Getting it (approximately) right (and center and left!): Reliability and uncertainty estimates for the comparative manifesto data

• General Model for estimating reliability and error in text-based policy time-series.
• Standard errors of measurement (SEMs) for Manifesto data for 40 countries.
• Exact Right–Left scale (RILE) errors and reliability.
• Extensions to country party and individual levels (MARPOR website http://wzb.eu).
• Length based error estimates shown as misleading.

The extensive estimates of party policy preferences produced by the CMP (Comparative Manifesto Project, now MARPOR) have proved robust and valid over a wide variety of research applications. But all estimates carry some error. We demonstrate that one of the two existing assessments of non-systematic error in the CMP data strongly overstates their reliability while the other understates it, leaving much potential for mis-estimation. We develop a new method which extends classical test theory and directly estimates overall data reliability; reliabilities and standard errors of measurement for each party system; and standard errors of measurement for each data-point. These should facilitate use of the CMP policy scores which are usually the only ones available for extended party and policy research.