If I'm reading the statistics in the study correctly [0], most of these 80 incidents (70%) were pulled from CCTV footage at a "night time drinking setting." Groups of 18 people on average standing around bars, presumably outside. In 20% of cases someone de-escalated, 5% of the time a bystander joined in to escalate the violent encounter, and 4% it was a mix of escalation/de-escalation. Presumably the rest of the time no one intervened.
This is not an ideal setting to analyze the bystander effect. People are drunk. People know each other. People may have just met but started bonding over alcohol (and cigarettes, if we are outside the bar). Seems more like a study of the behavior of drunk people than anything else.
Fair warning, I'm not great at interpreting statistics nor reading sociological studies, I prefer to read about medicine. I'm open to being corrected.
[0] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6790599/