Home Mathematics



RQ2: Do Interviewer Behaviors Account for Changes in Survey Length over the Course of the Data Collection Period?We now turn to the question of whether the observed changes in interviewer behaviors explain the changes in survey length over the course of the data collection period. To answer this question, we examine whether the interview order coefficient predicting survey length changes in magnitude as groups of behaviors are included in the model (Aneshensel 2013, p. 184; mediation models for each behavior individually are included in Online Appendix 20C). As seen in Table 20.3, interviewer behaviors only partially explain the change in interview length over the course of the field period. Each group of behaviors TABLE 20.3 Log(Interview Order) Coefficients Predicting Length of Interview with Interviewing Behaviors and Percent Change from Model with No Behaviors
Note: +p<.10, *p<.05, **p<.01, ***p<.001, ****p<.0001; negative percent reductions indicate an increase in the coefficient. reduces the interview order coefficient by about 1450%. The largest reduction in the coefficient for interview order comes with the inclusion of the inefficiency behaviors in WLT1, reducing the interview order coefficient by 52.4%. However, when all of the interviewer behaviors are included in the same model, this same magnitude reduction in the interview order coefficient is not observed, especially in WLT1. In WLT1, inclusion of the standardized behaviors increases the learning effect on length of interview, whereas the other behaviors explain the learning effect. Thus, the combined effects "cancel out" in the overall model. In sum, these 15 interviewer behaviors partially mediate the learning effect, but do not completely account for changes in the length of interview over the course of the field period. Variance ComponentsThere is significant variation across interviewers and respondents in the length of the interview. As shown in Table 20.4, the interviewer behaviors examined here explain between 21% and 32% of the variance in interview length at the interviewer level and between 42% and 54% of the variance in interview length at the respondent level. The inclusion of standardized behaviors alone actually increases the variance at the interviewer level in both studies, as does the inclusion of only inefficiency behaviors in WLT1. Nonstandardized behaviors explain the most variation in length across interviewers in both studies and across respondents in WLT2.

<<  CONTENTS  >> 

Related topics 