of Description the Socio-economic Panel (SOEP)
Simplified Model Jan Goebel Date: 3. Dec. 2006
1 Overview
The SOEP is a wide-ranging representative longitudinal study of private households. It provides information on all household members, consisting of Germans living in the Old and New German States, Foreigners, and recent Immigrants to Germany. The Panel was started in 1984. In 2005, there were nearly 12,000 households, and more than 21,000 persons sampled. Some of the many topics include household composition, occupational biographies, employment, earnings, health and satisfaction indicators. The SOEP started in 1984 with two different Samples (called A and B), one for households where the household head holds a German citizenship, and one (oversampled) for households heads with a NonGerman citizenship. Over the years additional samples where added, as described in table 1. Tab 1: SOEP Sub-Samples Description
Sample A B C D E F G H Year Sample of start Size (HH) 1984 1984 1990 1994/5 1998 2000 2002 2006 4528 1393 2179 522 1067 6052 1224 ~ 1000 Sample Region Description
West-Germany West-German Sample West-Germany Over sampling of foreigners East-Germany East-German sample Germany Immigrants since 1984 Germany Supplementary Sample Germany Supplementary Innovation Sample Germany Oversampling of households with high income Germany Refreshment Sample
2 Data Structure
2.1 Principles
The SOEP data is not one rectangular data file, like most other surveys. The normal SOEP distribution a user gets consist of 265 different data files. The data is collected annually, by personnel interviews with each household member aged 16 or older and one additional interview with the household head. Therefore the lion's share of the SOEP data files are either on personnel or on household level, and give information about one year of observation. In fact the bigger part of the SOEP can be treated as annually cross-sectional time series data for persons and households. However, to benefit from the full potential of the longitudinal features the
SOEP Group also provides two special dataset which ensure the linkage between persons and households over time. This central datasets are called HPFAD (for households) and PPFAD (for persons), stemming from the German word Pfad for path. Additional to the datasets which reflect directly the underlying interviews, the SOEP Group provides each year some datasets, which contain generated data. This data encloses a core set of longitudinally checked variables (mostly variables which will be often used by researchers). To keep the model simple other datasets the SOEP encloses are not mentioned here, like never changing Biography information for all persons collected at different years or data on events in spell format. Figure 1 shows the principle structure of the SOEP, at least for some of the main datasets. For a better understanding please have at first a look at the next section about naming conventions.
2.2 Naming conventions
The data is collected annually starting in 1984 and one year of collection is called wave. These waves are arranged alphabetically, i.e. the year 1984 is called wave A, year 1985 wave B and so on until 2005 which called wave V. Therefore the following conventions hold for most of the cross-sectional datasets: Tab 2: Dataset naming conventions
Digit 1 2 Meaning Wave or Year Indicator (A for 1984, B for 1985 ...), e.g. the A in AP Unit of Analysis (H for households and P for Persons), e.g. the P in BH
following Description of the dataset if not purely retrieved from the Person or Household questionnaire, GEN e.g. for generated variables in CHGEN
The following Figure 1 shows the principle structure and the possible linkage options between the dataset focused here. In principle the dataset on personnel level can all be linked with the help of a never changing person ID, called (persnr, i.e. Person number). The household can be linked with the help of the household ID. However, households are not stable over time (household split for example because of move outs of children), therefore you need a wave- or year-specific ID to match households and persons. To ensure the possibility to track also households over time you need an additional ID, the so called original household number, called hhnr. Please note that this variable (hhnr) is NOT unique at the household level, except for the first wave in 1984. This special household number do NOT change over time. Newly created households can not get a household ID which was in use by any other household within the SOEP at any time. This makes it possible to create another household ID to match all persons with their corresponding household history (match PPFAD with HPFAD). This variable is called hhnrakt and consists of the last valid household ID given to this household. A household with is not realized within one year gets a value of -2 (not valid) on the current household ID.
Time (Years)
Wave T 2003
persnr
Wave U 2004
Wave V 2005
PPFAD
TP
persnr
TPGEN
UP
UPGEN
VP
VPGEN
Personal Level
persnr
hhnrakt
thhnr
thhnr
uhhnr
uhhnr
2003
2004
2005
HPFAD
thhnr
TH
THGEN
UH
UHGEN
VH
VHGEN
thhnr
Household Level
uhhnr v hhnr
Fig. 1: Principle Data Structure (simplified)
3 Description of Variables
Apart from the normal description1 of variables a proper SOEP documentation need some additional features. The two main features needed are 1. 2. Track variable names of corresponding variables over time Link to construction variables from other datasets or/and levels or/and years
The problem or complexity is that a model for this must be very flexible, because everything you can think of is possible within the SOEP.
3.1 Example 1: LFS05
Variable Label: Labour Force Status in Year 2005. Corresponding Variable Names in Years 2004 is LFS04, in Year 2003 LFS03, ...
1 Under a normal description i subsume things like variable names and labels, frequencies, as well as question text or sample targeted.
Construction Variables:
Dataset VPGEN:
ERWTYP05 (constructed from Variables VP10 and VP24 from dataset VP)
Dataset VP:
VP05, VP06, VP07, VP08, VP10, VP7201, VP7202, VP7203
Dataset PPFAD:
GEBJAHR
3.2 Example 2: AHINC05
Variable Label: Adjusted monthly household net income in Year 2005. Corresponding Variable Names in Years 2004 is AHINC04, in Year 2003 AHINC03, ... Construction Variables:
Dataset UHBRUTTO:
uhhgr
Dataset VHBRUTTO:
vhhgr
Dataset UPBRUTTO:
upzug, ustell
Dataset VPBRUTTO:
vpzug, vstell
Dataset UPGEN:
HINC04 (constructed from Variable UH4801 from dataset UH)
Dataset VPGEN:
HINC05 (constructed from Variable VH5101 from dataset VH)
Dataset UP:
up09, up5901, up5902, up6801, up6802, up6803, up6804, up6805, up6806, up6807, up6808, up6809, up6810, up6811, up6812, up6813, up6814, up6815, up6816, up6817, up6818, up6819, up6820, up6821, up6822, up6823 vp10, vp7101, vp7102, vp8001, vp8002, vp8003, vp8004, vp8005, vp8006 vp8007, vp8008, vp8009, vp8010, vp8011, vp8012, vp8013, vp8014, vp8015 vp8016, vp8017, vp8018, vp8019, vp8020, vp8021, vp8022, vp10303 vp10304 vp79 GEBJAHR, TODJAHR, UNETTO, VNETTO
Dataset VP:
Dataset PPFAD: