828 International Organization members of the dyad are in their level of development.But treaties among coun- tries of a similar level of development-especially at the higher end-are consid- erably less likely.In the reported analyses we exclude"developed dyads"from the sample in order to minimize the bias from estimates derived from"irrelevant dyads."44 Otherwise our sample includes all independent states,as identified by Gleditsch and Ward.45 Event history methods offer a convenient way to incorporate time dependence in models of policy or innovation adoption.Our formulation is slightly more com- plicated than most since the unit of analysis is the country dyad and the model includes variables measured for one or the other member of the dyad as well as for the dyad itself.We estimate the following equation: yu.:=axi.+BZi..+8Vu.+pWyi-1+8u where yu is the number of years without a BIT between countries i (host)and j (home),X is a vector of conditions that affect country i's calculations,Z is a vec- tor of conditions that affect country j's calculations,V is a vector of characteris- tics of the relationship between countries i and j,and Wy*is a vector of spatial lag terms in which a count of BITs among other host countries in the previous year (y*)is weighted by various measures of their distance (W)to country i(see our discussion of spatial lags below).We estimate this equation with a Cox pro- portional hazard model,a useful estimator when one does not have strong assump- tions about the effect of time on the baseline hazard. Data and Measures Our dependent variable is the number of years a dyad goes without a treaty,marked by the year of a treaty's signing,rather than the year in which it enters into force.46 We reason that the signing not only approximates the moment during which a government deliberates over the treaty but is also the more important event for 44.We exclude dyads in which both members are classified as "high income countries"by the World Bank [that is,dyads in which both members have a GDP per capita of over $6,000 (in 1987 SUS)in a given year],thus excluding 125 BITs in the analysis. 45.Gleditsch and Ward 1999.Restricting our sample to those states (and their dates of existence) identified by Gleditsch and Ward means that we exclude eighteen BITs listed in the UNCTAD data, such as the fourteen to which Hong Kong was party as well as those involving states such as the United Arab Emirates and Slovenia,which occur several years prior to the dates that Gleditsch and Ward list the states as independent.The relatively inclusive sample ensures that we will encounter missing data,particularly for smaller countries.We experiment with several methods of managing missing data,none of which alter the substantive findings.For the results below,we have estimated missing values of time-varying covariates with decade means where possible and appropriate.Such extrapolations buy a more inclusive sample at the expense of potentially underestimated standard errors; see King et al.2001. 46.While approximately forty dyads have entered into second and,in one case,third treaties,we predict the duration until the first treaty.members of the dyad are in their level of development+ But treaties among coun￾tries of a similar level of development—especially at the higher end—are consid￾erably less likely+ In the reported analyses we exclude “developed dyads” from the sample in order to minimize the bias from estimates derived from “irrelevant dyads+”44 Otherwise our sample includes all independent states, as identified by Gleditsch and Ward+ 45 Event history methods offer a convenient way to incorporate time dependence in models of policy or innovation adoption+ Our formulation is slightly more com￾plicated than most since the unit of analysis is the country dyad and the model includes variables measured for one or the other member of the dyad as well as for the dyad itself+ We estimate the following equation: yij, t
aXi, t bZj, t dVij, t rWyt1 * «ij where yij is the number of years without a BIT between countries i ~host! and j ~home!, X is a vector of conditions that affect country i’s calculations, Z is a vec￾tor of conditions that affect country j’s calculations, V is a vector of characteris￾tics of the relationship between countries i and j, and Wy * is a vector of spatial lag terms in which a count of BITs among other host countries in the previous year ~ y * ! is weighted by various measures of their distance ~W! to country i ~see our discussion of spatial lags below!+ We estimate this equation with a Cox pro￾portional hazard model, a useful estimator when one does not have strong assump￾tions about the effect of time on the baseline hazard+ Data and Measures Our dependent variable is the number of years a dyad goes without a treaty, marked by the year of a treaty’s signing, rather than the year in which it enters into force+ 46 We reason that the signing not only approximates the moment during which a government deliberates over the treaty but is also the more important event for 44+ We exclude dyads in which both members are classified as “high income countries” by the World Bank @that is, dyads in which both members have a GDP per capita of over $6,000 ~in 1987 $US! in a given year#, thus excluding 125 BITs in the analysis+ 45+ Gleditsch and Ward 1999+ Restricting our sample to those states ~and their dates of existence! identified by Gleditsch and Ward means that we exclude eighteen BITs listed in the UNCTAD data, such as the fourteen to which Hong Kong was party as well as those involving states such as the United Arab Emirates and Slovenia, which occur several years prior to the dates that Gleditsch and Ward list the states as independent+ The relatively inclusive sample ensures that we will encounter missing data, particularly for smaller countries+ We experiment with several methods of managing missing data, none of which alter the substantive findings+ For the results below, we have estimated missing values of time-varying covariates with decade means where possible and appropriate+ Such extrapolations buy a more inclusive sample at the expense of potentially underestimated standard errors; see King et al+ 2001+ 46+ While approximately forty dyads have entered into second and, in one case, third treaties, we predict the duration until the first treaty+ 828 International Organization