¹ The WTO, like international law more generally, does not offer an explicit definition of third parties. See Chinkin, Christine, Third Parties in International Law (Oxford: Clarendon, 1993), 7Google Scholar. Instead, it contrasts third (or “other”) parties with the complainant(s) and defendant and permits these members to “reserve rights,” a status that means they can directly participate in dispute proceedings, as we detail below. The key is not to confuse third parties with co-complainants (that is, members who join the complaint as fellow plaintiffs) or with amicus curiae participants (that is, “friend of the court” submissions by nongovernmental organizations), whose role and effects on dispute settlement outcomes are qualitatively distinct.

² Bagwell, Kyle and Staiger, Robert W., The Economics of the World Trading System (Cambridge: MIT Press, 2002)Google Scholar.

³ Bagwell, Kyle and Staiger, Robert W., “Multilateral Trade Negotiations, Bilateral Opportunism, and the Rules of GATT/WTO,” Journal of International Economics 63, no. 1 (2004)CrossRefGoogle Scholar.

⁴ Carmody, Chi, “Of Substantial Interest: Third Parties under GATT,” Michigan Journal of International Law 18, no. 4 (1997)Google Scholar; Lanye, Zhu, “The Effects of the WTO Dispute Settlement Panel and Appellate Body Reports,” Temple International and Comparative Law Journal 17, no. 1 (2003), 235;Google ScholarLayton, Duane W. and Miranda, Jorge O., “Advocacy before World Trade Organization Dispute Settlement Panels in Trade Remedy Cases,” Journal of World Trade 37, no. 1 (2003), 95–96.Google Scholar

⁵ See, for example, WTO Documents TN/DS/W/38, TN/DS/W/25, and TN/DS/W/41. (This and later citations to official WTO documents use only the unique WTO document number. Most are publicly available at http://docsoniine.wto.org/.)

⁶ See Alter, Karen J., “European Governments and the ECJ,” International Organization 52, no. 1 (1998)Google Scholar; Garrett, Geoffrey, Keleman, R. Daniel, and Schulz, Heiner, “The European Court of Justice, National Governments, and Legal Integration in the European Union,” International Organization 52, no. 1 (1998)Google Scholar; Heifer, Laurence R. and Slaughter, Anne-Marie, “Why States Create International Tribunals: A Response to Professors Posner and Yoo,” California Law Review 93 (May 2005)Google Scholar.

⁷ This is such a strong point of consensus that the solicitor general has been called the “tenth justice.” See Bailey, Michael A., Kamoie, Brian, and Maltzman, Forrest, “Signals from the Tenth Justice: The Political Role of the Solicitor General in Supreme Court Decision Making,” American Journal of Political Science 49, no. 1 (2005)Google Scholar; McGuire, Kevin T., “Repeat Players in the Supreme Court: The Role of Experienced Lawyers in Litigation Success,” Journal of Politics 57, no. 1 (1995)Google Scholar; Johnson, Timothy R., Wahlbeck, Paul J., and Spriggs, James E II, “The Influence of Oral Arguments on the U.S. Supreme Court,” American Political Science Review 100, no. 1 (2006)Google Scholar.

⁸ Clifford J. Carrubba, Matthew Gabel, and Charles Hankla, “Judicial Behavior under Political Constraints: Evidence from the European Court of Justice” (Manuscript, Emory University, Atlanta, Ga., 2006). See also Kelemen, R. Daniel, “The Limits of Judicial Power: Trade-Environment Disputes in the GATT/WTO and the EU,” Comparative Political Studies 34, no. 6 (2001)CrossRefGoogle Scholar; Geoffrey Garrett and James McCall Smith, “The Politics of WTO Dispute Settlement” (Manuscript, UCLA, 2002).

⁹ Rosas, Allan, “Joinder of Parties and Third Party Intervention in WTO Dispute Settlement,” in Weiss, Friedl, ed., Improving WTO Dispute Settlement Procedures: Issues and Lessonsfrom the Practice of Other International Courts and Tribunals (London: Cameron, May 2000)Google Scholar; Smith, James McCall, “WTO Dispute Settlement: The Politics of Procedure in Appellate Body Rulings,” World Trade Review 2, no. 1 (2003)Google Scholar.

¹⁰ Stasavage, David, “Open-Door or Closed-Door? Transparency in Domestic and International Bargaining,” International Organization 58, no. 4 (2004), 682CrossRefGoogle Scholar.

