APPENDIX: HUMAN RIGHTS AND PUBLIC SUPPORT FOR WAR

APPENDIX: HUMAN RIGHTS AND PUBLIC SUPPORT FOR WAR I. SUPPLEMENT TO HOW HUMAN RIGHTS COULD AFFECT SUPPORT FOR WAR... 3 FIGURE S1: RELATIONSHIP BETWEEN HUMAN RIGHTS (CIRI) AND DEMOCRACY... 3 FIGURE S2: RELATIONSHIP BETWEEN HUMAN RIGHTS (PTS) AND DEMOCRACY... 4 II. SUPPLEMENT TO RESEARCH DESIGN... 5 A. SURVEY EXPERIMENT... 5 B. MILITARISM... 9 C. INTERNATIONALISM... 10 D. CONSERVATISM... 10 E. ETHNOCENTRISM... 12 F. RELIGIOSITY... 12 G. DEMOGRAPHICS... 13 III. SUPPLEMENT TO EFFECT OF HUMAN RIGHTS ON SUPPORT FOR WAR... 15 TABLE S1: SUMMARY STATISTICS FOR OUTCOME, TREATMENTS, AND CONTROLS... 15 FIGURE S3: BALANCE BETWEEN TREATMENT AND CONTROL GROUPS... 16 TABLE S2: PROBIT ESTIMATES OF SUPPORT FOR WAR... 17 FIGURE S4: EFFECT OF HUMAN RIGHTS ON SUPPORT FOR WAR (PROBIT)... 18 TABLE S3: REGRESSION ESTIMATES OF SUPPORT FOR WAR (5-POINT SCALE)... 19 FIGURE S5: EFFECT OF HUMAN RIGHTS ON SUPPORT FOR WAR (5-POINT SCALE)... 20 IV. SUPPLEMENT TO EFFECTS OF HUMAN RIGHTS ON THE MEDIATORS... 20 TABLE S4: SUMMARY STATISTICS FOR MEDIATORS... 20 TABLE S5: REGRESSION ESTIMATES OF THE MEDIATORS... 21 FIGURE S6: EFFECT OF HUMAN RIGHTS ON FOUR MEDIATORS... 22 V. SUPPLEMENT TO EFFECTS OF THE MEDIATORS ON SUPPORT FOR WAR... 22 TABLE S6: EFFECT OF MEDIATORS ON SUPPORT FOR WAR... 23 VI. SUPPLEMENT TO ESTIMATED CAUSAL PATHWAYS... 24 FIGURE S7: PRODUCT OF COEFFICIENTS VS POTENTIAL OUTCOMES (LINEAR MODEL)... 26 FIGURE S8: PRODUCT OF COEFFICIENTS VS POTENTIAL OUTCOMES (PROBIT MODEL)... 26 FIGURE S9: CAUSAL MECHANISMS AS A PERCENTAGE OF... 27 TOTAL EFFECT OF DEMOCRACY... 27 VII. SUPPLEMENT TO POLITICAL CONSEQUENCES... 27 VIII. SUPPLEMENT TO PUBLIC BELIEFS ABOUT HUMAN RIGHTS... 28 FIGURE S10: PERCEPTIONS VS LATENT HUMAN RIGHTS PROTECTION SCORES, BY PARTY ID... 29 FIGURE S11: PERCEPTIONS VS LATENT HUMAN RIGHTS PROTECTION SCORES, BY POLITICAL INTEREST... 30 IX. SUPPLEMENT TO REAL AND HYPOTHETICAL SCENARIOS... 30 TABLE S7: REGRESSION ESTIMATES OF SUPPORT FOR WAR WHEN THE TARGET WAS IN AFRICA... 31 FIGURE S12: EFFECT OF HUMAN RIGHTS WHEN THE TARGET WAS IN AFRICA... 32 X. SUPPLEMENT TO EFFECTS IN OTHER COUNTRIES... 32 TABLE S8: REGRESSION ESTIMATES OF BRITISH SUPPORT FOR WAR... 33 FIGURE S13: EFFECTS OF HUMAN RIGHTS ON FOUR MEDIATORS IN THE U.K.... 34 FIGURE S14: EFFECTS OF FOUR MEDIATORS ON BRITISH SUPPORT FOR WAR... 34 1

XI. SUPPLEMENT TO EFFECTS IN OTHER CONTEXTS... 35 TABLE S9: REGRESSION ESTIMATES OF SUPPORT FOR WAR IN THE BALLISTIC MISSILE SCENARIO... 36 FIGURE S15: EFFECTS OF HUMAN RIGHTS, BY GEOGRAPHIC REGION... 37 FIGURE S16: EFFECT OF HUMAN RIGHTS ON FOUR MEDIATORS... 37 FIGURE S17: EFFECTS OF FOUR MEDIATORS ON SUPPORT FOR WAR... 38 FIGURE S18: CAUSAL MECHANISMS AS PERCENTAGE OF TOTAL EFFECT... 38 2

I. SUPPLEMENT TO HOW HUMAN RIGHTS COULD AFFECT SUPPORT FOR WAR Democracy and human rights are distinct not only logically but also empirically. In any given year, one can find many examples of democracies that violate human rights, and of autocracies with relatively good human rights records. Figure S1 shows the observed relationship between human rights and democracy for nearly all countries in the world from 1981 through 2010. Both graphs use the CIRI Physical Integrity Rights as an index of human rights, with higher values indicating greater respect for rights. The top graph uses the binary measure of democracy by Geddes, Wright, and Frantz (GWF); the bottom graph uses Polity scores, with higher values indicating more democratic countries. Within each graph, the area of each circle is proportional to the number of cases exhibiting that combination of human rights and democracy, and the dashed lines represent the midpoints of the axes. Figure S1: Relationship between Human Rights (CIRI) and Democracy 8 Physical Integrity Rights 6 4 2 0 Autocracy Democracy Democracy, as measured by GWF 8 Physical Integrity Rights 6 4 2 0-10 to -8-7 to -5-4 to -2-1 to 1 2 to 4 5 to 7 8 to 10 Democracy, as measured by Polity Note: Sample size was 4,029 when democracy was measured by GWF and 4,246 when democracy was measured by Polity. 3

