Research Methods in AL

📊 Analyze → Dimension Reduction → Optimal Scaling is one of the less often explained options in SPSS. I want to show where it can help us make sense of categorical and ordinal data in applied linguistics. Suppose we ask learners how often they use AI for L2 writing: never, occasionally, or often. Coding these responses as 1, 2, and 3 gives them an order, but it does not tell us that the gap between never and occasionally is the same as the gap between occasionally and often. Now consider preferred feedback type: direct correction, explanation, or guiding questions. Those categories have no inherent order. Optimal scaling lets SPSS assign numerical scores to categories as part of the analysis. For an ordinal variable, we can preserve the order while allowing the gaps to differ. For a nominal variable, the numbers we used to enter the categories do not impose an order on the analysis. The menu leads to three procedures, depending on the question we want to answer: 1. Multiple Correspondence Analysis (MCA) Use it to explore which response categories tend to occur together across several nominal variables. For example, we could map learners’ preferred AI modality, main writing activity, and preferred feedback type to see which combinations recur. In SPSS: All variables multiple nominal → One set. 2. Categorical Principal Components Analysis (CATPCA) Use it to summarise one set of variables that may include ordered and unordered responses. For example, we could ask whether a few components capture the main patterns across learners’ frequency of AI use, confidence in revising, and preferred feedback type. In SPSS: Some variable(s) not multiple nominal → One set. 3. Nonlinear Canonical Correlation Analysis (OVERALS) Use it to explore how two or more sets of variables relate to one another. For example, one set could describe learners’ AI use, while another describes their planning, monitoring, and revision practices. The sets are groups of variables measured on the same learners. In SPSS: Multiple sets. For questionnaire research, I would use CATPCA to explore and summarise response patterns. If my question is whether the items measure proposed latent constructs, I would use a factor model suited to the item responses. The choice begins with what we want to learn from the data. Dr. Hessameddin Ghanbar