Coding of survey responses is time-consuming, costly and error-prone. This presentation is about the endeavors of Statistics Sweden to assure the quality of the coding process. Work has been done on a wide front.
Verification of the coding is being implemented 2011 in all relevant surveys. For computer-assisted (manual) coding, quality control comprises of independent verification of the coding, conducted for at least five percent of the coded records in a survey. If the original coding and the verification coding differ, the 'correct' code is decided by an adjudication (reconciliation) process. If a coder's work contains frequent errors, that coder's work shall be verified and suitable training shall be given. For automatic coding, an analogous quality control will be made at least once every third year, in an appropriate survey. If the error rate is unacceptable, revisions of the dictionary should be carried out.
Moreover, an instruction for developing code frames (when not having standard classifications) has been established. All classifications and code frames are stored in a common repository. The different staff, in a wide sense, involved in the coding process have to cooperate in order to quality assure the process. To that end, the different roles have been pointed out clearly. Because the coding of open-ended responses is often subjective in nature, proper training is essential. The coders at Statistics Sweden are trained on how to use classifications, reference files etc. through awareness of rules regarding what should be included or excluded from a given code. The training is done through a web based application connected to a database with exercises, which are given in three degrees of difficulty.
A modern IT tool for computer-assisted coding has been developed. Support for coding decisions is given, working with a user-friendly interface. Functionality for handling access rights for original coders, verification coders and 'adjudicators' is also available.
Information on the coding process and required quality assurance measures is given in the intranet process information system. Appliance will be followed up via the directors of the subject matter and data collection departments.
Keywords: coding; computer-assisted coding; verification coding; reconciliation
Biography: Jörgen Svensson is a methodologist and works as a process owner for data processing and analysis at Statistics Sweden. He has a responsibility for supporting best practices and common tools in the areas of coding, editing, imputation, estimation, disclosure control etc. He will present recent work within the area of coding of survey responses at Statistics Sweden.