A large percentage of countries appear in the southeast quadrants, meaning that they are democratic but score poorly on human rights. Countries that were in this category many times during the sample period include Bangladesh, Brazil, Columbia, India, Indonesia, Israel, Mexico, Nepal, Nigeria, Pakistan, Peru, the Philippines, Sri Lanka, Turkey, Venezuela, and many others. Likewise, many countries occupy the northwest quadrant, meaning they are autocratic but tend to respect physical integrity rights. Examples of countries that were in this category for most of the sample period include Burkina Faso, Gabon, Gambia, Jordan, Kuwait, Mauritania, Oman, Singapore, Swaziland, Taiwan, Tanzania, and the United Arab Emirates, among others. Thus, democracy and human rights are not only conceptually distinct, but they can and often do diverge in practice. Figure S1 presented the relationship between human rights and democracy when human rights were measured using the CIRI database. Figure S2, below, presents complementary analyses based on the Political Terror Scale, which we reverse-coded to ensure that higher values would indicate greater respect for human rights. Figure S2: Relationship between Human Rights (PTS) and Democracy 4

II. SUPPLEMENT TO RESEARCH DESIGN In this section we present the text of the survey experiment that YouGov administered to U.S. respondents in 2012. The other experiments were nearly identical, with minor wording changes as described in the paper. A. SURVEY EXPERIMENT We are going to describe a situation the United States could face in the future. The situation is general, and is not about a specific country in the news today. Some parts of the description may seem important to you; other parts may seem unimportant. Please read the details very carefully. Do you agree to read the details very carefully, and then give your most thoughtful answers? m Yes m No new page Programming note: independently draw the following two random variables, and plug the values into subsequent pages of the survey. Human rights: The country does not violate human rights; it does not imprison or torture its citizens because of their beliefs. OR The country violates human rights; it imprisons or tortures some of its citizens because of their beliefs. Regime: The country is a democracy. The president, the legislature, and local councils are elected by the people. OR The country is not a democracy. The people do not have the power to choose the leader. 5

Here is the situation: A country is developing nuclear weapons and will have its first nuclear bomb within six months. The country could then use its missiles to launch nuclear attacks against any country in the world. [regime] The country has not signed a military alliance with the United States and does not have high levels of trade with the United States. [human rights] The country's nonnuclear military forces are half as strong as U.S. nonnuclear forces. The country's motives remain unclear, but if it builds nuclear weapons, it will have the power to blackmail or destroy other countries. The country has refused all requests to stop its nuclear weapons program, When you have finished reading the situation carefully, please check this box: q I have read the situation carefully. new page Just to review, the country... [regime] has not signed a military alliance with the U.S., and does not have high levels of trade with the U.S. [human rights] has nonnuclear forces half as strong as U.S. forces. has unknown motives, but if it builds nuclear weapons, it will have the power to blackmail or destroy other countries. has refused to stop its nuclear weapons program. By attacking the country's nuclear development sites now, the United States could prevent the country from making any nuclear weapons. Do you favor or oppose using the U.S. military to attack the country's nuclear development sites? m Favor strongly m Favor somewhat m Neither favor nor oppose m Oppose somewhat m Oppose strongly new page 6

Here is the situation again, for your reference: [regime] has not signed a military alliance with the U.S., and does not have high levels of trade with the U.S. [human rights] has nonnuclear forces half as strong as U.S. forces. has unknown motives, but if it builds nuclear weapons, it will have the power to blackmail or destroy other countries. has refused to stop its nuclear weapons program. If the U.S. does not attack the country s nuclear development sites, what are the chances that each of the following things will happen? new page 7

Again, for your reference: [regime] has not signed a military alliance with the U.S., and does not have high levels of trade with the U.S. [human rights] has nonnuclear forces half as strong as U.S. forces. has unknown motives, but if it builds nuclear weapons, it will have the power to blackmail or destroy other countries. has refused to stop its nuclear weapons program. If the U.S. does attack the country s nuclear development sites, what are the chances that each of the following things will happen? new page 8

Here is the situation one last time. The country [regime] has not signed a military alliance with the U.S., and does not have high levels of trade with the U.S. [human rights] has nonnuclear forces half as strong as U.S. forces. has unknown motives, but if it builds nuclear weapons, it will have the power to blackmail or destroy other countries. has refused to stop its nuclear weapons program. Do you think the U.S. has a moral obligation to attack the country s nuclear development sites? m The U.S. has a strong moral obligation to attack. m The U.S. has a weak moral obligation to attack. m The U.S. has no moral obligation to attack. Do you think it would be morally wrong for the U.S. military to attack the country s nuclear development sites? m It would be morally wrong for the U.S. to attack. m It would not be morally wrong for the U.S. to attack. B. MILITARISM To measure militarism, we coded responses to the following five questions such that larger values indicated greater support for militaristic policies; standardized each item to have a mean of zero and a variance of one; and took the average of the five standardized items. Please tell us how much you agree or disagree with this statement: The use of military force only makes problems worse. m Agree strongly m Agree somewhat m Neither agree nor disagree m Disagree somewhat m Disagree strongly Some people think that U.S. military force should never be used under any circumstances. They are at 1 on the scale below. Other people think there are many situations in which U.S. military force should be used to deal with problems. They are at 7 on the scale below. And, of course, other people have opinions in between. Where would you put yourself on this scale? 9

