01 · Research Seminar · 2026
Measuring emotion in streamer commentary
A contrastive study of emotional expression in the first ten minutes of AAA and indie gameplay videos on YouTube.
OwnershipIndividual research project
I designed and completed the study, from sampling and transcript preparation through analysis and reporting.
Research question
Does the amount of affective language used by streamers differ between AAA and indie games?
Approach
From question to evidence.
- 01
Built a sample of 80 English-language gameplay videos: 20 game titles, four videos per title, split evenly between AAA and indie games.
- 02
Standardized the observation window to the first ten minutes of active gameplay and removed highly recurring scripted lines shared across all four transcripts for a title.
- 03
Used LIWC-22 to generate affect-related language measures. A Python script selected the required columns and created additional per-100-word fields with a documented word-count formula, zero-word guard, and two-decimal rounding.
- 04
Used Jamovi—not Python—for the independent-samples t-test, descriptive checks, and regression including channel size.
Analysis boundary
Preparation in Python. Statistics in Jamovi.
YouTube captions
10-minute window
Scripted-line removal
LIWC measures
Python-prepared CSV
Jamovi tests
What the work found
In the submitted course analysis, the AAA group had a higher mean on the constructed Affect_per100w measure than the indie group (0.91 vs 0.67), t(78) = 2.21, p = .030, d = 0.50. Channel size was not a significant predictor in the regression (p = .521).
Limitations & next step
The project made the boundary between data preparation and statistical analysis explicit. Before reusing the pipeline, I would re-validate the units produced from LIWC output and strengthen the removal check for scripted speech. The result should be read as an exploratory course finding, not a general claim about all game streams.
Methods & tools used here
Evidence from the source work
Selected artifacts.