¹¹ Sebenius, James K, “Negotiation Arithmetic: Adding and Subtracting Issues and Parties,” International Organization 37, no. 2 (1983), 308Google Scholar. See also Koremenos, Barbara, Lipson, Charles, and Snidal, Duncan, “The Rational Design of International Institutions,” International Organization 55, no. 4 (2001), 782Google Scholar.

¹² We emphasize that the term “early settlement” does no? count early withdrawal of a complaint if the (protectionist) status quo ante is left in place.

¹³ See also Carrubba, Gabel, and Hankla (fn. 8).

¹⁴ Caldeira, Gregory A. and Wright, John R., “Organized Interests and Agenda Setting in the U.S. Supreme Court,” American Political Science Review 82, no. 4 (1988)CrossRefGoogle Scholar.

¹⁵ Michael Tomz, Judith Goldstein, and Douglas Rivers, “Membership Has Its Privileges: The Impact of GATT on International Trade” (Manuscript, Stanford University, 2005).

¹⁶ This and all subsequently cited figures on WTO dispute participation, escalation, and outcomes are derived from the data set described and analyzed at length below.

¹⁷ In principle, such adoption is not automatic, because a panel report can be rejected if both the complainant and defendant (and the rest of the WTO membership) agree not to approve it, but, needless to say, this is extremely unlikely.

¹⁸ Once again, we emphasize that our term “third party” (and our discussion here of third-party rights in WTO disputes) does not include any entity besides WTO member governments. Amicus briefs, for example, from private interest groups, do not receive the same privileges in WTO dispute settlement proceedings. Hence the impact of NGO amici is beyond the scope of this article.

¹⁹ We are aware of only 4 exceptions to this rule, out of 507 instances of third-party involvement in our data set described below. Moreover, each of these exceptions effectively proves the rule. For example, Thailand's intervention on the side of the complainants in the EC—Bananas dispute (DS158), while initiated only after the panel was established, was also rapidly withdrawn. In the other three instances, the governments in question attempted to join consultations but had their requests to do so denied; however, they later submitted briefs as official third parties at the panel stage (for example, Thailand and the Philippines in Turkey—Textile and Clothing Products, DS34, and the EC in Korea—Government Procurement, DS163).

²⁰ See WTO Document TN/DS/W/36.

²¹ “Complainants” are the plaintiffs of the case. They, not third parties, must file requests for consultations, identifying the disputed policy measure and invoking specific treaty provisions as the basis for challenging it. They also have procedural roles and rights that are distinct from those of third parties. The WTO accordingly distinguishes the two kinds of participants clearly in its documentation of each case.

²² Specifically, only 19 of the 507 requests by nondisputants to join consultations have been rejected. However, many of those rejected—as the examples in fn. 19 make clear—subsequently participate as formal third parties at the panel stage, utilizing the lower bar to such participation granted by the DSU's Article 10.2, described below. Because a third of all cases going to a panel end before a ruling is issued, such late-stage third-party participation can still significantly affect the prospects for early settlement.

²³ WTO Document WT/DS200/8.

²⁴ WTO Document DS200/12.

²⁵ WTO Document DS282.

²⁶ WTO Document DS22.

²⁷ Some validation of this assertion can be found in another statistic, suggested by Hakan Nordstrom. Specifically, certain disputes target policies not addressed toward any one identifiable good (flagged, as defined later, by our variable, Nonmerchandise Issue). In such cases, where third parties by definition have zero affected exports and thus no “substantial trade interest,” third-party cites of “systemic interests” are still no more likely (chi-squared test p=0.61) than they are in more typical cases bearing on particular products.

²⁸ See Porges, Amelia, “Settling WTO Disputes: What Do Litigation Models Tell Us?” Ohio State Journal on Dispute Resolution 19, no. 1 (2003), 158Google Scholar.

²⁹ Busch, Marc L., “Democracy, Consultation, and the Paneling of Disputes under GATT,” Journal of Conflict Resolution 44, no. 4 (2000)CrossRefGoogle Scholar.

³⁰ Bagwell and Staiger (fn. 3).

³¹ Bown, Chad, “Participation in WTO Dispute Settlement: Complainants, Interested Parties and Free Riders,” World Bank Economic Review 19, no. 2 (2005)CrossRefGoogle Scholar.

³² Garrett, Kelemen, and Schulz (fn. 6), 152.

