walk from each center. Among the attendees we randomly selected ​ 63​people * Banerjee: Department of Economics, MIT, 77 per center (equally divided between men and Massachusetts Avenue, Cambridge, MA 02139 (email: women) to take part in two subsequent screen- banerjee@mit.edu); La Ferrara: Bocconi University, via ings. For these two screenings we randomized Roentgen 1, 20122 Milano, Italy (email: eliana.laferrara@ centers into ​ 54​that showed Shuga (treatment) unibocconi.it); Orozco: The World Bank, 1818 H Street and ​26​that showed a “placebo” TV series (con- NW, Washington, DC 20433, and Oxford University (email: vorozco@worldbank.org). We thank Awa Ambra Seck, Sara trol). In both cases the screenings contained four Spaziani, and Silvia Barbareschi for excellent research assis- episodes each and were one week apart, during tance. Laura Costica and Edwin Ikuhoria did a superb job September–December 2014, immediately after as research and field coordinators. This study was funded the baseline. We collected the f ­ ollow-up survey by the Bill and Melinda Gates Foundation and the World Bank i2i Trust Fund. La Ferrara acknowledges financial sup- eight months after the baseline. port from ERC Advanced Grant ASNODEV (Contract no. We elicited information on attitudes toward 694882). GBV using the following questions. First, we †  Go to https://doi.org/10.1257/pandp.20191073 to visit asked if a husband is justified in forcing his wife the article page for additional materials and author disclo- to have sex when she does not want to. Second, sure statement(s). 1  For an evaluation of the impact of MTV Shuga on we asked if a man is justified in hitting or beating HIV-related outcomes, see Banerjee, La Ferrara, and Orozco ­ his wife if she (i) goes out without telling him; (2018). (ii) neglects the children; (iii) argues with him; 133 134 AEA PAPERS AND PROCEEDINGS MAY 2019 Table 1—Impact on GBV Indexes Dependent variable (Yt​​​​​​): Justify violence (count) Justify violence (dummy) Full sample Females Males Full sample Females Males (1) (2) (3) (4) (5) (6) Treated −0.131 0.004 −0.268 −0.023 0.017 −0.055 (0.088) (0.085) (0.121) (0.017) (0.019) (0.022) t−1​​​ Y​​​ 0.236 0.201 0.296 0.260 0.262 0.251 (0.017) (0.024) (0.027) (0.018) (0.025) (0.027) Controls Yes Yes Yes Yes Yes Yes Mean dependent variable in control group 0.626 0.645 0.609 0.263 0.269 0.257 Notes: The dependent variable in columns 1 ­ –3 is the number of instances in which the respondent considers forced sex or wife beating justifiable; in columns 4 ­ –6 is an indicator for whether such number is ​​greater than 0. Columns ­ 1–3 report estimated Poisson coefficients; columns 4 ­ –6 marginal probit coefficients. Standard errors in parentheses clustered at the screening center level. All regressions include town fixed effects and individual controls. (iv) refuses to have sex; (v) burns the food; (vi) and ​​δc ​ ​​​indicates town fixed effects.3 We estimate fails to prepare food on time; and (vii) refuses (1) using a Poisson model when the outcome is to have another child. We create an indicator for a count, and Probit when it is an indicator, clus- each response and construct two outcome vari- tering standard errors at the screening center ables. The first, “Justify violence (count),” is the (location ​ ) level. l​ sum of all indicators and ranges from ​ 8​ 0​to ​ . The Table 1 reports the estimates of ​​ β  ˆ ​ ​. The depen- second, Justify violence (dummy), is an indica- dent variable in columns ­ 1–3 is Justify violence tor for whether there is at least one instance in (count), while in columns 4 ­ –6 it is Justify vio- which the respondent considers forced sex or lence (dummy). A negative value of β  ​​ˆ ​ ​ indicates wife beating justifiable.2 an improvement in attitudes, i.e., less support for GBV. II. Impact In the full sample the effect of treatment on the number of instances in which violence is We estimate the following regression: justified is negative, consistent with the message of Shuga, but insignificant. There is, however, ​ 1​​  =  β  (1)  ​​yilc Treate​ ​ 0​​  + γ ​y​ dilc ilc0​​  substantial heterogeneity across genders: while for women the effect is a precisely estimated X​  ′ilc + ​   ​  ζ + ​   0​   δ​c​​  + ​ ε​ilc1​​​, zero, for men the coefficient is negative and significant at the 5 percent level.4 To assess the where ​​yilct ​ ​​​is the outcome for individual i ​​ liv- magnitude, we can consider the ­ incidence-rate ing in location l ​​in city c ​​, measured at baseline ratio (IRR) of Treated, that is the ratio of the t  = 0​ (​ ) and f ­ollow-up (​ t = 1​); ​ Treate​ ​ 0 ​​​ is dilc expected count for the treated group to that of an indicator for being assigned to watch Shuga;​​ the control, holding other variables constant.5 X​ ilc 0 ​​​is a vector of controls measured at ­ baseline; The IRR associated with the coefficient ​ −  0.268​ in column 3 is 0 , meaning that—ceteris ​.76​ 3  Controls X include: gender, age, education, enrolled in 2  While the above questions are widely used in the lit- school, single, Muslim, speaking Yoruba, speaking English erature, they may be affected by reporting bias, generating as main or second language at home, not living with one’s concerns of experimenter demand effects. We think the risk parents, household size, wealth, homeownership, and father is moderate in our setting. The main educational component or mother with more than secondary education. of Shuga was clearly about HIV, so the possible reporting 4  −0.255​ for The coefficients are virtually identical (e.g., ​ bias on GBV is likely similar across treatment and control. men) if we include a second order polynomial in the lag of Furthermore, in ongoing work we use item list techniques to the dependent variable. 5  estimate the impact of Shuga on violence experienced and The IRR is obtained by exponentiating the estimated we find similar results. Poisson coefficient. VOL. 109 ENTERTAINMENT, EDUCATION, AND ATTITUDES TOWARD DOMESTIC VIOLENCE 135 ­ aribus—for individuals exposed to Shuga the p Table 2—Memory and Identification with Characters, number of instances in which violence is justi- Summary Statistics fied is about three-fourths of that for individu- Full Diff als not exposed. The effect is found also at the sample Females p-val) Males (  ­ extensive margin (column 6): treated men have Shuga theme: Violence 0.077 0.067 0.087 0.018 a​ 6​percentage points lower probability of justi-   in relationships 21​per- fying forced sex or wife beating, that is a ​ Thought of Malaika 0.551 0.598 0.505 0.000 cent decrease over the mean of the control group. Thought of Nii 0.246 0.256 0.238 0.177 Remembers what 0.585 0.580 0.589 0.640 III.  Measuring Memory and Identification with   happened to Malaika Characters Remembers what 0.474 0.452 0.495 0.029   happened to Nii To shed light on the workings of edutainment, Identifies with Malaika — 0.094 — — in our ­ follow-up survey we collected detailed Identifies with Nii — — 0.014 — information on what viewers remembered about the plot and the characters. Questions about Notes: Share of respondents who comply with the state- Shuga were only asked to the treatment group, ­ -value for the test ments in each row. Column 4 reports the p as the control could not possibly know, so this that the difference between columns 2 and 3 is 0. part of the analysis is restricted to the treated sample. Table 2 reports descriptive statistics. First, we asked about the main themes of the We then showed respondents a picture of each show. About 8 ​ ​percent of respondents indicated character and asked what was the most signifi- “violence in relationships” as a main theme, cant thing they remembered about that charac- with men being slightly more likely to indicate ter. Remembers what happened to Malaika takes this than women. This is consistent with the lim- value one for respondents who mention relevant ited time dedicated to the GBV subplot in the facts (e.g., “her husband was beating her,” “she ­ three-hour season: most respondents correctly tried to have an abortion,” etc.) and zero for identified HIV as the “main” theme. those who do not remember. About ​ 58​percent We then prompted respondents about the of men and women remember what happened to two characters that embedded the GBV theme Malaika. Remembers what happened to Nii is ​1​ in Shuga: Malaika and Nii.  