Summary
A repeated survey is a survey carried out more than once, mostly with regular frequency, for example monthly, quarterly, or annually. Most surveys in a statistical office are repeated. Samples in a repeated survey may be independent over time, or the sample design may deliberately involve a unit at several occasions. The sample design should balance accuracy requests, which often imply considerable overlap between samples over time, and response burden. A sample design, where a business is selected each time during a period and then is not selected for a time period, has advantages for both accuracy and the respondent. Panel surveys and longitudinal surveys are particular cases of repeated surveys. There are other arrangements, for instance based on permanent random numbers. A repeated survey may use administrative data, either only or in combination with directly collected data.
Measures of change are normally an important part of the statistical output of a repeated survey, for example indices and in many cases also time series. Seasonal adjustment may be used for short-term statistics to make comparisons easier for the users. Usually there are time-related requests on the output, such as comparability over time and high accuracy in estimates of change. The changes over time are due both to population changes and changes in values of variables. The requests have implications for the survey design. Differences in definitions and methods between two points in time mostly have a negative effect on the comparability between the two sets of statistics. Considering comparability over time only, such differences should be avoided. It is in the nature of a repeated survey to use the same definitions, methods etc.
A break in the time series may become unavoidable for external or internal reasons, and it can be justified when the advantages outweigh the disadvantages. It is important to measure its size, if possible, and to inform the users in advance about the introduced changes and the break. When statistics from repeated surveys are published it is often the case that the statistics for one or more of the earlier time periods are revised. It is recommended to have a revision policy, preferably aligned with other statistics, both nationally and internationally.
The repetitive character of the survey gives possibilities to improve the statistical production process and the quality of the output by utilising both previously collected data and process data (paradata). These possibilities should be taken into account before the first production round and incorporated in the design to ensure that appropriate data, paradata, and metadata are collected and saved for future use. An imbedded experiment is an example of a method to study effects of a suggested change in advance and possibly avoid time series breaks or at least reduce the effects.
There are three major reasons for a separate description in the handbook of repeated surveys: the possibilities to make improvements over time, the possibilities to utilise previous data if a unit is included repeatedly in the survey, and issues related to time series breaks. This topic provides an overview of the specifics of repeated surveys. Most methods are already described in other parts of the handbook, for example methods for sampling and estimation. References are given to relevant modules and also to the general literature, mainly on specific issues. There are few books dedicated to repeated business surveys, perhaps a bit surprisingly, since business surveys often are regularly repeated surveys. The overview edited by Cox et al. (1995) has good coverage. Snijkers, Haraldsen, Jones, and Willimack (2013) describe how to design and conduct business surveys; a recent book.
To read the entire document, please access the pdf file (link under "Related Documents" on the right-hand-side of this page).
Your feedback is appreciated. Please send your remarks, suggestions for improvement, etc. to memobust@cbs.nl.