US Multinationals and Human Rights: A Theoretical and Empirical Assessment of Extractive vs. Non-Extractive Sectors

The consequences of multinational corporations (MNCs) for human rights protection are poorly understood. We propose that the motives and behaviour of MNCs vary across industries. Extractive firms go where the resources are, which creates a status quo bias among them when it comes to supporting repressive rulers. Moreover, low skills levels, environmental pollution and exploitative motives can fuel tensions which are subsequently suppressed by governments. The same is not true for non-extractive MNCs, which are less controversial and can exit more easily. Using US foreign direct investment (FDI) broken up into extractive and non-extractive in 114 host countries (1999–2009), we find support for these propositions. Extractive FDI is associated with negative effects on rights, but non-extractive FDI is positive after controlling for a host of relevant factors, including endogeneity. Our results are robust to the use of Extreme Bounds Analysis (EBA) and alternative sample sizes.

4 analysis of FDI and human rights. Blanton and Blanton (2009) show that not all types of firms are attracted by cheap labour and lax regulation. In fact, repressive regimes seem unable to attract foreign investment in high-skilled sectors and it is mostly extractive industries that invest in host countries with low human rights protection. Janz (2018) finds that the presence of multinational corporations in developing countries is positively connected to rights outcomes, but only when FDI is invested predominantly in industries with medium or high skills and technology levels, and with non-exploitative investment motives.
We build on this momentum to further study the effects of FDI on human rights protection across different industry sectors, asking: What is the impact of FDI on human rights in extractive vs. non-extractive industries? We focus on this distinction because there is a longstanding argument, and a wealth of firm-level study evidence, about negative impacts in extractive sectors (e.g. MacDonald & McLaughlin, 2003, Idemudia, 2009Eweje, 2006aEweje, , 2006bEweje, , 2009). Many local social communities and activists are willing to give up on potential economic benefits from FDI and oppose new extractive investment (Mutti et al., 2012;Dashwood, 2014).
Our theory is more differentiated than the common two schools of thought. Our main argument is that extractive FDI is location-bound, so that firms have little incentive to punish repressive host countries; rather, with low-skilled work, environmental pollution and a lack of sincere CSR efforts, firms may not only invest in already repressive host countries, but fuel tensions and aggravate human rights impact further. Empirically, we examine US FDI located within 114 developing countries (1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009), for which disaggregated data are available. We find that FDI in extractive sectors is indeed associated with negative effects on human rights, but nonextractive FDI has positive effects, after controlling for a range of relevant factors, including the possibility of endogeneity. The results are robust to sample size, estimation technique, and Electronic copy available at: https://ssrn.com/abstract=3344832 6 of local communities to demonstrate positive societal impact (Vogel, 2005). If local conflict and political violence arise, firms can disinvest from repressive regimes to avoid reputational damage and punish repressive governments (Barry, Clay & Finn, 2013;Vadlammanati, Janz & Oyvind, 2018). The pro-MNC school of thought therefore proposes that foreign investors play a positive role with important consequences for better human rights protection. This has also been 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 (Scherer & Palazzo, 2011;Matten & Crane, 2005;Westermann-Behaylo, Rehbein & Fort, 2015). A range of macro studies confirm this positive effect of FDI on rights (Cingranelli & Richards, 1999;Apodaca, 2001;Hafner-Burton, 2005;Kim & Trumbore, 2010).
In contrast, critics argue that MNCs can contribute to unrest among local groups by damaging the environment or abusing cheap labour (Christian Aid, 2004). Host governments aiming to provide for a stable investment environment might resort to suppression of the 'masses,' specifically organised labour (Smith, Bolyard & Ippolito, 1999;Hymer, 1979). State suppression of unrest or protests among the population can include violations of civil and political rights, such as the right to life, freedom from torture, and or protection of the security of the person. If firms decide to remain such host countries, they may become complicit in wrongdoings, providing further revenue to such governments (Wettstein, 2010;Clapham, 2006).
