This study assessed the presence of measurement invariance of familism and parental respect across four racial/ethnic groups in a sample of 10–15 year olds to inform whether meaningful results can be obtained using these measures in samples with ethnic/racial diversity. Overall, we found no evidence of substantial difference in measurement properties of the instruments across the four racial/ethnic groups, supporting the continued use of these measurement instruments in research in adolescents.
We tested whether stronger levels of invariance across four ethnic/racial groups of younger adolescents: Asians, African Americans, Hispanics, and whites would substantially worsen model fit. We found no evidence of substantial configural invariance, weak invariance or strong invariance for either scale. When we tested for strong invariance, we found small and non-significant differences. We then tested the difference between the means of the latent variables. For familism, we found that Hispanic respondents had higher means on the scale than other ethnicities, and that this group was significantly different from the white group, as may have been hypothesized, given the origins of the measure. For parental respect, the differences between means were larger, and were statistically significant; however the differences were not substantial. In addition, the highest latent variable mean scores were found in the Hispanic and African American groups, not the Asian group, which replicates previous findings in the literature [23
Because of demographic differences in study populations, no studies provide directly comparable data to the present study. Villareal and colleagues [19
] explored cultural comparability of a measure of familism across different Hispanic groups in the US, in both Spanish and English, in a population of adults aged 18 to 65 sampled by telephone from areas with a high population of Hispanics. They found support for invariance of their familism measure across language (Spanish versus English), and country of origin (USA, Mexico, or Latin America), suggesting that the measurement properties of familism were consistent across these groups. Schwartz [1
] examined differences in the factor structure of a measure of familism across Hispanic, white and African American university students. This research also found no differences in factor structure or mean score across these groups; however the relatively low sample size of 57 white respondents and 64 African American respondents limited the power of the study to find differences that may have existed. Recently Schwartz and colleagues [2
] carried out a much larger investigation of invariance for familism and parental respect in two samples. The first comprised 1000 university students and the second included over 10,000 university students (in study 2). In both samples, they compared the fit of models across four ethnic groups (Asian, African American, Hispanic, white) and found measurement equivalence across these groups. Since we found no measurement invariance in our young sample of adolescents, this suggests that similar measures of familism and parental respect scales can be applied to a diverse sample of younger populations.
Given the recent, drastic demographic shifts and historical migration patterns of racial/ethnic groups in the US, there may be racial/ethnic differences in birth country, timing of migration to the US, and acculturation among young US adolescents that do not exist in US adults [28
]. Thus, non-equivalence of cultural values in a diverse sample of younger adolescents might be expected. If we had found differences in measurement properties across groups, the appropriate comparisons of the relationship between cultural values and substance use risk among adolescents could not be made across those groups and this would have both theoretical and practical implications. Specifically, it is theoretically important to understand differences in the structure of cultural values between racial/ethnic groups in order to understand the cognitive processes that are employed when an individual decides how to answer a particular item. It is of practical importance to ensure that there is cultural comparability of measures across groups if the measures are to be employed in diverse populations of members of many groups to ensure that the same construct is being assessed in all individuals. Our results indicate that there is no evidence of substantively important invariance, suggesting no evidence of problems with bias or differences in measurement properties when employed in a diverse sample of middle school adolescents.
The study has a number of strengths. In particular, the size of the sample was over 5000 individuals, which provided a great deal of power to estimate parameters with a high degree of precision. In addition, we tested factorial invariance by race/ethnicity among a highly diverse sample of Hispanics, Asians, African Americans, and whites. To our knowledge, this study is the first to report such a finding in middle school aged youth, hence demonstrating the appropriateness of the measures for respondents of this age group and these racial/ethnic groups. However there are limitations that should be taken into account when interpreting results. First, although the overall sample size was large, the number of African American respondents was relatively low (n
230) compared to other groups. Second, the sample was racially/ethnically diverse, but represented youth from public schools across the Los Angeles metropolitan area, which may limit the generalizability of the results to the larger population of adolescents in the U.S; in particular Hispanic youth in the Los Angeles area are more likely to have family origins in Mexico than are youth in other areas of the US. In addition, the description of youth as ‘Asian’ may be reducing a diverse population into a single category which is useful descriptively, but includes many different cultural groups from a large and diverse geographical area and therefore may be an oversimplification. Further research might explore measurement properties of cultural value measures among sub-groups of Asians (or other groups). It is also important to note that differences may exist between first, second and third generations of immigrants – as cultural values can sometimes become diluted or incorporate values from other cultures over generations [24
We also found that responses to individual items were highly skewed, leading to ceiling effects as a majority of youth endorsed ‘strongly agree’ for all items. The analysis approaches that we employed do not assume a normal (or any other) distribution and therefore the large number of responses at the high end of the scale reduced our ability to discriminate individuals at that level and hence may reduce power [30
]. In future research, a change in the response scale or in item wording might improve the ability to discriminate at that level. Increasing the number of response categories has been shown to increase the responsiveness of a scale (e.g. [31
]), however increasing the number of responses can make the responses less meaningful, particularly in younger individuals. Rephrasing the items to make them more extreme (increasing the difficulty of agreement) might also be effective, for example “I expect my relatives to help me when I need them” might become “I definitely
expect my relatives to help me when I need them” (italics added for emphasis). However, care should be taken that such rewording does not alter the meaning of the items.
It is also possible that social desirability played a role in the ceiling effect. The confidentiality of the survey was emphasized, which should reduce socially desirable responding due to impression management. However, self-deception enhancement and denial may still have played a role [32
]. Whether socially desirable responding is influencing responses on these measurement instruments is an empirical question worthy of further investigation.