U.S. Multinationals and Human Rights: A Theoretical and Empirical Assessment of Extractive Versus Nonextractive Sectors

The consequences of foreign direct investment (FDI) for human rights protection are poorly understood. We propose that the impact of FDI varies across industries. In particular, extractive firms in the oil and mining industries go where the resources are located and are bound to such investment, which creates a status quo bias among them when it comes to supporting repressive rulers (“location-bound effect”). The same is not true for nonextractive multinational corporations (MNCs) in manufacturing or services, which can, in comparison, exit problematic countries more easily. We also propose that strong democratic institutions can alleviate negative impacts of extractive FDI on human rights (“democratic safeguard effect”). Using U.S. FDI broken up into extractive and nonextractive industries in 157 host countries (1999–2015), we find support for these propositions.1 Extractive FDI is associated with more human rights abuse, but nonextractive FDI is associated with less abuse, after controlling for other factors, including concerns about endogeneity. We find also that the negative human rights impact of extractive FDI vanishes in countries where democratic institutions are stronger. Our results are robust to a range of alternative estimation techniques.

the heterogeneity of reasons for why a firm might be complicit in human rights violations.
Most case studies, especially those examining controversial industries, or MNCs in areas of weak governance, find that human rights are not respected by firms (Idemudia, 2009). On the other side of the spectrum, striving for generalization, the quantitative literature on foreign direct investment (FDI) has investigated cross-country patterns, and most of the statistical studies support the "Washington Consensus" (Williamson, 1990(Williamson, , 2000, or liberal view, which is that MNCs have positive effects on government respect for human rights, with some confusion as to the direction of causality (Apodaca, 2001(Apodaca, , 2002Cingranelli & Richards, 1999;Hafner-Burton, 2005; W. Meyer, 1996). 2 The recent business literature has criticized quantitative studies for lumping all forms of FDI together and for ignoring the fact that industry-level characteristics are likely to influence firms' human rights impact (Fortanier & Kolk, 2007;Giuliani & Macchi, 2014). Industries differ along economic and social dimensions, such as production inputs and outputs, technologies, choice of location, and the "footlooseness" of firms (Godfrey et al., 2010). Since FDI has become increasingly diverse over the last decades, its global impact may not be easily determined using aggregate FDI data. Moreover, it is not clear at all how and in what ways firms might be directly responsible for violating the rights of people, and what portion of the blame might be assigned to the economic motives of companies relative to the political motives of host governments.
We revisit the question of FDI's impact on human rights by integrating insights from qualitative work, namely that industry sectors and host-country context matter. We build on two quantitative pioneering studies that are more refined than previous work. Janz (2018) finds that the presence of MNCs in developing countries is positively connected to rights outcomes, but only when FDI is invested in industries with medium or high skills and technology levels, such as services. Blanton and Blanton (2009) show that the ability of countries to attract FDI via lax regulation, or suppressed labor standards, only works for some industry sectors and not for all types of FDI. These studies provide preliminary evidence that effects from FDI might vary across industries, but we do not yet know the specific reasons why some FDI has negative impacts, or how host-country conditions influence the locational environments that firms choose. Given the risks attached to political instability, and the general aversion of companies to violence-based risks, why firms would decide to locate in high-risk environments must lie in locational advantages.
We ask, therefore, what is the impact of FDI on human rights across different industry sectors, and what differences exist if any between extractive (oil and mining) versus nonextractive industries (manufacturing and services)? We focus on this distinction because among qualitative case studies, there is a great deal of attention paid to the negative impacts in extractive sectors, which has amounted to the theory of the so called "resource curse" that drives autocratic and corrupt politics in resource-wealthy states (Eweje, 2006a(Eweje, , 2006b(Eweje, , 2009Idemudia, 2009;MacDonald & McLaughlin, 2003;Ross, 2001). Many local communities and social activists are willing to give up on potential economic benefits from FDI and oppose new extractive investment because they fear that extraction leads to human rights abuses in local communities, where foreign companies run rough shod over the interests and demands of people (Dashwood, 2014;Mutti et al., 2012).
Our argument is twofold: first, extractive FDI is location-bound, so that firms have little incentive to punish repressive host countries, given that they are the "legal" owners of these assets. Even if there are external incentives for firms to divest from repressive regimes due to voluntary industry standards and pressure from NGOs and the media; extractive firms are unlikely to lobby host-country governments for better human rights conditions or threaten with withdrawal because they can be easily replaced. Indeed, extractive MNCs are bound by the low-mobility features of their sector, and threats to leave would simply not be credible ("location-bound effect"). Moreover, when a company is specialized in extraction, recovering losses from the sunk costs will depend heavily on retaining access to the natural resources.
Second, we argue that foreign investors operate not only within industries but they are also subject to domestic country context. If firms decide to invest in-or remain in-repressive host countries, they may become complicit in wrongdoings because they provide revenue to such governments (Clapham, 2006;Wettstein, 2010). This negative view is particularly relevant for institutionally weak countries. In other words, host-country potential for handling dissent generated by an FDI project is dependent on strong institutional quality. 3 Even though the challenge of domestic governance gaps and institutional voids is a core concern in the Business and Human Rights field (Wettstein et al., 2019), there are hardly any quantitative studies on FDI and human rights that test for such conditional effects, with rare exceptions (Clark & Kwon, 2018;Wang, 2017). The studies that do exist do not distinguish between different industries. We propose that democratic institutions provide safeguards for human rights protection by keeping leaders (and MNCs) in check ("democratic safeguard effect"). The effectiveness of global voluntary codes of conduct, such as the 2011 U.N. Guiding Principles on Business and Human Rights (Ruggie, 2011), or the updated OECD (2011) Guidelines for MNCs, are likely to be influenced by the interplay between industry-sector characteristics and host-country institutions. We shed more light on this by examining this issue empirically.
