Modeling Conflict While Studying Dynamics

xiaocisan · 发表于 2009-11-24 19:29:54

Modeling Conflict While Studying Dynamics:A Response to Nathaniel Beck

Our exchange with Neal Beck and his colleagues (Oneal and Russett, 1997; Beck, Katz,
and Tucker, 1998; Oneal and Russett, 1999a), which culminates in this volume, has been
unusually productive. There has been progress in assessing the influence of economic
interdependence on interstate relations and in addressing methodological problems
associated with the analysis of binary dependent variables. Because we are international
relations scholars, not statisticians, we are primarily interested in the substantive
implications of our dialogue. Consequently, we first present in this response new tests of
liberal theory using data that have only recently become available. These confirm the
pacific benefits of economic interdependence and democracy, whether Beck抯 method,
the General Estimating Equation (GEE), or an alternative correction for temporal
dependence is used. Our basic finding that trade reduces conflict is thus no longer at
issue between us and Beck. In the process of testing liberal theory, we have tried to
become informed consumers of others?statistical expertise, so we conclude by sharing
some thoughts about addressing dependence in logistic analyses of pooled time series.

1 Confirming the Liberal Peace with Various Estimators and New Data

In this section, we report several new tests of the pacific benefits of economic
interdependence for the period 1886-1992. The opportunity for these analyses is
presented by Kristian Gleditsch抯 (2002) recent compilation of data regarding states?
bilateral trade and their gross domestic products (GDPs), 1950-92. Gleditsch has
supplemented the IMF抯 (1997) Direction of Trade statistics and the GDP data in the

1

Modeling Conflict While Studying Dynamics

Penn World Tables (Summers et al., 1995) with information from additional sources.
There are two particularly important advantages to using Gleditsch抯 data. First, it
includes observations from within the Soviet bloc during the cold war梒ases absent
from the IMF data. Second, by consulting additional sources, Gleditsch has been able to
find values for most of the other missing data and, where values are still unknown, there
is greater reason to believe that no trade existed. Thus, the debate regarding missing
values of trade in the Direction of Trade is, for practical purposes, eliminated.1
For the pre-World War II period, we use the economic data we have collected,
which have been discussed in detail elsewhere (Oneal and Russett, 1999b; Russett and
Oneal, 2001). The primary sources of our early trade data are The Statesman抯 Yearbook
(e.g., Epstein, 1913) and the League of Nations (various years), which were carefully
compared to Katherine Barbieri抯 (1999) bilateral trade data. Maddison (1995) provides
estimates of per capita GDPs for a large number of countries, which permit us to estimate
GDPs for many other countries as well. With the publication of Gleditsch抯 economic
data for the post-World War II period and our assembly of the data for the pre-World
War I and inter-war periods, we believe that the state-of-the-art dataset on interstate
commercial relations, 1885-1992, is at hand.
In addition to Gleditsch抯 new data regarding trade-related interdependence, we
also use Zeev Maoz抯 (1999) revised data on militarized interstate disputes (MIDs).
Maoz has shown that past efforts to produce dyadic variables from the Correlates of War
(COW) project抯 state-level MIDs file (Jones, Bremer, and Singer, 1996) are problematic
to a degree. For example, individual states on one side of a multilateral contest may
never have threatened, displayed, or used force against particular states on the opposing
side. Maoz has corrected this problem and others to produce more accurate data for
dyadic analyses. Following Beck抯 suggestion in Chapter 7A, we analyze the onset of
militarized disputes, rather than dispute involvement; but we return to this issue when
discussing our results.
There is little need to discuss our other variables. Their definitions and the
sources of our information have been explained in our earlier contribution to this volume.
We adopt the specification, based on the weak-link assumption, from several recent
publications (e.g., Oneal and Russett, 1999b, Russett and Oneal, 2001). We postulate

2

Modeling Conflict While Studying Dynamics

that the likelihood of the onset of a militarized dispute is a function of the lower
democracy score and the lower bilateral trade-to-GDP ratio (DEPENDL) in a dyad; the
logged ratio of the two states?military capabilities; indicators of whether the pair is
allied, whether they are contiguous, and whether either is a major power; and the
logarithm of the distance separating them. Our model has evolved from that of Bremer
(1992) and Maoz and Russett (1993). Bennett and Stam (2000); Hegre (2000); Mousseau
(2000); Gartzke, Li, and Boehmer (2001); Zeng and King (2001); Heagerty, Ward, and
Gleditsch (2002) among others have used similar specifications.
In Table 8B.1., we present the results of several estimations for the politically
relevant dyads, 1886-1992, using different corrections for dependence in the time series.
We focus on this subset of cases for two reasons. First, Beck concurs with us that
interdependence significantly reduces conflict when all dyads are analyzed; but it is
marginally insignificant when he limits his analysis to the politically relevant dyads, pairs
that are either contiguous or include a major power. This set of cases, therefore, warrants
further analysis. Beck acknowledges, however, that interdependence is significantly
associated with peace for the subset of contiguous dyads. Second, the politically relevant
dyads are the states that suffer an appreciable risk of interstate conflict.2 Indeed, as we
show in Chapter 8 and Beck confirms, the benefit of trade is greatest for those states most
prone to conflict梩he contiguous pairs. We follow Beck抯 example by confining our
discussion to the lower trade-to-GDP ratio (DEPENDL); but in all the tests, the lower
democracy score is very significant and substantively important. We report in Table
8B.1. the reduction in the probability of the onset of a dispute from a one standard-
deviation increase in interdependence, as a means of making the practical importance of
interdependence apparent. As before, this evaluation is made by reference to a 憈ypical?
dyad. For comparison, we present in the last column the reduction in risk associated with
the two states becoming allied.
The first row gives the results of a logistic regression analysis. In this and all our
other tests we report robust standard errors. These allow for unspecified correlation
within the dyadic time series. The estimated coefficient of the trade variable is
significant at the .001 level (one-tailed test). The reduction in the probability of conflict
from increasing the economic importance of trade is 38%, which is a little less than that