Human Rights as a Latent Variable

By Chris Fariss

Keith Schnakenberg and I are working on a paper in which we measure the unobservable level of respect for human rights.  We use the same Bayesian model that Shawn Treier and Simon Jackman use to measure the latent level of democracy in their 2008 paper that was published in the American Journal of Political Science.   As with the construction of GRE scores, the ordinal item-response (IRT) model explicitly models the measurement error that results when different component indicators are aggregated together.

Posterior densities for item discrimination parameters for individual physical integrity rights (300 draws). The item discrimination parameter represent the degree to which the item discriminates between states' along the latent human rights variable. Greater values along the x-axis signify greater discrimination by the item.

The data we use to estimate the IRT model is available from the CIRI Human Rights Data Project.  The model allows us to generate point estimates and credible intervals for the latent variable of interest.

Latent variables estimates for all 192 countries in the CIRI dataset in the year 2007. Blue dots are point estimates (posterior means) and red lines are 95% credible intervals.

For those interested, Simon Jackman has posted several slide-shows on his website that demonstrate the IRT model in action.  In our paper we also demonstrate a simple way to include the uncertainty from the estimates in models that include such a measure as an independent variable.  You can download a copy of our paper at SSRN (we will upload the paper soon).  For now, here is the current version of our abstract:

We use a Bayesian statistical technique to estimate latent human rights variables from the two ordinal Cingranelli and Richards (CIRI) human rights indices (physical integrity rights and empowerment rights).   The ordinal item-response (IRT) model explicitly models the measurement error that results when different component indicators are aggregated together.  The model therefore produces point estimates and credible intervals for latent physical integrity respect and latent empowerment respect.  We then present a simple method that allows measurement error to be included in the uncertainty of causal estimates.  These methods provide new middle ground in the lively human rights measurement debate.  Next, we demonstrate the utility of the new human rights scales by replicating a recent analysis that uses ordinal CIRI human rights variable as two of the main explanatory variables of interest.  We conclude by recommending the inclusion of the new human rights scales in studies that posit a causal effect of human rights abuse on some outcome of interest.  The continuous human rights variable generated as part of this analysis will be made publicly available by the authors.

Advertisements

3 responses to “Human Rights as a Latent Variable

  1. For someone who is more comfortable with “confidence” intervals, can you explain what a “credible” interval is?

  2. Wikipedia has a good article that defines the Bayesian credible interval and compares it to a confidence interval:
    http://en.wikipedia.org/wiki/Credible_interval

  3. Pingback: A Dynamic Ordinal Item Response Theory Model with Application to Human Rights Data | Human Nature Group

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s