We examine U.S. FDI located within 157 host countries (1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), for which disaggregated data are available. We find that extractive FDI is associated with negative effects on human rights, but nonextractive FDI shows positive effects, after controlling for a range of relevant factors, including the possibility of endogeneity. We further examine the conditioning effect of host-country institutions, finding that the negative effect of extractive FDI on human rights becomes insignificant in countries where institutions are stronger. The results are robust to sample size, estimation technique, and alternative specifications. The rest of the article is organized as follows: In the next section, we discuss the features of different types of FDI, in particular, extractive investment, and how they are linked to human rights within the wider country context. We then test the proposed links empirically in a cross-country time-series analysis. The conclusion outlines the implications of our findings for future research.

Extractive FDI, Human Rights, and Democratic Safeguards
Our overall argument rests on the assumption that firms which operate in the same sector will adjust their behavior in a similar way, which leads to institutional isomorphism and homogenisation (Chand & Fraser, 2006;DiMaggio & Powell, 1983;Griffin & Mahon, 1997). Each industry, such as natural resource extraction, manufacturing, or services, has a normative, regulative, and economic structure to which MNCs adapt their behavior (March & Olsen, 2006;Scott, 1995). For example, industries differ in their use of skill levels, location decisions, the competitive behavior among firms, or role of branding, and CSR efforts (Beschorner & Hajduk, 2017;Giuliani & Macchi, 2014;Godfrey et al., 2010;K. E. Meyer, 2004). While there is also withinindustry variation of companies' individual behavior, we propose that it is the unique features of industry sectors that shape overall patterns of human rights impact on host countries because most firms, by and large, face the same incentives shaping their behavior.
The focus of our theory lies in distinguishing between extractive versus nonextractive FDI because extractive sectors, such as oil or mining, have two distinct industry-related features that are different from other sectors. First, under the "natural resource-seeking motive," foreign investment in extractive sectors requires developing and controlling the sources of supply of raw materials, which are the primary commodities required for production. Second, extractive FDI involves huge capital investments. Therefore, investment of extractive FDI may not be able or willing to avoid investing in repressive regimes where resources are located; and divestment from host countries during times of political instability and repression is almost impossible because of high sunk costs and the location-bound nature of this type of FDI (Vernon, 1971). Moreover, global law recognizes even highly repressive regimes in control of natural resources as the legal owners of these assets according to the principle of "might makes right" (Wenar, 2016). Investments in extractive sectors are also more vulnerable to issues of domestic security, because a change in the status quo might mean replacement of leaders, negotiating a new deal, and perhaps complete or partial loss of monopoly rents. A similar logic has been used to explain why ruling elites in resource-wealthy countries avoid reforms-the fear of replacement and the loss of future monopoly rents (Acemoglu & Robinson, 2008). Extractive investments may, in other words, have a status quo bias because of insecurity of rights obtained for extracting natural resources. High sunk costs, thus, reduce the incentives of extractive FDI to sanction ruling elites who promise stability using repression to weaken opposition movements or local unrest. Simultaneously, governments facing dissent know these constraints, and can safely ignore extractive MNC preferences of not wanting to be associated with repression. Crucially, due to high sunk costs and the location-bound nature of extractive FDI, threats of withdrawal will not be credible, so that MNCs in the extractive sector have little leverage over repressive host countries.
This location-bound, high-sunk cost nature of extractive industries perpetuates well-known issues including human rights challenges. For example, the location of resources often involves forced displacement and a loss of livelihood for local communities, exploitation of low-skilled workers, and environmental pollution (Eweje, 2006a(Eweje, , 2006b. These issues have led to protests, sabotage, and subsequent security challenges across extractive sectors in the past (Dashwood, 2014;Mutti et al., 2012). A prominent example is the Niger Delta, where government security forces used violence against activists and local protesters (Frynas, 1998;Idemudia, 2009;MacDonald & McLaughlin, 2003). Indeed, a detailed list of cases on human rights violations registered under The Alien Tort Claims Act (ATCA) of 1789 of the U.S. government against U.S. firms operating in various developing countries reports that the majority of the cases are associated with extractive investments (see our summary in the Appendix A).
Our argument does not claim that there is no within-industry variation in the extractive sector, or that all extractive firms are bound to actively violate human rights. Firms do have agency in how they behave toward their employees, local communities, and the environment, and they are increasingly being held accountable via industry-level standards, multilateral treaties, and international law. For example, the OECD (2013) Due Diligence Guidance for Responsible Supply Chains of Minerals from Conflict-Affected and High-Risk Areas, and the OECD (2017) Due Diligence Guidance for Meaningful Stakeholder Engagement in the Extractive Sector, aim to support firms in reducing their negative impacts. The 2010 Conflict Minerals provision of the Dodd-Frank Act for U.S. companies, which now have to disclose the use of conflict minerals in their supply chains, has similar goals. Similarly, a crucial finding from the "naming and shaming" literature is that recent developments to hold firms accountable have created pressure on foreign investors to divest from repressive regimes to avoid reputational damage (Barry et al., 2013;Vadlamannati et al., 2018). This issue is also discussed within the business literature, where firms might be seen as political and powerful actors that can step in when states are unable to provide public goods to their citizens, lobby governments for improvements, and drive a "race to the top" (Matten & Crane, 2005;Scherer & Palazzo, 2011;Westermann-Behaylo et al., 2015).
