Correction for Measurement Error Using Survey Data
Comparative cross-national social and political surveys collect measures of social attitudes, political opinions, preferences and behaviours across countries.
These measures can contain error due to the measurement instruments. Ignoring measurement error results in biased estimates of the relationships between the observed variables: the relationships can be underestimated because the reliability and/or validity are low, or overestimated because of high correlated method effects when very similar measurement instrument are used. Either of these can lead to wrong conclusions.
When information about the reliability and validity of measurements is available (see Data Quality Assessment), it is possible to apply statistical techniques to correct for measurement error. The most straightforward way to correct for measurement error is by using multiple-indicator models under the framework of Structural Equation Modelling. However, due to the variety of topics it encompasses, the ESS often measures a concept of interest using only one question.
The Survey Quality Predictor (SQP) allows to predict the reliability and validity of survey questions. This information can be incorporated in statistical analyses to avoid bias.
- Correction for measurement error in multilevel models: Malcolm Fairbrother (Bristol University) and Diana Zavala-Rojas (ESS CST, UPF) are developing an algorithm which allows for correction for measurement error in multilevel models using R.
- ESS web application for correction for measurement error: researchers at UPF aim to make correction for measurement error easier for ESS data users by developing a web application using Shiny, an open source web application framework for R.
- Collaboration with substantive researchers in the application of statistical techniques to correct for measurement error: although methodological advances are increasing, collaboration between social and political scientists and methodologists is fundamental to spread the application of new tools. Substantive researchers who wish to learn and apply statistical techniques to correct for measurement error using ESS data can contact email@example.com for more information.