Errata for Karim and Beardsley (2016), Explaining Sexual Exploitation and Abuse in Peacekeeping Missions: The Role of Female Peacekeepers and Gender Equality in Contributing Countries, Journal of Peace Research 53 (1): 100-115. Errata Summary We have discovered a data management error in the calculations of the weighted averages of the records of gender equality in peacekeeping contributing countries. We mistakenly treated contributing countries with missing data as if their values on the gender-equality indicators were zero. 1 This led to weighted averages that were systematically lower for missions comprised of contributing countries with more missing data. Our intention, instead, was to only calculate the weights across the contributing countries for which data existed. Because of the higher amount of missing data for the indicator related to the ratio of girls to boys in primary school, the results pertaining to the relationship between that measure of gender inequality and allegations of sexual exploitation and abuse (SEA) are not statistically significant when the variable is coded as we intended. The results for the other two measures of gender inequality remain robust, in no small part because we have much less missing data for the female labor force participation variable and the physical protection of women variable. Moreover, in subsequent analyses that we have prepared for a forthcoming book (Equal Opportunity Peacekeeping (2017) Oxford University Press), we find additional evidence that connects the practice of gender inequality in contributing countries with allegations of SEA in peacekeeping missions. In those analyses, even before we discovered this error, we had chosen to not use data on the ratio of girls to boys in primary school in part because of the large amounts of missing data for this variable. We regret this error but remain firm in our conclusion that peacekeeping missions that consist of more peacekeepers from contributing countries with poorer records of gender equality are more likely to experience allegations of SEA. Comparison of the Original Results with the New Results The weighted averages of the practices of gender inequality in contributing countries were calculated in the original submission by first multiplying the values of the variables for each contributing country by the proportion of personnel that the contributing country has provided. The data were then aggregated to the mission level by summing across each contributing country to produce a weighted average. In the original submission, we calculated the proportion of personnel that each contributing country provided by using the actual total mission size in the denominator. Instead, we had intended to remove the personnel from contributing countries with missing data from the denominator since those countries would not be able to contribute to the weighted average. In other words, if we only have information for 9 of 10 contributing countries, then the weighted average 1 To be transparent, Kyle Beardsley bears the full weight of culpability for this coding mistake. We thank Matthias Ecker-Ehrhardt for beginning the discussion that led to the self discovery of this mistake. 1
should only factor in information from the 9 countries for which we have data. Since the values for each of the gender inequality measures are strictly positive, our weighted averages were biased downward when we had missing data, and the bias was increasing in the extent of the missingness. In this sense, a mission could appear to consist of more personnel from contributing countries with better records of gender equality than another mission when that is actually not the case and the difference is solely a product of having less missing data. How much missingness was in the data for these gender inequality variables? Of the 3340 contributor-mission-years in our data (prior to aggregation to the mission-year level), we had complete data for 97% of the observations related to the ratio of girls to boys in primary school, 99% if the observations related to the participation of women in the labor force, and 99% of the observations related to the physical protection of women index. The table below shows the means and standard deviations of the affected variables in the original submission and in the updated data. It is evident that the values for the primary school gender ratio are the most affected by the update, with a noticeable upward shift in the mean, and a more than 50% reduction in the size of the standard deviation. Avg. contributor primary school gender ratio Avg. contributor labor force gender ratio Avg. contributor physical protection of women index Original Mean Original SD Updated Mean Updated SD 95.32 6.94 97.80 3.08 51.80 7.93 51.80 7.93 2.90 0.74 2.90 0.74 Not surprisingly, the findings do not change much between the results in the original published article and the results using the modified data for the labor-force and physicalprotection-of-women measures of gender inequality. The findings related to the primaryschool gender ratio, however, do weaken in terms of statistical significance. The following presents the updated tables of coefficients and figures with the updated data. 2