However, there are features of the extractive industry that are unique among other FDI-high sunk cost and location-bound raw materials-which are likely to strongly influence the behavior of firms in problematic host environments. Even if some "good" extractive firms wanted to pressure repressive regimes to improve the rule of law and human rights protection, any exit threat in the extractive industry is not likely to be credible, as host governments are also aware of the low mobility of such investment. In sum, a combination of "bad behavior" of some-but surely not all-extractive MNCs, together with low mobility of such FDI and low credible exit threats, we expect this kind of FDI to have negative impacts on human rights protection. The examination of this argument is crucial because most recent developments in hard and mostly soft law to improve MNCs' behavior, as well as pressure from NGOs and the media, may only be effective to some extentthey may not change the industry's effect as a whole due to the nature of how raw material extraction works.
In comparison, nonextractive FDI, for example, service sectors or manufacturing, is more footloose relative to extractive FDI, and has fewer barriers against its choice of location (Kobrin, 2009). MNCs can divide up their activities in many locations and use outsourcing from multiple plants in their overall risk-reducing strategies (Blonigen et al., 2007) to hedge against risk emanating from one specific location. The sunk cost involved in nonextractive sectors can, at times, also be high; but nonextractive sectors are not generally bound to particular raw material locations, where the real value is in the ground. Nonextractive industries which have heavily invested in a host country may not have any interest to leave repressive regimes, but if they wanted to, they could. Any exit threat would be much more credible than in the extractive industry. Moreover, manufacturing and service FDI are likely to be more important where the "resource curse" effects of high dissent and bad governance are generally absent, simply because nonextractive FDI is deterred by political instability and corruption, which are often associated with the resource curse (Asiedu, 2006;Globerman & Shapiro, 2003). This means that not only is a good environment, such as democracy and good governance, attractive to nonextractive FDI, but this FDI is less biased toward the status quo and less location-bound, thereby having greater freedom to disassociate from human rights violations and threaten exit. This does not mean, however, that other kinds of FDI are absent where natural resources exist but that the policy environments in resource-dependent states, often characterized by high corruption and predatory regimes, are likely to attract less nonextractive FDI (Globerman & Shapiro, 2003;Wenar, 2016).
To sum up, extractive multinationals have to go where the resources are located, which creates a status quo bias among them when it comes to supporting repressive rulers. Under these conditions, FDI is likely to be far more aligned with state actions despite damage to reputations. The same is not true for nonextractive MNCs, which can exit more easily, or threaten credibly to do so. Thus, we propose to test the following main hypothesis: Hypothesis 1: "Location-bound effect": Ceteris paribus, extractive FDI is connected to worse government respect for human rights, in comparison with nonextractive sectors.
We argue further that this relationship can be expected to vary according to host-country institutions. U.N. Special Representative for Business and Human Rights (SRSG) John Ruggie (2008) has stated that the "root cause of the business and human rights predicament today lies in the governance gaps created by globalization-between the scope and impact of economic forces and actors, and the capacity of societies to manage their adverse consequences." (p. 189). Not all countries that receive extractive FDI are repressive regimes (e.g., Canada, Norway, and Australia), and their positive human rights record indicates that democratic institutions can significantly mitigate negative impacts. One of the most accepted theories in comparative human rights research is that democratic institutions increase the costs of using repressive behavior for leaders because they can be voted out of office. They also provide alternative mechanisms of control for citizens via participation, so that grievances are expressed at the ballot rather than in violent protests (Davenport, 2007). Domestic institutions of host countries that reduce instability can benefit from FDI. Democratic institutions can also mitigate the negative effects from FDI once it is present, keeping leaders' repressive tendencies in check, and shaping adequate policy responses that steer away from repressive action (e.g., during security crises around extraction sites, Wang, 2017). In fact, protests and instability may not even occur in the first place because democratic institutions provide for other ways for citizens to voice disagreement with MNC's, including the resort to legal action (Schrempf-Stirling & Wettstein, 2017). Clark and Kwon (2018) examine if the effect of FDI stock varies by regime type, showing that the harmful impact of foreign capital on human rights might be significantly weaker in democratic states. We extend this argument to extractive industries and propose that democratic institutions condition the effects of extractive FDI in a way that reduce human rights repression. Hypothesis 2: "Democratic safeguard effect": The effect of extractive FDI on human rights is conditional on democratic institutions, which reduce negative impacts from FDI.

Model Specifications
To examine our theoretical propositions, we apply panel data covering 157 countries (see Appendix B for list of countries) over the 1999-2015 (17 years) period for which we have U.S. FDI data. Since some of the data are not available for all countries for all years, our data set is unbalanced. We estimate: where PTS it is our outcome variable, which measures the degree to which a government practices the violations of physical integrity rights of citizens measured by the Political Terror Scale (PTS) index, α i is the intercept, H it− 1 is our main variables of interest, Z it− 1 are control variables, and γ γ γ γ i country dummies, δ δ year-specific dummies, and ω it is the error term.