While there is a wealth of support for this view from micro studies, macro studies generally fail to provide evidence for a global negative effect of FDI, with few exceptions (Smith, Bolyard & Ippolito, 1990;Clark & Kwon 2018). In fact, many macro studies find no significant link between FDI and human rights whatsoever (Mitchell & McCormick, 1988;Minkler & Sweeney, 2011;Sorens & Ruger, 2012;Cao, Greenhill & Prakash, 2013).
Electronic copy available at: https://ssrn.com/abstract=3344832 These mixed results stem from a fundamental theoretical and empirical shortcoming in most of the macro literature: the use of aggregate FDI data instead of differentiating between industries. Each industry, such as extractive, manufacturing, or services, has a normative, regulative and economic structure to which MNCs adapt their behaviour (March & Olsen, 2006;Scott, 1995). Firms which operate in the same sector will adjust their behaviour in a similar way, which leads to institutional isomorphism and homogenisation (DiMaggio & Powell, 1983;Chand & Fraser, 2006;Griffin & Mahon, 1997). Therefore, as we know from micro studies, industries differ in their use of skills levels, the competitive behavior among firms, or role of branding and CSR efforts (Godfrey, Hatch & Hansen, 2010;Meyer, 2004;Beschorner & Hajduk, 2017;Giuliani & Macchi, 2014). As Blanton (2006, 2009) have pointed out, the composition of FDI has changed over time towards much more diverse investment types; therefore, macro studies which employ aggregate FDI data lack differentiation. In fact, many macro studies on FDI are aware of this shortcoming and recommend for future research that "human rights behavior of multinationals may vary depending on sectoral differences" (Kim & Trumbore, 2010, pg. 732), "the sector invested in, determines whether foreign investment is beneficial for host countries" (Apodaca, 2002, pg. 902), and "[f]uture research might examine particular forms of FDI" (Sorens & Ruger, 2012, pg. 6).
Two macro studies have started incorporating these recommendations. Drawing on international business literature (Dunning 1988, Cohen 2007, Blanton and Blanton (2009) have integrated sectoral differences in their study on the role of human rights in investment decisions, arguing that market-seeking, high-skilled FDI favours host countries with higher respect for human rights. Janz (2018) examined how the presence of FDI in ten different industries can influence human rights protection in the host country. The study found that sectors with higher Electronic copy available at: https://ssrn.com/abstract=3344832 skills levels, especially in market-seeking industries, tend to be positively connected to rights outcomes, because such sectors are more likely to generate long-term positive spill-over effects on the economy and reduce tensions in the population. Janz (2018) and Blanton and Blanton (2009) provided important evidence that industry sectors matter, but neither focuses on the specific distinction between extractive and non-extractive FDI. Substantially extending their sectoral approach, we propose theory on effects of extractive versus non-extractive industries on state protection of human rights.

Theory: Extractive vs. Non-Extractive FDI effects
Under the 'resource-seeking motive', foreign investment in extractive sectors, such as oil or mining industries, requires developing and controlling the sources of supply of raw materials, primary commodities required for production. Extractive FDI requires huge capital investments in extraction, and divestment from host countries during times of political instability is almost impossible because of high sunk costs and the location-bound nature of this type of FDI (Vernon 1971). Therefore, investments in extractive sectors are more vulnerable to issues of domestic security because 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 resource extraction. High sunk costs may thus reduce the incentives of extractive FDI to sanction ruling elites who promise stability by using repression to weaken opposition movements or local unrest.
Moreover, extractive FDI seeking efficiency gains tries to exploit differences in factor endowments to gain economies of scale. In the process, firms may prefer a cheap and pliant 9 workforce. Thus, human capital efficiency and human rights conditions may matter less than in more advanced sectors (Narula & Dunning, 2000). When skills levels of workers are low, and labour is abundant, accusations of exploitation of workers are common (Eweje, 2006a(Eweje, , 2006b).
In addition, extractive industries often face allegations of environmental pollution and risking local people's health, which can lead to protests, sabotage and subsequent security challenges (Mutti et al., 2012;Dashwood, 2014). A prominent example is the Niger Delta, where government security forces used violence against activists and local protesters (MacDonald & McLaughlin, 2003;Idemudia, 2009;Frynas, 1998).