All things considered do you think it was a mistake to invade Iraq? m Yes m No m Not sure All things considered do you think it was a mistake to invade Afghanistan? m Yes m No m Not sure The federal budget deficit is approximately $1 trillion this year. If the Congress were to balance the budget it would have to consider cutting defense spending, cutting domestic spending (such as Medicare and Social Security), or raising taxes to cover the deficit. What would you most prefer that Congress do - cut domestic spending, cut defense spending, or raise taxes? m Cut Defense spending m Cut Domestic spending m Raise Taxes C. INTERNATIONALISM To measure internationalism, we coded responses to the following item such that higher values indicated higher support for a U.S. role abroad; and standardized the variable to have a mean of 0 and a standard deviation of 1. Please tell us how much you agree or disagree with this statement: The United States needs to play an active role in solving conflicts around the world. m Agree strongly m Agree somewhat m Neither agree nor disagree m Disagree somewhat m Disagree strongly D. CONSERVATISM To measure conservatism, we coded responses to the following questions such that larger values indicated more conservative attitudes; standardized each item to have a mean of zero and a variance of one; and took the average of the standardized items. Generally speaking, do you think of yourself as a.? 10

m Democrat m Republican m Independent m Other (open textbox) m Not sure If Democrat: Would you call yourself a strong Democrat or a not very strong Democrat? m Strong Democrat m Not very strong Democrat If Republican: Would you call yourself a strong Republican or a not very strong Republican? m Strong Republican m Not very strong Republican If Independent or Other: Do you think of yourself as closer to the Democratic or the Republican Party? m The Democratic Party m The Republican Party m Neither m Not sure How would you rate each of the following individuals and groups? [yourself] m Very liberal m Liberal m Somewhat Liberal m Middle of the Road m Somewhat Conservative m Very Conservative m Not sure Thinking about politics these days, how would you describe your own political viewpoint? (Randomly reverse answer options) m Very liberal m Liberal m Moderate m Conservative m Very Conservative m Not sure 11

With which party, if any, are you registered? m No Party, Independent, Declined to State m Democratic Party m Republican Party m Other Do you approve of the way each is doing their job [President Obama] m Strongly Approve m Somewhat Approve m Somewhat Disapprove m Strongly Disapprove m Not Sure E. ETHNOCENTRISM To measure ethnocentrism, we coded responses to the following items, such that larger values indicated more ethnocentric beliefs; standardized each item to have a mean of zero and a variance of one; and took the average of the standardized items. What do you think the U.S. government should do about immigration? Select all that apply. m Grant legal status to all illegal immigrants who have held jobs and paid taxes for at least 3 years, and not been convicted of any felony crimes. m Increase the number of border patrols on the US-Mexican border. m Allow police to question anyone they think may be in the country illegally. Affirmative action programs give preference to racial minorities in employment and college admissions to in order correct for past discrimination. Do you support or oppose affirmative action? m Strongly support m Somewhat support m Somewhat oppose m Strongly oppose Do you favor or oppose allowing gays and lesbians to marry legally? m Favor m Oppose F. RELIGIOSITY To measure religiosity, we coded the following four items such that higher values reflected greater religious observance; standardized each item to have a mean of zero and a variance of one; and took the average of the standardized items. 12

How important is religion in your life? m Very important m Somewhat important m Not too important m Not at all important Aside from weddings and funerals, how often do you attend religious services? m More than once a week m Once a week m Once or twice a month m A few times a year m Seldom m Never m Don t know People practice their religion in different ways. Outside of attending religious services, how often do you pray? m Several times a day m Once a day m A few times a week m Once a week m A few times a month m Seldom m Never m Don t know Would you describe yourself as a born-again or evangelical Christian, or not? m Yes m No G. DEMOGRAPHICS We used the items below to construct demographic variables: male (1 = male, 0 = female); white (1 = white, 0 = nonwhite), age/100 (equivalent to age in centuries), and education (0 = high school or less, 1/3 = some college, 2/3 = college degree, 1 = postgraduate). Are you male or female? m Male m Female What racial or ethnic group best describes you? m White m Black or African-American m Hispanic of Latino m Asian or Asian-American 13

m Native American m Middle Eastern m Mixed Race m Other (open textbox) In what year were you born? [Open integer textbox] What is the highest level of education you have completed? m Did not graduate from high school m High school graduate m Some college, but no degree (yet) m 2-year college degree m 4-year college degree m Postgraduate degree (MA, MBA, MD, JD, PhD, etc.) 14

III. SUPPLEMENT TO EFFECT OF HUMAN RIGHTS ON SUPPORT FOR WAR To estimate the effect of human rights on support for war, we created a dependent variable that was coded 100 if the respondent thought the U.S. military should attack, and was coded 0 otherwise. Table S1 provides summary statistics for the dependent variable, the randomized treatments, and attitudinal and demographic control variables. Table S1: Summary Statistics for Outcome, Treatments, and Controls Variable Mean Std Dev Min Max Outcome Support for strike 49.16 50.01 0.00 100.00 Treatments Human rights (HR) 0.48 0.50 0.00 1.00 Democracy 0.49 0.50 0.00 1.00 HR Democracy 0.24 0.42 0.00 1.00 Controls Militarism 0.00 0.71-1.53 1.24 Internationalism 0.01 1.00-1.58 1.76 Conservatism 0.00 0.86-1.48 1.41 Ethnocentrism 0.00 0.72-1.15 1.04 Religiosity 0.00 0.83-1.38 1.37 Male 0.47 0.50 0.00 1.00 White 0.74 0.44 0.00 1.00 Age/100 0.52 0.16 0.18 0.94 Education 0.42 0.33 0.00 1.00 15

