Note III
Small bids and a contested prediction
Among the more durable ideas to come out of observational research on couples is the bid: a small conversational opening, often trivial in content, that invites a response. A remark about something outside the window is a bid. The observational category of interest is not the remark but what happens next — whether the other person takes it up, gives it minimal acknowledgement, or does not register it.
The term is associated with John Gottman, who from the 1980s onward recorded couples in laboratory settings at the University of Washington, coding facial expression, physiological measures and speech turn by turn. The recording apparatus produced an unusually detailed corpus, and the descriptive findings from it are the part of the programme that has held up best. Coders could distinguish reliably between interactions in which conversational openings were taken up and interactions in which they were not, and that distinction varied systematically between couples.
The accuracy figures
What travelled furthest, though, was not the description. It was a set of numbers: claims that observing a couple for a short period allowed their eventual separation to be predicted with accuracy in the region of ninety per cent or higher. Papers by Gottman and Robert Levenson in 1992 and again in 2000, and a 1999 paper by Sybil Carrère and Gottman on the opening minutes of a conflict discussion, reported figures of this kind, and the figures were repeated widely outside the research literature.
In 2001, Richard Heyman and Amy Slep published a paper in the Journal of Marriage and Family setting out the statistical objection. The models producing those accuracy rates had been fitted to the same samples whose outcomes they were then reported as predicting. A model with enough parameters, tuned against a known outcome in a small sample, will describe that sample very closely; the question is how it performs on data it has not seen. Heyman and Slep tested exactly this and found that predictive performance dropped substantially under cross-validation. Their argument was not that the underlying observations were wrong. It was that the accuracy figures described a fit rather than a prediction, and should not have been reported as the latter.
The critique was aired outside the journals as well. In 2010, the journalist Laurie Abraham published an examination of the prediction claims in Slate, tracing how the figures had been generated and how they had come to circulate as established fact. The samples involved were also small — dozens of couples, not thousands — and drawn largely from one region of the United States.
What is left
The distinction worth holding onto is between two kinds of statement. That trained observers can code small conversational exchanges consistently, and that couples differ measurably in how those exchanges go, is supported by the recordings. That such a coding permits a reliable forecast about any particular couple is not, and the researchers who raised the objection were making a point about statistical procedure that has since been widely accepted.
This pattern is common enough in relationship research to be worth naming on its own. A careful observational finding attracts a quantified claim; the quantified claim is what gets repeated; and the qualification travels a fraction of the distance. Reading the original papers alongside their published critiques is usually the only way to see where one ends and the other begins.
Sources
- Gottman, J. M., & Levenson, R. W. (1992). Marital processes predictive of later dissolution: Behavior, physiology, and health. Journal of Personality and Social Psychology, 63(2).
- Carrère, S., & Gottman, J. M. (1999). Predicting divorce among newlyweds from the first three minutes of a marital conflict discussion. Family Process, 38(3).
- Gottman, J. M., & Levenson, R. W. (2000). The timing of divorce: Predicting when a couple will divorce over a 14-year period. Journal of Marriage and Family, 62(3).
- Heyman, R. E., & Slep, A. M. S. (2001). The hazards of predicting divorce without crossvalidation. Journal of Marriage and Family, 63(2).
- Abraham, L. (2010). Can you really predict the success of a marriage in 15 minutes? Slate, 8 March 2010.