Emphasis on emotions in student learning: Analyzing relationships between overexcitabilities and the learning approach using Bayesian MIMIC modeling

Abstract

The aim of this study is to investigate interrelationships between overexcitability and learning patterns from the perspective of personality development according to Dabrowski’s theory of positive disintegration. To this end, Bayesian structural equation modeling (BSEM) is applied which allows for the simultaneous inclusion in the measurement model of all, approximate zero cross-loadings and residual covariances based on zero-mean, small-variance priors, and represents substantive theory better. Our BSEM analysis with a sample of 516 students in higher education yields positive results regarding the validity of the model, in contrast to a frequentist approach to validation, and reveals that overexcitability – the degree and nature of which is characteristic of the potential for advanced personality development, according to Dabrowski’s theory – is substantially related to the way in which information is processed, as well as to the regulation strategies that are used for this purpose and to study motivation. Overexcitability is able to explain variations in learning patterns to varying degrees, ranging from weakly (3.3% for reproduction-directed learning for the female group) to rather strongly (46.1% for meaning-directed learning for males), with intellectual overexcitability representing the strongest indicator of deep learning. This study further argues for the relevance of including emotion dynamics – taking into account their multilevelness – in the study of the learning process.

Keywords:

• Bayesian structural equation modeling
• overexcitabilities
• Dabrowski’s theory of positive disintegration
• learning patterns
• confirmatory factor analysis with covariates
• Mplus

Notes

1. Drawing on Bayes theorem, the formula for the posterior distribution P(θ|z) of the unknown parameter θ given the observed data z can be expressed as:

where P(θ) stands for the prior distribution of the parameter, reflecting substantive theory or the researcher’s prior beliefs, and P(z|θ) is referred to as the distribution of the data given the parameter, which represents the likelihood (Kaplan & Depaoli, 2012; Kruschke, Aguinis, & Joo, 2012; Levy, 2011; Zyphur & Oswald, 2015). Omitting the marginal distribution of the data P(z) in the formula, reveals the proportionality of the unnormalized posterior distribution to the product of the likelihood and the prior distribution (Kaplan & Depaoli, 2012; Levy, 2011). The uncertainty regarding the population parameter value, as indicated by the variance of its prior probability distribution, is influenced by the observed sampling data, yielding a revised estimate of the parameter, as reflected in its posterior probability distribution (Kaplan & Depaoli, 2012).

2. Bayesian estimation makes use of Markov chain Monte Carlo (MCMC) algorithms to iteratively draw random samples from the posterior distribution of the model parameters (Muthén & Muthén, 1998–2015). The software program Mplus uses the Gibbs algorithm (Geman & Geman, 1984) to execute MCMC sampling. MCMC convergence of posterior parameters, which indicates that a sufficient number of samples has been drawn from the posterior distribution to accurately estimate the posterior parameter values (Arbuckle, 2016), is evaluated via the potential scale reduction (PSR) convergence criterion (Gelman et al., 2014; Gelman & Rubin, 1992). When a single MCMC chain is used, the PSR compares variation within and between the third and fourth quarters of the iterations. A PSR value of 1.000 represents perfect convergence (Kaplan & Depaoli, 2012; Muthén & Muthén, 1998–2015).

3. In Bayesian parameter estimation, the term “significant” is used by the authors to indicate that the 95% Bayesian credibility interval of a particular parameter did not cover zero. The Bayesian credibility interval can be retrieved directly from the percentiles of the posterior probability distribution of the model parameters. Using the posterior distribution percentiles, it is possible to determine directly the probability that a population parameter value is situated within a specific interval. If the posterior probability interval of a particular parameter does not contain zero, the null (condition) can be rejected as implausible, and as a consequence, the parameter is considered significant (which is indicated by a one-tailed Bayesian p-value below .05). A hypothesis testing perspective was also used in assessing model fit (Levy, 2011).