Figure S3 shows that the control group (scenarios in which the country violates human rights) and the treatment group (scenarios in which the country respects human rights) were balanced with respect to the other randomized treatment, democracy; and with respect to attitudinal and demographic variables that might affect support for war. Figure S3: Balance between Treatment and Control Groups Democracy Attitudes Militarism Internationalism Conservatism Ethnocentrism Religiosity Demographics Male White Age/100 Education -.2 0.2.4.6.8 1 Violates HR Respects HR In the paper, we used linear regression models to estimate the effect of human rights on support for war. If we had used probit models instead of linear regressions, our conclusions would have remained the same. Table S2 gives the parameter estimates from three probit regressions, and Figure S4 gives the average treatment effects implied by column 3, which contains the most complete set of explanatory variables. 16

Table S2: Probit Estimates of Support for War Variable Model (1) (2) (3) Treatments Human rights -0.43 ** -0.50 ** -0.43 ** (0.07) (0.07) (0.10) Democracy -0.43 ** -0.36 ** (0.07) (0.10) Human rights -0.14 Democracy (0.14) Controls Militarism 0.54 ** 0.54 ** (0.07) (0.07) Internationalism 0.11 ** 0.11 ** (0.04) (0.04) Conservatism -0.04-0.04 (0.07) (0.07) Ethnocentrism 0.34 ** 0.34 ** (0.08) (0.08) Religiosity 0.04 0.04 (0.05) (0.05) Male 0.06 0.06 (0.08) (0.08) White -0.21 * -0.21 * (0.09) (0.09) Age/100-0.25-0.25 (0.24) (0.24) Education -0.36 ** -0.36 ** (0.12) (0.12) Intercept 0.18 ** 0.83 ** 0.80 ** (0.05) (0.15) (0.15) Note: The table gives estimated coefficients and robust standard errors from probit regressions, in which the dependent variable was coded 1 if the respondent supported a military strike and 0 otherwise. The sample size in each column was 1,430. Asterisks indicate the level of statistical significance. * p<0.05; ** p<0.01. 17

Figure S4: Effect of Human Rights on Support for War (Probit) Effect if not democratic -0.14 Effect if democratic Average effect -0.19-0.17-0.3-0.2-0.1 0.0 Effect on Probability of Supporting War The results remained the same when we measured support for a military strike as a scale with five values: 0 if strongly opposed (12% of respondents), 25 if somewhat opposed (21% of respondents), 50 if neither favored nor opposed (18% of respondents), 75 if somewhat favored (29% of respondents), and 100 if strongly favored (20% of respondents). The mean value of this dependent variable was 56, with a standard deviation of 33. Table S3 below shows three linear regression models based on this dependent variable. The models show that, on average, positive information about the human rights practices of a country reduced support for war by about 13 points on the 100-point scale. Figure S5 shows the average treatment effects, based on column 3 of Table S3. 18

Table S3: Regression Estimates of Support for War (5-Point Scale) Variable Model (1) (2) (3) Treatments Human rights -12.8 ** -12.6 ** -11.4 ** (1.7) (1.5) 2.0 Democracy -9.5 ** -8.4 ** (1.5) (2.0) Human rights -2.4 Democracy (3.0) Controls Militarism 14.1 ** 14.1 ** (1.4) (1.4) Internationalism 3.0 ** 3.0 ** (0.8) (0.8) Conservatism 0.1 0.0 (1.4) (1.4) Ethnocentrism 7.5 ** 7.5 ** (1.6) (1.6) Religiosity 2.0 2.0 (1.0) (1.0) Male 0.2 0.3 (1.6) (1.6) White -5.4 ** -5.4 ** (2.0) (2.0) Age/100-7.4-7.5 (4.9) (4.9) Education -13.0 ** -13.0 ** (2.4) (2.4) Intercept 62.2 ** 79.8 ** 79.3 ** (1.2) (2.9) (3.0) Note: The table gives estimated coefficients and robust standard errors from linear regressions in which the dependent variable was coded 0, 25, 50, 75, or 100, based on the respondent s level of support for a military strike. The sample size in each column was 1,430. Asterisks indicate the level of statistical significance. * p<0.05; ** p<0.01. 19

Figure S5: Effect of Human Rights on Support for War (5-Point Scale) Effect if not democratic -11 Effect if democratic -14 Average effect -13-20 -10 0 Effect on support for war Note: The dots show the effects of each randomized treatment; the horizontal lines are 95 percent confidence intervals. IV. SUPPLEMENT TO EFFECTS OF HUMAN RIGHTS ON THE MEDIATORS Table S4 gives descriptive statistics for the four mediators: perceptions of threat, morality, success, and cost. Table S4: Summary Statistics for Mediators Variable Mean Std Deviation Minimum Maximum Threat 56 29 0 100 Morality 55 40 0 100 Success 59 27 0 100 Cost 58 23 0 100 Note: The sample size for each mediator was 1,430. In the paper, we presented the average effect of human rights on each mediator (Figure 2). The focus on average effects was not only concise but also empirically justified. To calculate the effects, we regressed each mediator on the randomized treatments, while controlling for attitudinal and demographic variables. As Table S5 shows, below, the interaction between human rights and democracy was never statistically significant. 20