Isolated locations of extractive FDI can exacerbate these problems. Raw materials are often located in remote areas, so that and integration and interaction with the host society can be rather limited (Cohen, 2007, pg. 66-71). Resource-seeking firms have been accused of maintaining a 'silo mentality' and a lack of integration with society other than ensuring the security of their operations (MacDonald & McLaughlin, 2003). It is not surprising that CSR initiatives of extractive industries are often seen as 'greenwashing' for PR purposes, rather than an earnest attempt to address societal impact (Hilson, 2012;Hamann & Kapelus, 2004). It is not surprising that companies in the oil and mining industry tend to overly stress their positive impact on growth, possibly to 'make up' for accusations of harmful societal impacts (Fontanier & Kolk, 2007;Böhling, Murguía & Godfrid, 2019). Indeed, a detailed list of human rights violations cases registered under The Alien Tort Claims Act (ATCA) of 1789 of the US against US firms operating in various developing countries reports that the majority of the cases are indeed associated with extractive investments (see appendix for details).
In contrast, non-extractive FD, for example, service sectors or manufacturing, is more footloose and has fewer barriers to its choice of location (Kobrin, 2009). MNCs can also divide Electronic copy available at: https://ssrn.com/abstract=3344832 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. Skeptics of globalization argue that this footlooseness reduces the ability of a state to control FDI and demand better deals. Moreover, the competition among states to attract this footloose capital may lead to the erosion of a state's tax base since MNCs will demand the best rates (Rodrik, 1997;Stiglitz, 2002). Some non-extractive industries might also rely on cheap labour or might pollute the environment, even if to a lesser degree than extractive FDI. Recall, however, that earlier dependency theories of mal-development argued that FDI is bad because of sunk costs, which is not as prevalent as in extractive industries. Moreover, manufacturing and service FDI likely to be in locations where the 'resource curse' effects of high dissent and bad governance are generally absent. This means that not only is a good environment, such as democracy and good governance, attractive to non-extractive FDI, but this FDI is more free to leave rather than be biased towards the status quo, thereby having to be complicit in human rights violations.
To sum up, extractive multinationals have to go where the resources are, which creates a status quo bias among them when it comes to supporting repressive rulers. In addition, low skills levels, environmental pollution and exploitative motives can further fuel tensions which may be subsequently suppressed by governments. The same is not true for non-extractive MNCs, which can exit more easily, and some non-extractive industries rely on medium and high-skilled work, which fuels less tensions. We therefore propose our main hypothesis: Hypothesis: Ceteris paribus, extractive FDI is connected to worse government respect for human rights, in comparison to non-extractive sectors.

Data & Research Design
Our dependent variable, the human rights performance of states, is from the Cingranelli and Richards Human Rights Database (Cingranelli & Richards 1999). This measure is widely used in similar studies and is available annually from 1981 onwards for 195 countries. The source of information used for coding the index is from the US State Department's annual country reports on Human rights practices and from Amnesty International annual reports. The coding for each variable for each country year is evaluated by at least two trained coders 1 . In this paper, we focus on the "physical integrity rights index" (PIR hereafter), which is an additive index constructed from measures of on 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. The PIR data are based on the human rights practices of governments and any of its agents, such as police or paramilitary forces, which are state agencies.
Our main independent variable of interest is disaggregated FDI from US firms in US$ million provided by the BEA database on Balance of Payments and Direct Investment Position. 2 The BEA defines US direct investment abroad as either ownership or control of 10 percent or more of a foreign business enterprise, voting securities or the equivalent 3 . The data is recorded in historical cost basis and hence cannot be identifiable in constant nor current prices (Blanton & Blanton 2009). The BEA data captures US FDI position abroad disaggregated by industry sector 12 from 1989 onwards. It should be noted that the industry classification of US FDI by the BEA was converted from Industrial Classification (SIC) to North American Industry Classification System (NAICS) in 1998, so that we use the data from 1999 onwards. As mentioned above, we use accumulated FDI stock divided by GDP to capture the relative power of MNCs compared to the size of the host economy. The stock of accumulated investment is a better measure of the historical power of MNCs rather than measures of flow, which are highly volatile and do not capture properly the presence of MNCs in a host economy (Bornschier and Chase-Dunn 1985;de Soysa and Oneal 1999). Investments in extractive sectors include oil, petroleum, mining and minerals. We collect the data on extractive and non-extractive US FDI for 114 developing countries based on the availability of data and controls variables. In particular, we consider the share of extractive investments and non-extractive US investments relative to the host country's GDP to capture the penetration of various forms of US investments. We expect that it is mainly the penetration of extractive investments vis-à-vis non-extractive investments that are be associated with human rights violations in host countries.
