Merging SOEP Datasets: biopupil, youthl, bioparen

FDZ SOEP
, , , ,
1

Dataversion

v41eu

Data Set

biopupil, youthl, bioparen

Please describe your problem

Dear team,

I am currently merging three SOEP datasets:

  1. First dataset- biopupil: Mental health variables (Strengths and Difficulties Questionnaire, SDQ) and school grades measured at age 12.

  2. Second dataset - youthl: Educational outcomes measured at age 17.

  3. Third dataset - bioparen: Parental education and other family background information.

My research question: What is the association between mental health at age 12 and educational outcomes at age 17?

I’m writing with three questions:

  1. Could you please let me know whether using these three datasets together makes sense for this analysis? Should any dataset be replaced?
  2. After merging the first two datasets (biopupil and youthl), I noticed that the sample size decreases by more than 60%. Does this level of attrition seem reasonable, or could it indicate an issue with my merging procedure?
  3. When I merge all three datasets, I find that only 65 additional observations are kept in 2006, while the entire 2007 birth cohort disappears. This makes me wonder whether bioparen is comprehensive for these cohorts, or whether there are differences in coverage that I should be aware of.

I would greatly appreciate any advice or suggestions.

Best regards,

Derrick

_________________________

For reference, the birth year distributions are as follows:

# age 12 sample

> table(data_select$birthy)
2002 2003 2004 2005 2006 2007
606 608 563 643 628 742

# after merging first and second datasets:

> table(merged_data_2_final$birthy)

2002 2003 2004 2005 2006 2007
305 333 219 202 160 210

# after merging all three datasets:

>table(merged_data_2_final_withparent$birthy)
2002 2003 2004 2005 2006
305 333 219 202 65

2
Copilot said:

Dear Derrick,

Thank you for using the SOEP data for your research question. Regarding your questions:

  1. biopupil has been integrated into youthl together with all current youth-related topics. This means that youthl already contains the self-assessment variables that you found in biopupil. The corresponding variable names are ylh0036 to ylh0061. Since youthl is a long-format dataset, you can find both the information provided by the children at age 12 and the educational outcomes of the same individuals in different survey years (syear) within youthl. In addition to bioparen, you can obtain information on the parents as individual-level data. For this purpose, I would also recommend using pl and pgen, particularly for educational qualifications. The linkage between children and parents can be found through the pointer variables in kidlong.

  2. Based on the above, I do not think that merging these two datasets is necessary. However, you are likely to encounter issues related to panel attrition.

  3. Please use the parents’ own information from pl and, if helpful, the generated educational information available in pgen for parental educational attainment and related characteristics.

If you have any further questions, please do not hesitate to contact me.

Best regards,

Jana Nebelin

2 „Gefällt mir“