The role of gender culture in predicting fathers’ time-use: Evidence from subnational disparities in Switzerland

Abstract

The present study explores how much time fathers spend on physical and interactive childcare as a function of cultural differences between the 26 Swiss cantons. Two features are important: (1) the subnational design makes the dependency on culture salient and (2) the random forest models quantify the dependency even if it is complex and highly group-specific. The analysis is based on data from the Labour Force Survey about fathers of younger (n = 6985) and older children (n = 6932). Gender culture is construed as family and care ideals aligned on a spectrum between ‘male breadwinner/female part-time care provider’ and ‘dual breadwinner/external care’. The gender culture index turns out to be a high-ranking predictor of fathers’ time spent on interactive care, but not of time spent on physical care. The relation between gender culture and fathers’ time-use is positive with a slight U-shape. Moreover, the models show that 16 socio-demographic variables, together with gender culture, only predict a small part of the variation in fathers’ time-use.

Keywords:

• father involvement
• time use
• gender culture
• Swiss cantons
• random forests

Notes

1. The notion of complexity is similar to the assumption of configurational complexity frequently discussed in comparative research (Ragin, 2008).

2. The full questionnaire can be obtained online. For questions on childcare, see for example pages 160–162 in the 2007 questionnaire (FSO, 2010).

3. The question was not about typical days but only about the preceding day. Interviews on Mondays were randomly about the preceding Saturday or Sunday, so that weekdays are equally represented in the sample.

4. Each random forest was built with 1000 trees, since the errors in prediction (MSE) converge long before. Variable importance measures come from the cforest implementation in the R package party (Hothorn, Bühlmann, Dudoit, Molinaro, & Laan, 2006). Based on conditional inference trees, this implementation better deals with a variable selection bias in variable-importance measures (Strobl et al., 2009). Partial-dependence plots and portion of variance explained were obtained with the randomForest implementation from the R package randomForest (Liaw & Wiener, 2002). Comparing the results on variable importance from cforest with those from randomForest, gender culture is relegated to the 11th rank for interactive care of older children. Despite this drop, it is still judged more important than, for example, urbanity or partner’s education or employment level.

Additional information

Funding

This research was funded by the Swiss National Science Foundation (SNSF) [Project Number 406040_129250]