Inhaltszusammenfassung

The prediction of an individual divorce risk has become a frequent topic, with rates between 80 and 90%. The goals of this study are to demonstrate the influence of base rates in a sample (vs. in the population) and furthermore to show the need for cross validations to augment reliability. Based on cross-sectional archive data sets (Swiss Household Panel SHP: N = 5156; US-insurance company data N = 9941), two samples were spilt for model development and cross validation. In each sample extrem...The prediction of an individual divorce risk has become a frequent topic, with rates between 80 and 90%. The goals of this study are to demonstrate the influence of base rates in a sample (vs. in the population) and furthermore to show the need for cross validations to augment reliability. Based on cross-sectional archive data sets (Swiss Household Panel SHP: N = 5156; US-insurance company data N = 9941), two samples were spilt for model development and cross validation. In each sample extreme groups of most satisfied and most unsatisfied married persons were drawn, in order to match the group size and thus to over represent the divorced, Next, backward logistic regression was used to find predictor sets. These were used in cross validations with and without base rate corrections, using the whole samples and the extreme groups. The influence of base rates on various parameters of prediction was demonstrated. Taking cross validations into account, the rate of correctly predicted divorces (positive predictive value) lies between 20 and 40%. It is concluded that we need to be more cautious and accurate in the interpretation of prediction rates of divorce, especially with print media and clients. » weiterlesen» einklappen

Autoren