Use this URL to cite or link to this record in EThOS:
Title: Applications of call record data to nonresponse bias adjustments
Author: Hanly, Mark J.
ISNI:       0000 0004 5923 7176
Awarding Body: University of Bristol
Current Institution: University of Bristol
Date of Award: 2015
Availability of Full Text:
Access from EThOS:
Call record data describe the process of survey recruitment, and include the timing and outcome of interviewers' contact attempts. These are a convenient source of auxiliary information which can potentially be used to correct for nonresponse bias in household surveys. However, call records are distinct from traditional forms of auxiliary information: they are not a fixed characteristic of the household; and by their nature they are longitudinal and non-rectangular. These issues complicate their use, and existing applications of call records to post-survey adjustments are limited. I delineate three potential uses of call record data: (i) as predictor variables in a model of the response outcome; (ii) as predictors in a model for the survey items; and (iii) to use in a joint model for the survey variables and response process. Approaches (i) and (ii) are implemented under the 'missing at random' framework of design-based weighting and multiple imputation respectively. Related to (i), I investigate sequence analysis as a potentially more efficient way of summarising complex call records. Approach (iii) leads to an event-history model for the response process. This is estimated jointly with incomplete survey data, and in doing so the missing at random assumption is relaxed. The analysis is applied to data from a current, large-scale household survey: The Irish Longitudinal Study on Ageing. Incorporating call record data improved post-survey adjustments, but only for a limited number of variables, particularly estimates related to marital status. Throughout the analysis I offer methodological insights which have implications beyond the setting of nonresponse adjustments, most notably, the role of costs in the sequence analysis algorithm and the question of imputing in a hierarchical setting when the number of level-one units is unknown. Future research should explore approaches which utilise call record data to minimise bias during fieldwork.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available