Statistical Discourse Analysis

Identifying triggers of target actions within individual or social processes across time and their links to higher-level outcomes requires modeling individual (and group) characteristics and sequences of actions. One such method is statistical discourse analysis (SDA). SDA can model (a) pivotal moments that radically change subsequent processes and divide the time points into distinct time periods, (b) effects of previous actions (or their sequences) on target actions, and (c) influences at various levels (turn, time period, individual, group, organization, province, country, etc.). SDA addresses difficulties involving data (unit of analysis, coding, inter-rater reliability, missing data, parallel conversations, breakpoints, time periods, statistical power), dependent variables (discrete variables, infrequency bias, nested data, multiple dependent variables), and explanatory variables (variables at earlier turns, cross-level interactions, indirect multilevel mediation, serial correlation, false positives, odds ratios, robustness). To show how sequences of processes are related to higher-level group outcomes, we add the group outcome as an independent variable. For example, we can test how social metacognitive actions (e.g., agree, rudely disagree) affect the likelihood of correct, new ideas (micro-creativity) and justifications, using 3,296 turns of talk by 80 students in 20 groups working on an algebra problem. A rude disagreement often triggered another rude disagreement, which yielded less micro-creativity. In groups that solved the problem or after a wrong idea however, a rude disagreement yielded greater micro-creativity. After a student with a higher mathematics grade spoke, more justifications followed; this effect differed across time periods. SDA requirements include specification of explanatory variables, independent and identically distributed residuals, and a minimum sample size (20 units at the highest level).