オンライン学習履歴データの統計的取り扱いについて

The aim of this paper is to discuss the appropriate statistical treatment of online learning log data that are presently viewed as informative and useful resources to improve foreign language teaching practices in higher education. However, current research methodologies in foreign language teaching research that include experimental designs, null hypothesis significance testing, and psychological scaling are not attuned to utilizing disordered online learning log data. This paper indicates that online learning log data should be treated as “feature values” like in data mining rather than manifest variables under a certain psychometric latent variable model, since the data substantially precedes the theoretical derivation of the construct in measurement. The rationale of this treatment is based on relatively new pragmatic, utilitarian, and consequential perspectives on foreign language teaching research, unlike orthodox research techniques that are strongly supported by cognitivism. Furthermore, this paper underscores the importance of examining mathematical properties of online learning log data. Typically, online learning log data follow non-normal distributions, such as (a) binomial distribution, (b) Poisson distribution, (c) geometric distribution, (d) negative binomial distribution, (e) log-normal distribution, (f) Gamma distribution, (g) Weibull distribution, and (h) ex-Gaussian distribution. Due to these distributional properties, data analysts are concerned about visualizing the empirical distributions of the given online learning log data by histograms or kernel density estimation, and then fitting specific probability density functions to the given data using the maximum likelihood estimation method or other estimation methods. This paper demonstrates the suggested statistical treatment of online learning log data with numerical examples from the author’s teaching experiences.