The PTS index measures the amount violations of personal integrity rights, such as politically motivated execution, torture, forced disappearance, unlawful imprisonment, and discrimination based on political and religious beliefs (Gibney & Dalton, 1996). The PTS data are generated from two sources of information-namely, from Amnesty International country reports and annual country reports supplied by the United States State Department. The PTS scale is widely used and shows high correspondence with other measures of state repression. The PTS is defined as follows: 1. If countries are under secure rule of law, political imprisonment and torture are rare, and the political murders are extremely rare. 2. If imprisonment for nonviolent political activities is limited, torture and beating are exceptional, and political murder rare. 3. If political imprisonment is extensive, execution and political murder may be common, and detention for political views is acceptable. 4. If the practices of level 3 are expanded to a larger segment of population, murders and disappearances are common, but terror affects primarily those who interest themselves in political practices and ideas. 5. If level of terror is population wide, and decision makers do not limit themselves by which they pursue private and ideological goals.
Therefore, a higher value denotes more violations of human rights. We take the average of the two measures from the Amnesty International and the U.S. State Department to avoid any subjective biases, especially in the data from the State Department as suggested by Qian and Yanagizawa (2009) and Poe and colleagues (2001). 4 This index is a commonly accepted measure for human rights protection and has been used widely in the literature (Davenport & Armstrong, 2004;Davenport & Nordas, 2013;Vadlamannati et al., 2018;Walker & Poe, 2002). The mean of the PTS index during 1999-2015 is 2.68 with a standard deviation of 1.09.
H it-1 captures our main variables of interest, which are lagged by one year to reduce bias from simultaneity. These include (a) total U.S. FDI stock per capita (log), and (b) disaggregated U.S. FDI per capita (log) provided by the U.S. Bureau of Economic Analysis (BEA) database on Balance of Payments and Direct Investment Position. 5 The BEA defines U.S. direct investment abroad as either ownership or control of 10% or more of a foreign business enterprise, voting securities or the equivalent. 6 The data are recorded in historical cost basis and hence cannot be identifiable in constant nor current prices (Blanton & Blanton, 2009). The BEA data capture the U.S. FDI position abroad disaggregated by industry sector from 1989 onwards. It should be noted that the industry classification of U.S. FDI by the BEA was converted from Industrial Classification (SIC) to North American Industry Classification System (NAICS) in 1998. Therefore, we use the data from 1999 onwards for consistency. As mentioned above, we use FDI per capita and log the data to address the problem of skewness. We collect the data on extractive (oil, petroleum, mining, and minerals) and nonextractive (manufacturing and services) U.S. FDI for 157 developing countries based on the availability of data at the BEA. We expect that it is mainly the penetration of extractive investments visà-vis nonextractive investments that are associated with more human rights violations in host countries. Since most countries have some of both types of investments, we estimate their independent effects on human rights.
The vector of control variables (Z it ) includes other potential determinants of human rights which we obtain from the extant literature on the subject. We follow the pioneer studies of Poe and Tate (1994) and Poe and colleagues (1999) and other comprehensive evaluations on determinants of repression (Carey & Poe, 2004;Landman, 2006). We avoid the trap of "garbage-can models" or "overfitting" by keeping our models simple (Achen, 2005;Schrodt, 2014). We adopt the conservative strategy of accounting only for key factors that may directly associate with our main independent variables, exploring several other variables in robustness checks. Accordingly, we control for economic development by including per capita income (log) in US$ constant prices taken from the World Development Indicators 2018 (WDI). Country size (population, log) is an important control because governance and rights protection is more challenging in larger countries, which may also attract higher values of FDI due to market size. Next, we include a measure of regime type based on the Polity IV data (Jaggers & Gurr, 1995). We create a set of dummies for democracy and autocracy using the Polity IV index above +6 representing full democracy, which takes the value 1, and 0 otherwise. 7 Likewise, we include an autocracy dummy, which takes the value 1 if the Polity IV index is below −6, and 0 if not. The reference category is made up of imperfectly institutionalized regimes, referred to as anocracies (Fearon & Laitin, 2003). In addition, we include a variable measuring civil war that takes the value 1 if there is armed conflict between an organized rebel group and a state where at least 25 deaths have occurred in a single year, and 0 otherwise (Gleditsch et al., 2002). Naturally, an ongoing civil war is likely to affect the degree of state repression. We also include a count of the number of years of civil peace to distinguish between immediate post-war situations and the history of peaceful conditions since these factors affect our independent variables (Beck et al., 1998).
In addition, we include oil export dependency, which is related to repression due to the so called "resource curse" (de Soysa & Binningsbø, 2009). Following Michael Ross (2001), we construct an oil production dummy taking the value 1 if oil and gas production exceed $100 per capita, and 0 if not. It is noteworthy that we lag all controls by one year to avoid simultaneity bias. Finally, we also include a lagged dependent variable (LDV) in our models. There are two reasons for its inclusion. First, an LDV also captures regional diffusion and spill-over effects that are not directly observed (Neumayer, 2005). Second, it is theoretically plausible that bureaucratic decisions associated with the organs of state repression use past decisions to repress or not in present circumstances, so that this behavior can be quite sticky (Poe et al., 1999). The descriptive statistics are presented in Appendix C. For more details on data definitions and sources, see Appendix D.
