Abstract

To date, most cross-country comparisons of children's subjective well-being have been conducted using single-item scales. Despite multi-item scales being more powerful for this purpose, they have seldom been tested on children when comparing results among more than 4 countries. Moreover, with very few exceptions, international comparisons have mostly been carried out using samples of children aged 12 or over and it is therefore uncertain how the scales available might work among younger populations, even if some scales have been tested in a few countries. We tested 3 psychometric scales on a sample of over 34,000 children from 15 countries aged mostly 10 and 12: the SLSS, the BMSLSS and the PWI-SC. We used the pooled database to identify models with a good fit by means of Confirmatory Factor Analysis, providing construct validity for each of the three scales for this set of countries. The comparability of the scales among countries was tested using Multi-group Confirmatory Factor Analysis to assess to what extent it is valid to make cross-national comparisons. Our results suggest that it is acceptable to compare correlations and regressions between most of the countries in our survey using each of these measures, with only a few exceptions. Some of the models using the specific modified SLSS version adopted in this research displayed promising results due to the fact that their correlations and regressions appeared to be comparable among all countries in the sample. However, mean scores for the overall indexes are only comparable among countries in some cases using partial intercept constraints. Two Multi-group Structural Equation Models including the three correlated multi-item psychometric scales plus two single-item scales (Overall Life Satisfaction and Overall Happiness Scale) displayed good fit indexes with constrained loadings for all countries, both for the 10 and 12-year-old samples. This result suggests that subjective well-being comparability increases among countries when using the five psychometric scales all together. With semi-partial constrained loadings and intercepts, fit statistics suggest that the means of this overall model can cautiously be compared among all countries: comparable items and not strictly comparable items were identified. Correlations among the psychometric scales and regressions of the multiple-item scales on the single-item scales clearly show different patterns among countries and variations according to age group, suggesting a high diversity of interrelations among these measures depending on age and different language and socio-cultural contexts.