Measurement errors in official statistical surveys |
|
Convenor | Mr Anton Karlsson (Statistics Iceland ) |
Coordinator 1 | Mr Øyvin Kleven (Statistics Norway) |
One proposed solution for increasing response rates in surveys is to offer more than one mode in the data collection phase, possibly reducing non-response bias. The main problem of this approach is the possibility of measurement error differently affecting responses by the data collection method used. This presentation describes a split ballot test of collecting data for the Icelandic Labour Force Survey using a telephone interview, a web-questionnaire or mix of both modes. The main goal is to examine to which extent different modes can reduce possible non-response bias, without increasing the likelihood of measurement error in
In the context of the reform of household statistics Germany plans to adopt an infra-annual rotation pattern for the Labour Force Survey, which will consequently reduce the distances between the interviews and increase the burden of households. Currently independent interviewing is used for conducting household surveys. Especially in panel studies respondents are confused by confronting them with recurring questions for several waves. Therefore the Federal Statistical Office of Germany is planning to implement dependent interviewing for the German microcensus.To test the feasibility of the technique the FSO carried out a pretest.The presentation focuses on results and challenges.
Measurement effects (MEs) are a problem in mixed-mode (MM) surveys threatening accuracy of MM estimates and may need to be adjusted after data collection. Our approach is based on the potential outcomes framework predicting outcomes of modes deemed more accurate, but not observed for all respondents. We suggest using re-interview data collected in addition to the MM data as auxiliary information, because available data from registers is typically weak. The viability of this approach is evaluated by simulation taking into account various factors, such as the size of the re-interview sample, costs, and efficiency of adjustments.
For many surveys measurement errors is the most damaging source of error. Measurement errors may be difficult to detect unless they lead to illogical responses. One approach to detect measurement errors is to use auxiliary variables. Some textbooks mention the danger of a social desirability effect in reporting voting in election surveys (a higher chance of misreporting if one did not participate in the election and). In this presentation we illustrate this approach for bias exploration using the Norwegian Election Survey data. The claimed turnout in the survey can be checked individually against the true head-count from the electoral