Table S5: Regression Estimates of the Mediators Variable Mediator Threat Morality Success Cost Treatments Human rights -7.8 ** -11.8 ** -0.5 0.5 (1.8) (2.4) (1.8) (1.6) Democracy -6.0 ** -8.1 ** -1.7 1.0 (1.7) (2.4) (1.8) (1.6) Human rights -3.5-2.4-2.5 1.9 Democracy (2.6) (3.6) (2.7) (2.3) Controls Militarism 10.7 ** 18.6 ** 10.8 ** -7.4 ** (1.2) (1.6) (1.3) (1.1) Internationalism 0.7 4.8 ** 1.7 * -1.5 * (0.7) (0.9) (0.7) (0.7) Conservatism 0.6-0.2-0.6-0.8 (1.3) (1.7) (1.3) (1.1) Ethnocentrism 8.0 ** 8.5 ** 4.4 ** -3.4 ** (1.5) (2.0) (1.5) (1.3) Religiosity 2.8 ** 3.9 ** -0.1 0.9 (0.9) (1.2) (0.9) (0.8) Male -8.7 ** 0.7 2.2-3.3 ** (1.4) (1.9) (1.4) (1.2) White -6.0 ** -5.0 * -1.4 2.0 (1.6) (2.4) (1.7) (1.5) Age/100 15.0 ** -8.7 15.8 ** -1.7 (4.1) (5.9) (4.3) (3.8) Education -14.4 ** -13.2 ** -0.8 0.8 (2.0) (2.8) (2.1) (1.8) Intercept 70.5 ** 78.3 ** 53.1 ** 57.1 ** (2.6) (3.5) (2.7) (2.4) Note: Estimated coefficients and robust standard errors from linear regressions. In each column, the dependent variable is a mediator, scaled from 0 to 100. The sample size in each column was 1,430. Asterisks indicate the level of statistical significance. * p<0.05; ** p<0.01. 21

Figure S6 provides a graphical interpretation of the regression models. The figure shows that human rights exerted a fairly consistent effect on each mediator, regardless of whether the target was a democracy or an autocracy. Figure S6: Effect of Human Rights on Four Mediators Effect on Threat If not democratic If democratic Average effect Effect on Morality If not democratic If democratic Average effect Effect on Success If not democratic If democratic Average effect Effect on Cost If not democratic If democratic Average effect -20-15 -10-5 0 5 Estimated Effect V. SUPPLEMENT TO EFFECTS OF THE MEDIATORS ON SUPPORT FOR WAR Figure 3 in the paper presented the effects of four mediators on U.S. support for a military strike. We obtained the estimates by regressing support for a military attack on all four mediators, while controlling for the randomized treatments and myriad attributes of the respondent. Table S6, below, presents the complete set of regression estimates. 22

Table S6: Effect of Mediators on Support for War Variable Coefficent Std. Error Mediators Threat 46.7 4.9 ** Morality 52.1 3.6 ** Success 15.5 4.0 ** Cost -19.5 4.6 ** Treatments Human rights (HR) -4.4 2.7 Democracy -4.4 2.8 HR Democracy -1.1 3.8 Controls Militarism 0.7 1.9 Internationalism 0.4 1.0 Conservatism -1.8 1.9 Ethnocentrism 2.1 2.2 Religiosity -2.1 1.3 Male 4.8 2.1 * White -0.8 2.6 Age -13.8 6.6 * Education 1.8 3.2 Intercept 5.6 6.3 Note: Estimated coefficients and robust standard errors from a linear regression, in which the dependent variable was 100 if the respondent supported a military strike and 0 otherwise. The sample size was 1,430. Asterisks indicate the level of statistical significance. * p<0.05; ** p<0.01. 23

VI. SUPPLEMENT TO ESTIMATED CAUSAL PATHWAYS We estimated the strength of each causal pathway by computing the product of regression coefficients (Baron and Kenny 1986). This method is easy to understand and useful for models with multiple mediators. Other methods have been gaining traction in political science, including work by Imai, Keele, Tingley, and Yamamoto (2011) that situates causal mediation within a potential-outcomes framework. To assess the robustness of our findings, we adapted the Imai et al. approach to our unique experimental design: a model with two treatment variables and four mediators. For each individual i, let H # be a treatment indicator that takes a value of 1 when i was asked about a country that respected human rights, and 0 when i was asked about a country that violated human rights. Similarly, let D # be a treatment indicator that is 1 when i was asked about a democracy, and 0 when i was asked about an autocracy. Use Y # (h, d) to denote i s support for a military strike under treatment conditions H # = h and D # = d. Our experiment had four combinations of treatment conditions, implying four potential outcomes for each individual: Y # (1,1), Y # (1,0), Y # (0,1), and Y # (0,0). We randomly presented each individual with one of these four scenarios, and therefore observed only one of the four potential outcomes for each individual. In addition to measuring the respondent s support for a military strike, we measured four mediators (threat, morality, success, and cost), which for convenience we will index as k = {1,2,3,4}. For each person in our panel, let M # 5 h, d represent the value of mediator k when the target had attributes h and d. With four experimental conditions, there were four potential values of each mediator: M # 5 (1,1), M # 5 (1,0), M # 5 (0,1), and M # 5 (0,0). For any given individual, though, we observed only one of these four values. For each individual i, the effect of human rights transmitted via mediator k (controlling for the level of democracy) is defined as ν # 5 (d) = Y # 1, d, M # 5 1, d, M # 75 1, d Y # 1, d, M # 5 0, d, M # 75 1, d. The first term on the right hand side is i s support for a military strike when the target respected human rights and had democracy value d, and all the mediators (both k and k, meaning not k ) were as if the country respected human rights and had democracy value d. The second term is identical, except that mediator k is as if the country violated human rights but remained at democracy level d. ν # 5 (d) is the difference between an observable quantity and a counterfactual one. The minuend, Y # 1, d, M # 5 1, d, M # 75 1, d, simplifies to Y # (1, d), the observable response to a scenario in which the adversary respected human rights and had democracy level d. The subtrahend, Y # 1, d, M # 5 0, d, M # 75 1, d, on the other hand, is hypothetical. It represents the preference i would have expressed if she read about a country that respected human rights and had democracy level d, but perceived mediator k as if that same country had violated human rights. Because the subtrahend is a counterfactual, ν # 5 (d) is not directly observable for any member of the sample. Fortunately, one can estimate the causal pathways with the following algorithm: (1) 24