³³ Weiss, Friedl, “Third Parties in GATT/WTO Dispute Settlement Proceedings,” in Denters, E. and Schrijver, N., eds., Reflections on International Lav)from the Low Countries (Cambridge, Mass.: Kluwer Law, 1998)Google Scholar; Covelli, Nick, “Member Intervention in World Trade Organization Dispute Settlement Proceedings after EC—Sardines: The Rules, Jurisprudence, and Controversy,” Journal of World Trade 37, no. 3 (2003).Google Scholar

³⁴ Smith (fn. 9), 75, 85.

³⁵ Rosas (fn. 9).

³⁶ WTO Document DS87, 7.136, emphasis added.

³⁷ WTO Document DS70, 7.47, footnote 4, emphasis added.

³⁸ WTO Document DS114/R, 7.82, emphasis added.

³⁹ Garrett and Smith (fn. 8).

⁴⁰ Kelemen (fn. 8).

⁴¹ Carrubba, Gabel, and Hankla (fn. 8).

⁴² Of course, the main protagonists' own domestic audiences may also induce posturing. Regardless, the issue is that even when settlement is otherwise feasible, the addition of third parties is likely to make negotiations more complicated.

⁴³ Stasavage (fn. 10), 673.

⁴⁴ Fearon, James, “Signaling Foreign Policy Interests: Tying Hands versus Sinking Costs,” Journal of Conflict Resolution 41, no. 1 (1997)CrossRefGoogle Scholar.

⁴⁵ Groseclose, Tim and McCarty, Nolan, “The Politics of Blame: Bargaining before an Audience,” American Journal of Political Science 45, no. 1 (2001)CrossRefGoogle Scholar.

⁴⁶ Porges (fn. 28), 176.

⁴⁷ William Davey warns, in particular, against the recommendation that panels and the AB be required to address all the points raised by third parties, since this “could potentially lead to abuse where a third party views the case completely differently than the parties.” See Davey, “The WTO Dispute Settlement Mechanism,” Illinois Public Law and Legal Theory Research Papers Series, no. 03-08 (University of Illinois, 2003), 15.

⁴⁸ Waincymer, Jeff, WTO Litigation: Procedural Aspects of Formal Dispute Settlement (London: Cameron May, 2002), 339Google Scholar. See also Smith (fn. 9), 85.

⁴⁹ WTO Document DS302/R, 4.351.

⁵⁰ WTO Document DS285.

⁵¹ As Porges astutely observes, “In the absence of a stakeholder with a financial interest in early settlement, the government may have no incentive to have a bargaining position oriented toward settlement”; see Porges (fn. 28), 155.

⁵² See Milgrom, Paul and Roberts, John, Economics, Organization, and Management (Englewood Cliffs, N.J.: Prentice-Hall, 1992)Google Scholar; Hongbin Cai, “Uncertainty and Commitment in Multilateral Bargaining” (Manuscript, University of California, Los Angeles, 2002); Speier, Kathryn E., “Settlement with Multiple Plaintiffs: The Role of Insolvency” Journal of Law, Economics, and Organization 18, no. 2 (2002)Google Scholar; Manzini, Paola and Ponsati, Clara, “Stakeholders in Bilateral Conflict,” Mathematical Social Sciences 50, no. 2 (2005)Google Scholar.

⁵³ Davis, Christina L., “International Institutions and Issue Linkage: Building Support for Agricultural Trade Liberalization,” American Political Science Review 98, no. 1 (2004), 155Google Scholar. As John H.Jackson argues, the WTO treaty texts discourage “compensation” as a solution to a dispute, calling instead for remedies that strictly eliminate the violation itself; see Jackson, , “Editorial Comment: International Law Status of WTO Dispute Settlement Reports: Obligation to Comply or Option to 'Buy Out'?” American Journal of International Law 98, no. 1 (2004)Google Scholar.

⁵⁴ Hudec, Robert E., Enforcing International Trade Law: The Evolution of the Modern GATT Legal System (Salem, N.H.: Butterworth Legal Publishers, 1993)Google Scholar.

⁵⁵ Whether we have data on the policy outcome (that is, these 120 cases versus the 82 with unknown policy outcomes) is not collectively a function of the variables in our analyses to follow, as verified by a probit regression of observability against all our subsequent covariates. The only significant predictor of missing data on the policy outcome is how recently the dispute occurred. The missing data thus introduce no sample selection bias into the early settlement results to follow.