We estimate pooled Time Series Cross-Section (TSCS) regressions across a sample of 114 developing countries. The model to be estimated for both economy-wide effects and sectorspecific effects is: Where PIRit represents the Physical Integrity Rights index in country i in year t, Hit denotes hypothesis variables, namely extractive FDI and non-extractive US FDI relative to host country's GDP. In order to control for potential feedback from human rights we lag it by one year. Also, it Electronic copy available at: https://ssrn.com/abstract=3344832 13 is plausible to assume that the effects of FDI on human rights are not contemporaneous. Rather there might be a substantial lag before the actual effects are realized. While, υt are time fixed effects, νi are country fixed effects and ωit is an error term. Like others, we also include lagged dependent variable (LDV) in all our models. There are two reasons for its inclusion. First, a lagged dependent variable 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, Tate & Keith, 1999).
The baseline models are estimated using ordered probit models since the outcome variable is rank ordered (0-8), including time dummies. We also run pooled OLS models (POLS henceforth). The pooled data are susceptible to having highly correlated data between and across panels that could lead to optimistic standard errors (Beck & Katz, 1995). We use the Newey-West method which allows us to compute an AR1 process for autocorrelation and obtain Huber-White corrected robust standard errors, which are robust to heteroscedasticity (Newey & West, 1987). In both these methods (ordered probit and Newey-West POLS) we do not include country fixed effects because some of the variables (ethnic fractionalization and legal heritages) are 'time invariant'. Usage of two-way fixed effects will not only be collinear with time-invariant or largely time-invariant regressors, but also generate biased estimates (Beck, 2001). However, we drop the time invariant variables from our models and perform two-way fixed effects because accounting for unit (country) heterogeneity is an additional robustness check since TSCS results can be sensitive to specification (Wilson & Butler, 2007). We estimate Huber-White corrected robust standard errors, a method which is robust to heteroskedasticity and serial correlation (Wiggins, 1999).
Electronic copy available at: https://ssrn.com/abstract=3344832 14 The vector of control variables (Zit) 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, Tate and Keith (1999) and other comprehensive evaluations on determinants of repression (Landman, 2006). Accordingly, the models control the effects of income by including GDP (logged), measured in constant 2000 US dollars, growth rates, and the log of total population. To measure the political regime in power, we include the regime type score provided by Freedom House. Ongoing conflicts and other terror events in a country affect the degree of state repression (Dreher, Gassebner & Siemers, 2010), so that we include a civil war variable taking 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). We also include the count of the number of years of civil peace so as to distinguish between immediate post-war situations. In addition, we include oil export dependency by computing the share of oil, gas and mineral exports in total exports from the World Trade Organization (WTO). Since oil export dependency explains the so called 'resource curse' (Ross 2001;de Soysa & Binningsbo, 2009;Sachs and Warner 2001) we account for this factor independently of the amount of US FDI. Additionally, we include some time-invariant variables accounting for the degree of fractionalization in a society, since some claim that ethnic and religious differences can precipitate social unrest (Easterly, 2006). The data is from Fearon and Laitin (2003). Finally, we include measures of the legal heritage of countries by including dummy variables, separately for British, Socialist, French, and German legal heritages (La Porta et al., 1998).
It is quite possible that our key explanatory variablesthe US FDI measuresare endogenous to having less human rights violations. That is, it might be that governments committed to respecting human rights can attract more FDI as Blanton (2006, 2009) report. For example, the expectation of instability arising out of dissent and uprising could deter new FDI flows into the country. Not taking this endogeneity into account would induce bias in our estimate of the effect of FDI on human rights. We thus control for reverse feedback effects by replicating the baseline models using the system-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. We are not aware of an instrument variable estimator for an ordinal score dependent variable when the error term is serially correlated and heteroscedastic.