1. Using all cases, estimate a regression model of support for a military strike. In this model, E(Y # ) = αt # + βm # + γx #, where the vector T # summarizes the treatments that i actually received (H #, D #, H # D # ) and α is a vector of associated coefficients; M # is a vector of mediators (M # A, M # B, M # C, M # D ) with coefficients β, and X # is a vector of control variables with coefficients γ. The estimates from the model are presented in Table S6, above. 2. Using all cases, regress each mediator on the randomized treatments and demographic control variables, i.e., E(M # 5 ) = δt # + ωx #. The parameter estimates from these models are presented in Table S5, above. 3. For each i, a. Use the mediator models to predict M # 5 (h, d) for each mediator k, for all four possible combinations of h and d, conditional on the respondent s true demographic attributes, X #. This step addresses the missing data problem by imputing the values we would have observed, if we could have administered all four treatment conditions to every person in the sample. b. Use the regression model in Table S6 to predict Y # 1, d, M # 5 1, d, M # 75 1, d, where M # 5 1, d and M # 75 1, d are imputed values of the mediators. Use the same regression model to predict Y # 1, d, M # 5 0, d, M # 75 1, d, where M # 5 0, d and M # 75 1, d are imputed values of the mediators. c. Compute ν # 5 d = Y # 1, d, M # 5 1, d, M # 75 1, d Y # 1, d, M # 5 0, d, M # 75 1, d for both levels of d, and then average over the two values of d to get ν # 5 = ν # 5 d = 0 + ν # 5 d = 1 /2. 4. Compute the sample-wide average, ν 5 = A J J 5 #KA ν #. When all the models are linear, the potential outcomes approach should produce the same conclusions as the simpler product-of-coefficients method. When some models are nonlinear, however, the estimates could diverge. To assess the robustness of our findings, we first apply the potential outcomes framework in a purely linear setting, and then apply it to a nonlinear setting that uses probit regression to analyze a dichotomous version of our dependent variable. As expected, the two approaches agreed when all models were linear (Figure S7), but differed slightly when we used a nonlinear probit regression (Figure S8). In the latter case, the estimated importance of the threat pathway increased from 26 to 33 percent of the total effect, but in general our substantive conclusions remained the same. 25

Figure S7: Product of Coefficients vs Potential Outcomes (Linear Model) Threat 26 26 Morality 41 41 Success Cost 2 1 2 2 Other 30 30 0 20 40 60 Percentage of Total Effect Product of Coefficients Potential Outcomes Figure S8: Product of Coefficients vs Potential Outcomes (Probit Model) Threat 26 33 Morality 41 40 Success 2 2 Cost 2 3 Other 23 30 0 20 40 60 Percentage of Total Effect Product of Coefficients Potential Outcomes 26

Finally, we estimated the mechanisms through which the other treatment, democracy, affected support for war. Recall that democracy reduced support by approximately 14 percentage points, on average. We estimate that 25% of this effect arose by changing perceptions of threat, and an additional 34% arose by changing perceptions of morality. In contrast, success and costs played only minor roles in the causal chain (see Figure S9, below). These findings accord with Tomz and Weeks (2013), who found that democracy pacifies the public primarily by changing perceptions of threat and morality, not by raising expectations of costs or failure (850). Figure S9: Causal Mechanisms as a Percentage of Total Effect of Democracy VII. SUPPLEMENT TO POLITICAL CONSEQUENCES The section of the paper titled Political Consequences presents the effects of human rights by the respondent s level of political engagement. We used the following items to measure political attentiveness, political activism, and voting records. Some people seem to follow what's going on in government and public affairs most of the time, whether there s an election going on or not. Others aren t that interested. Would you say you follow what's going on in government and public affairs...? m Most of the time m Some of the time m Only now and then m Hardly at all m Don t know During the past year did you (Check all that apply) m Attend local political meetings (such as school board or city council) m Put up a political sign (such as a lawn sign or bumper sticker) m Work for a candidate or campaign m Donate money to a candidate, campaign, or political organization m None of these 27

Have you ever run for elective office at any level of government (local, state or federal)? m Yes m No Did you vote in the 2008 General Election? m No m I usually vote, but did not in 2008 m I am not sure m Yes. I definitely voted. VIII. SUPPLEMENT TO PUBLIC BELIEFS ABOUT HUMAN RIGHTS We designed a survey to gauge Americans perceptions of the human rights records of 57 countries. Our list included countries with varying population sizes, geographic regions, and human rights records. Each subject was randomly assigned to one of 5 groups of 10 countries with approximately equal numbers of residents. We took this approach to ensure that subjects were not comparing the human rights records of countries with markedly different population sizes. After rating the 10 countries in their population group, we asked each subject to rate China and the U.S., plus one of the following five countries: Bangladesh, India, Mexico, Nigeria, or Russia. Thus, each subject gave their impressions about the human rights records of 13 countries. Selecting Countries for the Survey To construct the 5 groups of 10 similarly-sized countries that we randomly assigned to our subjects, we used the following procedure. 1) We built a dataset of all countries between 2008-2013 for which the Schnakenberg and Fariss (2014) Latent Human Rights Protection (LHRP) scores were continuously observed. We chose these dates to correspond with the wording of our survey, which asked about human rights in the last 10 years. 2013 was the most recent available year for the LHRP scores. 2) We discarded all countries with populations lower than 5 million people, according to 2015 CIA population estimates. We also set aside the 3 countries with populations of 300 million or more China, India, and the US and treated them separately. 3) We divided the remaining 118 countries into 5 population quintiles. 4) Within each of the 5 population groups, we sorted countries according to the average of their LHRP scores from 2006-2013. We then selected 10 countries from each group: the three countries with the best average human rights scores, the three countries with the worst average human rights scores, and the three countries whose human rights scores were closest to the median for the sample as a whole. Finally, we selected a tenth country in each group to improve geographic representativeness, adding one country to each group that had just missed the cutoff for worst, median, or best human rights. Finally, to ensure coverage of large and important countries that were not already 28