⁵⁶ The data on policy outcomes in these cases are taken from Busch, Marc L. and Reinhardt, Eric, “Developing Countries and GATT/WTO Dispute Settlement,” Journal of World Trade 37, no. 4 (2003)Google Scholar.

⁵⁷ We were unable to code the direction of support for a number of third parties, but only in cases not reaching the ruling stage. Accordingly, missing data for this variable have no effect on the ruling direction model estimated below.

⁵⁸ For more information about sources and coding of these variables, see Busch and Reinhardt (fn. 56).

⁵⁹ See, for example, the contributions to Petersmann, Ernst-Ulrich and Pollack, Mark A., eds., Transatlantic Economic Disputes: The EU, the US, and the WTO (Oxford: Oxford University Press, 2003)CrossRefGoogle Scholar.

⁶⁰ These are calculated for the year the complaint was first filed, using constant (1995) U.S. dollars, from the World Bank's World Development Indicators: http://devdata.worldbank.org/dataonline/.

⁶¹ Palmeter, David and Mavroidis, Petros C., Dispute Settlement in the World Trade Organization: Practice and Procedure, 2nd ed. (Cambridge: Cambridge University Press, 2004), 164.CrossRefGoogle Scholar

⁶² For this and the legal complexity variable below, we draw heavily on the wonderful new World Bank-financed data set, Henrik Horn and Petros C. Mavroidis, The WTO Dispute Settlement Data Set, version 1.0 (January 31, 2006), www.worldbank.org; it identifies each specific legal provision of the WTO agreements cited in each complaint's request for consultations and any subsequent requests for panel establishment. The legal provisions counted in the Discriminatory Measure dummy include GATT Articles I and XXIV; Article V of the General Agreement on Trade in Services (GATS); Article XXIV of the Anti-Dumping Agreement; and the so-called Enabling Clause.

⁶³ Specifically, as cited in the complainant's request for consultations and, where applicable, request(s) for a panel. Number of Articles Cited ranges from 1 to 39, averaging 7, in our data set.

⁶⁴ Conducted by the authors in Geneva, Switzerland, May 2003 and May 2006.

⁶⁵ See Bailey, Kamoie, and Maltzman (fn. 7), 79; Solberg, Rorie Spill and Ray, Leonard, “Capacity, Attitudes, and Case Attributes: The Differential Success of the States before the United States Courts of Appeals,” State Politics and Policy Quarterly 5, no. 2 (2005), 155–56CrossRefGoogle Scholar, 159.

⁶⁶ Davis (fn. 53).

⁶⁷ Porges (fn. 28), 155.

⁶⁸ The product codes affected in each dispute are listed in Horn and Mavroidis (fn. 62). We used the codes to obtain trade volume data at the six-digit level from COMTRADE, unstats.un.org/unsd/com-trade/. Only fifteen disputes involved products at a more detailed level than six digits, so the limits on precision necessitated by reliance on COMTRADE do not skew our trade data. In keeping with standard practice, we replace trade values of zero with tlOOO to feasibly compute logarithms. Fifteen of our 202 disputes challenged export promotion (rather than import limitation) measures. For these cases, Log Merchandise Trade reflects the defendant's exports to the world of the specified product.

⁶⁹ It takes on a value of 100 for disputes not naming a specific merchandise product(s).

⁷⁰ Bown (fn. 31).

⁷¹ The Stata data set, command files, and output log for all the main analyses as well as sensitivity tests below are available from the authors at http://userwww.service.emory.edu/~erein/.

⁷² Because we observe dispute outcomes even when no third party participates, and because third-party intervention itself 'serves as a causal variable affecting dispute outcomes, this counterargument is about endogeneity, not sample selection. For these two reasons, the Heckman selection model is not an appropriate reference point in designing an empirical approach to the problem.

⁷³ Throughout, figures in brackets represent 95 percent confidence interval bounds.

⁷⁴ The only variable (Article XXIT) that is statistically significant in model l's third-party equation and in our later settlement models has a positive effect in both. Thus, the kinds of cases that draw third parties also tend to settle more often, thereby raising, not lowering, the bar for tests of our settlement hypothesis.

⁷⁵ A chi-squared test on these figures yields p < 0.001. A similarly supportive pattern holds for whether a case went all the way to a ruling, as the figure shows. The figure presents the data on ruling issuance chiefly because they are available for the full 202 observations, whereas settlement is not. However, recall that cases ending before a ruling do not universally reflect early settlement: judging from the 120 whose policy outcomes we do observe, a third are withdrawn without substantial concessions by the defendant.

