Future research could remedy this limitation by sorting agreements to a more detailed level than is possible for the excitement of TIF data. A particularly promising approach could result from the detailed content of individual agreements. Analysis of the text of the agreements would allow researchers to distinguish, for example, bilateral tax treaties aimed at avoiding double taxation, which are often concluded in the form of contracts, and other types of tax treaties. The databases necessary to carry out these studies exist, footnote 105, but it has not been possible to consistently read and classify several thousand international agreements. Recent advances in computerized text analysis could prove successful in overcoming this restriction. Indeed, scientists have already begun to use text analyses to evaluate a limited number of international agreements, such as preferential trade agreements or bilateral investment agreements. B, footnote 107 and the same method can be used to examine international agreements more broadly. Particularly promising methods are thematic modeling and grouping, which allow the researcher to define an unlimited number of categories in which a document can be grouped according to its content. This method would allow the automated and granular categorization of international agreements. A more nuanced categorization, acquired through such an analysis, could further divide our understanding of the role of the treaty and the potential consequences that the abandonment of the treaty would have on the landscape of international agreements of the United States. 98 It cannot be ruled out that the importance of agreements within a category of experts varies more than the importance of agreements between categories of specialists. This possibility is discussed below at 87-88. As explained in the text, as international agreements can be repealed at any time, survival periods are of an ongoing nature.
However, since survival periods are measured only once a year when the TIF is published, the data can be described as continuous data grouped by year. For truly continuous data in which an event can occur at any time, the Cox Proportion Hazard ModelFootnote 117 has emerged as a preferred choice by researchers, notes 118, as it is a semi-parametric model based on a few hypotheses. The popularity of this model is due to the fact that it can be estimated without making parametric assumptions about the base risk rate. For example, the researcher is not obliged to consider that survival decreases over time, exponentially or in other predefined ways at constant speed.