We thus follow Miguel, Satyanath and Sergenti (2004), Eichengreen and Leblang (2008) in estimating linear ordered probability models, which provide consistent estimates. The other reason why we use GMM is because of Nickell bias problem, as OLS tends to be inconsistent to the inclusion of a lagged dependent variables and the presence of fixed effects simultaneously in a short panel like ours (Nickell, 1981).
The GMM results are based on a two-step estimator implemented by Roodman (2006) in Stata 11. 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 the measures of lagged dependent variable, extractive and non-extractive FDI as endogenous and all other variables as strictly exogenous. As before, we include time dummies in the GMM regressions. In order to minimize the number of instruments in the regressions, we collapse the matrix of instruments as suggested in Roodman (2006).

Empirical Results
The results of regression estimate in assessing the impact of US FDI by sector on human rights performance are presented in table 1. In column 1, we find that extractive FDI has a statistically significant negative impact on PIR, net of all the controls. The negative effect suggests that an increase in extractive investments increases the level of violations of physical integrity rights.
These results are robust to the inclusion of a lagged dependent variable (see column 2, table 1).
In both columns, the negative effects are significantly different from zero at 1% level. The negative and statistically significant effect of extractive FDI is also robust to entering legal heritage dummies (see table 1). In column 3 we replace extractive FDI with non-extractive FDI, which shows a positive and significant effect on respect for human rights. These positive effects of non-extractive FDI are significantly different from zero at the 5% level. These effects remain robust when we include the lagged dependent variable (see column 4, table 1). Notice that the non-extractive FDI shows a robustly positive association with respect for human rights across the columns 5 and 6 (table 1). Thus, even when extractive FDI is accounted for, the level of nonextractive FDI has a strong positive effect on rights, net of all the control variables, such as income per capita and the level of democracy. These positive effects of non-extractive FDI and negative effects of extractive FDI when included together in the same models remain significantly different from zero at the 5% and 1% levels respectively. We come back to the issue of measuring the substantive impact from the ordered probit models later. Overall, the results support the proposition that MNCs operating in extractive sectors do not seem to have the same incentives and motives for punishing bad governments; rather, extractive activity tends to be associated with worse human rights.
Electronic copy available at: https://ssrn.com/abstract=3344832 ==== Table 1 about here ==== To understand the magnitude of our results in ordered probit models, we compute the marginal effects at the mean of extractive and non-extractive FDIs. It is noteworthy that marginal effects in ordered probit are not straightforward to interpret. We follow Dreher, Gassebner and Siemers (2010) and compute estimated probabilities before and after a shock of one standard deviation of extractive and non-extractive FDIs on PIR reported in table 2. Accordingly, nonextractive FDI has a higher impact compared to extractive FDI. The estimated probability of observing the PIR index values of 1, 2 and 3 (at the mean of all variables) are 0.7%, 5% and 11%, respectively, while index values 6, 7 and 8 occur with a predicted probability of 19%, 7% and 0.3%. However, after an increase in extractive FDI by one standard deviation these predictions get substantially higher, i.e., 1.0%, 6.3% and 13.2% for low PIR index values 1, 2 and 3. While they decrease for the high PIR index values 6, 7 and 8 to 15.4%, 4.5% and 0.1%, respectively (see table 2). Similarly, after an increase in non-extractive FDI by one standard deviation the predictions get marginally lower for low PIR index values, i.e., 0.5%, 4.1% and 9.8% for 1, 2 and 3, while they increase for the high PIR index values 6, 7 and 8 to 20%, 8% and 0.3%, respectively (see table 2). These effects are not only statistically significant, but also quantitatively important.
