Other Estimates (Updated March 12, 2014)

The model that generates the UDS also produces a number of other useful quantities. These estimates can assist researchers interested in understanding, and evaluating, the various existing measures that form the foundation of the UDS. The model provides estimates of each rater's reliability, estimates of the extant measures' score cutoffs, all normalized to a single latent scale, and estimates of democracy experts' perceptions of individual democracy scores along the same underlying scale.

We furnish all data in compressed comma separated format. Common software, such as gzip and WinZip, can decompress the files. Users may download the estimates in summary format, or may opt to obtain 1000 or 5000-draw samples from the posterior distributions of the parameters.

Please note that we updated these scores on March 12, 2014. The new scores, while similar to those previously available, reflect a comprehensive update to the UDS that extends the scores through 2012 and adds a new measure of democracy to the underlying analysis. To help users assess the impact of the update, we have replicated figures 4 and 5 from the original article, using the new data. Previous releases of the scores, including the version originally reported in the companion article, are available in our archive.

Rater Reliability

An especially useful byproduct of the model used to generate the UDS is the estimation of the constituent raters' measurement error variance. These estimates provide a metric with which one can judge the relative reliability of the ten democracy scores used in the creation of the UDS.

Error Variance Summary Scores

Error Variance 1000 Draw Sample

Error Variance 5000 Draw Sample

Summary and posterior sample files are formatted as follows:

Summary File

Posterior Sample Files

Column	Description
rater	The measure the error variance comes from
mean	Error variance posterior mean
sd	Error variance posterior standard deviation
median	Error variance posterior median
pct025	Error variance posterior 2.5 percentile
pct975	Error variance posterior 97.5 percentile

Column	Description
rater	The measure the error variance comes from
mean	Error variance posterior mean
sd	Error variance posterior standard deviation
sigmasq1	The first draw from the posterior distribution of the error variance
sigmasq2	The second draw from the posterior distribution of the error variance
...
sigmasqn	The nth draw from the posterior distribution of the error variance

Rater Cutoffs

The democracy measures underlying the UDS all provide ordinal or continuous scores, rating country-years as more or less democratic. Unfortunately, these scales are not directly comparable and it is difficult, for example, to understand what a score of five on the Polity IV scale means on Freedom House's metric. The rater cutoff estimates found here are all normalized to the same underlying latent scale, making these sorts of comparisons possible.

Cutoff Summary Scores

Cutoff 1000 Draw Sample

Cutoff 5000 Draw Sample

Summary and posterior sample files are formatted as follows:

Summary File

Posterior Sample Files

Column	Description
rater	The measure the cutoff comes from
cut	The ordinal level of the cutoff, starting from 1
mean	Cutoff posterior mean
sd	Cutoff posterior standard deviation
median	Cutoff posterior median
pct025	Cutoff posterior 2.5 percentile
pct975	Cutoff posterior 97.5 percentile

Column	Description
rater	The measure the cutoff comes from
cut	The ordinal level of the cutoff, starting from 1
mean	Cutoff posterior mean
sd	Cutoff posterior standard deviation
gamma1	The first draw from the posterior distribution of the cutoff
gamma2	The second draw from the posterior distribution of the cutoff
...
gamman	The nth draw from the posterior distribution of the cutoff

Rater Perceptions

The model underlying the UDS generates estimates of how the constituent raters perceive the level of democracy in given country-years, on the scale of the UDS. These estimates are useful for identifying especially unusual ratings in the extant measures. We provide perception estimates for all 13 of the measures upon which the UDS are based.

Arat Summary Scores	Arat 1000 Draw Sample	Arat 5000 Draw Sample
Bollen Summary Scores	Bollen 1000 Draw Sample	Bollen 5000 Draw Sample
BLM Summary Scores	BLM 1000 Draw Sample	BLM 5000 Draw Sample
EIU Summary Scores	EIU 1000 Draw Sample	EIU 5000 Draw Sample
F. House Summary Scores	F. House 1000 Draw Sample	F. House 5000 Draw Sample
Hadenius Summary Scores	Hadenius 1000 Draw Sample	Hadenius 5000 Draw Sample
Mainwaring Summary Scores	Mainwaring 1000 Draw Sample	Mainwaring 5000 Draw Sample
Munck Summary Scores	Munck 1000 Draw Sample	Munck 5000 Draw Sample
PACL Summary Scores	PACL 1000 Draw Sample	PACL 5000 Draw Sample
Polity Summary Scores	Polity 1000 Draw Sample	Polity 5000 Draw Sample
Polyarchy Summary Scores	Polyarchy 1000 Draw Sample	Polyarchy 5000 Draw Sample
PRC Summary Scores	PRC 1000 Draw Sample	PRC 5000 Draw Sample
Vanhanen Summary Scores	Vanhanen 1000 Draw Sample	Vanhanen 5000 Draw Sample

Summary and posterior sample files are formatted as follows:

Summary Files

Posterior Sample Files

Column	Description
country	Country name
year	Observation year
cowcode	Correlates of War project country code
mean	Rater perception posterior mean
sd	Rater perception posterior standard deviation
median	Rater perception posterior median
pct025	Rater perception posterior 2.5 percentile
pct975	Rater perception posterior 97.5 percentile

Column	Description
country	Country name
year	Observation year
cowcode	Correlates of War project country code
mean	Rater perception posterior mean
sd	Rater perception posterior standard deviation
t1	The first draw from the posterior distribution of the rater perception
t2	The second draw from the posterior distribution of the rater perception
...
tn	The nth draw from the posterior distribution of the rater perception