Coding Results. Contrary to our practice for the previous three sets of "tactic" variables, we did not halt our coding for the "propagandizing" variables to conserve resources to finish the project. We continued to code these variables, because we regarded them as more important to later analysis. As disclosed in Tables 7.4 to 7.7, the percentages of parties that were coded on these four variables ranged from approximately 60 to 80. BV632, "operating party schools," proved to be the most resistant to coding, being also the rarest of the propagandizing activities by far--witness the low means for BV632 in comparison with the others. The most common activity was "passing resolutions"--an activity hardly ever neglected--followed closely by "publishing position papers." A clear majority of the parties also operated their own mass media (code 2 on BV631), which typically meant publishing a newspaper.

Our coding of these variables was clearly affected by the amount and quality of information in our files. The correlations between the BV and AC codes for these variables all hover around .60, informing us that the coders were usually very confident in coding parties which "frequently performed the activity" (code 2) but were often unsure about recording that a party "never engaged in the activity" (code 0). The coding problem lies more in assigning the low frequency code than the high one. If a party frequently engages in an activity, writers usually can be counted on to report it. Therefore parties are not usually scored 2 unless the literature mentions the activity with regularity. But, if a party does not perform a given activity, writers seldom go out of their way to document the observation. If the literature fails to mention an activity--such as the operation of a party newspaper, it presents the coder with a dilemma. Should he conclude that the party does not operate a newspaper because it was not mentioned, or should he omit coding the variables because of inadequate information? In point of fact, our coders were free to do either, depending on their feel for the situation after reading the literature. But, if they coded the variable primarily from an absence of comment on the existence of a party newspaper, then they were instructed to apply a low AC code to their judgment (usually an AC of 3) to flag the inferential basis of their code for BV632. It is this feature of the coding process that produces the high correlations between the BV and AC codes for all the activity (tactics) variables. If one buys the assumption that an absence of information about an activity coin-

go to page 86