The results of OLS regressions with Newey-West standard errors and two-way OLS fixed effects are presented in table 3. As seen there, the basic results remain the same as that reported for ordered probit regressions. The effect of extractive FDI is negative on government respect for human rights. A unit increase (1 percent) in the extractive FDI as a share of GDP reduces Electronic copy available at: https://ssrn.com/abstract=3344832 government respect for physical integrity by roughly 2.11%. A standard deviation increase in extractive FDI/GDP would decrease PIR by roughly 0.20% which is about 10% of a standard deviation of PIR (see column 1, table 3). These negative results remain significantly different from zero at the 1% level even when including a lagged dependent variable (see column 2, table 3). The non-extractive FDI as a share of GDP has a non-negligible effect, where a standard deviation increase in the non-extractive FDI/GDP increases respect for human rights by 9% which is roughly about 5% of a standard deviation of PIR, and by almost 40% at the maximum value of non-extractive FDI/GDP (see column 3, table 3). These results also remain robust to the inclusion of a lagged dependent variable (see column 4). In columns 5 and 6 we estimate both extractive and non-extractive FDIs jointly. There remains a negative effect of extractive FDI and a positive impact of non-extractive FDI simultaneously, which are significantly different from zero at 1% and 5% level. The results remain robust to the inclusion of a lagged dependent and all other control variables. Finally, when estimating the two-way fixed effects where the timeinvariant variables drop out (see column 7 and 8 in table 3), we find that the effects of extractive FDI turn insignificant, while the positive effects of non-extractive FDI remain significantly different from zero at 5% level, even after including the lagged dependent variable. Thus, even when estimating the within-country variance, the hypothesis that non-extractive US FDI has positive effects on human rights is upheld. Naturally, when it comes to the issue of extractive FDI and human rights, it is the cross-sectional effects that are theoretically interesting, thus the insignificant effect when estimating the within country effects is unproblematic.
Interestingly, with the ordered probit and the OLS regressions, the control variables are highly consistent with those reported by others. There is a positive relationship between economic development (per capita GDP) and human rights. The results on the rate of economic Electronic copy available at: https://ssrn.com/abstract=3344832 growth largely remain insignificant. Like others, we find that countries with a larger population have more violations of rights. This effect is consistent across the methods displayed in all models. We find significant positive effects of ethnic fractionalization on human rights in all the methods, results that are consistent with those who argue that high fractionalization make states safer (de Soysa & Binningsbø, 2009;Landman & Larizza, 2009). Conflicts cause higher violations of human rights as others also report (Poe & Tate, 1994;Poe, Tate & Keith, 1999;Dreher, Gassebner & Siemers, 2010). Likewise, civil peace years are associated with lower incidences of human rights abuse. With respect to legal heritage, relative to Scandinavian legal origin, we find that other legal origins remain insignificant, while countries with German legal origin show positive effects on human rights. The variable for oil exporters in general is largely insignificant when US FDI in extractive activity is in the model. As expected, in all the models irrespective of estimation technique, democracy is associated with less human rights violations.
Note that our democracy indicator is from Freedom House, which reports low values for democracy and higher values for autocracies. Interestingly, our main results on extractive and non-extractive FDIs remain highly significant despite the inclusion of several statistically highly significant controls.

==== Tables 2 and 3 about here ====
Next, we examine our models controlling for possible endogeneity between human rights and types of FDI. As discussed earlier, we make use of System GMM method to control for reverse causality. Table 4 presents the results of the models estimated using GMM. The Hansen test and the Arellano-Bond test do not reject the GMM specifications at conventional levels of significance across the columns. The Hansen J-Statistic clearly shows that the null-hypothesis of Electronic copy available at: https://ssrn.com/abstract=3344832 exogeneity cannot be rejected at the conventional level of significance. As can be seen, extractive FDI is negative and significantly different from zero at the 1% level in all the columns. Notice that even after controlling for potential feedback from PIR, the coefficient value of column 1 of table 4 reduced from -0.91 in the baseline models in table 1, to -1.63. Likewise, in column 2, we find that non-extractive FDI has a significant positive impact on PIR at the 10% confidence level. Finally, in the column 3 we find the results to be robust when both types of FDI are estimated jointly (see table 4). These results highlight an interesting point. First, they show that the size of the coefficient for both forms of FDI are marginally higher in GMM than in non-GMM regressions, that is, the values are higher when the potential feedback effect of human rights on FDI is controlled. The effects are also substantively important. A standard deviation increase in extractive FDI leads to a decrease in PIR by 0.15 points in comparison to 0.09 points in non-GMM estimations. Likewise, a standard deviation increase in non-extractive FDI leads to increase in PIR by 0.10 points in comparison to 0.05 points in non-GMM estimations. This is clearly not negligible given that the PIR is coded on a 0-8 scale, with a standard deviation of 2.1.