included in one of the five groups, we asked all respondents to rate China and the US, and randomly assigned each respondent to rate Bangladesh, India, Mexico, Nigeria, or Russia. Survey Procedures We recruited a representative sample of 2,051 Americans in April 2017 using Survey Sampling International. The survey began by telling subjects, Some countries almost never kill, torture, or imprison their own citizens because of their beliefs. Other countries very often kill, torture, or imprison their own citizens because of their beliefs. And, of course, there are countries in between. On the next few pages, we would like your opinions about several countries We then continued: We will ask how often in your opinion the governments of those countries kill, torture, or imprison their own citizens because of their beliefs For each country being rated, we displayed a map of the region. The survey read: The map below shows the country of [Country], which is located in [Region]. Over the past 10 years, how often do you think the government of [Country] has killed, tortured, or imprisoned its own citizens because of their beliefs? Respondents chose among four options: Almost Never, Sometimes, Often, and Very Often. The results are shown in Figure 5 of the paper. Additional Analyses We also checked the accuracy of perceptions across the political spectrum. Figure S10 displays the relationship between perceptions and LHRP scores for Republicans, Democrats, and Independents. The correspondence is high regardless of political affiliation. Figure S10: Perceptions vs Latent Human Rights Protection Scores, by Party ID 4 Democrats Corr = 0.81 4 Independents Corr = 0.80 4 Republicans Corr = 0.77 Perceptions of HR 3 2 Perceptions of HR 3 2 Perceptions of HR 3 2 1 1 1-4 -3-2 -1 0 1 2 3 4 Latent HR Score -4-3 -2-1 0 1 2 3 4 Latent HR Score -4-3 -2-1 0 1 2 3 4 Latent HR Score 29

We also investigated the accuracy of perceptions by level of political interest. Figure S11 shows that the correlation between perceptions and expert scores is strong not only among people with the highest interest in politics, but also among people less interested in politics. Figure S11: Perceptions vs Latent Human Rights Protection Scores, by Political Interest 4 Low Interest Corr = 0.76 4 Medium Interest Corr = 0.80 4 High Interest Corr = 0.82 Perceptions of HR 3 2 Perceptions of HR 3 2 Perceptions of HR 3 2 1 1 1-4 -3-2 -1 0 1 2 3 4 Latent HR Score -4-3 -2-1 0 1 2 3 4 Latent HR Score -4-3 -2-1 0 1 2 3 4 Latent HR Score IX. SUPPLEMENT TO REAL AND HYPOTHETICAL SCENARIOS In order to ensure that subjects did not infer that our vignette was about specific countries such as Iran or North Korea, we replicated our experiment on MTurk but told subjects that the country pursuing nuclear weapons was located in Africa. Table S7 shows the results of three regressions estimating the effect of our main treatment, human rights, on support for a military strike. The regressions show that respect for human rights reduced public support for war by around 24-25 percentage points, on average. Moreover, as in our other studies the impact of human rights did vary significantly based on whether the country was democratic or not. 30

Table S7: Regression Estimates of Support for War when the Target was in Africa Variable Model (1) (2) (3) Treatments Human rights -25.2 ** -24.1 ** -24.9 ** (3.5) (3.3) (4.5) Democracy -9.2 ** -9.9 * (3.3) (4.8) Human rights 1.5 Democracy (6.6) Controls Militarism 14.3 ** 14.3 ** (2.1) (2.1) Internationalism 4.0 * 4.0 * (1.8) (1.8) Conservatism -5.3 * -5.3 * (2.5) (2.5) Ethnocentrism 9.9 ** 10.0 ** (3.3) (3.3) Religiosity 6.0 ** 6.0 ** (2.3) (2.3) Male -0.6-0.5 (3.4) (3.4) White 0.8 0.9 (4.3) (4.3) Age/100-30.2 * -30.3 * (14.5) (14.6) Education -11.7 * -11.8 * (5.8) (5.8) Intercept 59.5 ** 78.3 ** 79.6 ** (2.5) (6.9) (7.0) Note: Estimated coefficients and robust standard errors from linear regressions, in which the dependent variable was 100 if the respondent supported a military strike and 0 otherwise. The sample size in each column was 736. Asterisks indicate the level of statistical significance. * p<0.05; ** p<0.01. 31

We used the estimates from Column 3 to produce Figure S12, which summarizes the conditional and unconditional effects of human rights when the target was in Africa. Figure S12: Effect of Human Rights when the Target was in Africa Effect if not democratic Effect if democratic Average effect -25-23 -24-40 -30-20 -10 0 Effect on support for war (%) X. SUPPLEMENT TO EFFECTS IN OTHER COUNTRIES We also assessed whether our findings generalize to other countries. We fielded our experiment in the UK in March 2014 using Survey Sampling International. Table S8 shows the results of three regressions estimating the effect of our main treatment, human rights, on British support for a military strike. We used the estimates in Column 3 to calculate the conditional and average effects of human rights displayed in Figure 6 of the paper. 32

