Sage Journals: Discover world-class research

Abstract

The relationship between natural resource wealth and civil conflict remains unclear, despite prolonged scholarly attention. Conducting a meta-analysis—a quantitative literature review—can help synthesize this broad and disparate field to provide clearer directions for future research. Meta-analysis tools determine both the aggregate effect of natural resources on conflict and whether any particular ways in which variables are measured systematically bias the estimated effect. I conduct a meta-analysis using sixty-nine studies from sixty-two authors. I find that there is no aggregate relationship between natural resources and conflict. Most variation in variable measurement does not alter the estimated effect. However, measuring natural resource wealth using Primary Commodity Exports and including controls for mountainous terrain and ethnic fractionalization all do significantly impact the results. These findings suggest that it may be worth exploring more nuanced connections between natural resources and conflict instead of continuing to study the overall relationship.

Keywords

Natural resource wealth civil conflict meta-analysis quantitative literature review measurement heterogeneity

Introduction

Many theoretical mechanisms plausibly link natural resource wealth to civil conflict (Ross, 2006). Scholars have devoted considerable effort to understanding both the overall relationship between natural resources and conflict and the effect of any one kind of resource on different measures of conflict (Lujala et al., 2005). Recent research has striven to improve the measurement of natural resource wealth and to theoretically differentiate between different types of natural resources that may be more likely to lead to conflict. While both of these strategies do remove some underlying theoretical and empirical noise, a more systematic stocktaking of existing literature can illuminate robust and persistent regularities that have public policy implications. Practitioners want to prevent future conflict, and any broad themes that can be derived from decades of existing scholarship will help achieve this goal.

Conducting a meta-analysis can help harness the empirical power of previous research and structure further inquiry into the natural resources and conflict relationship. Meta-analyses are quantitative literature reviews that combine the results of existing, similar scholarly work (Stanley, 2001). A meta-analysis allows for the estimate of the aggregate (or overall) effect of natural resources on conflict that can then be corrected for publication bias—a phenomenon wherein research counter to conventional wisdom is less likely to get published. The ensuing moderator analysis is specifically designed to accommodate substantial divergence regarding the best ways to theorize about and measure the independent and dependent variable, as is the case with natural resources and conflict (Stanley, 2008). These meta-analysis tools can, therefore, evaluate both the aggregate effect of natural resources on conflict and highlight any variable specifications that produce systematically different results.

I use data from sixty-nine studies to conduct a meta-analysis of natural resources on conflict. I show that there is no aggregate effect between natural resources and conflict. Studies measuring natural resource wealth using oil, forest resources, and minerals and conflict using onset, incidence, duration, and intensity all produce similarly null effect estimates. However, effect estimates that measure natural resources using Primary Commodity Exports (PCE) and those using controls for mountainous terrain and ethnic fractionalization are fundamentally different.¹ These findings suggest that future research would do well to focus on studying factors that may condition the relationship between natural resources and conflict instead of continuing to develop new techniques to more precisely estimate the presence of an aggregate relationship.

Natural resources, conflict, and meta-analysis

A selective reading of the vast literature on the relationship between natural resources and conflict results in wildly different conclusions depending on which studies are included. Most researchers propose that natural resources and conflict are highly correlated, and both qualitative and quantitative studies of different varieties confirm this proposition (Ross, 2004). The typical theoretical argument provided to link natural resources and conflict is that “resources have ‘fueled’ a given conflict” (Ross, 2004: 340). This implies that increased natural resource wealth causes more conflict.

An alternative hypothesis presented by Bodea (2012) and others is that natural resource wealth gives the government more money to spend on security, either by increasing the size of security forces or by redistributing resource wealth to citizens to reduce their interest in rebelling. This implies that increased natural resource wealth reduces the number and intensity of conflicts.

Strong research designs find a clear positive (e.g., Urdal, 2005), negative (e.g., Bodea, 2012), or no relationship (e.g., Arezki et al., 2015) between natural resources and conflict. Adding to this complexity, scholars are interested in both the overall relationship and specific theoretical mechanisms that may explain why certain types of natural resources influence certain measures of conflict. Lootable resources, for example, may provoke conflict while non-lootable resources have no effect (Lujala et al., 2005).

