Measurement Invariance: Comparing the Latent Motivation to Learn Construct Across Subgroups

• In Event: Advancing Methods for Evaluating Construct Validity in STEM Education Research

Abstract

The purpose of this study was to apply an alignment optimization (AO) method to examine the validity and sub-group means of the previously established latent motivation to learn construction, among the adults in the USA.

Theoretical framework
Motivation to learn (MtL) is critical for promoting participation in adult education and training. However, MtL cannot be directly observed, and thus, the previous studies have identified the four-item latent MtL construct and shown the validity evidence across sub-groups, such as those with different sociodemographic characteristics. However, the individual characteristics were separately examined, and therefore, less is known about the validity at the intersections of multiple characteristics (e.g., older women with lower educational attainment and non-STEM occupations). Arguably the most common approach to investigate the validity across subgroups is the invariance test using multigroup confirmatory factor analysis (MGCFA). Yet, the invariance test with MGCFA often faces methodological challenges (e.g., many parameters need to be equal; estimation of comparable latent means) when examining many sub-groups.

Data sources
Data were obtained from the 2012/2014/2017 USA Program for the International Assessment of Adult Competencies (PIAAC) restricted-use file. PIAAC provides nationally representative data on the basic skills as well as the detailed sociodemographic characteristics, education
participation, and skill use. This study examined the sample of adults aged between 25 and 65 (n = 8,050). The AO method, which is a machine-learning technique to fit the
exploratory MGCFA and estimate the latent means of sub-groups, was adopted to examine the validity of latent MtL construct across 16 sub-groups defined by the 5-year age group, gender, and type of occupation (STEM vs. non-STEM).

Results
The AO method showed that a fair comparison of MtL across 16 sub-groups of interest is
feasible. The latent mean comparisons revealed that younger (25-34, 35-44) men with STEM occupations are more likely to have greater MtL than other sub-groups. Also, both older (age,
45-54 and 55-65) men and women with non-STEM occupations were the least
motivated sub-groups. Several sub-groups with higher MtL may continue benefiting from adult education and training participation, while other sub-groups with lower MtL may be left behind over time and experience growing social inequality. The observed differences in MtL by sub-groups provide the empirical evidence to support the development of intervention and policies to address not only existing education disparities but also cumulative education inequality through differing MtL over the adult life course.

Scientific or scholarly significance of the study
Despite the increasing interest in diversity and intersectionality in education research, the validation and comparison of a latent construct like MtL across many sub-groups is often challenging with the common methodological approach. This study demonstrated the usefulness and applicability of the AO method with respect to the comparisons of latent constructs across a large number of sub-groups.

Author

• Takashi Yamashita, University of Maryland - Baltimore County
  Presenting Author