interviewers on the variables central to the cluster were combined. The intercorrelations (coefficients of concordance) for the three patterns were .69, .44, and .37 respectively. All three are significant at the one percent level, showing strong homogeneity within clusters. The ranks of the interviewers on variables in each cluster were then averaged, making it possible to assign a rank to each interviewer on each cluster. The rank correlations between the clusters are .00, .10, and .37 respectively. They seem to represent well-defined different aspects of interviewer performance.
Three measures--the reporting of knowledge of superstitions (real or fictitious) and the loss of information through neglect--do not fit into any pattern. It will be recalled that for these variables we were the most dubious whether they should be used as criteria at all. Neglect looked like a residual category when the loss of information was considered and in the analysis of interviewer contribution, knowledge of fictitious superstitions was the only measure where significance did not reach the five percent level. They will therefore not be used as performance measures.
In the previous sections we have discussed the interviewer from two points of view: general evaluation of the interviewer as a person, and a reconstruction of the value of the interviewer's work from the completed questionnaire. Here we shall relate the two approaches. Do interviewer characteristics predict performance? This question is important in the understanding of the interviewing process and also in the administration of survey research. We shall therefore investigate first closely the relation of interviewer traits to the different performance measures; later, we shall summarize their meaning for the patterns of performance.
First, let us consider the relationships of interviewer traits with performance on the influence cluster (Table 26). Autonomy related positively with all four groups, significantly so with those measures relating to the rural low-educated respondents, that is, autonomous interviewers are less likely to obtain belief in superstitions. We have assumed before that this cluster of performance represents an interviewer tendency to make the respondent answer "yes" regardless of the question. We see now that the more autonomous interviewer is less likely to establish with the respondent a relation in which the latter feels that the best answer is agreement with any question. Instead, she keeps a sufficient distance from the respondent to enable the latter to react only to the questions. The difference between autonomous and non- autonomous interviewers shows itself more clearly with low-educated rural respondents than with more sophisticated urban ones.
The second cluster, accuracy, shows little relation with autonomy (Table 27). The interviewer trait most closely associated with accuracy and conscientiousness in performance is adjustment to a work group, as defined by the scores in the Guilford-Martin test. This test was designed to spot potential troublemakers in industrial and business organizations; conversely, a high score on this test identifies people who are likely to follow instructions conscientiously, the quality shown by this cluster.
One measure in this pattern which does not show the same relations with interviewer traits is accuracy among the rural low-education respondents. In discussing the pattern we expressed some doubts as to whether this variable fitted in, and its relation to interviewer traits seems to confirm these doubts. It shows high correlations with intellectual ability and motivations, but a small negative correlation with the Guilford-Martin test. To obtain reliable answers from these respondents seems to require abilities other than compliance in following instructions, presumably a quick grasp of a new situation and willingness to follow it through.
In the third cluster we do not find any consistent pattern (Table 28). All of the trait variables show, however, high correlations with some of the performance variables, although in different directions. Probing certainly is an important part of the interviewer's performance. Whether an interviewer probes seems to depend on a combination of circumstances or on interviewer traits which were not directly measured.
Let us now summarize the main points of our inquiry into interviewing performance. Table 29 shows the correlation between the four interviewer traits--intellectual ability, autonomy, group adaptation (compliance), and motivation--and the three types of performance--influence, accuracy, and probing. In addition, we consider the two intermediate criteria, which were obtained early in the survey. In accordance with their intermediate position they are shown on both sides, with interviewer traits, and again with interviewer performance.
The performance criteria are related only to the personality measures. Influence is correlated with autonomy, and accuracy with compliance (the Guilford-Martin test). On the other hand, the criterion measures which could be obtained, apart from the questionnaire, do not relate to personality: the number of errors in the first interviews is a function of intellectual ability and the supervisors' rating is a function of motivation.
Another way of looking at the table is to consider the variables which do not correlate. In the long run, intellectual ability and motivation made little difference in performance, while more subtle personality factors which may determine the interaction with the respondent did so. In their turn, these factors had little weight in the early evaluation of the interviewer.
This contrast may look paradoxical, but it leads to stressing the importance of the survey operation as a long-drawn out process. The advantage of the more intelligent interviewer, with higher education, a quick grasp