Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Sabine Zinn, Ariane Würbach, Hans Walter
Steinhauer, and Angelina Hammon
Attrition And Selectivity of tHe nePS StArting coHortS: An overvieW of tHe PASt 8 yeArS
nePS Survey Paper no. 34Bamberg, March 2018
nePS Survey PAPerS
Survey Papers of the German National Educational Panel Study (NEPS) at the Leibniz Institute for Educational Trajectories (LIfBi) at the University of Bamberg The NEPS Survey Paper Series provides articles with a focus on methodological aspects and data handling issues related to the German National Educational Panel Study (NEPS). The NEPS Survey Papers are edited by a review board consisting of the scientific management of LIfBi and NEPS. They are of particular relevance for the analysis of NEPS data as they describe data editing and data collection procedures as well as instruments or tests used in the NEPS survey. Papers that appear in this series fall into the category of 'grey literature' and may also appear elsewhere. The NEPS Survey Papers are available at https://www.neps-data.de (see section “Publications“). Editor-in-Chief: Corinna Kleinert, LIfBi/University of Bamberg/IAB Nuremberg Contact: German National Educational Panel Study (NEPS) – Leibniz Institute for Educational Trajectories – Wilhelmsplatz 3 – 96047 Bamberg − Germany − [email protected]
A ri on and Selec vity of the NEPS Star ng Cohorts: AnOverview of the Past 8 Years
Zinn, S., Würbach, A., Steinhauer, H. W., & Hammon, A.
Leibniz Ins tute for Educa onal Trajectories
E-mail address of lead author:
methoden@li i.de
Bibliographic data:
Zinn, S., Würbach, A., Steinhauer, H.W. & Hammon, A. (2018). A ri on and Selec vity of theNEPS Star ng Cohorts: An Overview of the Past 8 Years (NEPS Survey Paper No. 34). Bamberg,Germany: Leibniz Ins tute for Educa onal Trajectories, Na onal Educa onal Panel Study.
NEPS Survey Paper No. 34, 2018
Zinn, Würbach, Steinhauer, & Hammon
A ri on and Selec vity of the NEPS Star ng Cohorts: An Overview of the Past 8 Years
AbstractThis ar cle documents the number of target persons par cipa ng in the panel surveys of theNa onal Educa onal Panel Study (NEPS) as well as the number of respondents who temporar-ily dropout and of those leaving the panel (a ri on). NEPS comprises panel surveys with sixmutually exclusive star ng cohorts covering the complete life span. Sample sizes, numbers ofpar cipants and temporary as well as final dropouts and par cipa on rates are reported in de-tail for each wave and for subsamples, if applicable. Sample par culari es, such as the conver-sion of temporary dropouts into final ones, are elaborated on. All figures presented are derivedfrom the corresponding Scien fic Use Files (SUFs) published by February 1, 2018. Selec vitydue to a ri on (i.e., final dropouts) is studied. For this purpose, we examine how a ri on dis-torts the NEPS samples with respect to relevant design variables (such as stra fica on criteria)and panel member characteris cs (like sex and birth year). In detail, we study the panel statusof each panel member, that is being part of the panel or having dropped out finally, along allof the panel waves with respect to star ng cohort and popula on specific characteris cs. Weconclude this ar cle with some recommenda ons for dealing with the detected selec on biasin sta s cal analyses.
NEPS Survey Paper No. 34, 2018 Page 2
Zinn, Würbach, Steinhauer, & Hammon
1 Introduc on
This ar cle documents the number of target persons par cipa ng in the panel surveys of theNa onal Educa onal Panel Study (NEPS) as well as the number of respondents who temporar-ily dropout and of those leaving the panel (a ri on). We introduce discrete me event historymodels as proper means to study panel a ri on and selec vity in NEPS. For this purpose, weconsider all of the six NEPS star ng cohorts and their corresponding Scien fic Use Files (SUFs)published by February 1st, 2018. NEPS is a na onwide study gathering informa on about theeduca onal trajectories of people residing in Germany. To cover the complete life span withrespect to significant educa onal transi ons, it surveys target persons from six mutually exclu-sive star ng cohorts:
Star ng Cohort 1 (SC1) children born between February and July 2012,
Star ng Cohort 2 (SC2) children whose enrollment in school was expected in 2011 tobe in 2012,
Star ng Cohort 3 (SC3) students in grade 5 in regular and special schools in school year2010/11,
Star ng Cohort 4 (SC4) students in grade 9 in regular and special schools in school year2010/11,
Star ng Cohort 5 (SC5) freshman in 2010/11 at universi es and universi es of appliedsciences,
Star ng Cohort 6 (SC6) adults born between 1944 and 1986 living in Germany.
Detailed informa on on the objec ves, the composi on, and the contents of NEPS is givenin Blossfeld, Roßbach, and von Maurice (2011). The popula on and the sampling design ofall star ng cohorts is described in very detail in Würbach, Zinn, and Aßmann (2016) for theSC1, Steinhauer, Zinn, Gaasch, and Goßmann (2016) for the SC2, Steinhauer and Zinn (2016a)for the SC3, Steinhauer and Zinn (2016b) for the SC4, Zinn, Steinhauer, and Aßmann (2017)for the SC5, and Hammon, Zinn, Aßmann, and Würbach (2016) for the SC6. Up to now, thefollowing SUFs have been released, see h ps://www.neps-data.de/:
SC1: Waves 1 to 4 from 2012 to 2015 (SUF version 4.0.0),
SC2: Waves 1 to 6 from 2011 to 2015 (SUF version 6.0.0),
SC3: Waves 1 to 7 from 2010 to 2015 (SUF version 7.0.0),
SC4: Waves 1 to 9 from 2010 to 2015 (SUF version 9.1.0),
SC5: Waves 1 to 9 from 2010 to 2015 (SUF version 9.0.0),
SC6: Waves 1 to 7 from 2009 to 2016 (SUF version 8.0.0).