Many studies include either oil or PCE as the independent variable and civil war onset as the dependent variable. However, disparate findings using these specifications have led to broadening definitions of natural resources (including forest and mineral resources) and of conflict (including incidence, intensity, and duration). Some believe that this heterogeneity makes these studies inherently incomparable. Additionally, recent developments such as Geographic Information System (GIS) technology and more robust statistical methods have led some to suggest that only certain studies are high enough quality to inform the relationship between natural resources and conflict.

Conducting a meta-analysis will help us better understand both the overall relationship between natural resources and conflict and whether any particular methods of measurement or ways in which analyses were conducted influence the estimated effect. Meta-analysis has often been suggested (Zigarell, 2011), but is rarely used in political science. Scholars are wary about including diverse studies that have heterogeneous conceptions of the independent and dependent variable, different units of analysis, and may be of suspect quality (see Supplemental Material (SI.1)). Those meta-analyses that have been conducted limit themselves to incorporating studies of a preferred type—whether published, of subjectively high quality, or that define variables in a certain way. The result is that these meta-analyses cherry pick studies to include.

Lau et al. (1999: 854) concisely address this criticism saying, “diversity is not a problem in meta-analysis as long as such diversity can be coded and taken account in the analysis.” Leveraging diversity is the strength of meta-analysis. Djankov and Murrell (2002) suggest thinking of a meta-analysis as a statistical literature review: studies about similar topics are evaluated with proper consideration of the different ways authors approach a common question. In particular, meta-analysis is well suited to consider studies that differ in the ways they measure and specify their independent and dependent variables, as is the case with the relationship between natural resources and conflict (Doucouliagos, 1995).

Data collection

To collect relevant work about natural resources and conflict for the meta-analysis, I searched Harzing’s “Publish or Perish” and Google Scholar for all papers containing the words “natural resource,” “conflict,” “regression,” and various synonyms published from 1998 to 2017. From the results of this search, I read all abstracts and selected potentially relevant articles. I cross-referenced these articles’ citations to develop the most complete and systematic search of the literature. Relevant studies needed to contain regression analysis with a natural resource measure as an independent variable and a conflict measure as the dependent variable; some studies used more sophisticated designs that had to be excluded because of non-comparability (see Supplemental Material (SI.2)). The unit of analysis had to be at either the country-year level or the geographic grid square-year level. The final sample includes sixty-nine studies from sixty-two different groups of authors.

Even the most comprehensive computer assisted literature search will miss some important studies on any particular topic (Rosenthal and DiMatteo, 2001). However, “the central findings of meta-analysis are remarkably robust to marginal changes in the population of studies” so we should not be overly concerned with missing a few relevant studies (Stanley and Doucouliagos, 2012: 16, 31).

Using these studies, I code effect sizes using the Fisher z value, which adjusts the $t$ statistic of an estimate by the sample size. The standard measurement for the quality of each effect size estimate is the precision, $P = \frac{1}{SE}$ or the inverse of the standard error. I use these measures to determine the presence of an aggregate effect and to test for publication bias.

I also code a series of moderator variables about each study’s regression specifications. Moderator variables include dummy variables accounting for different definitions of the independent and dependent variables and some controls (see Supplemental Material (SI.3)). Thus, there are dummy variables for whether a particular study measures conflict onset, incidence, intensity, or duration and for natural resource wealth measured using PCE, oil, timber, or minerals. I also code commonly used controls whose inclusion is believed to theoretically influence the estimated effect. Moderator variables are relatively broadly defined with the goal of capturing theoretical differences in the ways in which different scholars develop regression specifications. Subtle disagreements about how to measure a particular variable are ignored due to limited degrees of freedom and myriad possible minor differences in measurement (Stanley and Doucouliagos, 2012: 30). I test whether moderator variables systematically bias the estimated effect. This is an indicator of whether different measurement strategies are capturing the same underlying relationship between natural resources and conflict.

Analysis

I introduce and apply several meta-analysis techniques that allow me to aggregate and analyze the heterogeneous ways in which the natural resources-conflict relationship is studied. Influence analysis ferrets out estimates using incompatible variable definitions and those of substantively different quality. Tests for publication bias address remaining research quality concerns by correcting the estimated aggregate effect based on unpublished work. Moderator analysis quantifies whether the remaining variation in variable measurement is significant. These analyses can be easily performed using common statistical software.² Supplemental Material (SI.4) contains more details about each step in the analysis.