Original Table 2: SEA accusations in military contingents, negative binomial regression Variables Model 1 Model 2 Model 3 Female ratio balance in PKO mission -33.79** -17.73* -19.10 (16.40) (9.202) (16.69) Avg. contributor primary school gender ratio -0.0547** (0.0234) Avg. contributor labor force gender ratio -0.0990** (0.0436) Avg. contributor physical protection of women index 1.149 (1.014) Size of military contingent in PKO 0.000237*** 0.000185*** 0.000190*** (4.04e-05) (3.14e-05) (6.20e-05) GDP per capita in host country -0.000304-0.000406-0.000236 (0.000455) (0.000423) (0.000550) Index of Mass Rape in the Previous War -0.746-0.786* -0.768 (0.534) (0.426) (0.579) Constant 6.604*** 6.633*** -2.401 (1.674) (1.917) (4.751) ln(alpha) -0.413-1.245-0.216 (1.036) (1.453) (1.130) Observations 80 80 80 *** p<.01; ** p<.05; * p<.1 Standard errors in parentheses 3
Modified Table 2: SEA accusations in military contingents, negative binomial regression Variables Model 1 Model 2 Model 3 Female ratio balance in PKO mission -28.68** -17.73* -19.12 (14.33) (9.201) (16.67) Avg. contributor primary school gender ratio -0.0198 (0.128) Avg. contributor labor force gender ratio -0.0990** (0.0436) Avg. contributor physical protection of women index 1.146 (1.007) Size of military contingent in PKO 0.000227*** 0.000185*** 0.000190*** (4.07e-05) (3.13e-05) (6.19e-05) GDP per capita in host country -0.000352-0.000406-0.000236 (0.000523) (0.000423) (0.000549) Index of Mass Rape in the Previous War -0.800-0.786* -0.768 (0.581) (0.426) (0.579) Constant 3.598 6.633*** -2.389 (11.55) (1.917) (4.728) ln(alpha) -0.207-1.245-0.217 (1.107) (1.453) (1.130) Observations 80 80 80 *** p<.01; ** p<.05; * p<.1 Standard errors in parentheses 4
Original Table 3: SEA accusations in military contingents, negative binomial regression Variables Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Female ratio balance in PKO mission Avg. contributor primary school gender ratio Avg. contributor labor force gender ratio Avg. contributor physical protection of women index Size of military contingent in PKO GDP per capita in host country Index of Mass Rape in the Previous War Avg. contributor Polity score Avg. contributor GDP per capita -35.28** -24.53*** -3.465-24.22-12.04-17.87 (17.59) (8.796) (13.38) (21.20) (12.12) (17.46) -0.0439* -0.0636* (0.0257) (0.0345) -0.0941*** -0.0931** (0.0324) (0.0466) 3.485*** 0.428 (1.195) (2.224) 0.000247*** 0.000199*** 0.000166*** 0.000203*** 0.000171*** 0.000190** (3.40e-05) (3.01e-05) (3.28e-05) (6.36e-05) (4.60e-05) (6.35e-05) -0.000381-0.000445-0.000166-0.000141-0.000361-0.000249 (0.000419) (0.000331) (0.000262) (0.000579) (0.000484) (0.000571) -0.835-0.876** -0.811** -0.648-0.740-0.753 (0.516) (0.345) (0.331) (0.630) (0.500) (0.591) 0.133 0.162* 0.538*** (0.160) (0.0943) (0.138) -8.82e-05-2.59e-05-3.93e-05 (9.64e-05) (3.33e-05) (8.82e-05) Constant 5.145*** 5.832*** -13.53** 7.601*** 6.334*** 0.296 (1.745) (1.789) (5.281) (2.489) (2.194) (8.731) ln(alpha) -0.771-1.532-1.569*** -0.340-1.135-0.184 (0.932) (0.982) (0.568) (1.133) (1.617) (1.121) Observations 80 80 80 80 80 80 *** p<.01; ** p<.05; * p<.1 Standard errors in parentheses 5
Modified Table 3: SEA accusations in military contingents, negative binomial regression Variables Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Female ratio balance in PKO mission Avg. contributor primary school gender ratio Avg. contributor labor force gender ratio Avg. contributor physical protection of women index Size of military contingent in PKO GDP per capita in host country Index of Mass Rape in the Previous War Avg. contributor Polity score Avg. contributor GDP per capita -36.18** -24.34*** -4.205-10.54-12.03-17.87 (14.34) (8.898) (13.56) (19.80) (12.12) (17.47) -0.219 0.0963 (0.155) (0.113) -0.0941*** -0.0931** (0.0330) (0.0466) 3.419*** 0.414 (1.236) (2.198) 0.000244*** 0.000198*** 0.000165*** 0.000193*** 0.000171*** 0.000190** (3.12e-05) (3.03e-05) (3.45e-05) (6.54e-05) (4.60e-05) (6.34e-05) -0.000398-0.000441-0.000167-0.000292-0.000361-0.000249 (0.000333) (0.000334) (0.000275) (0.000589) (0.000484) (0.000570) -0.862** -0.866** -0.783** -0.799-0.740-0.753 (0.398) (0.348) (0.347) (0.699) (0.500) (0.591) 0.335** 0.157 0.525*** (0.152) (0.0980) (0.142) -8.18e-05-2.59e-05-3.99e-05 (5.19e-05) (3.33e-05) (8.78e-05) Constant 21.36 5.832*** -13.27** -7.399 6.334*** 0.347 (14.31) (1.817) (5.496) (10.38) (2.194) (8.617) ln(alpha) -1.121-1.485-1.444** -0.240-1.135-0.183 (1.221) (0.964) (0.570) (1.191) (1.617) (1.122) Observations 80 80 80 80 80 80 *** p<.01; ** p<.05; * p<.1 Standard errors in parentheses 6
