capacity utilization rates, or concentration ratios and political contributions.22 Unfortunately, the
industry-level time series of data necessary for our analysis is not readily available and would require
a substantial commitment to data collection that is beyond the scope of the current project. Thus
we attempt to account for these political-economic features with our SECTOR, DEFENDANT and
MEASURE controls, with the associated caveat that this is a limitation of the current study.
This data collection approach leaves us with 174 trade dispute observations for which we have
sufficient data for the t − 1 to T + 3 period. Tables 1 and 2 illustrate the summary statistics and
other features of the data used in the estimation.
3.3 Additional Estimation Issues
We conclude this section with a brief discussion of potential omissions from our data set and the
associated implications for the estimation. While we are essentially looking at the relevant population
of dispute settlement activity under the GATT/WTO, this is admittedly only a sample of
GATT/WTO-inconsistent policies that governments undertake. There are surely other examples of
GATT/WTO-inconsistent policies that have not been brought to the dispute settlement provisions
and thus aren’t part of our data set. While Horn et al. (1999) have provided evidence that there does
not appear to buy a bias in trade dispute initiation under the WTO; nevertheless, here we discuss
briefly the characteristics of potential unreported activity and comment on whether its omission is
likely to introduce bias into the estimation and thus affect the interpretation of our results.
First consider the typical characteristics of GATT/WTO-inconsistent activity that would take
place without being reported. A potential plaintiff (i.e., a country facing a deterioration of market
access conditions of a trading partner, relative to what it expected) would not initiate a dispute even
in the face of a GATT/WTO-inconsistent policy if the expected gain from dispute initiation were
smaller than the cost. This is more likely if (i) the gains to the plaintiff from the defendant actually
removing the GATT/WTO-inconsistent policy are small, (ii) the probability of the defendant actually
removing the GATT/WTO-inconsistent policy is small, and (iii) the costs to the plaintiff for pursuing
a case are large.23 We consider the implication of each issue in turn.
First, assume that our data set systematically omits GATT/WTO-inconsistent activity that, if removed, would lead to only small welfare gains for the plaintiff, which is likely if the value of trade
at stake in the potential dispute were very small. It is unlikely that omitting such GATT/WTOinconsistent
activity would bias our results, given that we are estimating the impact of various determinants
on the growth rate of imports in equation (1) and not on the level.
Next assume that our set of trade disputes omits GATT/WTO-inconsistent activity that faces
a low probability of removal. The theory suggests that a potential defendant will be less likely
to remove a policy (and therefore it will fail to liberalize imports) when the costs of the dispute
settlement process are not high enough to induce it to comply with its obligations. If our sample
of data systematically omits observations that are less likely to result in liberalization because the
retaliatory or international obligation costs facing the defendant are too low, this would lead our
results us to underestimate the impact of these variables overall.
Finally consider omissions of GATT/WTO-inconsistent activity due to high costs to the plaintiff
pursuing the case. First note that the pecuniary costs facing a plaintiff for merely initiating a dispute
are not large. However, the political costs for initiating a dispute may be large and particularly
important for poor plaintiffs facing potential defendants on whom they may be reliant for bilateral
aid, military assistance, etc. Furthermore, the pecuniary costs of following through with the litigation
are substantial and may be prohibitive for certain poor plaintiff countries, suggesting endogeneity
between SMALL PLAINTIFF and PANEL GUILT. Nevertheless, because equation (1) is not affected
by income levels, it is unclear that such a data omission would otherwise influence the interpretation
of our results.