⁷⁶ Note that the data are essentially cross-sectional and do not call for panel techniques. To be sure, the disputes occur in a sequence (albeit a greatly overlapping one), but the eight-year period in the data set is not long enough to exhibit any notable temporal variation. Specifically, a set of year dummies are not collectively significant when added to any of this article's models. Furthermore, the findings about the impact of third parties are qualitatively identical even if models 2,3, and 4 alternatively used Newey-West standard errors robust to first-order autocorrelation across disputes sorted by the WTO'S sequential DS codes.

⁷⁷ As revealed by heteroskedastic probit estimates, the error variances in models 2, 3, and 4 are potentially a function of dispute stakes (specifically, Log Merchandise Trade). Hence we report hetero-skedastic-consistent robust standard errors.

⁷⁸ Auxiliary regressions of each explanatory variable on all the others yield R-squareds around 0.10 to 0.20, with none higher than 0.45 except for the correlated stakes variables, Log Merchandise Trade and Nonmerchandise Issue, at a moderate value of 0.58.

⁷⁹ What is more, Third Party and Systemic Interests retain their negative coefficients and statistical significance in models 2, 3, and 4 if we (1) use trade data from the year of the complaint, instead of from the prior year, for Log Merchandise Trade; or (2) add a dummy variable for antidumping cases, which some say are less amenable to settlement (which, it turns out, is statistically incorrect).

⁸⁰ It is commonly assumed that a regressor's endogeneity automatically biases ordinary estimates. That assumption is wrong. Estimates are biased only if the unexplained portions of the endogenous variables (third parties and early settlement) are correlated; whether this is so is a matter for appropriate testing.

⁸¹ The explanatory power for our instrumental variable is validated by model l's results. Its exoge-neity is likewise supported by its zero coefficient if added to model 3 (noted above) or by its bivariate correlation with Early Settlement of just -0.04.

⁸² We choose model 3 in particular, out of the three settlement models, because its version of Third Party matches that predicted by model 1. To keep matters simple, we drop Systemic Issues from the equation because it is a derivative of third-party involvement.

⁸³ Specifically, because both the first- and second-stage equations (of the number of third parties and early settlement, respectively) are estimated by maximum likelihood, we use Murphy-Topel IV standard errors. Greene, William H., Econometric Analysis, 4th ed. (Upper Saddle River, N.J.: Prentice Hall, 2000), 133–37Google Scholar; Hardin, James W., “The Robust Variance Estimator for Two-Stage Models,” Stata Journal 2, no. 3 (2002), 254–55Google Scholar.

⁸⁴ The Stata program and output used for this IV procedure and test are bundled with the data set available from the authors at http://userwww.service.emory.edu/~erein/. It is also worth noting that, using this IV technique, Third Party's coefficient becomes even more negative than it is in Table 3, model 3.

⁸⁵ Model 7's estimates were produced using Stata 9.2's heckprob command with the difficult technique(bfgs) maximization option.

⁸⁶ With all other variables at their sample means, the marginal probability of a pro-complainant ruling is 0.45 [0.32, 0.60] if the complaint cites just one article and does not make a nonviolation claim. By citing ten articles (not an extreme value) and including a nonviolation claim, however, the complaint's chances of victory drop to just 0.03 [0.00, 0.26].

⁸⁷ Bagwell and Staiger (fn. 3).

⁸⁸ Mavroidis, Petros C., The General Agreement on Tariffs and Trade:A Commentary (Oxford: Oxford University Press, 2005)Google Scholar.

⁸⁹ Marc L. Busch, “Overlapping Institutions, Forum Shopping, and Dispute Settlement in International Trade,” International Organization (forthcoming).

⁹⁰ Tomz, Goldstein, and Rivers (fn. 15).

⁹¹ For example, Caldeira and Wright (fn. 14).

⁹² McGuire (fn. 7); McGuire, Kevin T., “Explaining Executive Success in the U.S. Supreme Court,” Political Research Quarterly 51, no. 2 (1998)Google Scholar.

⁹³ Bailey, Kamoie, and Maltzman (fn. 7); Spriggs, James F. II and Wahlbeck, Paul J., “Amicus Curiae and the Role of Information at the Supreme Court,” Political Research Quarterly 50, no. 2 (1997)CrossRefGoogle Scholar. Carrubba, Gabel, and Hankla (fn. 8) also make this argument.