Original Table 4: SEA accusations in police contingents, negative binomial regression Variables Model 10 Model 11 Model 12 Female ratio balance in PKO mission -2.707** -1.211-2.818** (1.136) (1.195) (1.249) Avg. contributor primary school gender 0.0205 ratio (0.0596) Avg. contributor labor force gender ratio -0.115*** (0.0410) Avg. contributor physical protection of women index -0.546 (0.577) Size of police contingent in PKO 0.000435** 0.000430*** 0.000484*** (0.000212) (0.000139) (0.000172) GDP per capita in host country -0.00158*** -0.00187*** -0.00166*** (0.000484) (0.000398) (0.000507) Index of Mass Rape in the Previous War -0.771* -0.266-0.790** (0.446) (0.424) (0.360) Constant 0.498 6.941*** 4.410* (5.570) (1.936) (2.459) ln(alpha) -1.692-4.375-1.718 (3.165) (15.00) (3.111) Observations 64 64 64 *** p<.01; ** p<.05; * p<.1 Standard errors in parentheses 7
Modified Table 4: SEA accusations in police contingents, negative binomial regression Variables Model 10 Model 11 Model 12 Female ratio balance in PKO mission -3.164** -1.238-2.806** (1.267) (1.173) (1.246) Avg. contributor primary school gender 0.0538 ratio (0.0644) Avg. contributor labor force gender ratio -0.112*** (0.0398) Avg. contributor physical protection of women index -0.536 (0.590) Size of police contingent in PKO 0.000422* 0.000431*** 0.000483*** (0.000227) (0.000139) (0.000173) GDP per capita in host country -0.00153*** -0.00185*** -0.00165*** (0.000457) (0.000399) (0.000507) Index of Mass Rape in the Previous War -0.801* -0.269-0.786** (0.441) (0.423) (0.363) Constant -2.652 6.801*** 4.365* (6.145) (1.911) (2.497) ln(alpha) -1.596-4.644-1.732 (3.002) (19.80) (3.160) Observations 64 64 64 *** p<.01; ** p<.05; * p<.1 Standard errors in parentheses 8
Original Figure 2 Expected count of SEA for military contingents 0 5 10 15 0.02.04.06.08.1 Proportion of women in the PKO military contingents Note: p=0.039 in a two-tailed test. Modified Figure 2 Expected Count of SEA for Military Contingents 0 5 10 15 0.02.04.06.08.1 Proportion of Women in the PKO Military Contingents Note: p=0.045 in a two-tailed test. 9
Original Figure 3 Expected count of SEA for police contingents 0.5 1 1.5 0.05.1.15.2.25 Proportion of women in the PKO police contingents Note: p=0.017 in a two-tailed test. Modified Figure 3 Expected Count of SEA for Police Contingents 0.5 1 1.5 2 0.05.1.15.2.25 Proportion of Women in the PKO Police Contingents Note: p=0.012 in a two-tailed test. 10
Original Figure 4 Expected count of SEA for military contingents 0 2 4 6 8 10 70 80 90 100 Weighted average of the contributing countries' primary school female ratio Note: p=0.019 in a two-tailed test. Modified Figure 4 Expected Count of SEA for Military Contingents 0 200 400 600 800 1000 70 80 90 100 Weighted Average of the Contributing Countries' Primary School Female Ratio Note: p=0.877 in a two-tailed test. 11
Original Figure 5 Expected count of SEA for military contingents 0 2 4 6 8 10 40 50 60 70 80 Weighted average of the contributing countries' female labor participation Note: p=0.023 in a two-tailed test. Modified Figure 5 Expected Count of SEA for Military Contingents 0 2 4 6 8 10 40 50 60 70 80 Weighted Average of the Contributing Countries' Female Labor Participation Note: p=0.023 in a two-tailed test. 12
Original Figure 6 Expected count of SEA for military contingents 0 5 10 15 1 2 3 4 Weighted average of the contributing countries' security index for women Note: p=0.167 in a two-tailed test. Modified Figure 6 Expected Count of SEA for Military Contingents 0 5 10 15 1 2 3 4 Weighted Average of the Contributing Countries' Security Index for Women Note: p=0.255 in a two-tailed test. 13
Original Figure 7 Expected count of SEA for police contingents 0.5 1 1.5 2 40 50 60 70 80 Weighted average of the police contributing countries' female labor participation Note: p=0.003 in a two-tailed test. Modified Figure 7 Expected Count of SEA for Police Contingents 0.5 1 1.5 2 40 50 60 70 80 Weighted Average of the Police Contributing Countries' Female Labor Participation Note: p=0.005 in a two-tailed test. 14