Taken together all of the SUFs comprise in total 72 studies. Table 1 gives an overview of allof these studies inclusively (NEPS internal) study numbers, study me, survey periods, panelwaves, and survey mode. In each study, survey ques onnaires have been administered in oneof the following survey modes:
NEPS Survey Paper No. 34, 2018 Page 3
Zinn, Würbach, Steinhauer, & Hammon
• CATI: computer assisted telephone interview,
• CAPI: computer assisted personal interview,
• CAWI: computer assisted web interview,
• PAPI: paper and pencil interview.
Some studies allowed respondents to choose betweenmodi, while other studies assigned themrandomly. In few studies special groups of respondents where assigned to a par cular surveymode to increase the likelihood of par cipa on. For example, SC6 panel members who couldnot be interviewed on the phone (via CATI) were automa cally assigned to the CAPI mode.
Generally, target persons are surveyed in two different contexts, either in groups such as testgroups in schools or universi es or individually, for examplewhen interviewedon the telephoneor personally at home. Comprehensive details on this and the NEPS studies in general are givenat the web page of the NEPS data.1
Besides ques onnaires, NEPS also administers competence tests to gather informa on on thedevelopment of knowledge, skills and other competencies relevant for educa onal processesanddecisions. There are domain-general tests such as cogni ve func oning anddomain-specifictests such competencies in mathema cs and reading. In Table 1 waves with tests are markedby a star. Note that target persons at younger ages, i.e. in SC1 and in SC2 from 2011 to 2013,are tested but ques onnaires are answered by their parents. At later ages (i.e., in SC2, SC3 andSC4), both, parents and target persons, are interviewed.
Table 1: A ribu on of studies to star ng cohorts and panel waves.
Wave Time Study Number Mode Period
Star ng Cohort 11⋆ 6-8 months B04 CAPI 2012/132⋆ 16-17 months B05 CATI & CAPI 20133⋆ 25-27 months B91 CAPI 20144⋆ 37-39 months B100 CAPI 2015
Star ng Cohort 21⋆ 4-5 years A12 PAPI 20112⋆ 5-6 years A13 PAPI 20123⋆ Grade 1 A14/A14A PAPI 20134⋆ Grade 2 A15/A15_L1 PAPI 2013/145⋆ Grade 3 A89 PAPI 2014/156⋆ Grade 4 A97/B103 PAPI 2015/16
Star ng Cohort 31⋆ Grade 5 A28/A56/A63 PAPI 2010/112⋆ Grade 6 A29/A57 PAPI 2011/123⋆ Grade 7 A30/A30A/A58 PAPI 2012/134 Grade 8 A31, A59 PAPI 2013/145⋆ Grade 9 A94 PAPI 2014/151See h ps://www.neps-data.de/en-us/datacenter/dataanddocumenta on.aspx.
NEPS Survey Paper No. 34, 2018 Page 4
Zinn, Würbach, Steinhauer, & Hammon
6 Grade 9 A98 PAPI 20157⋆ Grade 10 B106/A99 (CATI&CAWI)/PAPI 2015/16
Star ng Cohort 41⋆ Grade 9 A46/A60/A67/A83/A86 PAPI 20102⋆ Grade 9 A47/A61/A68/A84/A87 PAPI 20113⋆ Grade 10 A48/A62/A69/A85/A88/B37 PAPI/CATI 2011/124 Grade 10 B38 CATI 20125⋆ Grade 11 A49/B39 PAPI/CATI 2012/136 Grade 11 B40 CATI 20137⋆ Grade 12 A50/B41 PAPI/CATI 2013/148 Grade 13 A96/B93 PAPI/CATI 2014/159 Grade 13 B109/B109_O (CATI/CAPI) & CAWI 2015
Star ng Cohort 51⋆ 1st study year B52 CATI 2010/112 2nd study year B54 CAWI 20113 2nd study year B55 CATI 20124 3rd study year B56 CAWI 20125⋆ 3rd study year B59 CATI 20136 4th study year B58 CAWI 20137⋆ 4th study year B94 CATI 20148 5th study year B95 CAWI 20149 5th study year B111 CATI 2015
Star ng Cohort 62 23-65 years B72 CATI/CAPI 2009/103⋆ 24-66 years B67 CAPI/CATI 2010/114 25-67 years B68 CATI/CAPI 2011/125⋆ 26-68 years B69 CAPI/CATI 2012/136 27-69 years B70 CATI/CAPI 2013/147⋆ 28-70 years B97 CAPI/CATI 2014/158 29-71 years B115 CATI/CAPI 2015/16(i) Study numbers star ng with ‘A’ mark studies conducted at schools while study numbers star ng with ‘B’indicate studies conducted via telephone interview, at home, or online. (ii) ⋆ marks waves with competencetests. (iii) A forward slash separa ng survey modes indicates that two modes were offered exclusively and a ‘&’indicates that persons were interviewed by two modes (e.g. because of add-on studies).(iv) In SC1, parents areinterviewed about their children. (v) In the SC2 Waves 1 to 5, children are tested only and not interviewed.(vi) In SC5, test rounds are assigned study numbers, namely B53 in Wave 1, B57 in Wave 5, and B90 in Wave 7.(vii) SC6 starts with Wave 2 since one subsample of the SC6 builds upon the ALWA study(cf. h p://fdz.iab.de/en/FDZ_Individual_Data/ALWA.aspx) which by design cons tutes the first wave of SC6.
The remainder of this ar cle is structured as follows: first, we detail the number of par cipantsand temporary as well as final dropouts along all of the panel waves and star ng cohorts. Sec-ond, we present the results of the selec vity analyses in which we study how a ri on affectsthe composi on of the NEPS samples. We conclude with some recommenda ons for dealingwith the detected selec on bias in sta s cal analyses.