Influence analysis

An influence analysis helps determine whether any one estimate substantively changes the estimated effect of natural resources on conflict. Influential estimates are too far afield to be included in the meta-analysis, especially if such estimates use different measurement strategies from the mode (Stanley and Doucouliagos, 2012: 42).

The influence analysis involves running standard tests for influential observations on estimate $t$ values and precision (Viechtbauer and Cheung, 2010). Influential estimates do tend to measure natural resources and conflict differently than most studies. Measures using open forest cover per capita, gold production per capita, and conflict related deaths have low precision, but extremely high positive estimated effect sizes. Though all valid approaches, forest cover is quite different from traditional measures of forest timber resources and gold production is only a subset of traditional measures of mineral wealth. Additionally, conflict deaths are not necessarily the same as conflict intensity. These conceptualizations of natural resources and conflict are qualitatively different than the others in the analysis, warranting their exclusion.

Note, however, that almost all of the heterogeneity in variable measurement remains after removing influential observations. Conflict is measured by onset, intensity, incidence, and duration; natural resource wealth is measured by PCE, timber, mineral, and oil resources. The heterogeneity that remains is beneficial to the analysis because it allows me to examine the impact of different modeling choices on the aggregate effect. Measuring only the effect of oil on conflict onset cannot answer broader questions about the underlying relationship between natural resources and conflict.

Publication bias and aggregate effect

Traditional meta-analyses often produce a funnel plot showing the distribution of effect size estimates plotted with their precision. This method is effective at discerning the aggregate effect only in the absence of publication bias. To avoid publication bias, some scholars select what they believe are the highest quality and most unbiased studies to include in the meta-analysis. Analyzing only these selected studies is counterproductive because it trades measurable publication bias for uncontrolled researcher bias.

Three regression tests have been developed in economics to determine the presence of publication bias in a meta-analysis and, further, to indicate if there is a genuine aggregate effect when accounting for publication bias. These tests are considered the standard for detecting and correcting for publication bias (Lakens et al., 2016). I introduce these methods below, suggesting that their use is preferable to attempts to include only studies deemed to be of high quality.

Funnel Asymmetry Test (FAT)

The FAT uses regression analysis to determine if an estimated effect is positively correlated with its standard error (Costa-Font et al., 2011). This is a test of publication bias because we expect that estimates and their standard errors are independent when publication bias is not present.³ We are concerned, therefore, with determining the relationship between estimated effect sizes and their standard errors, which can be accomplished by using a weighted least squares (WLS) regression model (Carter and McCullough, 2014).

Equation (1) shows this WLS model where $t_{i}$ is the $t$ value of estimate $i$ , $\frac{1}{{SE}_{i}}$ is the precision of $i$ , $α$ is the intercept, $β_{1}$ is the slope, and $ε_{i}$ is normally distributed random error (Moreno et al., 2009). Publication bias systematically alters or shifts the estimate of the aggregate effect. This can be written empirically using the conditional expectation of a truncated normal distribution, $E ({effect}_{i}) = μ + S E_{i} \cdot λ (c)$ , where $λ (c)$ is the inverse Mills ratio and $μ$ is the center of the distribution without publication bias. The Mills ratio “censors” the normal distribution such that it skews the underlying aggregate effect by adding publication bias. Heckman developed the Heckman correction as a way to statistically adjust for the Mills ratio’s bias. The Heckman correction can be estimated using a Taylor polynomial, giving us

t_{i} = α + β_{1} (\frac{1}{{SE}_{i}}) + ϵ_{i}

(1)

We can test the hypothesis that there is no publication bias by testing whether estimates and standard errors are independent, which is equivalent to testing whether a funnel graph is symmetric. That is, $H_{0}$ : $α = 0$ . Intuitively, $t_{i}$ is the $t$ value or the estimate divided by the standard error. By conducting a $t$ test on $α$ , we are testing $e f f e c t_{i} = α \cdot {SE}_{i}$ where $\frac{e f f e c t_{i}}{{SE}_{i}} = t_{i}$ . If $α$ is not zero, there is a systematic relationship between the standard error and the effect estimate, indicating non-independence and publication bias (Costa-Font et al., 2011).

