What does artificial intelligence offer beyond traditional statistical models such as regression analysis for investigating household behavior, particularly the factors that trigger marital separation and marital dissolution?
Together with Bruno Arpino (University of Florence) and Marco Le Moglie (Catholic University of Milan), we analyzed data from over 2,000 German married or cohabiting couples. These couples were followed for an average of 12 years by the annual GSOEP survey (German Socioeconomic Survey). panel), more than 900 ended up in separation.
By employing machine learning approaches (especially random survival forests), this procedure independently discovered the relationships between the various elements contained in the database. More than 40 factors were considered in this case, ranging from age to education level, health status to psychological characteristics. A large amount of raw data was fed into ML, and although no precise hypotheses were formulated, it indicated health collapse as simply the event of interest. Combined, the algorithm showed the impact of each element in the data. The variables posing the greatest threat to union stability were identified with 70% accuracy (which exceeds the 50% predictive power achieved by traditional regression methods).
ML was not only able to discover the factors behind a couple’s breakup, but was also able to use this knowledge to predict the couple’s demise in advance. This is also because rather than sending all available data to the ad-hoc algorithm, half was used to direct the algorithm itself, and the validity of the results was verified on the other half of the dataset.
The results of the analysis are very interesting, especially because the ML method allows us to weigh the relative importance of different factors that lead to breakups. Factors that were particularly influential in previous studies, such as unemployment and a partner’s higher education and income, are no longer relevant here.
The four major risk factors that emerged from this study, in descending order, were personal satisfaction, women’s paid work load, some personality factors, and age.
The strongest predictor of separation is personal satisfaction. Obviously a couple won’t last long if both partners are dissatisfied. Less obvious is that when women are very happy in their marriages and men are less so, marital stability is significantly reduced, while the opposite effect is less clear. When women work many hours outside the home, they are at higher risk of separation or divorce, even when men are more involved in household chores (although this result is not new, existing literature suggests that (depending on greater agency and independence) of working women.
In terms of personality traits, high extroversion in men (classically associated with high infidelity) and low openness in women, who are less able to adapt to the changes brought about by cohabitation, are more strongly associated with couple demise. It is a feature that Also, low levels of loyalty in both partners (understood in everyday life as organizational skills and, if low, as disorganized and unable to honor promises) are also not conducive to being together. . However, too high or too low levels of neuroticism can also be a problem. This result can be interpreted as the fact that suffering from excessive anxiety, jealousy, guilt, worry and anger clearly complicates relationships.
This is especially true for women, while those who do not feel this type of emotion may interpret the personality trait as uninteresting in their partner (in this case men). However, no personality combinations were identified that were more strongly associated with relationship breakdown. Finally, when considering age, very young couples tend to be more volatile, while for women she’s over 40, relationship stability increases, but not for men. .
ML analytics is not without limits. The big problem in this case is that it only mentions Germany and gives very little detail about the psychological aspects of the two partners. However, from a methodological point of view, this study demonstrates the great potential of ML techniques in demographic and sociological research in general, monitoring and analyzing a large number of predictors, and automatically predicting linear or nonlinear relationships. It emphasizes the ability to find targets. Finding an additive or non-additive relationship between these factors and the outcome of interest yields better accuracy and more robust estimates against collinearity than commonly used methods.
Courtesy of Bocconi University