NEPS Survey Paper No. 34, 2018 Page 5
Zinn, Würbach, Steinhauer, & Hammon
2 Par cipants, Dropouts, and A ri on
NEPS surveys target persons togetherwith relevant context persons such as parents, educators,and teachers, where it applies that is in at younger ages in SC1, SC2, SC3 and SC4. This ar cle,however, focuses on the target persons only. Informa on on context persons are providedelsewhere, for example, at the web page of the NEPS data. In the subsequent, a target personis considered to be a par cipant when that person has provided some informa on on him- orherself during a study.2
Ini ally for each star ng cohort a gross sample had been established comprising all of the unitsdrawn to be part of the panel survey. In SC1, SC5, and SC6 the whole gross sample has beenadministered in Wave 1, and each of its members had been asked for panel consent during thefirst wave. All respondents with posi ve consent form the panel cohort of the correspondingstar ng cohort at Wave 1. On the contrary, in SC2, SC3, and SC4 the panel consent had beenobtained before the first wave, thus, the sample administered in Wave 1 already cons tutedthe panel sample. In other words, the people asked to par cipate in the first waves cons tutedifferent samples: in SC1, SC5, SC6 the gross sample, and in SC2, SC3, SC4 the panel sampleat Wave 1. At the start of a specific wave, the panel sample of each star ng cohort consists ofall individuals who ini ally gave their panel consent and did not refuse further par cipa on, orare defined as final dropout because of one of the following two reasons: (i) con nuous non-par cipa on over a period of two years3 or (ii) a response code in a previous study defined tobe an a ri on event. These response codes are:
• respondent refuses par cipa on in general / permanent dele on of address / withdrawpanel consent (for target person),
• death of target person,
• target person already surveyed,
• respondent refuses new address (for target person),
• target person cannot be surveyed / permanently sick or disabled,
• communica on impossible / respondent does not speak enough German / no commu-nica on possible in one of the languages offered,
• respondent refuses par cipa on in general / permanent dele on of all of the data /with-draw panel consent (for target person).
Some mes not all of the members of the panel sample are administered in each panel wave.There are two main reasons for this. First, ques onnaires or tests cannot be administeredbecause of missing contact informa on. This occurs mainly in highly mobile popula ons suchas students gradua ng from school and leaving home for further training or studying. Second,by design only specific subgroups are considered in a wave, for example, only students of aspecific field. Persons who were administered in a study but did not par cipate and who arenot a final dropout are regarded as temporary dropout. Note that final dropouts can occur2In SC1 and in SC2 Wave 1-4 this informa on stems from one parent. In SC2 Wave 1-4 also informa on on thetarget provided by the teacher determines the child as par cipant.
3For reasons of panel stability and because of specific study interests, the rule was adapted from me to me,i.e., not applied consistently in all studies and star ng cohorts. More informa on on this can be found in thestudy methods reports published together with the SUFs.
NEPS Survey Paper No. 34, 2018 Page 6
Zinn, Würbach, Steinhauer, & Hammon
within and between studies: within waves a ri on results from an accordant response code,and between waves a ri on arises because of ac ve refusal or con nuous non-par cipa onover a period of two years. 3
Subsequently, the dis nct panel samples of NEPS are described, broken down by star ng co-hort, panel wave, administered sample, number of par cipants and temporary dropouts aswell as final dropouts within and between waves. In, SC2-SC6 sampling par culari es allow forthe deriva on of design specific subsamples which are considered in our presenta on. Theseare:
Star ng Cohort 2K1_AUG The augmenta on sample of Wave 3. These children were surveyed and
tested in the Grades 1 to 4 (Waves 3-6) in elementary schools, but werenot surveyed or tested in Kindergarten ins tu ons in Waves 1 and 2.
KIGA_IND The group of Kindergarten children, who were tested only in Kindergartenin Waves 1 and 2. These children did not move to an elementary schoolsampled in advance and par cipa ng. While the children are temporarydropouts by design un l Wave 6 the parents were s ll asked for par cipa-on. In Wave 6 these children are surveyed and tested again (at home).
KIGA_PANEL The group of Kindergarten children being surveyed and tested in Kinder-garten inWaves 1 and 2 and transi oned to elementary schools sampled inadvance and par cipa ng. In Wave 3 they have been surveyed and testedtogether with the children of subsample K1_AUG in the Grades 1 to 4.
Star ng Cohort 3G7_AUG The augmenta on sample of Wave 3. These children were surveyed and
tested in the Grades 7 to 10 (Waves 3-6) in school or at home when theyhave le school or the school withdrew par cipa on consent for NEPS.They were not surveyed or tested in Grade 5 or Grade 6 (Waves 1 and 2).
G5 The original panel sample. These children were surveyed and tested in theGrades 5 to 10 (Waves 1-6) in school or at homewhen they have le schoolor the school withdrew par cipa on consent for NEPS.
Star ng Cohort 4 (Waves 3 to 8)4ACA All students educated at a secondary school.VOC All students and persons in voca onal training or in the German transi on
system.
Star ng Cohort 55TEA Freshman students studying for a teacher degree.UNI Freshman students at universi es without TEA.AUN Freshman students at universi es of applied sciences without TEA.PR Freshman students at private universi es.
4Beware that in SC4Wave 1-2 all of the students are surveyed and tested in school, thus in the academic context.At first in Wave 3, students le secondary school to start voca onal training or to enter the German transi onsystem. InWave 9, all SC4 panelmembers have le secondary school, yielding a sample of persons all surveyedand tested individually.
5The subsamples of the SC5 are made up by its explicit strata.
NEPS Survey Paper No. 34, 2018 Page 7
Zinn, Würbach, Steinhauer, & Hammon
Star ng Cohort 6ALWA Persons from the ALWA sample who agreed to par cipate in NEPS.NEPS1 Persons born in the years 1944-86 who gave panel consent during NEPS
Wave 1.NEPS3 The augmenta on sample of NEPS Wave 3 comprising persons born in the
years 1944-86 who agreed to par cipate in NEPS.