Additional Evidence from Equal Opportunity Peacekeeping Since the submission for publication of this article in the Journal of Peace Research, we have conducted additional analyses for a forthcoming book entitled Equal Opportunity Peacekeeping, to be published in early 2017 by Oxford University Press. In chapter 5 of the book, we conduct similar analyses that use the labor-force-participation and physicalprotection-of-women indicators of the practice of gender equality in contributing countries. We chose not to use the primary-school ratio even before catching this data mistake in part because of the missing data problem and in part because primary school gender ratios are more removed from the gender norms surrounding peacekeeping personnel than the other indicators. In addition to examining labor force participation and the physical protection of women, we added a measure of the practice of gender inequality in contributing countries for the book analyses: the weighted average of whether the contributing countries have adopted National Action Plans (NAPs) as part of the implementation of UNSC Resolution 1325. We argue that, although NAPs have struggled to bring about drastic societal reforms, the sorting of countries according to whether or not they have undertaken the effort to draft a NAP can provide an additional perspective into which countries are amenable to gender reforms to the security and political sectors. In addition to adding the NAP measure, the models that we present in the book incorporate three additional modifications. First, we have used Israel s GDP figures for the UNDOF mission the GDP entries pertaining to UNDOF were listed as missing in the original publication. Second, we have replaced Cohen s indicator of mass rape with an indicator of widespread sexual violence from the SVAC dataset that is a successor of the earlier Cohen data. Third, we have used a random effects negative binomial estimator instead of a basic negative binomial estimator with clustered standard errors. Since each mission is likely to have heterogenous baseline levels of accusations of SEA and potentially different levels of underreporting of SEA, a random effects setup is desirable with these data and can more directly strip out a major source of autocorrelation in the data. Our findings confirm that in the updated models women s participation in the labor force, the physical protection of women, and the adoption of 1325 NAPs in contributing countries are associated with reduced levels of allegations of SEA. The following is a table of coefficients from the random effects negative binomial regressions that we employ in Chapter 5 of Equal Opportunity Peacekeeping. We also show the substantive effects in figures similar to those in the article. Note that the table and figures use SEAHV instead of SEA in the book, we emphasize that sexual harassment and sexual violence are part of what counts as SEA, and so we refer to the SEA allegations more precisely as allegations of sexual exploitation, abuse, harassment and violence (SEAHV). SEAHV accusations in military contingents, random effects negative binomial regression 1 2 3 Proportion of women in PKO mission -11.52-16.64-6.176 (18.17) (17.08) (16.03) 15
Avg. contributor women labor force partic. -0.141*** (0.0541) Avg. contributor physical protection of women index 1.513** (0.750) Avg. contributor NAP adoption -3.746** (1.546) Size of military contingent in PKO 0.000200* 6.71e-05 0.000145*** (0.000110) (6.35e-05) (5.13e-05) GDP per capita in host country -8.94e-06 2.31e-05-1.32e-05 (7.43e-05) (7.70e-05) (7.25e-05) Widespread sexual violence 0.275-0.0153-0.164 (0.842) (0.875) (0.831) Constant 6.994** -4.152 0.873 (2.745) (2.655) (1.058) ln(r) 1.483 0.851 1.169* (0.927) (0.564) (0.637) ln(s) -0.0546-0.249-0.0829 (0.739) (0.632) (0.640) Observations 85 85 85 *** p<.01; ** p<.05; * p<.1 Standard errors in parentheses Expected Count of SEAHV for Military Contingents 0 2 4 6 8 40 50 60 70 80 Weighted Average of the Contributing Countries' Female Labor Participation Note: p=0.009 in a two-tailed test. 16
Expected Count of SEAHV for Military Contingents 0 2 4 6 8 1 2 3 4 Weighted Average of the Contributing Countries' Security Index for Women Note: p=0.044 in a two-tailed test. Expected Count of SEAHV for Military Contingents 0 2 4 6 0.2.4.6.8 1 Proportion of Military Personnel from NAP Countries Note: p=0.015 in a two-tailed test. 17