Conclusion and Research Proposal

The analysis of the decision and the outcome of moving out can not be concluded with clear results. We tried to find differences across non-cognitive characteristics which determine the choice to move out. The main problem we faced, was that our data set had no clear identification possibility for all students finishing tertiary education. Therefore we had to rely on the individuals followed from their highs school graduation to their tertiary education degree. This caused a biased sample, which possible suffers from the following deficiencies: First the share of people leaving home is lower(Selection and Social Desirability Bias). Second the treatment and the control group only exists of a fraction of the people graduating of college(Identification Bias). On this sample we still proofed the structural patterns, which where already stated in the literature, but results beyond this layer where very poor. By including personality parameters we again faced a huge cut in the sample, as those where only measured in the middle of the 2000s, except the first measure of Locus of control in 1999. Therefore besides proofing the by the literature indicated structural and social patterns, we only proofed an impact of Agreeableness on the the choice. We tried to proxy for the dimension of the break in social environment by a probit regression on the decision to leave the federal home state. But due to the selection bias, we only got a very small treatment group and therefore couldn't identify any significant parameters. For a closer look at that topic, the GSOEP could be in the more distance future a appropriate data source, as the DIW increases right now the sample size every year and therefore the group, for which the identification is possible, will grow and maybe drive the impact of the biases back.
In the wage regression, we faced the same problems with our sample. But as a lot of individuals in our sample are not yet integrated in the labor market, the GSOEP could be already in the near future appropriate. Also we assumed that the new social environment and therefore larger network of a individual leaving, could have a positive impact. Due to the short time of research we couldn't include the geographical data of every household and therefore couldn't integrate a measure of the dimension of the networks. 
This caused a biased sample which suffers at the following possible deficiencies: First the treatment and the control group only exists of a fraction of the people graduating of college.(Identification bias) Second the share of people leaving is smaller due to the hazard of following up children left the household (Selection Bias). Third individual's, moved out, but didn't declare this to the registration office will make false statements( social desirability) and therefore the share of people moved out will be lower.    
We find for the decision to move out, beyond the by the literature predicted social background parameters, only the big five facet Agreeableness as a significant factor. We face some very strong biases we can't control for and therefore can't really 

Appendix

No cite w\cite{2017}