Replication Data for: TIP for Tat: Political Bias in Human Rights Trafficking Reporting

Human trafficking affects millions of people globally, disproportionately harming women, girls, and marginalized groups. Yet one of the main sources of data on global responses to trafficking, the Trafficking in Persons (TIP) Report, is susceptible to biases because report scores are tied to political outcomes. The literature on human rights measurements has established two potential sources of bias. The first is the changing standards of accountability --- where more information and increased budgets change the standard to which countries are held over time. The second is political biases in reports, which are amended to comply with the interests of the reporting agency. In this paper, we examine whether either of these biases influence the TIP Reports. As opposed to other country-level human rights indicators, the State Department issues both narratives and scores which incentivizes attempts to influence the scores based on political interests. Using a supervised machine learning algorithm we examine how narratives are translated into scores, if scores are biased, and disentangle whether bias stems from changing standards or political interests. We find that the TIP Report scores are more influenced by political biases than changing standards.人口贩运在全球范围内影响数百万民众，对女性、女童及边缘群体造成不成比例的伤害。然而，作为全球应对人口贩运相关数据的主要来源之一，《人口贩运问题报告》（Trafficking in Persons (TIP) Report）却易受偏差影响，因报告评分与政治结果直接挂钩。人权评估领域的既有研究已明确两类潜在偏差来源：其一为问责标准的动态变化——随着信息增多与预算投入增加，各国所遵循的问责基准会随时间推移发生改变；其二为报告中的政治偏差，即报告内容会被调整以契合报告机构的自身利益。本研究旨在探究这两类偏差是否会对TIP报告产生影响。与其他国家级人权指标不同，美国国务院会同时发布评估叙事与评分结果，这一机制催生了基于政治利益干预评分的动机。本研究借助监督式机器学习算法，分析评估叙事如何被转化为评分，验证评分是否存在偏差，并厘清偏差究竟源自问责标准的动态变化，还是政治利益驱动。研究结果表明，相较于问责标准的动态变化，TIP报告评分受政治偏差的影响更为显著。