Table S8: Regression Estimates of British Support for War Variable Model (1) (2) (3) Treatments Human rights -13.0 ** -12.2 ** -10.9 ** (2.5) (2.4) (3.4) Democracy -10.2 ** -8.9 * (2.4) (3.5) Human rights -2.6 Democracy (4.8) Controls Militarism 11.0 ** 11.0 ** (1.5) (1.5) Internationalism 5.6 ** 5.5 ** (1.3) (1.3) Conservatism 1.1 1.2 (1.3) (1.3) Ethnocentrism 6.2 ** 6.1 ** (1.5) (1.5) Religiosity 0.8 0.8 (1.5) (1.5) Male -2.8-2.8 (2.4) (2.4) White 5.6 5.6 (4.3) (4.3) Age/100-47.1 ** -47.1 ** (8.3) (8.3) Education -10.2 ** -10.2 ** (3.7) (3.7) Intercept 45.0 ** 71.8 ** 71.2 ** (1.9) (5.5) (5.7) Note: Estimated coefficients and robust standard errors from linear regressions, in which the dependent variable was 100 if the respondent supported a military strike and 0 otherwise. The sample size in each column was 1,450. Asterisks indicate the level of statistical significance. * p<0.05; ** p<0.01. 33

Next, we investigated the mechanisms through which these treatments arose. Using the same procedures we employed for the United States, we decomposed the effects of human rights into four mechanisms: threat, morality, success, and cost. Figure S13 displays the average effect of respect for human rights on perceptions of threat, morality, success, and cost, and Figure S14 displays the effects of the same four mediators on British support for war. Figure S13: Effects of Human Rights on Four Mediators in the U.K. Effect on Threat Effect on Morality Effect on Success Effect on Cost -12-6 -4 3-20 -15-10 -5 0 5 Estimated Effect Figure S14: Effects of Four Mediators on British Support for War Threat 52 Morality 45 Success Cost -14 17-40 -20 0 20 40 60 Effect on Support for War (%) By joining these two parts of the causal chain, we produced Figure 7 of the paper, which presents the importance of each causal mechanism in the UK as a percentage of the total effect. The estimates in Figure 7 were based on the product-of-coefficients method, but we obtained the same conclusions when we estimated the mechanisms within a potential outcomes framework using probit models. 34

XI. SUPPLEMENT TO EFFECTS IN OTHER CONTEXTS To test the effects of human rights in other contexts, we developed a new experiment that differed from previous ones in four ways. First, in place of nuclear weapons, we described a country that was developing conventional ballistic missiles that could strike other counties in its neighborhood. Second, we randomly located the country in a specific geographic region. Third, we lengthened the time to weapons acquisition from six months to one year. Finally, we portrayed the country as much weaker than the United States. The new scenario said: A country in [region] is developing ballistic missiles, which will become operational within a year. The missiles, which will carry conventional explosives, could be used to attack other countries in [region], where region was Africa, Asia, Eastern Europe, or South America. We also explained that The country s military is much weaker than the U.S. military. As in our earlier studies, we independently randomized human rights and democracy, and provided non-randomized information about military alliances and trade. We administered the experiment to 1,211 U.S. adults, who were recruited by Lucid in April 2018. We then analyzed the data using the same procedures we had employed for earlier experiments. Table S9 shows the results of three regressions estimating the effect of our main treatment, human rights, on support for war in the ballistic missile scenario. We used the estimates in Column 3 to calculate the conditional and average effects of human rights displayed in Figure 8 of the paper. 35

Table S9: Regression Estimates of Support for War in the Ballistic Missile Scenario Variable Model (1) (2) (3) Treatments Human rights -24.7 ** -23.1 ** -24.9 ** (2.7) (2.5) (3.7) Democracy -11.0 ** -12.7 ** (2.5) (3.8) Human rights 3.4 Democracy (5.0) Asia 0.9 0.8 (3.5) (3.5) Eastern Europe 5.9 6.0 (3.6) (3.6) South America -0.8-0.9 (3.5) (3.5) Controls Militarism 10.2 ** 10.2 ** (1.9) (1.9) Internationalism 5.7 ** 5.7 ** (1.3) (1.3) Conservatism 3.1 3.1 (2.2) (2.2) Ethnocentrism -3.4-3.3 (2.7) (2.7) Religiosity 5.8 ** 5.8 ** (1.6) (1.6) Male 0.7 0.7 (2.6) (2.6) White 1.6 1.6 (3.1) (3.1) Age/100-25.6 ** -25.6 ** (7.9) (7.9) Education -11.0 ** -11.1 ** (3.8) (3.8) Intercept 47.8 ** 66.0 ** 66.9 ** (2.0) (4.9) (5.1) Note: Estimated coefficients and robust standard errors from linear regressions, in which the dependent variable was 100 if the respondent supported a military strike and 0 otherwise. The sample size in each column was 1,211. Asterisks indicate the level of statistical significance. * p<0.05; ** p<0.01. 36

In the ballistic missile experiment, we randomly situated the country in one of four regions: Africa, Asia, Eastern Europe, or South America. Figure S15, below, shows that the effects of human rights were substantively and statistically similar, regardless of the region in which the country was located. Figure S15: Effects of Human Rights, By Geographic Region As in our other studies, we disaggregated the effect of human rights in our ballistic missiles experiment into four causal mechanisms: threat, morality, success, and cost. Figure S16 summarizes the effect of human rights on each of the four mediators; Figure S17 shows the effect of each mediator on public support for war; and Figure S18 shows the importance of each mechanism as a percentage of the total effect. As in our other studies, human rights affected support for war by altering perceptions of threat and morality, but not by changing beliefs about the probability of success or the cost of the operation. Figure S16: Effect of Human Rights on Four Mediators 37

Figure S17: Effects of Four Mediators on Support for War Figure S18: Causal Mechanisms as Percentage of Total Effect 38