The sample of SC2 consists of three subsamples:
K1_AUG The augmenta on sample of Wave 3. These children were survey and testedin the Grades 1 to 4 (Waves 3-6) in elementary schools, but were not surveyedor tested in Kindergarten ins tu ons in Wave 1 and Wave 2.
KIGA_IND The group of Kindergarten children, who were only tested in Kindergartensin Wave 1 and Wave 2. They are temporary dropouts by design un l Wave 6,when they are surveyed and tested again (at home).
KIGA_PANEL The group of Kindergarten children, who were surveyed and tested in Kinder-gartens in Wave 1 and Wave 2 and transi on to elementary schools surveyedin Wave 3. There they have been surveyed and tested together with thechildren of subsample K1_AUG in the Graded 1 to 4.
The sample of SC3 is made up by two subsamples:
K7_AUG The augmenta on sample of Wave 3. These children were surveyed and testedin the Grades 7 to 10 (Waves 3-6) in school or at home when they have le schoolor the school withdrew par cipa on consent for NEPS. They were not surveyed ortested in Grade 5 or Grade 6 (Waves 1 and 2).
K5 The original panel sample. These children were surveyed and tested in the Grades5 to 10 (Waves 1-6) in school or at home when they have le school or the schoolwithdrew par cipa on consent for NEPS.
In the Waves 3 to 8 the sample of SC4 can be composed into two dis nct groups of persons
ACA All students educated at a secondary school.VOC All students and persons in voca onal training or in the German transi on system.
In the Waves 1 and 2 this decomposi on of the SC4 sample does not apply since all of thestudents are surveyed and tested in school (namely in Grade 9), thus in the academic context.At Wave 9 all members of the SC4 have le the school context, yielding a sample of persons allsurveyed and tested individually (i.e., at home, via telephone, or web-based). The sample ofSC5 can be disaggregated into four subsamples made up by its explicit strata
TEA Freshman students studying for a teacher degree.UNI Freshman students at universi es of applied sciences without LA.AUN Freshman students at universi es of applied sciences without LA.PR Freshman students at private universi es.
Finally, the SC6 sample consists of
ALWA Persons from the ALWA sample who agreed to par cipate in NEPS.NEPS1 Persons born in the years 1944-86 who gave panel consent during Wave 1.NEPS3 The augmenta on sample of Wave 4 comprising persons born in the years 1944-88
who agreed to par cipate in NEPS.
NEPS Survey Paper No. 34, 2018 Page 8
Zinn, Würbach, Steinhauer, & Hammon
The figures of SC1 and SC2 are given in the Tables 3 and 4. The Tables 5 and 6 summarizethe numbers of SC3 and SC4, and the Tables 7 and 8 present the numbers of SC5 and SC6.Par cipa on rates are calculated as the ra o between the size of the administrated sampleand the number of par cipants. The Figures 1 to 6 illustrate the panel progress of all star ngcohorts graphically.
2.1 Star ng Cohort 1
The NEPS SC1 (Newborns) started with a gross sample size of 8483 persons (cp. Table 3). InWave 1, 3481 interviews could be realized corresponding to a par cipa on rate of 41.0%. Thepanel cohort reduced to 3431 (par cipa on rate 40.4%) since 42 par cipants gave no panelconsent in Wave 1, and 8 par cipants withdrew their panel consent before Wave 2. The num-bers of Wave 2 are reported separately for CATI and CAPI mode. In the parent interview (CATI)we recorded 2849 respondents, the corresponding par cipa on rate is 83.0%. Addi onally,direct measurements and another parent interview were applied to a random subsample ofthe SC1 panel cohort in Wave 2. In total, 1893 persons were asked for par cipa on and 1510cases could finally be realized corresponding to a par cipa on rate of 79.8%.
0
1000
2000
3000
W1 W2 W3 W4
Cas
es participantstempor. dropoutfinal dropout
Figure 1: Size of Panel Cohort SC1 along Waves.
Among the 2616 realized interviews inWave 3, 2609 are valid (par cipa on rate 79.5%). Seveninterviews are considered invalid due to technical problems during the survey. InWave 4, 2480interviews were realized, but two interviews had been conducted from interviewers withoutapproval for execu on. The data from these two interviews were regarded as not exploitableand thus regarded as temporary dropouts. The corresponding par cipa on rate is 78.8%. Dueto con nuous non-par cipa on over a period of two years 143 of the 541 temporary dropoutsare converted to final dropouts between Waves 4 and 5. Figure 1 displays these numbers,where the height of each bar gives the ini al number of targets with valid panel consent.
NEPS Survey Paper No. 34, 2018 Page 9
Zinn, Würbach, Steinhauer, & Hammon
Table3:
Pane
lProgressS
C1.
Wave
Sub-
Pane
lNot
Administered
Parcipa
nts
Parcipa
onTempo
rary
Fina
lDropo
uts
Fina
lDropo
uts
sample
Coho
rtad
ministered
Rate
Drop
outs
(inWave)
(betwee
nWaves)
1Total
--
8483
3481
0.41
00
5002
50
2Total
3431
034
3128
620.83
446
810
149
CATI
3431
034
3128
490.83
048
010
248
CAPI
3431
1538
1893
1510
0.79
834
043
21
3Total
3281
032
8126
090.79
553
913
35
4Total
3143
031
4324
780.78
854
112
414
7No
tes:‘-’
does
nota
pply.
NEPS Survey Paper No. 34, 2018 Page 10
Zinn, Würbach, Steinhauer, & Hammon
We see that the amount of temporary dropouts remains stable across the panel waveswhereasthe number of final dropouts is adding up, of course.
2.2 Star ng Cohort 2
The NEPS SC2 (Kindergarten) started in 2010with a panel cohort comprising 3007 Kindergartenchildren whose school enrollment was expected to be in the school year 2012/13 (cp. Table 2).In the firstwave, 2949 Kindergarten children par cipated togetherwith their parent. The corre-sponding par cipa on rate is 98.1%. Wave 2 consists of 2727 par cipants yielding an iden calpar cipa on rate as in Wave 1.