Malaika is a young when respondents mention things like “he was woman who is enrolled in university and is mar- beating his wife,” “he did not want to use birth ried to Nii, a businessman who is very jealous 0​when they do not remember. control,” etc. and ​ and limits his wife’s freedom. Nii would like to About ​ 50​percent of the men and 4 ​ 5​percent of have a child, while Malaika wants to finish her the women remember facts about Nii. studies first. They argue over this and Malaika Finally, we showed pictures of the characters hides that she is taking contraceptives, until the on a tablet and asked respondents which of those moment Nii finds out and beats her. Toward the people they “saw themselves as.” We denote end of the series, Malaika gets pregnant and tries these variables as Identify with Malaika (for to procure herself an abortion, ending up in the women) and Identify with Nii (for men). Only​ hospital with serious complications. 1.4​percent of the men identify with Nii—not We showed respondents pictures of the char- surprisingly, as he was a negative character in acters in Shuga and asked: “Since you saw the the show. About ​ 9​percent of the women identify movie, have there been times during which with Malaika. you remembered a specific character or scene from the program? Which ones?” The vari- IV.  Memory, Identification, and GBV ables Thought of Malaika and Thought of Nii in Table 2 take value one for respondents who According to psychological theory, the effects indicated Malaika/Nii, and zero otherwise. of the edutainment should be mediated by the Approximately 6 ​ 0​percent of women and ​50​per- extent to which viewers’ attention is captured cent of men report that they thought of Malaika by the characters and viewers can take the (significantly different), while for Nii the shares characters’ perspective (Singhal and Rogers are ​26​ and ​24​ percent. 2012). Therefore, we estimate the relationship 136 AEA PAPERS AND PROCEEDINGS MAY 2019 Table 3—Memories of the Show and GBV Attitudes Dependent variable: Justify violence (count) Full sample Females Males (1) (2) (3) (4) (5) (6) Shuga theme: Violence in relationships −0.153 −0.098 −0.115 −0.068 −0.194 −0.157 (0.153) (0.156) (0.224) (0.217) (0.160) (0.162) Thought of Malaika −0.429 −0.432 −0.287 −0.292 −0.791 −0.696 (0.084) (0.087) (0.103) (0.101) (0.158) (0.166) Thought of Nii −0.307 −0.280 −0.141 −0.129 −0.620 −0.525 (0.107) (0.107) (0.121) (0.123) (0.171) (0.167) Remembers what happened to Malaika −0.401 −0.332 −0.264 −0.223 −0.602 −0.493 (0.075) (0.078) (0.105) (0.104) (0.157) (0.162) Remembers what happened to Nii −0.390 −0.321 −0.215 −0.173 −0.645 −0.541 (0.087) (0.084) (0.121) (0.121) (0.169) (0.156) Identification with Malaika 0.025 0.009 (0.217) (0.213) Identification with Nii 0.613 0.582 (0.451) (0.353) Controls No Yes No Yes No Yes Notes: The dependent variable is the number of instances in which the respondent considers forced sex or wife beating justi- fiable. The table reports estimated Poisson coefficients. Each coefficient is from a different regression, where the independent variable of interest is the one listed in each row. All regressions include the lagged dependent variable and town fixed effects; columns 2, 4, and 6 also include individual controls. between the proxies for memory and identifica- violence as a main theme. Viewers who say that tion described above and respondents’ attitudes they have occasionally “thought about”  either toward GBV at f ­ ollow-up. This relationship can- character (Thought of Malaika/Nii) and those not be interpreted as causal because the extent who remember specific facts about them to which someone remembers or identifies with (Remembers what happened to Malaika/Nii) Malaika and Nii is likely correlated with their display significantly lower support for domes- attitudes toward violence. However, it should tic violence, with most coefficients being sig- be noted that we always control for the baseline nificant at the 1 percent level. The effects are value of the dependent variable. Also, we will quite large, with IRR’s of ​ 0.75​for women who compare estimates when including or not individ- thought of Malaika (column 4) and ​ 0.59​for ual controls, to gauge the extent of the potential men who thought of Nii (column 6). This means endogeneity. Our results are reported in Table 3. that the number of instances in which violence Each coefficient in the table is estimated is justified by women and men who thought of from a different regression, where the depen- the respective character is, in order, about three- dent variable is Justify violence (count) and fourths and two-thirds of that of respondents the independent variable of interest is the who did not. The effects are qualitatively sim- binary variable listed by row.6 Town fixed ilar at the extensive margin (not