ACSPRI Conferences, RC33 Eighth International Conference on Social Science Methodology

Font Size:  Small  Medium  Large

A threefold comparison of comparisons to cross-validate moral attitude items: two measurements in three survey series across four time points

Tilo Beckers

Building: Law Building
Room: Breakout 2 - Law Building, Room 026
Date: 2012-07-10 11:00 AM – 12:30 PM
Last modified: 2011-12-20


Each cross-national comparative study usually relies on a limited amount of empirical evidence, often just one dataset, ideally with several measurements for each theoretical construct. But not every data set provides several items for each construct. Taking the example of moral attitudes towards same-sex sexual relationships, I compare analyses from hierarchical linear models, first, between two measurements, the first asking about the wrongness of “sexual relations between two adults of the same sex” (International Social Survey Programme/ISSP 1991, 1994, 1998, 2008) and the second framing this attitude as the justifiability of “homosexuality” (applied in the European Values Study/EVS 1990, 1999 and 2008 as well as in the World Values Survey/WVS 1995). Thus, for each time point of the ISSP another survey series can be used to cross-validate the statistical results. As the time points of two surveys in one country are relatively similar, only little deviation should be attributed to this aspect although period effects (e.g. impacts of event and news) cannot be excluded by principle. As the comparison is threefold by including the longitudinal perspective, it is furthermore possible to compare the stability of equivalence or deviations in results for each several countries across time. Besides this, the comparison is not limited to univariate parameters and a selection of theoretically relevant correlations but also includes a comparison of the coefficients of two-level hierarchical linear models for the time points 1999 and 2008. I compare the intra-class correlation coefficients as well as the direction, ranking order and magnitude of effect sizes. The study proves that “comparing comparisons for cross-validation” is a valuable tool for researchers using secondary data sources. Thereby, a strategy is offered for researchers to circumvent the possible pitfalls of using just one indicator and to improve the validity in cross-national research designs. Most importantly, researchers avoid premature conclusions about their often limited analyses and are able to qualify their results based on deviations found in the statistical analyses. And this is possible even in cases where other proper techniques such as multi-group CFA comparisons are not possible due to the lack of multiple indicators in one data set.