0
1000
2000
3000
W1 W2 W3 W4 W5 W6
Cas
es
(a) KIGA (n = 3007)
0
2000
4000
6000
W3 W4 W5 W6
Cas
es
participantstempor. dropoutfinal dropoutnot administered
(b) K1_AUG (n = 6341)
Figure 2: Size of Panel Cohort SC2 along Waves.
In Wave 3 in the school year 2012/13, an augmenta on sample of Grade 1 students (K1_AUG)was drawn and asked for par cipa on. This augmenta on sample is related to the sample ofKindergarten children by the elementary schools to which they pass. The augmenta on grosssample contains 19205 students. In total, 6917 students provided panel consent and are fol-lowed up through their me in elementary school and beyond. A small propor on of thesestudents cons tutes the Kindergarten children who have already been surveyed inWave 1 and2 (576 students in KIGA_PANEL). Among the sample with panel consent, 6733 par cipated inthe survey and tes ng of Wave 3 corresponding to a par cipa on rate of 97.3%. Kindergartenchildren who did not pass to a NEPS school6 are assigned to the field of individual retracking(KIGA_IND). By design, they are not interviewed and tested un l Wave 6 when they are sup-posed to be in Grade 4. Accordingly, fromWave 3 up to Wave 5 they are defined as temporarydropouts. Among the 6340 realized interviews in Wave 4 (par cipa on rate is 96.1%), 5801cases belong to K1_AUG and 539 cases to KIGA_PANEL. In Wave 5, 5799 interviews were real-ized, 5296 cases in the K1_AUG subsample and 503 in the subsample KIGA_PANEL.
6A NEPS school provided consent for par cipa ng in NEPS, i.e., here students could be survey and tested in theirschool context.
NEPS Survey Paper No. 34, 2018 Page 11
Zinn, Würbach, Steinhauer, & Hammon
Table4:
Pane
lProgressS
C2.
Wave
Sub-
Pane
lNot
Administered
Parcipa
nts
Parcipa
onTempo
rary
Fina
lDropo
uts
Fina
lDropo
uts
sample
Coho
rtad
ministered
Rate
Drop
outs
(inWave)
(betwee
nWaves)
1Total
3007
030
0729
490.98
147
110
2Total
2996
215
2781
2727
0.98
154
01
3Total
9336
2419
6917
6733
0.97
318
40
5K1
_AUG
6341
063
4161
760.97
416
50
2KIGA
_IND
2419
2419
--
--
-3
KIGA
_PAN
EL57
60
576
557
0.96
719
00
4Total
9331
2733
6598
6340
0.96
123
226
23K1
_AUG
6339
296
6043
5801
0.96
021
725
15KIGA
_IND
2416
2416
--
--
-2
KIGA
_PAN
EL57
621
555
539
0.97
115
16
5Total
9282
3118
6164
5799
0.94
120
416
177
K1_A
UG
6299
669
5630
5296
0.94
118
514
941
KIGA
_IND
2414
2414
--
--
-31
KIGA
_PAN
EL56
935
534
503
0.94
219
125
6Total
9044
554
8490
6943
0.81
811
8036
769
4†K1
_AUG
6109
6160
4854
620.90
342
516
118
6†KIGA
_IND
2383
458
1925
998
0.51
873
519
249
7†KIGA
_PAN
EL55
235
517
483
0.93
420
1411
†
Notes:‘-’
does
nota
pply.
†Allparen
talrefusalsu
nlW
ave6areinclud
edinthesenu
mbe
rsbe
causeforthe
target
person
sthe
ycomeinto
effecta
tfirstw
henleaving
theelem
entary
scho
ol,w
hich
occursbe
fore
Wave7.
NEPS Survey Paper No. 34, 2018 Page 12
Zinn, Würbach, Steinhauer, & Hammon
The overall par cipa on rate in Wave 5 is 94.1%. All students are asked for par cipa on inWave 6, including those from subsample KIGA_IND. In sum, 6943 students are tested and sur-veyed yielding a par cipa on rate of 81.8%. Among these, 5462 students belong to the K1_AUGsubsample, 483 to the KIGA_PANEL subsample, and 998 students are part of the subsampleKIGA_IND. The number of final dropouts in Wave 6 is far higher for KIGA_IND as compared tothe other two subsamples. This might be due to the fact that this par cular subsample wasnot surveyed for three years. The KIGA_IND subsample was tested and surveyed individuallyin Wave 6. In contrast, students of K1_AUG and KIGA_PANEL are tested and surveyed theirins tu onal context. We see a considerable decrease in the panel cohort size when the schoolcontext was le in Wave 7 and all students together with their parents were tested and sur-veyed individually. In each subsample, the increase in the final dropouts between the Waves 6and 7 is very high. This issue is mainly a ributable to the summa on of all parent withdrawalsof the previous studies. Un l Wave 6 the affected target persons could be surveyed and testedin spite of parental withdrawal. However, in Wave 7 all students transi oned to the individualfield, i.e. ques onnaires and tests are passed at home. That is, in case of an exis ng parentwithdrawal, surveying have had to be abandoned. As a result 526 target persons have had tobe excluded from the panel sample though they were s ll willing to par cipate. Figure 2 vi-sualizes these numbers, where the height of each bar gives the ini al number of targets withvalid panel consent..
2.3 Star ng Cohort 3
The SC3 panel cohort (Grade 5) comprises the two subsamples G5 and G7_AUG. The G5 sub-sample has been established in 2010. Its gross sample consist of 11563 Grade 5 students. Twoyears later, in 2012, the SC3 sample was enriched by the G7_AUG augmenta on sample. Forthis purpose, 3944 Grade 7 students had been drawn and asked for par cipa ng in NEPS.
0
2000
4000
6000
W1 W2 W3 W4 W5 W6 W7
Cas
es
(a) G5 (n = 6112)
0
500
1000
1500
2000
W3 W4 W5 W6 W7
Cas
es
participantstempor. dropoutfinal dropoutnot administered
(b) G7_AUG (n = 2205)
Figure 3: Size of Panel Cohort SC3 along Waves.