The baseline models are estimated using the generalized least squares fixed-effect (FGLS) estimator. The pooled data are susceptible to having highly correlated data between and across panels that could lead to optimistic standard errors (Beck & Katz, 1995). Using FGLS over a simple ordinary least squares (OLS) allows estimations in the presence of AR (1) autocorrelation within panels and cross-sectional heteroscedasticity across the panels. As mentioned, we control for country and year-specific fixed effects. We also use ordered logit models because the outcome variable is rank ordered (1)(2)(3)(4)(5). In ordered logit estimations, we include only year-specific dummies because of including two-way fixed effects in nonlinear estimations may be problematic due to the well-known "incidental parameter problem" (Lancaster, 2000;Wooldridge, 2002).

Endogeneity Concerns
It is quite possible that our key explanatory variables-U.S. FDI and its disaggregated measures-are endogenous to having fewer human rights violations. That is, it might be that governments committed to respecting human rights attract more FDI as suggested by Blanton (2006, 2009). For example, the expectation of instability arising out of dissent and uprising could deter foreign investors. Not taking this endogeneity into account would induce bias in our estimate of the effect of FDI on human rights. We thus control for endogeneity concerns by replicating the baseline estimations using the system-generalized method of moments (GMM) estimator as suggested by Arellano and Bond (1991), Arellano and Bover (1995), and Blundell and Bond (1998), which is considered most appropriate in the presence of endogenous regressors. The dynamic panel GMM estimator exploits an assumption about the initial conditions to obtain moment conditions that remain informative even for persistent data. Second, and the most important reason why we use the GMM estimator is because of the Nickell (1981) bias problem, as FGLS estimations tends to be inconsistent to the inclusion of an LDV and the presence of two-way fixed effects simultaneously in a short panel like ours.
The GMM results are based on a two-step estimator implemented by Roodman (2006). We apply the Sargan-Hansen test on the validity of the instruments used (amounting to a test for the exogeneity of the covariates) and the Arellano-Bond test of second-order autocorrelation, which must be absent from the data in order for the estimator to be consistent. We treat both LDV and our hypotheses variables (namely, total U.S. FDI, extractive and nonextractive FDI) as endogenous, and the other variables as strictly exogenous. As before, we include time dummies in the GMM regressions. To minimize the number of instruments in the regressions, we collapse the matrix of instruments as suggested by Roodman (2006).

Interaction Effects
Next, we examine whether the negative effects of FDI on human rights are more prevalent in countries with weak institutions. This is an important consideration because extractive activities by firms are also prevalent in institutionally strong countries like Australia, Canada, Norway, among others where pressures for maintaining status-quo may not mean that human rights are violated. To test this proposition, we introduce an interaction term: where β β ( ) H iv it × −1 is the interaction term between extractive FDI per capita (log) and the conditioning variable measuring the "quality of institutions" in a host country namely, the Polity regime type index (β β iv it −1 ) in country i during year t−1. The interaction term is lagged by 1 year. We use Polity II index which is coded on a −10 to +10 scale, where the directionality of the index runs from strict autocracy to full democracy. The index is derived from coding of the following key components, which capture the strength of institutions in a polity; namely, competitiveness and openness of executive recruitment, constraints on the power of chief executive and competitiveness of political participation. These interaction effects will allow us to understand whether the negative effects of extractive FDI are concentrated in countries with weak institutions. Once again, we use the FGLS two-way fixed-effect estimator and generate marginal plots to assess the substantive effects of the conditional results.

Empirical Results
Tables 1 to 4 present our regression results. Table 1 displays the results estimated using FGLS two-way fixed-effect models, which examine the relationship between human rights abuse and U.S. FDI by sector, controlling for other determinants which are added in a stepwise manner. Table 2 presents the replication of the baseline models using ordered logit estimators, and we report the marginal effects from ordered logit estimations in Figure 1. Table 3 presents the results addressing endogeneity concerns on the effects of U.S. FDI and disaggregated measures on human rights violations using the simulated generalized method of moments (SGMM) method. Finally, Table 4 reports the results for extractive FDI in interaction with our institutional measure, and we present the conditional plot for this result in Figure 2.       (1)   Table 2.
Note. FDI = foreign direct investment.

Figure 2. Extractive U.S. foreign direct investment (FDI), institutions, and marginal effect on human rights violations (PTS index).
Note. PTS = Political Terror Scale.
As seen from column 1 in Table 1, the impact of total U.S. FDI per capita on human rights violations is statistically insignificant. When other controls are added in the model, total U.S. FDI has a statistically significant negative coefficient, independently of all the controls. Since higher values of the PTS index indicate more rights violations, the negative coefficient for total U.S. FDI suggests that more FDI is related to better rights protection, similar to other studies on FDI and human rights (Apodaca, 2001(Apodaca, , 2002Cingranelli & Richards, 1999;Hafner-Burton, 2005; W. Meyer, 1996). As seen in column 2 in Table 1, at the mean value of U.S. FDI per capita (log) (1.12), there is a 0.06-point decline in the PTS index, independent of other control variables. A standard deviation increase in FDI per capita (log) (1.56) above the mean value lowers the PTS index by roughly 0.17 points, which is significantly different from 0 at the 1% level. In column 3, we replace total U.S. FDI with two disaggregated measures namely, extractive and nonextractive FDI per capita (log). We find that extractive FDI has a positive impact on PTS index, net of all the controls including nonextractive FDI, which is statistically significant at the 1% level. The positive effect suggests that an increase in extractive investments increases the level of human rights violations. In the same model, we find contrary results related to nonextractive FDI, which is also significantly different from 0 at the 5% level. The substantive effects suggest that a standard deviation increase in extractive FDI per capita (log) is associated with an increase in PTS index by 0.1 points. However, one standard deviation increase in nonextractive FDI per capita (log) is associated with a decrease in PTS index by 0.23 points, which is double the effect of extractive FDI. Thus, even when extractive FDI is accounted for, the level of nonextractive FDI has a strong positive effect on human rights protection. These effects of nonextractive FDI and extractive FDI, when included together in the same model, remain significantly different from zero at the 5% and 1% levels, respectively as shown in column 4. The substantive effects of nonextractive FDI continue to be higher than that those of extractive FDI, net of all the control variables. Finally, our results on extractive vis-à-vis nonextractive FDI on human rights are robust to the inclusion of an LDV in column 5 of Table 1. Thus, even when estimating the within-country variance with an LDV, the hypothesis that extractive FDI reduces and nonextractive FDI increases human rights protection is upheld. Overall, these results support the argument for a negative "location-bound effect" (hypothesis 1) associated with extractive investment. These results support studies focused on mining projects that show that company-community conflicts increase where mining projects locate (Berger et al., 2017).
