Foreign direct investment and repression: An analysis across industry sectors

ABSTRACT

The impact of foreign direct investment (FDI) on repression in developing nations is still disputed. Some argue that FDI improves economic development and exports human rights values. Others criticize the exploitation of cheap labor and resources, which may lead to tensions and government oppression. Previous studies have employed aggregate FDI data with conflicting results. Alternatively, I propose that the effects depend on what kind of FDI enters a country. I build a sectoral framework to discuss how skills and technology levels, as well as the motivation for FDI, can mediate the impact. I then examine the link in a panel data analysis (1983–2010) in 121 countries, integrating sectoral FDI in several resource, manufacturing, and service industries. The results show that investment in high-skilled and high-tech sectors has positive effects. The results are robust across several measures for repression, and when accounting for sector size, regional and time effects.

 :

•  

Acknowledgments

An earlier version of this article was presented at the 2013 Annual Convention of the International Studies Association. The author is grateful for the comments of the conference participants, as well as valuable suggestions from Christopher Adolph, Robert G. Blanton, Markus Gehring, Pieter van Houten, Nathan M. Jensen, Noel P. Johnston, Todd Landman, Paasha Mahdavi, Peter Nunnenkamp, Alex Sutherland, and three anonymous reviewers. Patricia Abaroa from the US Bureau of Economic Analysis explained the US investment data on numerous occasions; Chris Fariss kindly provided his Latent Scores.

Funding

This research was partially funded by the Centre for International Studies at Cambridge University, Trinity Hall, Cambridge, and the Cambridge European Trust.

Notes on contributor

Nicole Janz is an Assistant Professor at the School of Politics and International Relations at the University of Nottingham. Her research interests include globalization, human rights and corruption.

Notes

1. There are no cross-country measures for direct human rights violations conducted by multinational corporations (MNCs) in host countries. Rather, the existing indices are suitable for assessing the levels of government repression and rights violations by state actors.

2. See reports on the Business and Human Rights Resource Centre Web site, for example, http://business-humanrights.org/en/oil-pollution/human-rights-impacts-of-oil-pollution-nigeria [9 June 2016].

3. Exceptions are Smith et al. (1999) and Timberlake and Williams (1984).

4. One could also make the opposite case and argue that expropriation of MNCs, and subsequent MNC withdrawal, might result in hindering human rights (Johnston et al. 2015).

5. There are no efficiency-seeking and low-skilled FDI industries included in this study due to data availability. I can, therefore, not fully test the skills variation argument within efficiency-seeking FDI.

6. Cingranelli, David L. and David L. Richards. 2008. The Cingranelli-Richards Human Rights Dataset Version 2008.03.12. http://www.humanrightsdata.com/p/data-documentation.html.

7. Political Terror Scale 1976–2010. Retrieved from the Political Terror Scale Web site: http://www.politicalterrorscale.org/ [25 October 2011].

8. Latent Human Rights Protection Scores (Version 2), provided by Christopher Fariss by e-mail. Updates of this measurement are available from http://humanrightsscores.org/. See Fariss (2014) and further also Schnakenberg and Fariss (2014).

9. As Blanton and Blanton (2015) have also noted, data for sectoral FDI invested in developing nations in the form of flow data are not available from, for example, the World Bank, US BEA, UNCTAD, or OECD. Therefore, a robustness check using data for sectoral FDI flow is not possible.

10. Direct investment abroad is defined by the US Bureau of Economic Analysis (BEA) as “[o]wnership or control, directly or indirectly, by one U.S. person, or entity, of 10 percent or more of the voting securities of an incorporated foreign business enterprise or an equivalent interest in an unincorporated foreign business enterprise.” See http://www.bea.gov/glossary/ [9 June 2016].

11. UNCTAD used to provide similar data for a fee; this service is now discontinued. The OECD provides data for sectoral outward FDI but only for OECD host countries and not for developing nations, which are the focus of this study.

12. While some studies (Meyer 1996; Nunnenkamp and Spatz 2004) have argued that US FDI in a host country can serve as proxy for its total inward FDI position, I hesitate to make this claim.

13. Adolph, Quince, and Prakash (2016) demonstrate that a country's labor culture and labor regulation can influence other states via trade relations. In particular, African exports to China can influence labor practices and worker rights in Africa.

