A project analyzing panel data from the SOEP (Socio-Economic Panel) of the DIW.
From left to right: Distributions for Openness, Agreeableness, Conscientiousness, Extraversion and Neuroticism in the SOEP population (2005-2019)
In the past decades, mental health has become a growing concern across the world. Historically, the research, diagnosis and treatment of mental health conditions has drastically lagged behind corresponding efforts made to tackle physical health conditions. More recently, awareness for the challenges mental disorders pose has risen, however investment in both research and treatment still lags behind the investments made to tackle physical disorders. The Covid19 pandemic and the related mitigation measures have further excacerbated this problem over the last months. Social as well as physical activities are often associated with lower risks for mental health events. However, a lot of these analyses often do not take personalitiy characteristics into account This project has the goal of developing an analysis approach based on Panel-Data from the SOEP, the largest Panel dataset available in Germany, to establish whether the described activities effect the risk for mental health events differently in different personality subsets of the population.
Left: Relationship between Age and Mental Health, Right: Relationship between Age and Physical Health, split for men and women.
In this study, the influence of smoking, alcohol and a range of behaviours on the mental health are assessed in individual models, while controlling for age, sex, wealth and physical health. Both age and sex have a clear impact on mental health. Both wealth and physical health have also been reported to influence mental health. Within and between subject modelling was explored, the choice was made to model between subjects as both the Big Five scales as well as the categorization of the behavioural items remained quite consistent over time and not a lot of variation in the behaviour of individuals between survey years could be observed. To assess potential differences in the effect size and direction between different personality traits, individual models were run on stratified sub-populations. For each test item or scale, POLS (Pooled Ordinary Least Square) regressions were run on the whole population. In addition, for each personality trait, an individual model was run with the data from the 20 % sub-population with the lowest and the 20 % sub-population with the highest score on each respective trait as well as the remaining 60 % of the population. All five personality traits were added as controls when the model was run on the entire population, on the stratified populations the personality trait for which the stratification was performed was removed as a control variable. If not otherwise specified, all analyses were run as Python scripts, some of the pre-processing was performed with Stata.
Three different types of models have been run, the type of model is described in the name of the script:
- combined models: Here items from the SOEP panel describing similar behaviour (e.g. drinking beer, drinking wine, drinking spirits) have been combined to a scale.
- selected models: Here the items describing similar behaviours have been retained as individual variables
- change models: Here the behaviour changes of individuals have been modeled
A detailed description of all analyses performed can be found in the report.