For the natural resources and conflict data, I use Equation (1) to produce model 1 in Table 1. The intercept or constant term, $α$ , is significant and positive, indicating publication bias in the estimates and that the estimated effect of natural resources on conflict is overly positive.

Table 1.

FAT-PET and PEESE tests for natural resources and conflict.

	Dependent variable:
	$t$ value
	(1: FAT-PET)	(2: PEESE)
Constant ( $α$ )	1.268*** (0.105)
Precision ( $β_{1}$ )	−0.003*** (0.001)	0.0001(0.001)
SE ( $β_{2}$ )		0.0002(0.001)
Observations	497	497
R ²	0.018	0.0002

p<0.1; **p<0.05; ***p<0.01. Model 1 detects publication bias via the significance of the intercept and determines if a true effect exists by the significance of precision. Precision in model 2 provides the best estimate of the aggregate effect. R² shows that the overall effect is substantively unsubstantial.

Precision Effects Test (PET)

After rejecting the FAT’s null hypothesis of no publication bias, we are interested in determining if there is an aggregate effect after correcting for publication bias. This can be done by testing whether the precision is significantly related to the $t$ value. In essence, we subtract out the constant term representing the publication bias from the $t$ score and determine if the resulting $t$ score is related to the precision (Stanley and Doucouliagos, 2012: 62; Costa-Font et al., 2011).

The PET’s null hypothesis is that there is no aggregate effect when looking at the significance of the coefficient of precision in Equation (1). In our case, the coefficient of precision in Table 1, model 1 is significant and negative. This means that there is an aggregate negative effect of natural resources on conflict. That said, we still have no idea whether this aggregate effect is practically meaningful, nor have we developed an estimate for the size of the effect. The PEESE model does just that.

Precision-Effect Estimate with Standard Error (PEESE)

Equation (1) produces a biased estimate for the aggregate effect of natural resources on conflict because it suffers from publication bias even when corrected (Carter and McCullough, 2014). If we know that there is an aggregate effect from the PET, we can use the PEESE to determine its magnitude. Equation (2) presents the PEESE model:

t_{i} = β_{1} (\frac{1}{{SE}_{i}}) + β_{2} {SE}_{i} + ϵ_{i}

(2)

When an aggregate effect exists, Stanley and Doucouliagos (2014) and Carter and McCullough (2014) show that the PEESE provides the best available estimate of that aggregate effect.

The PEESE model differs from the PET model in that it regresses the effect size on the squared standard error. To see this, recall that $t_{i} = \frac{e f f e c t_{i}}{{SE}_{i}}$ , so we can multiply Equation (2) through by ${SE}_{i}$ . The thinking behind the development of the PEESE is that studies with higher precision will exert more influence on the fit of Equation (2). As the standard error of estimates increases, publication bias gradually increases such that the relationship between the effect and the standard error develops into a parabola, as is modeled in Equation (2). This parabola will fit the data better than the line estimated in the PET because it accounts for increasing levels of publication bias as the effect size increases.

It is important to note that much of the research behind publication selection and estimating aggregate effects is based on the general theory of the FAT and PET and then refined using simulations. Though the PEESE does have the above explanation for why it is a good estimator of the aggregate effect, its power comes through its ability to outperform other methods in simulations.

In our case, we notice that Table 1, model 2 produces Equation (2) for the natural resources and conflict data. The coefficient on precision is $0.00001$ , substantively insignificant. Thus, removing the publication bias reveals that there is no aggregate relationship between natural resource wealth and conflict.

Moderator analysis

While the aforementioned methods correct for publication bias and provide an estimate of the aggregate effect, moderator analysis helps disentangle whether any particular model specifications systematically cluster together. Moderator variables that consistently and significantly correlate with Fisher z values are considered influential. Influential moderators signal that studies including these moderators are significantly different compared to studies excluding them. Once influential moderator variables are flagged, we can return to theoretical explanations to determine why the inclusion of said moderators might bias the aggregate effect.