NEPS Survey Paper No. 34, 2018 Page 13
Zinn, Würbach, Steinhauer, & Hammon
Table5:
Pane
lProgressS
C3.
Wave
Sub-
Pane
lNot
Administered
Parcipa
nts
Parcipa
onTempo
rary
Fina
lDropo
uts
Fina
lDropo
uts
sample
Coho
rtad
ministered
Rate
Drop
outs
(inWave)
(betwee
nWaves)
1Total
6112
061
1257
780.94
533
40
13
2Total
6099
060
9955
370.90
856
11
8
3Total
8295
082
9572
770.88
298
929
10G5
6090
060
9051
310.84
393
029
10G7
_AUG
2205
022
0521
460.97
359
00
4Total
8256
082
5667
180.81
415
0533
580
G560
510
6051
4783
0.79
012
4919
580
G7_A
UG
2205
022
0519
350.87
825
614
0
5Total
7643
076
4357
780.75
616
2524
00
G554
520
5452
4001
0.73
412
7317
80
G7_A
UG
2191
021
9117
770.81
135
262
0
6Total
7403
074
0355
860.75
517
3978
5G5
5274
052
7439
200.74
312
9262
5G7
_AUG
2129
021
2916
660.78
344
716
0
7Total
7320
241
7079
5491
0.77
615
4246
20G5
5207
150
5057
3924
0.77
611
0429
16G7
_AUG
2113
9120
2215
670.77
543
817
6No
tes:Num
bersforW
ave7arediffe
rent
from
thenu
mbe
rsinthecurren
tSUFversion7.0.1.
Correctednu
mbe
rswillbe
availableinthene
xtSU
Fversion.
NEPS Survey Paper No. 34, 2018 Page 14
Zinn, Würbach, Steinhauer, & Hammon
In sum, 6112 students (i.e., 52.9%) of the G5 gross sample and 2205 students of the G7_AUGgross sample (i.e., 55.9%) provided valid panel consent. Table 5 details the SC3 panel progress,separately for the two samples G5 and G7_AUG. Its third column gives the panel cohort sizeat the beginning of each wave. The columns four and five show the number of students whohad been administered an interview and those who had not. Then, in the columns six to ninethe number of par cipants, temporary, and final dropouts at the end of each wave are given.The last column contains the number of students ac vely refusing further par cipa on in theSC3 panel study. The basically same informa on is provided by Figure 3, where the height ofeach bar gives the ini al number of students with valid panel consent. From both, Table 5 andFigure 3, the large number of students finally dropping out a er Wave 4 is no ceable. This isbecause 578 students in special-need schools were dismissed from the panel.
2.4 Star ng Cohort 4
The gross sample of the SC4 (Grade 9) consists of 26868 students. Of these, 16425 students(61.1%) provided valid panel consent. Table 6 gives details on the SC4 panel progress separatedby its two subsamples ACA (academic track) and VOC (voca onal track). The table provides thepanel cohort size at the beginning of each wave together with the number of students whohad been administered and those who had not. For students who had been administered thefollowing columns give the corresponding status (par cipant, temporary, and final drop out) atthe end of each wave. The last column gives the number of students ac vely refusing furtherpar cipa on in the panel study.
Figure 4 displays the numbers of table 6 graphically. Note that the height of each bar gives theini al number of students with valid panel consent. In the Waves 1 and 2, all students are inACA. From Wave 3 to Wave 8 the students in the academic track (ACA) are located at top ofthe graphic, whereas the students in the voca onal track (VOC) are shown at the bo om of thegraphic. Over me, more and more students leave school for voca onal educa on.
0
5000
10000
15000
W1 W2 W3 W4 W5 W6 W7 W8 W9
Cas
es
participants ACAtempor. dropout ACAnot administeredfinal dropouttempor. dropout VOCparticipants VOC
Figure 4: Size of Panel Cohort SC4 along Waves.
NEPS Survey Paper No. 34, 2018 Page 15
Zinn, Würbach, Steinhauer, & Hammon
Table6:
Pane
lProgressS
C4.
Wave
Sub-
Pane
lNot
Administered
Parcipa
nts
Parcipa
onTempo
rary
Fina
lDropo
uts
Fina
lDropo
uts
sample
Coho
rtad
ministered
Rate
Drop
outs
(inWave)
(betwee
nWaves)
1Total
1642
50
1642
516
106
0.98
131
90
0
2Total
1642
50
1642
515
215
0.92
612
100
61
3Total
1636
48
1635
614
011
0.85
722
3411
10
ACA
013
815
1195
10.86
518
4222
0VO
C8
2541
2060
0.81
139
289
0
4Total
1625
314
440
--
--
75
ACA
1379
3-
--
--
3VO
C64
718
1313
510.74
545
57
2
5Total
1624
113
216
109
1298
20.80
626
4448
34
ACA
063
0557
680.91
552
215
1VO
C13
298
0472
140.73
621
2246
83
6Total
1575
496
35-
--
-60
2AC
A62
89-
--
--
1VO
C33
4661
1953
920.88
166
760
1
7Total
1569
218
515
507
1183
00.76
331
2155
637
ACA
053
3347
360.88
859
25
22VO
C18
510
174
7094
0.69
725
2955
115
8Total
1509
913
1813
781
9871
0.71
634
0051
015
51AC
A0
688
610
0.88
575
316
VOC
1318
1309
392
610.70
733
2550
715
35
9Total
1303
80
1303
890
440.69
432
6273
212
64No
te:‘-’do
esno
tapp
ly.
