Does government protection of human rights improve? The answer depends crucially on how we model it. Fariss (2014) proposed a statistical approach that produced latent scores suggesting an overall improving trend.
Method: We replicated his analysis but replaced actual data with randomly generated indicators for common violations like torture and political imprisonment, mimicking the effect of his specific weighting strategy.
We found identical improving trends using this fake-data replication. However, when we relaxed Fariss' assumptions to allow all human rights violation types equal influence on the latent scores, that positive trend vanished entirely.