In determining whether a moderator variable is influential, robustness is key. No perfect model specification exists, so utilizing many approaches is the best strategy (Feld and Heckemeyer, 2011). I use a general to specific approach that involves starting by regressing all moderator variables on the Fisher z value and removing the least significant moderator until all remaining variables are significant. The main model is a simple WLS regression with author fixed effects and weights of $\frac{1}{{SE}^{2}}$ . I check the robustness of these results with three other model specifications, individual regressions for each moderator variable, and non-general to specific approaches (see Supplemental Material (SI.4)).

Four moderator variables are consistently significant across these model specifications. PCE is the only independent or dependent variable specification that systematically biases effect estimates. Using PCE significantly decreases the estimated effect size. PCE is a broad category, capturing more than just natural resources. It includes exports of food, beverages, raw materials, metal, oil, and other energy forms. Not all of these components are traditionally considered “natural resources,” especially food and beverage exports. Further, exports do not fully account for primary commodity production, which does not fully account for the wealth of natural resources in a country. The significance of PCE also means that the heterogeneity in measuring conflict and minerals, oil, and timber all produce substantively similar estimates despite different theoretical mechanisms.

Two control variables are also influential: mountainous terrain and ethnic fractionalization. Including rough terrain as a control increases the estimated effect size. Mountainous terrain may be multicollinear with natural resources such that including it as a control changes the estimated effect. We can imagine that oil and mineral deposits are high in many mountainous areas; accounting for this relationship could alter the estimate of the effect of natural resources on conflict.

Ethnic fractionalization decreases the estimated effect size. Consistent with recent evidence from Wimmer et al. (2009), ethnicity and disputes between ethnic groups might be driving conflict and natural resources may interact with ethnicity to exacerbate this effect.

Finally, a dummy variable for whether the study was peer reviewed increases the estimated effect. This is consistent with the publication bias tests that show the tendency for peer reviewed articles to produce higher effect estimates compared to working papers.

Discussion and conclusion

This meta-analysis finds that there is no aggregate effect of natural resources on conflict. I introduce several econometric techniques to show that peer reviewed work tends to estimate a positive effect when none exists. Influence analysis assures that these results are not unduly biased by a small number of studies. The theoretical mechanisms linking different types of natural resources and kinds of conflict may lead to systematically biased estimates. In general, I show that this is not the case. PCE does emerge as a natural resource variable unlike other, more traditional methods of measuring resource quantity or production.

The significance of the mountainous terrain and ethnic fractionalization moderator variables suggests the presence of a more nuanced relationship between natural resources and conflict. Humphreys (2005), Wegenast and Basedau (2014), Wimmer et al. (2009), and others have highlighted the conditioning effect that ethnic diversity and state presence (which could be proxied using mountainous terrain) can have on the relationship between natural resources and conflict.

Many researchers are leveraging increased technological capacity to study the relationship between natural resources and conflict using more precise data and smaller units of analysis. The results of this meta-analysis suggest that a more productive avenue for future research is to return to theory and think carefully about conditioning effects. Further investment in estimating the aggregate effect is unlikely to change the overall non-relationship between different types of natural resource wealth and different measures of conflict.

The meta-analysis techniques introduced here allow researchers to obtain a comprehensive picture of complex and heterogeneously studied relationships. Not only can we estimate an aggregate effect between two variables and correct that estimate for publication bias, but we can also utilize the full range of model specifications used by different scholars to determine if any one systematically biases the estimated effect. Topics prone to debates over statistical models and variable definitions are now prime targets for meta-analysis.

Supplemental Material

Supplemental_Information – Supplemental material for A meta-analysis of natural resources and conflict

Supplemental material, Supplemental_Information for A meta-analysis of natural resources and conflict by William O’Brochta in Research & Politics

Footnotes

Acknowledgements

I especially thank and am extremely appreciative of the guidance and support I received from T.D. Stanley. Guillermo Rosas and Santiago Olivella helpfully assisted with methodological advice. Editor-in-Chief Kristian Skrede Gleditsch, Matthew Gabel, and three anonymous reviewers provided valuable suggestions.

Declaration of conflicting interest

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

William O’Brochta

Supplemental materials

The supplemental files are available at http://journals.sagepub.com/doi/suppl/10.1177/2053168018818232. The replication files are available at

Notes

Carnegie Corporation of New York Grant

This publication was made possible (in part) by a grant from the Carnegie Corporation of New York. The statements made and views expressed are solely the responsibility of the author.