NEPS Survey Paper No. 34, 2018 Page 16
Zinn, Würbach, Steinhauer, & Hammon
Hence, the number of students in the top part (ACA) declines, whereas the number of studentsin the bo ompart (VOC) increases. InWave 9 all students have le school and thus dis nguish-ing ACA and VOC is not any longer necessary. From both, Table 6 and Figure 4, some numbersare no ceable. First, in Wave 4 and Wave 6 the majority of students had not been adminis-tered. This is because these two waves were targeted only at students in the voca onal trackwho had par cipated in the previous wave (Wave 3 and Wave 5) to keep in touch. Second, inWave 8 a large number of students had not been administered. These aremainly students fromspecial-need schools, for whom further financing was unclear. However, star ng from Wave 9financing was secured again and the majority of these students repar cipated. The large num-ber of final dropouts a er Waves 8 and 9 is caused by conver ng temporary dropouts to finalones because of con nuous nonpar cipa on over a period of two years. Due to this, 1396 stu-dents were defined as final dropouts and removed from the panel sample a er Wave 8, andanother 1246 students a er Wave 9.
2.5 Star ng Cohort 5
For SC5 (First-Year Students), in total 31082 freshmen with valid contact informa on could berecruited at private and public universi es and universi es of applied science. These cons -tute the SC5 gross sample. From these, 17910 persons took part in Wave 1 and gave theirpanel consent. This corresponds to 57.6% of the administered cases and is the panel cohort ofSC5. The remaining cases are ascribed to the final dropouts of Wave 1. Table 7 details the SC5panel progress separated by its four subsamples TEA (freshman studying for a teacher degree),UNI (freshman at universi es without TEA), AUN (freshman at universi es of applied scienceswithout TEA), and PR (freshman at private universi es). In the Wave 1 competence tests, onlyone third (33.2%) of the panel cohort took part. In the Waves 2-9, par cipa on rates fluctuatebetween 58.8% and 73.5%. We find that the par cipa on rates in the CAWIs (Waves 4, 6, and8) is generally lower than those in the CATIs conducted earlier in the same year (Waves 3, 5,and 7).
0
5000
10000
15000
W1 W2 W3 W4 W5 W6 W7 W8 W9
Cas
es
participantstempor. dropoutfinal dropoutnot administered
Figure 5: Size of Panel Cohort SC5 along Waves.
NEPS Survey Paper No. 34, 2018 Page 17
Zinn, Würbach, Steinhauer, & Hammon
Table7:
Pane
lProgressS
C5.
Wave
Sub-
Pane
lNot
Administered
Parcipa
nts
Parcipa
onTempo
rary
Fina
lDropo
uts
Fina
lDropo
uts
sample
Coho
rtad
ministered
Rate
Drop
outs
(inWave)
(betwee
nWaves)
1Total
--
3108
217
910
0.57
60
1317
20
TEA
--
7864
5555
0.70
60
2309
0UNI
--
1190
480
240.67
40
3880
0AU
N-
-74
6038
940.52
20
3566
0PR
--
3854
437
0.11
30
3417
0
1T†
Total
1791
00
1791
059
490.33
211
955
60
TEA
5555
055
5520
210.36
435
313
0UNI
8024
080
2427
150.33
853
072
0AU
N38
940
3894
1115
0.28
627
781
0PR
437
043
798
0.22
433
90
0
2Total
1790
40
1790
412
273
0.68
556
0427
13TEA
5552
055
5238
390.69
117
058
2UNI
8022
080
2256
090.69
923
9815
8AU
N38
930
3893
2510
0.64
513
803
3PR
437
043
731
50.72
112
11
0
3Total
1786
49
1785
513
113
0.73
545
6118
134
TEA
5542
255
4042
530.76
812
3750
11UNI
7999
479
9558
410.73
120
7777
11AU
N38
873
3884
2701
0.69
611
3548
10PR
436
043
631
80.72
911
26
2
4Total
1764
90
1764
911
202
0.63
564
3215
19TEA
5481
054
8136
950.67
417
824
2UNI
7911
079
1150
030.63
329
026
12AU
N38
290
3829
2219
0.58
016
055
5
NEPS Survey Paper No. 34, 2018 Page 18
Zinn, Würbach, Steinhauer, & Hammon
Table7con
nued
:Pan
elProg
ress
SC5.
Wave
Sub-
Pane
lNot
Administered
Parcipa
nts
Parcipa
onTempo
rary
Fina
lDropo
uts
Fina
lDropo
uts
sample
Coho
rtad
ministered
Rate
Drop
outs
(inWave)
(betwee
nWaves)
PR42
80
428
285
0.66
614
30
0
5Total
1761
50
1761
512
694
0.72
146
2629
53
TEA
5475
054
7541
860.76
512
1772
0UNI
7893
078
9356
150.71
121
5112
70
AUN
3819
038
1925
820.67
611
4889
3PR
428
042
831
10.72
711
07
0
5T†
Total
1731
70
1731
787
670.50
685
455
59TEA
5403
054
0329
070.53
824
951
17UNI
7766
077
6639
630.51
038
012
29AU
N37
270
3727
1687
0.45
320
382
10PR
421
042
121
00.49
921
10
3
6Total
1725
30
1725
310
183
0.59
070
4723
7TEA
5385
053
8533
520.62
220
294
1UNI
7735
077
3545
940.59
431
2615
5AU
N37
150
3715
1975
0.53
217
364
1PR
418
041
826
20.62
715
60
0
7T†
Total
1722
316
623
600
446
0.74
313
024
2TEA
5380
5323
5743
0.66
714
00
UNI
7715
7372
343
261
0.76
168
140
AUN
3710
3552
158
111
0.70
338
92
PR41
837
642
310.73
810
10
7Total
1719
727
4114
456
9611
0.66
544
2641
921
13TEA
5380
2741
2639
1924
0.72
965
263
566
UNI
7701
077
0151
330.66
723
8718
198
0
NEPS Survey Paper No. 34, 2018 Page 19
Zinn, Würbach, Steinhauer, & Hammon
Table7con
nued
:Pan
elProg
ress
SC5.