The results on the control variables are consistent with those reported by other studies. The level of income largely remains statistically insignificant. We find that countries with a larger population have more violations of rights. However, this effect is not consistent across all the models. We suspect that both income and population change slowly over time and hence are correlated with the fixed effects. This could be the reason why both variables are highly significant at the 1% level in ordered logit estimations in Table 2 where country fixed effects are not controlled. As expected, in all the models irrespective of estimation technique (in Tables 1 and 2), democracy is associated with fewer human rights violations. Likewise, civil peace years are associated with lower incidences of human rights abuse and civil conflicts with higher violations of human rights, as others also report (Poe et al., 1999;Poe & Tate, 1994). The variable for oil exporters in general is largely insignificant when estimating the models with the linear estimator in Table 1. Interestingly, our main results on extractive and nonextractive FDIs remain highly significant despite the inclusion of several statistically significant controls.
Next, due to the rank-ordered nature the dependent variable PTS, we reestimate the models using ordered logit controlling for year-specific dummies. These results are reported in Table 2.
As before, total U.S. FDI per capita (log) is negative and significantly different from 0 at the 1% level in both column 1 and in column 2, where other controls are included. Likewise, we also find in column 3 and 4 that extractive FDI increases human rights violations while nonextractive FDI decreases violations. These results are largely similar to those reported in our baseline estimations in Table 1. Our results of U.S. FDI's impact on human rights, thus, remain robust to utilizing alternative estimation techniques.
To understand the magnitude of our results in the ordered logit models, we compute the marginal effects at the mean of both extractive and nonextractive FDI per capita (log). It is noteworthy that marginal effects in ordered logit are not straightforward to interpret. We follow Dreher and colleagues (2009) and compute estimated probabilities before and after a shock of one standard deviation of extractive and nonextractive FDIs on the PTS (see Figure 1). Accordingly, nonextractive FDI has a higher impact compared to extractive FDI. The estimated probability of observing the PTS index values of 1, 2, and 3 (at the mean of all variables) are 4%, 25%, and 19%, respectively, while index values 4 and 5 occur with a predicted probability of 3% and 0.3%. However, after an increase in nonextractive FDI per capita (log) by one standard deviation, these predictions get higher (i.e., 6% and 29% for low PTS index values 1 and 2, fewer or no human rights violations). They decrease for the higher PTS index values 3, 4 and 5% to 13.5%, 1% and 0.1%, respectively. Similarly, after an increase in extractive FDI by one standard deviation, the predictions get marginally lower for low PTS index values (i.e., 3% and 23% for 1 and 2) while they increase for the high PTS index values 3, 4 and 5% to 20%, 3% and 0.4%, respectively. These effects are not only statistically significant, but also quantitatively important.
Next, we examine our models controlling for possible endogeneity between human rights and types of FDI. As discussed earlier, we make use of the System-GMM method. Table 3 presents the results of the models estimated using GMM. The Hansen test and the Arellano-Bond test do not reject the SGMM specifications at conventional levels of significance across the columns, indicating that endogeneity may not be a major problem. The Hansen J-Statistic clearly shows that the null-hypothesis of exogeneity cannot be rejected at the conventional level of statistical significance. As can be seen from column 1, total U.S. FDI per capita (log) is negative and significantly different 0 at the 1% level. These results remain robust to the inclusion of various control variables in column 2. In column 3 and 4, we include both extractive and nonextractive measures of U.S. FDI per capita (log). As seen, extractive FDI is positive and significantly different from 0 at the 1% level in columns 3 and 4. Notice that even after controlling for potential feedback from the PTS index, the coefficient value of column 5 in Table 1 (in which an LDV is also included) is reduced only marginally. Likewise, we find that nonextractive FDI per capita (log) has a significant negative impact on the PTS index at the 1% confidence level. These results highlight an interesting point. First, they show that the size of the coefficient for both forms of FDI are, by and large, similar in GMM and non-GMM regression estimates, and they suggest that endogeneity may not be a major problem.

Conditional Effects
Thus far, we examined the direct effect of FDI on human rights violations. Next, in Table 4, we examine whether the effects of extractive FDI, in particular, are conditional on the level of institutional quality. We introduce an interaction term between extractive FDI and the Polity index, which is our measure of institutional quality in Table 4.
