Developing a short version of the Comprehensive Happiness Scale for early childhood teachers: A hybrid machine learning approach

Objective: This study develops and validates a shortened version of the Comprehensive Happiness Scale for early childhood teachers using a hybrid machine learning approach. Methods: Secondary data on 212 early childhood teachers were used to develop and validate a comprehensive 53-item happiness scale. First, the full 53-item data were fed through unsupervised learning to stratify the latent population for item responses. Next, candidate shortened versions with different item counts, (3-25 items), were created to investigate concurrent validity. The shortened instruments’ validity was compared using Kappa coefficients and AUC values calculated by supervised learning. Results: First, after inputting early childhood teachers’ response data on all 53 items, the entire sample (N = 212, m = 3.59, sd = 0.40) was stratified into three latent groups through unsupervised learning. These groups were identified to have lower overall composite happiness, with particularly low levels of “satisfaction and engagement with the organization” and “leisure and physical well-being” (N = 52, m = 3.11, sd = 0.18), higher overall composite happiness with particularly high levels of “psychological well-being” (N = 50, m = 4.11, sd = 0.25); and an intermediate group (N = 110, m = 3.58, sd = 0.17) similar to the total population mean and distributed between the other two groups' composite happiness levels. Second, the shortened 17-item instrument was adequate to categorize the low (Kappa = 0.98, AUC = 1.00), high (Kappa = 0.98, AUC = 1.00), and intermediate (Kappa = 0.85, AUC = 0.98) groups in terms of comprehensive happiness. Conclusions: A shortened 17-item instrument was developed to measure early childhood teachers’ comprehensive happiness. This shortened instrument can be used to more easily and accurately measure the happiness level of infant and toddler teachers and support their comprehensive happiness.