Checks on Robustness -Further disaggregation of non-extractive FDI
We examine the robustness of our main findings in the following ways. First, it is noteworthy that our non-extractive FDI measure is made up of investments in manufacturing and service sectors. We thus break up the non-extractive FDI by industry, i.e. manufacturing and service FDI, to further assess the propositions. We find that FDI in both manufacturing and services are positively associated with human rights protection compared to extractive FDI which continues Electronic copy available at: https://ssrn.com/abstract=3344832 to show a negative relationship with the human rights. Although we find positive effects of manufacturing and service sector FDI on human rights, the results for manufacturing FDI become insignificant once we control for endogeneity using GMM. However, the significant positive effects of service sector FDI are upheld in GMM estimations along with the negative effects of extractive FDI.
Second, it can be argued that the negative effects of extractive FDI can be driven by the large investments by US companies in some of the oil rich countries. To examine the sensitivity of the results on extractive FDI, we drop all the Organization of the Petroleum Exporting Countries (OPEC hereafter) from our sample 4 . Despite dropping these countries from our sample, we still find strong significant negative effects of extractive FDI on human rights violations. These results remain robust to the inclusion of a lagged dependent variable, alternative estimations techniques such as Newey-West OLS and fixed effects and after controlling for endogeneity using GMM. Third, we re-estimate all the models by dropping the variables for which we could not find any statistical significance. These include GDP growth rate and oil exports share. With these modifications in the list of variables, our results on extractive and non-extractive FDI on human rights are upheld and remain robust. These robustness results are not shown here and are provided upon request.

EBA Results
We examine the sensitivity of our main variables (i.e. forms of US FDI) on various permutations and combinations of controls by employing Extreme Bounds Analysis 5 (EBA hereafter) 22 proposed by Leamer (1983) and Levine and Renelt (1992). The EBA enables us to examine whether the proposed variables are robust determinants of human rights, independently of which additional variables are included in the set of control variables. In order to perform EBA estimations we use a similar approach as Levine and Renelt (1992). Thus, the general form of the regression which is usually estimated in EBA is: Where, y is PIR, vector C includes "commonly accepted" explanatory variables, in our case it is the lagged dependent variable. E is a vector containing the variables of interest which includes our measures on FDI. The vector Z contains up to three possible additional explanatory variables (as in Levine & Renelt, 1992) which, according to the broader literature, are related to the dependent variable. We include all our controls in vector Z and  is the error term. The EBA test for a variable in E states that if the lower extreme bound for δE -i.e., the lowest value for δE minus two standard deviations -is negative, while the upper extreme bound for δEi.e., the highest value for δE plus two standard deviationsis positive, the variable E is not robustly related to y. This criterion of Leamer (1983) was criticized by McAleer, Pagan and Volker (1985) and Sala-i-Martin (1997) as being too stringent. Sala-i-Martin et al. (1997) then proposed an alternative criterion based on the cumulative distribution function (CDF) of the estimated coefficients which are significant at the 5% level. We thus follow Gassebner, Lamla and Vreeland (2009) and Dreher, Sturm and Vreeland (2009) in reporting the percentage of the regressions in which the coefficient of the variable in vector E is statistically different from zero at the 5%-level (i.e. % sign column). The CDF (0), according to Gassebner, Lamla and Vreeland Electronic copy available at: https://ssrn.com/abstract=3344832 (2009) represents the proportion of the cumulative distribution function lying on each side of zero. It indicates the larger of the areas under the density function either above or below zero, i.e., whether this happens to be CDF (0) or 1-CDF (0). So CDF (0) always lies between 0.5 and 1.0 (Dreher, Sturm & Vreeland, 2009).