In column 1, only the interaction term is included without any control variables. We do however control for nonextractive FDI. While in column 2, we include all other control variables, column 3 also includes an LDV. As seen in columns 1 to 3, the interaction term is negative but statistically insignificant. Interestingly though, our extractive FDI per capita (log) measure on its own (i.e., when the institutions measure is 0) has a positive and statistically significant effect on human rights violations. Likewise, the effect of institutions (Polity index) on human rights violations is negative when extractive FDI per capita is 0, which is statistically significant at the 1% level. It is important to note that the interpretation of the interaction term even in linear models is not straightforward. Consequently, a simple t-test on the coefficient of the interaction term is not sufficient to examine whether the interaction term is statistically significant, or otherwise (Ai & Norton, 2003).
We rely on marginal plots for assessing the nature of the conditional effects. The interactive effect is best assessed with a margins plot which depicts the magnitude of the interaction in Figure 2. To calculate the marginal effect of extractive U.S. FDI per capita (log) on the PTS index, we take into account both the conditioning variable (Polity index) and the interaction term, and we display graphically the total marginal effect conditional on the Polity index. The y-axis of Figure 2 displays the marginal effect of U.S. extractive FDI per capita (log), and the marginal effect is evaluated on the Polity index variable on the x-axis. Note that we include the 95% confidence interval in Figure 2.
In line with our theoretical expectations, extractive FDI per capita (log) increases the PTS index (at the 95% confidence level at least) when the Polity index is lower than the score of 6. In other words, the marginal effects are significant and positive when the upper bound of the confidence interval is above zero. The marginal effects suggest that extractive FDI increases the PTS index by 0.1 points when the Polity index score is 6. Consider a country which has a Polity index score of −10 (a strict autocracy). A point increase in U.S. extractive FDI per capita (log) in a country with a Polity score of −10 would increase the PTS index by almost 0.15 points (i.e., FDI is connected with more human rights abuse at high levels of autocracy). Interestingly, the margins plot also shows that the effect of extractive FDI on human rights violations becomes statistically insignificant once the Polity IV index of a host country is above 6 (i.e., in a country with stronger democratic institutions). These results suggest that the negative effects of extractive FDI on human rights are more prevalent in host countries with weak institutions, supporting our hypothesis 2, which proposed a "democratic safeguard effect." This might also be one of the main reasons why FDI in extractive industries in countries, such as Australia, Canada, Norway, and Chile, is not associated with the repression of human rights.

Checks on Robustness
We examine the robustness of our main findings in the following ways. First, we exclude high-income OECD countries, which seldom have bad human rights records. It is quite plausible that the results might be driven by these countries. The new results (without high-income OECD countries, reported in Table A in Supplemental Material) 8 remain robust and mirror those reported in our baseline estimations.
Second, we use the PTS in its disaggregated version, which is based on only Amnesty International reports, to avoid any form of biases which are likely to be present in the PTS index based on sources by the U.S. State Department (Qian & Yanagizawa, 2009). Our new results with the Amnesty International PTS index (reported in Table B, Supplemental Material) still show strong significant effects for extractive FDI and the abuse of human rights. These results remain robust to the inclusion of an LDV and alternative estimations techniques, such as ordered logit.
Third, following de Soysa and Vadlamannati (2011Vadlamannati ( , 2012, we run all our results using an alternative measure for human rights, the physical integrity rights index (PIR hereafter), sourced from the Cingranelli and Richards (1999) Human Rights Database. This measure is available annually from 1981 onwards until 2011 for 195 countries. 9 The source of information used for coding the index are the same as for the PTS, but the PIR data are collated somewhat differently. 10 The PIR is an additive index constructed from measures of torture, extrajudicial killing, political imprisonment, and disappearances. It ranges from 0, meaning no government respect for these four human rights to 8, or full government respect for these four human rights. Once again, our results (in Table C in Supplemental Material) based on PIR index as dependent variable remain robust.
Fourth, we estimate our baseline models by excluding outliers in all our FDI variables, as extreme values could influence our main findings. Excluding the outliers aggregated and disaggregated for U.S. FDI per capita from the sample does not change our main results substantially (Table D in  Supplemental Material). Fifth, we use an alternative method of operationalization of our main variable of interest. Instead of per capita FDI, we use FDI measures as a share of gross domestic product (GDP), which is an alternative proxy of penetration of FDI into the host economy, and FDI measured in US$ million (log). Our baseline specifications estimated using these alternative FDI measures remain robust and, in the direction, reported above ( Table E in Supplemental Material).
Sixth, Clark and Kwon (2018) propose that any negative effects of FDI on human rights might be diminishing over time, mostly because there is now less primary sector FDI within overall FDI, which may have driven the negative effects of FDI. We extend this argument by proposing that even for problematic extractive FDI, improvements over time are plausible because new guidelines and voluntary codes for the extractive industries aim at changing foreign investors' impact on the ground. 11 We test this proposition by interacting both measures of FDI, extractive and nonextractive FDI per capita (log), with year-specific dummies. We report these results in Supplemental Material Table F, and in Figure 3 here. We find that the negative human rights effects of extractive FDI tend to increase over time, rather than diminish. The positive effects of nonextractive FDI on human rights protection tend to increase over time. We present a graphical representation of coefficients of both variables from interactions with year dummies in Figure 3. As seen there, the effects of extractive and nonextractive FDI on human rights continue to diverge significantly over time. This supports existing criticism that suggests that initiatives in soft and hard law against negative impacts in the extractive sectors do not go far enough in creating tangible changes on the ground (Schrempf-Stirling & Wettstein, 2017). Since such initiatives have become stronger in recent years, it might be that improvements may take decades and that they do not yet show in the data.
