The research, and what it does not show
This is where a lot of marriage material stops being honest, so here is the whole thing. The four behaviours come out of real observational work on real couples. The famous accuracy figures attached to them do not survive proper testing, and we are not going to quote them at you.
What the research says
Gottman JM, Levenson RW (1992). Marital processes predictive of later dissolution: behavior, physiology, and health. Journal of Personality and Social Psychology, 63(2), 221 to 233 doi
Couples were filmed arguing and their interaction was coded second by second, alongside measures of what their bodies were doing. Patterns of criticism, defensiveness, contempt and withdrawal distinguished couples whose marriages went on to deteriorate. This is descriptive observational research on how conflict looks, and that is the claim it supports.
Heyman RE, Slep AMS (2001). The hazards of predicting divorce without crossvalidation. Journal of Marriage and Family, 63(2), 473 to 479 doi
This is the central methodological criticism, and it is the reason no percentage appears in this lesson. The very high accuracy figures came from models fitted to the same couples whose outcomes they were then said to predict. Tested on independent data, accuracy fell substantially. The famous number is not a prediction rate for any new couple, including yours.
Stanley SM, Bradbury TN, Markman HJ (2000). Structural flaws in the bridge from basic research on marriage to interventions for couples. Journal of Marriage and Family, 62(1), 256 to 264 doi
Finding out what distinguishes distressed couples from happy ones does not automatically tell you what to teach couples. The authors argue that this step has often been taken too quickly in the field. We publish this next to the four behaviours rather than after them.