Polity data through 2003 was taken directly from the Polity Project web site. To augment it, we added the cases of Luxembourg and Iceland from the 2000 version which were excluded in the last release. They were assigned the same scores in 2001-2003 as for 2000.
Polity uses three “standardized authority codes” which are treated as missing data in their main 21-point Polity1 measure: foreign “interruption” or occupation (-66), “interregnum” or anarchy resulting from the collapse of central political authority [usually in civil war] (-77), and “transitions” in which authority patterns are difficult to characterize because they are undergoing change (-88). Because missing data reduces cases and causes particular problems for time-series analysis, it is routine for most analysts to replace many of these missing data codes with valid ones. Polity itself offers some guidance to the various ways that might be done, but encourage analysts to choose on the basis of their particular needs. They use one approach to calculate a measure they call Polity2: they recode “interregnum” as a neutral 0 code and then interpolate “transitions” using the previous and subsequent codes, but retain “interruptions” as missing data.
Our preferred measure, which we designate Polity3, combines the approaches used by various analysts. First, we identify missing data codes that are bracketed by identical regimes before and after the periods of missing data - these we recode to the previous (and subsequent) valid codes. For example, Denmark was a +10 from 1915-1939 and again 1945-2003. The years 1940-1944, coded as -66 in Polity1, are recoded as +10 in Polity3. Second, we retain as missing all other interregnums and interruptions, eschewing Polity2's recoding of interregnums as a neutral 0, which seriously misrepresents some regimes and creates additional problems in interpolation for others. Third, we replace “transitions” (-88) with linear integer-rounded interpolations from the last valid code to the next valid one, following the Polity2 procedure. Because this is done without the prior assignment of 0 scores to “interregnums”, however, we avoid some perverse codings of Polity2. For example, the Batista regime in Cuba was scored a -9 from 1955-1958 and the Castro regime was scored a -7 from 1961-2003 in Polity1. 1960 is judged an “interregnum” and 1961 a “transition”. Polity2's algorithm assigns a 0 in 1960 and then interpolates 1961 to -4, which suggests a much greater discontinuity (real change) than was actually present. Cuba was NOT notably less autocratic in 1960-1961 than before and after this period, as Polity2 records it. In our process, Polity3 retains a missing data code for 1960 and interpolates 1961 to -8. This process results in more missing data, but avoids inaccurate estimation.