The data set consists of 1380 records, but some of the records are deficient due to missing field values. The missing data fields had been indicated by a value of -9. If those records, in which some of the 29 statements had been left unanswered, are discarded, 1127 records remain for analysis. This data set was chosen. If all those records containing any missing fields would have been discarded, only 946 records would have remained.
In the analyses the opinions were divided into three fuzzy classes: "Agree", "No firm opinion" and "Disagree" as presented in figure 5. A value of 2 or 3 resulted in a membership of 0.745 to class "No firm opinion" and 0.255 to either "Agree" (3) or "Disagree" (2). This resulted in 87 new variables. Because the strong opinions are usually the most interesting properties, the "No firm opinion" class memberships were omitted from most training data sets. This results in somewhat clearer groups of neurons, which represent firm opinions.