References

Arezki

Bhattacharyya

Mamo

(2015) Resource discovery and conflict in Africa: What do the data show? Working Paper WPS/2015–14, Center for the Study of African Economies, Oxford, UK.

Bodea

(2012) Natural resources, weak states and civil war: Can rents stabilize coup prone regimes? Technical report (Working paper). World Bank Policy Research Working Paper No. 6071. Washington, DC: World Bank, pp. 1–40.

Carter

McCullough

(2014) Publication bias and the limited strength model of self-control: Has the evidence for ego depletion been overestimated? Frontiers in Psychology 5. Available at: https://doi.org/10.3389/fpsyg.2014.00823

Costa-Font

Gemmill

Rubert

(2011) Biases in the healthcare luxury good hypothesis?: A meta-regression analysis. Journal of the Royal Statistical Society: Series A (Statistics in Society) 174(1): 95–107.

Djankov

Murrell

(2002) Enterprise restructuring in transition: A quantitative survey. Journal of Economic Literature 40(3): 739–792.

Doucouliagos

(1995) Worker participation and productivity in labor-managed and participatory capitalist firms: A meta-analysis. Industrial and Labor Relations Review 49(1): 58–77.

Feld

Heckemeyer

(2011) FDI and taxation: A meta-study. Journal of Economic Surveys 25(2): 233–272.

Humphreys

(2005) Natural resources, conflict, and conflict resolution: Uncovering the mechanisms. Journal of Conflict Resolution 49(4): 508–537.

Lakens

Hilgard

Staaks

(2016) On the reproducibility of meta-analyses: Six practical recommendations. BMC Psychology 4(1): 1–10.

10.

Lau

Siegelman

Heldman

, et al. (1999) The effects of negative political advertisements: A meta-analytic assessment. American Political Science Review 93(4): 851–875.

11.

Lujala

Gleditsch

Gilmore

(2005) A diamond curse?: Civil war and a lootable resource. Journal of Conflict Resolution 49(4): 538–562.

12.

Moreno

Sutton

Ades

, et al. (2009) Assessment of regression-based methods to adjust for publication bias through a comprehensive simulation study. BMC Medical Research Methodology 9(2). Available at: https://doi.org/10.1186/1471-2288-9-2

13.

Rosenthal

DiMatteo

(2001) Meta-analysis: Recent developments in quantitative methods for literature reviews. Annual Review of Psychology 52: 59–82.

14.

Ross

(2006) A closer look at oil, diamonds, and civil war. Annual Review of Political Science 9(1): 265–300.

15.

Ross

(2004) How do natural resources influence civil war? Evidence from thirteen cases. International Organization 58(1): 35–67.

16.

Stanley

(2001) Wheat from chaff: Meta-analysis as quantitative literature review. Journal of Economic Perspectives 15(3): 131–150.

17.

Stanley

(2008) Meta-regression methods for detecting and estimating empirical effects in the presence of publication selection. Oxford Bulletin of Economics and Statistics 70(1): 103–127.

18.

Stanley

Doucouliagos

(2012) Meta-Regression Analysis in Economics and Business. New York: Taylor and Francis.

19.

Stanley

Doucouliagos

(2014) Meta-regression approximations to reduce publication selection bias. Research Synthesis Methods 5(1): 60–78.

20.

Urdal

(2005) People vs. Malthus: Population pressure, environmental degradation, and armed conflict revisited. Journal of Peace Research 42(4): 417–434.

21.

Viechtbauer

Cheung

MWL

(2010) Outlier and influence diagnostics for meta-analysis. Research Synthesis Methods 1(2): 112–125.

22.

Wegenast

Basedau

(2014) Ethnic fractionalization, natural resources and armed conflict. Conflict Management and Peace Science 31(4): 432–457.

23.

Wimmer

Cederman

Min

(2009) Ethnic politics and armed conflict: A configurational analysis of a new global data set. American Sociological Review 74(2): 316–337.

24.

Zigarell

(2011) Of publishable quality: Ideas for political science seminar papers. PS: Political Science and Politics 44(3): 629–633.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.29 MB