We estimate the EBA with the ordered probit method controlling for time fixed effects with robust standard errors. These results based on close to 1000 regression estimates are reported in three sets in table 5. As seen from table 5, in all three sets we find our FDI measures to be robust determinants of PIR, with CDF (0) being above 0.9. While extractive FDI has a negative impact on PIR, the non-extractive FDI is positive. These results remain the same when included jointly as seen from set 3 from table 5. In addition, most of the control variables are strongly related to PIR.

Conclusion
In this paper, we have argued that most macro studies on the effects of FDI on human rights fail to account for sectoral differences. Existing studies either propose that FDI supports repressive regimes and fuels tensions, or that FDI enhances development and growth, thereby improving human rights. Using aggregate FDI, statistical analyses have produced a mixed bag of results (e.g. Sorens & Ruger, 2012;Apodaca, 2001;Hafner-Burton, 2005;Smith, Bolyard & Ippolito, 1990). Scholars from the business literature have long argued that a micro approach to the effects of multinational corporations on society would be more futile because it allows for differentiation. In particular, industry-level characteristics can mediate firms' human rights Electronic copy available at: https://ssrn.com/abstract=3344832 impact (Fontanier & Kolk, 2007;Giuliani & Macchi 2014). Following two pioneer studies which have taken an industry approach to assess FDI effects on the macro level (Blanton & Blanton 2009; Janz 2018), we have divided FDI into extractive versus non-extractive FDI, arguing that extractive FDI is location-bound and may not punish bad rulers, and even fuel tensions which might require governments to suppress 'the masses'. CSR efforts of oil and mining multinationals are therefore seen as critical (e.g. Idemudia, 2009;Eweje, 2009). In contrast, nonextractive FDI is more footloose and can exit when regimes use political violence against the population, and this form of FDI is not exclusively reliant on low-skilled labour or subject to tensions from environmental damage to the same degree as extractive FDI.
In our analysis, we demonstrated that effects from US FDI located in 114 developing countries (1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009) depend indeed on the industry. Extractive FDI is associated with negative effects on human rights, and non-extractive FDI has positive effects, after controlling for a range of relevant factors, including the possibility of endogeneity, and checking the robustness of these findings using different models. We have therefore provided evidence that industry sectors matter, in particular, when we compare extractive and non-extractive FDI.
For future research, we propose that additional insights from different industries are embedded into the macro literature. For example, Giuliani and Macchi (2014) have argued that the home country factor, i.e. advanced versus emerging country, could further influence the effects of MNCs on human rights. FDI comes increasingly from non-Western countries, such as China, and the global impact from such investment on human rights is still unclear.
For the micro literature, we have provided robust and significant evidence that negative externalities from oil and mining FDI are not just isolated incidents or anecdotal stories, but that a worrisome global pattern across countries and time exists that remains highly relevant for Electronic copy available at: https://ssrn.com/abstract=3344832 25 policy makers and activists alike. Further case-studies on the micro level could examine the pathways discussed here, in particular, how MNCs react and interact with repressive regimes, and what kind of regulation and activism it would take for MNCs to give up their location and disinvest from rogue states.
To sum up, our theory and empirical results have integrated literature from political science, international business studies, and business ethics, which have previously remained separate. We have combined theory and existing insights from micro studies to refine and differentiate arguments from the two schools of thought among macro studies, therefore building a sectoral bridge between the different fields and methodological approaches. Our differentiation between extractive and non-extractive FDI showed that differentiation is key to move the macrodebate forward, while leaving many opportunities for micro approaches to further explain and extend our findings.
Electronic copy available at: https://ssrn.com/abstract=3344832 Notes: Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Electronic copy available at: https://ssrn.com/abstract=3344832   Notes: Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Electronic copy available at: https://ssrn.com/abstract=3344832 Notes: Results based on 298 (set 1) and 300 (set 2) and 378 (set 3) regression combinations respectively estimated using Ordered Probit with time fixed effects. 'Average Beta' and 'Average Standard Error' report the unweighted average coefficient and standard error, respectively. '% Sign.' refers to the percentage of regressions in which the respective variable is significant at least at the 5% level. 'CDF-U' is the unweighted CDF as detailed in the text. The threshold to consider a variable robust is 0.9. 'Lower Bound' and 'upper Bound' give the lowest and highest value of point estimate minus / plus two standard deviations.