14. Some observations for US FDI stock were negative. The reason is that the investment position abroad is measured as the net financial claims that US companies have on their foreign affiliates. A negative position means that US companies are in a net liability position toward their foreign affiliates, which can occur, for example, if the US parent companies received loans from their foreign affiliates (Personal correspondence with BEA, February 13, 2012). In this article, similar to Blanton and Blanton (2009), all negative or zero values of FDI/GDP were recoded to 1 before logging, which then logs to 0.

15. For example, judicial independence (Abouharb, Moyer, and Schmidt 2013) or treaty membership (Brysk and Jimenez 2012; Sandholtz 2012).

16. Data for a greater disaggregation into sector size of each particular industry sector examined here are not available.

17. For the ordered logit models, I use the functions lrm() and robcov() from the R package “rms” Version 4.3–0, which produces the same results as the corresponding STATA command ologit with the cluster() modification (STATA Version 13.0). For OLS with PCSE, I use the functions plm() and vcovBK() from the R package “plm” Version 1.3.1, which produces the same results as the STATA command xtpcse with the pairwise specification.

18. A second, alternative explanation of the negative coefficients for total global FDI refers to the three outcome measures. Only the coefficients for CIRI and PTS are negative and significant, while the latent scores remain insignificant. Only the latent scores correct for more monitoring of violations in recent years. In fact, the robustness section will show that, for the CIRI outcome variable, the negative effect of FDI vanishes once temporal variation is taken into account.

19. In the main models, I use the “polity2” scale, which includes values of −10 (most autocratic) to 10 (most democratic). Davenport and Armstrong (2004) have argued that effects of democracy may only take place after a certain threshold. Following Davenport and Armstrong (2004) and Young (2009), I use a binary composition that inserts dummy variables for each of the democracy levels into the models. I find, consistent with the authors, that there may be a threshold effect above a democracy level of 8 in some cases. The coefficients, signs, and goodness of fit are not affected by this (see Tables A20 and A21 in the appendix).

20. The repression literature has suggested that civil war is much more prevalent and crucial than international war in its impact on rights protection. When replacing my conflict measure, which combines both domestic and international war following Apodaca (2001), with separate variables for major and minor civil wars, the results are not affected (see Tables A18 and A19 in the online appendix).

21. Since global sectoral data located in non-OECD nations are not available from UNCTAD, OECD, or the World Bank, models employ only US FDI in this section.

22. The BEA classified FDI data up until 1998 according to the Standard Industrial Classification (SIC); FDI data starting from 1999 were classified according to the North American Industry Classification System (NAICS). Most sectors employed here remain practically the same and can be examined for the full timeline. However, petrol and mining FDI are available only for the two separate time periods (petrol is only available until 1998; mining is only available from 1999). The BEA notes the following: “For earlier years, petroleum is shown as a separate major industry group because petroleum-related activities accounted for a major portion of all direct investment activity; however, their relative importance has declined significantly in recent years, reducing the need for a separate group.” See: http://www.bea.gov/scb/account_articles/international/0899iid/box2.htm [22 April 2016].

23. US FDI total divided by UN FDI total multiplied by 100 (per country-year); lagged by one year as all independent variables.

24. I thank an anonymous reviewer for making this point. Peter Nunnenkamp and Robert G. Blanton provided helpful feedback on how to deal with this issue.

25. I thank an anonymous reviewer for this suggestion.

26. This is preferred to adding time-fixed effects, which would overspecify the model in light of the small sample size.

27. Investment in food sectors turns, for the two-year lag, to a positive sign and then becomes insignificant. Food is the least stable sector across the robustness checks. An explanation could be that it depends on country characteristics not accounted for by the control variables; when adding country and region dummies, food FDI becomes insignificant in all models.

28. Power analysis for a multiple regression indicates that models including country dummies are least able to detect small effects. For example, for the latent scores as an outcome, the model with country dummies includes 118 predictors and has good power to detect a medium-sized effect, but it has little power to detect small effect sizes (Cohen's f2 = .02), leaving about a 70 percent chance of a Type II error. Since the effect sizes of the sectoral FDI data are rather small across all models employed here, the country dummy model is least favorable.

29. In terms of statistical power, the sample size in the main sectoral models (see Table 4, Models 1–3) is large enough to detect small effects. When sector size controls are included (Table 4, Models 4–6), only medium-size effects can be detected so that chances of missing smaller effects are higher (see power calculations in Tables A30 and A31).

30. I thank an anonymous reviewer for this suggestion.

31. An exception is one model, where wholesale trade has a significant positive coefficient (which was previously mostly negative), that could be explained by the fact that we now have quite different country-year compositions in the sample, including OECD members and not just developing nations, as in all the other models here.