reported), with effects are always included, while socioeco- these viewers having an 8 ​ 0​percent lower ​ ​to 1 nomic controls (the same as in Table 1) are probability of justifying GBV. included in ­ even-numbered columns but not in Remembering facts that happened to the ­odd-numbered ones. two characters is also significantly correlated We find that eight months after viewing Shuga, with outcomes at ­ follow-up. The correspond- attitudes toward GBV are not s ­ignificantly ing IRR’s are around ​ 0.8​and ​ 0.6​for women different for viewers who indicate domestic ­ and men, respectively. The probability of justi- fying violence is about ​ 7​to ​9​percentage points lower for these viewers (estimates not reported). 6  We obtain entirely consistent results when using “Justify Interestingly, identification with the characters is violence (dummy)” as an outcome. not significantly correlated with attitudes toward VOL. 109 ENTERTAINMENT, EDUCATION, AND ATTITUDES TOWARD DOMESTIC VIOLENCE 137 GBV, except perhaps for men: men who iden- Encouraging the Adoption of Iron-Fortified tify with Nii are, not surprisingly, less bothered Salt.” NBER Working Paper 21616. by domestic violence. Note that all coefficients Banerjee, Abhijit, Eliana La  Ferrara, and Vic- in Table 3 are very similar with and without the tor Orozco. 2018. “The Entertaining Way to inclusion of observable individual characteristics. Behavioral Change: Fighting HIV with MTV.” In interpreting these estimates recall that Unpublished. we found a zero treatment effect on women. Berg, Gunhild, and Bilal Zia. 2017. “Harnessing Combined with the fact that greater recall predicts Emotional Connections to Improve Financial more negative attitudes toward GBV even among Decisions: Evaluating the Impact of Finan- women, this suggests the possibility of reverse cial Education in Mainstream Media.” Jour- causality that is not fully purged by controlling nal of the European Economic Association 15 for observable individual characteristics (includ- (5): 1025–55. ing baseline attitudes). Note, however, that the Chong, Alberto, and Eliana La  Ferrara. 2009. effect of recall is much larger for men, and it is “Television and Divorce: Evidence from Bra- for men that treatment has a causal effect. One zilian ‘Novelas.’” Journal of the European way to interpret these results is that the stronger Economic Association 7 (2/3): 458–68. effect of recall on men reflects the causal effect Jensen, Robert, and Emily Oster. 2009. “The of memory on their reaction to Shuga, though we Power of TV: Cable Television and Women’s acknowledge that this is speculative. Status in India.” Quarterly Journal of Econom- ics 124 (3): 1057–94. V. Conclusions Kearney, Melissa S., and Phillip B. Levine. 2015. “Media Influences on Social Outcomes: The Despite the fact that domestic violence was a Impact of MTV’s 16 and Pregnant on Teen secondary theme of MTV Shuga, we have shown Childbearing.” American Economic Review that (random) exposure to this educational TV 105 (12): 3597–3632. series induced an improvement in men’s atti- La Ferrara, Eliana, Alberto Chong, and Suzanne tudes toward women eight months later. Using Duryea. 2012. “Soap Operas and Fertility: Evi- a detailed set of measures of viewers’ memories dence from Brazil.” American Economic Jour- of the characters and identification with them, nal: Applied Economics 4 (4): 1–31. we also find that attitudes toward GBV signifi- Ravallion, Martin, Dominique van de Walle, Puja cantly improve for men and women who report Dutta, and Rinku Murgai. 2015. “Empowering occasionally thinking about the characters Poor People through Public Information? Les- and who remember specific facts about them. sons from a Movie in Rural India.” Journal of Identification with the characters seems to play Public Economics 132 (C): 13–22. a lesser role. These findings call for a deeper Singhal, Arvind, and Everett Rogers. 2012. Enter- analysis of the links between attention, empathy tainment-Education: A Communication Strat- and the policy impact of edutainment programs. egy for Social Change. London: Routledge. World  Health  Organization. 2013. Global and REFERENCES Regional Estimates of Violence against Women: Prevalence and Health Effects of Inti- Banerjee, Abhijit, Sharon Barnhardt, and Esther mate Partner Violence and Non-partner Sexual Duflo. 2015. “Movies, Margins and Marketing: Violence. Geneva: WHO.