1 Using traditional control systems to reveal changes










- Slides: 10

1 Using traditional control systems to reveal changes when introducing new data collection instruments Sindre Børke (sbe@ssb. no) Senior Adviser Division for Data Collection Methods, Statistics Norway 1

Paper contra web Questionnaires Information model (survey definitions) Web questionnaire Paper questionnaire Websolution Survey database Data from paper questionnaires Editing Process Data from web questionnaires 2

Outsourced editing gives better data • Household census: average 6, 2 and 3, 4 errors • Hotel statistics: average 1. 7 and 0. 7 errors • Hotels, prize per room: 41% contra 36% with error 3

Without controls, more errors in web? • Not in household census 2001 • Hotels, number of rooms let out: 12% contra 18% error • (PPI: 1. 27% contra 1. 40% outside the defined range) • Road goods transport: Different patterns to be explained 4

Behind – or on - the screen • Following the respondent’s filling-in process, and the warnings presented during this process 5

Errors in incoming web and paper version of the hotel statistics questionnaire 6

Errors detected in the web and paper version of the hotel statistics questionnaire 3. 0 16. 7 18. 1 25. 2 One error message More error messages 7

Transfer of electronic data Information model (survey definitions) Individual Supporting adaptions software Web questionnaire Basic software Enterprise database Editing Process Data from Enterprise system Paper questionnaire Websolution Survey database Data from paper questionnaires Data from web questionnaires 8

Transfer of electronic data • Allready in wide use for accounts and salary data in Norway. • Presents challenges in development and upkeeping – Supporting software from the software vendors – Individual adaptions at the respondent’s • Looking into one case with traditional controls, finding better results than from paper questionnaires • Requires different control/editing, due to different set of possible errors and different consequences when errors occurs 9

In short: • Few surveys seems to have data to observe possible changes in quality • Those data we find seem to tell that there are changes • Possible changes in data that are not part of a control can not be revealed with the instrument we use in this paper • This field must be further investigated, and give feed-back/forward to the editing procedure and to questionnaire developing 10