Taken together with our findings on the impact of weak institutions, future research should investigate the lack of improvement over time in more detail, and in particular, assess what is driving the worsening of extractive FDI effects versus increasingly good impact of nonextractive FDI. The fact that global consumers, such as in the garment and fashion industry, do not have ways of knowing and punishing bad behavior in the extractive sector is possibly a strong explanation. Finally, we re-estimate all the models by dropping the variables for which we could not find any statistical significance. These include per capita income, autocracy, oil exports share dummy, and peace years. With these modifications, our results on extractive and nonextractive FDI on human rights are upheld and remain robust. These robustness results are not shown here but are reported in Supplemental Material. Overall, these findings suggest that our results are robust not only to the size of the sample and alternative methods of operationalization of our main variable of interest but they are also robust to alternative estimation techniques.

Conclusion
Most studies on the question of FDI and human rights report positive effects, supporting the liberal contention that multinational companies bring benefits to host countries, including improvements in governance. Case-study-based research, as well as studies of civil war and armed violence, often implicate MNCs in human rights violations and armed violence within host countries.
Our study provides evidence that the impact of U.S. FDI on human rights varies across industries. In particular, we propose that extractive firms in the oil and mining industries are bound to the location of resources, which creates a status quo bias among them when it comes to explicit, or tacit, support when dealing with repressive rulers ("location-bound effect"). We find a negative effect of extractive U.S. FDI on human rights protection. The same is not true for nonextractive MNCs in manufacturing or services, which is connected to greater protection of human rights. We also find that strong democratic institutions can condition the negative impacts of extractive FDI on human rights ("democratic safeguard effect"), so that the effect of FDI is zero on human rights where the constraints on the executive are strong.
With these findings, we addressed an important gap in existing statistical studies on FDI and human rights. Most studies have simply used total FDI instead of disaggregating FDI. We think that sharper theory needs to focus on the heterogenous nature of FDI and its impacts on political and economic life. Recent statistical (large-N) studies on FDI suggest increasing conflict when FDI locates in mining activity, particularly in Africa (Berger et al., 2017). It stands to reason then that extractive FDI should be associated with human rights violations, given that governments have a strong interest in stamping out dangerous dissent, whether companies desire such crackdowns or not. Many existing studies have stressed that "human rights behavior of multinationals may vary depending on sectoral differences" (Kim & Trumbore, 2010, p. 732), proposing that "the sector invested in, determines whether foreign investment is beneficial for host countries" (Apodaca, 2002, p. 902), and that future research should examine "particular forms of FDI" (Sorens & Ruger, 2012, p. 6). Building on a new development in the quantitative literature toward more differentiation (Blanton & Blanton, 2009;Janz, 2018), we employ disaggregated industry-specific data to assess impacts on human rights. Our study also shows that it is important to integrate host-country conditions. Crucially, the findings suggest that qualitative case studies, alongside NGO and media reports, do not merely amount to anecdotal evidence, but they reflect larger patterns of negative impacts from extractive industries across countries and time. Future studies might probe more carefully how exactly FDI in the extractive sector balances out its obligations to multilateral initiatives relative to the economic losses from societal dissent.
The implications for policy and voluntary standards are clear: we need to see the development of stronger democratic safeguards; otherwise, efforts of individual firms may not produce tangible improvements-even if some firms want to do better. Such initiatives as the Extractive Industry Transparency Initiative (EITI) have come a long way toward facilitating transparency and providing a platform for company-community conflict resolution (Lujala et al., 2017). Finally, we find it worrying that the negative effects of nonextractive FDI on human rights protection tend to increase over time, rather than improve, as we showed in our robustness tests. This finding is preliminary, and not much time has passed since the Ruggie framework and voluntary guidelines in the extractive industry are adopted; but the finding supports criticism of soft and hard law, which suggests that these initiatives do not go far enough (Schrempf-Stirling & Wettstein, 2017)-in particular, in areas with weak democratic institutions. Global hard law initiatives might also do well to question the current norms that legalize the "might is right" principle when it comes to natural resources, which would give firms a massive incentive to act in ways that will constrain their support for repressive regimes (Wenar, 2016). One might also research the many ways in which local communities close to extractive activity might act as a constraint on nationallevel politics, and how bargaining failures between MNCs and local communities might be prevented for getting better win-win solutions.
By integrating the industry-sector level with domestic country context, we open up new avenues for statistical as well as qualitative case study work. Future quantitative research should explore further boundary conditions under which improvements of FDI on human rights can be expected, while qualitative evidence on the firm-level is equally necessary to trace under which circumstances policy changes can be more effective in protecting the rights of local communities. Why, for example, do some firms internalize a greater sense of social responsibility than others? How much leverage can companies have over governments intent on political survival, especially given the fierce competition for natural resources into the future? We encourage mixed-methods research designs for overcoming the existing disconnect between statistical analyses at the macro level and single case studies focused on the micro level. 7. Note that estimating the models with the original Polity IV index does not alter our results substantially. 8. The supplemental file will be available in our replication materials upon request. 9. One main reason why we do not use this measure in our main analysis is that the data are available only until 2011. 10. For more on construction of the data set and coding rules, see CIRI Human Rights Data project, retrieved February 27, 2019, from: http://www.humanrightsdata.com. 11. We thank an anonymous reviewer for pointing this out.