Surveys are frequently used to study humans scores on latent factors, including human values, attitudes and behavior. Often such a study includes a comparison of groups of individuals or countries at one or multiple points in time (i.e., a cross-sectional or a longitudinal comparison). If latent variable scores are to be meaningfully compared across groups, countries or time, the measurement structures underlying these latent factors should be stable, that is ‘invariant’. Many studies examining ‘measurement invariance’ (MI) of survey instruments have shown that the invariance assumption is very hard to meet. In particular, strict forms of measurement invariance such as scalar invariance, or at least partial scalar invariance, often do not hold. Failure to achieve strict MI across groups may certainly be expected in large-scale international comparative (survey) research including many latent factors or a large number of groups to be compared, like the European Social Survey or the Pisa Study. When a strict form of MI does not hold (across relevant groups and/or time points) it is clear that individuals respond(ed) differently to survey items making it impossible to compare latent factor means in a meaningful (i.e., a valid) way. As such, potential bias (caused by measurement non-invariance) obstructs the comparison of regression coefficients or latent factor means.
Recent developments in the field of Bayesian statistics provided new tools for assessing MI and imposing strict or more relaxed forms of MI. Using Bayesian structural equation models (BSEM) the strictness of (ideal) forms of MI may be relaxed. In particular, exact zero constraints on the cross-group differences between all relevant measurement parameters (e.g., factor loadings and indicator intercepts) may be substituted by ‘approximate’ zero constraints. Instead of forcing intercepts to be exactly equal across groups, a substantive prior distribution (around zero) is used to bring the parameters closer to one another while allowing for some ‘wiggle room’ (i.e., some deviation from zero is allowed for).
Since approximate MI was developed only recently, much research is still needed to explore its possibilities and its limitations. Insights from additional research is beneficial as it will help researchers to make informed choices about applying full, partial, or approximate MI. Therefore, we aim to provide a forum for a discussion of MI covering different areas:
(1) simulation studies which are designed to: (a) demonstrate the possibilities and limitations of the different methods of MI (i.e., full v. partial v. approximate MI), and (b) to determine which prior specification in which model is to be advised in case Bayesian statistics is used;
(2) real data applications in different research areas dealing with measurement invariance over groups, countries and/or time points;
(3) tutorial papers with suggestions how to use the different forms of MI;
(4) philosophical papers discussing MI in general and approximate MI in particular.
The Research topic will start with a kick-off meeting where all potential contributors are welcome to share ideas on potential papers. This expert workshop will be organized at Utrecht University in The Netherlands. After the kick-off event the authors are asked to submit their papers, which will be reviewed by experts in the field. When the Research Topic is published a conference will be organized on the topic of approximate measurement invariance where all contributors to the Research Topic are welcome to present their work. This conference is open for all researchers interested in the topic and will be advertised in the different research areas of the contributors.
Surveys are frequently used to study humans scores on latent factors, including human values, attitudes and behavior. Often such a study includes a comparison of groups of individuals or countries at one or multiple points in time (i.e., a cross-sectional or a longitudinal comparison). If latent variable scores are to be meaningfully compared across groups, countries or time, the measurement structures underlying these latent factors should be stable, that is ‘invariant’. Many studies examining ‘measurement invariance’ (MI) of survey instruments have shown that the invariance assumption is very hard to meet. In particular, strict forms of measurement invariance such as scalar invariance, or at least partial scalar invariance, often do not hold. Failure to achieve strict MI across groups may certainly be expected in large-scale international comparative (survey) research including many latent factors or a large number of groups to be compared, like the European Social Survey or the Pisa Study. When a strict form of MI does not hold (across relevant groups and/or time points) it is clear that individuals respond(ed) differently to survey items making it impossible to compare latent factor means in a meaningful (i.e., a valid) way. As such, potential bias (caused by measurement non-invariance) obstructs the comparison of regression coefficients or latent factor means.
Recent developments in the field of Bayesian statistics provided new tools for assessing MI and imposing strict or more relaxed forms of MI. Using Bayesian structural equation models (BSEM) the strictness of (ideal) forms of MI may be relaxed. In particular, exact zero constraints on the cross-group differences between all relevant measurement parameters (e.g., factor loadings and indicator intercepts) may be substituted by ‘approximate’ zero constraints. Instead of forcing intercepts to be exactly equal across groups, a substantive prior distribution (around zero) is used to bring the parameters closer to one another while allowing for some ‘wiggle room’ (i.e., some deviation from zero is allowed for).
Since approximate MI was developed only recently, much research is still needed to explore its possibilities and its limitations. Insights from additional research is beneficial as it will help researchers to make informed choices about applying full, partial, or approximate MI. Therefore, we aim to provide a forum for a discussion of MI covering different areas:
(1) simulation studies which are designed to: (a) demonstrate the possibilities and limitations of the different methods of MI (i.e., full v. partial v. approximate MI), and (b) to determine which prior specification in which model is to be advised in case Bayesian statistics is used;
(2) real data applications in different research areas dealing with measurement invariance over groups, countries and/or time points;
(3) tutorial papers with suggestions how to use the different forms of MI;
(4) philosophical papers discussing MI in general and approximate MI in particular.
The Research topic will start with a kick-off meeting where all potential contributors are welcome to share ideas on potential papers. This expert workshop will be organized at Utrecht University in The Netherlands. After the kick-off event the authors are asked to submit their papers, which will be reviewed by experts in the field. When the Research Topic is published a conference will be organized on the topic of approximate measurement invariance where all contributors to the Research Topic are welcome to present their work. This conference is open for all researchers interested in the topic and will be advertised in the different research areas of the contributors.