Wave
Sub-
Pane
lNot
Administered
Parcipa
nts
Parcipa
onTempo
rary
Fina
lDropo
uts
Fina
lDropo
uts
sample
Coho
rtad
ministered
Rate
Drop
outs
(inWave)
(betwee
nWaves)
AUN
3699
036
9922
770.61
612
6715
552
2PR
417
041
727
70.66
412
020
45
8Total
1466
51
1466
486
290.58
860
2411
1TEA
4751
047
5129
330.61
718
171
0UNI
6540
165
3939
450.60
325
877
0AU
N30
220
3022
1546
0.51
214
733
1PR
352
035
220
50.58
214
70
0
9Total
1465
31
1465
210
096
0.68
943
2123
591
9TEA
4750
047
5034
300.72
212
5268
276
UNI
6533
165
3245
220.69
219
3674
411
AUN
3018
030
1818
980.62
910
3981
214
PR35
20
352
246
0.69
994
1218
Notes:‘-’
does
nota
pply.
†Testsw
erecond
uctedinow
nstud
iesd
isnctfrom
thesurvey
interviews.In
Wave1an
d5thetesng
phaseen
deda
ertheinterview
phase,an
dinWave7tesng
was
cond
uctedbe
fore
theinterviewss
tarted
.
NEPS Survey Paper No. 34, 2018 Page 20
Zinn, Würbach, Steinhauer, & Hammon
In Wave 7, the oversampling part of the TEA subsample has not been administered (i.e., 15.9%of the Wave 7 panel sample) because at this me its further financing was not secured. How-ever, it was again star ng with Wave 8. In Wave 7, for the first me study members are consid-ered as final dropouts because of con nuous nonpar cipa on over a period longer than twoyears. As a consequence, the propor on of people dropping out from the sample (betweenstheWaves 7 and 8) is no ceably higher than in the waves before. Because of the same reason,a er Wave 9 a large propor on of temporary dropouts was declared to be final dropouts. Inthe Waves 1, 5, and 7 competence tests took place. The Wave 7 competence test was onlyadministered to a par cular subgroup of the panel cohort, namely to 600 business administra-on students. Compared to the par cipa on in the Wave 5 tes ng (50.6% of the administered
cases), par cipa on in the Wave 7 tes ng was high, i.e. 74.3% of the administered cases. InWave 9, five years a er study start, most students graduated and/or le university. Thus, theirpropensity to take (further) part in a student sample likely declines. Figure 5 displays the num-bers of table 7 graphically. Note that the height of each bar gives the ini al number of studentswith valid panel consent, that is, the 17910 students who took part in Wave 1 and gave theirpanel consent.
2.6 Star ng Cohort 6
The sample of the SC6 (Adults) consists of three subsamples: the par cipants of theALWA studywho agreed to con nue to par cipate in NEPS (ALWA), the newly drawn individuals of the firstNEPS wave (NEPS1)7 and the individuals of the refreshment sample in the third wave of theNEPS (NEPS3). Table 8 details the SC6panel progress separatedby its subsamples ALWA,NEPS1,and NEPS3. The column ”Not administered“ involves individuals who did not ac vely withdrawtheir panel consent, but who could not be contacted anymore (e.g., because of missing validcontact informa on).
0
2500
5000
7500
10000
12500
W1 W2 W3 W4 W5 W6 W7
Cas
es
(a) ALWA/NEPS1 (n = 11932)
0
1000
2000
3000
4000
5000
W3 W4 W5 W6 W7
Cas
es
participantstempor. dropoutfinal dropoutnot administered
(b) NEPS3 (n = 5208)
Figure 6: Size of Panel Cohort SC6 along Waves.
7In the SUF, the first NEPS wave is denoted as Wave 2.
NEPS Survey Paper No. 34, 2018 Page 21
Zinn, Würbach, Steinhauer, & Hammon
Table8:
Pane
lProgressS
C6.
Wave
Sub-
Pane
lNot
Administered
Parcipa
nts
Parcipa
onTempo
rary
Fina
lDropo
uts
Fina
lDropo
uts
sample
Coho
rtad
ministered
Rate
Drop
outs
(inWave)
(betwee
nWaves)
2Total
8997
027
009
1164
90.43
119
2713
433
1381
ALWA
8997
089
9765
720.73
019
2749
810
97NEP
S1-
018
012
5077
0.28
20
1293
528
4
3Total
1219
50
1219
593
230.76
425
6630
651
1ALWA
7402
074
0256
390.76
315
8218
151
1NEP
S147
930
4793
3684
0.76
998
412
50
4Total
1139
00
2850
114
112
0.49
518
0612
583
414
ALWA
6714
067
1453
800.80
110
2331
120
4NEP
S146
760
4676
3524
0.75
478
336
921
0NEP
S3-
017
111
5208
0.30
40
1190
30
5Total
1550
425
515
249
1169
60.76
721
1314
400
ALWA
6199
361
9648
800.78
875
755
90
NEP
S140
978
4089
3100
0.75
854
844
10
NEP
S352
0824
449
6437
160.74
980
844
00
6Total
1380
925
113
558
1063
90.78
523
5456
552
8ALWA
5637
114
5523
4555
0.82
581
415
416
1NEP
S136
4811
935
2928
470.80
752
016
211
4NEP
S345
2418
4506
3237
0.71
810
2024
925
3
7Total
1246
522
1244
397
700.78
517
7190
234
4ALWA
5208
252
0641
890.80
573
728
010
9NEP
S132
5310
3243
2604
0.80
338
525
482
NEP
S340
0410
3994
2977
0.74
564
936
815
3
NEPS Survey Paper No. 34, 2018 Page 22
Zinn, Würbach, Steinhauer, & Hammon
Table8con
nued
:Pan
elProg
ress
SC6.
Wave
Sub-
Pane
lNot
Administered
Parcipa
nts
Parcipa
onTempo
rary
Fina
lDropo
uts
Fina
lDropo
uts
sample
Coho
rtad
ministered
Rate
Drop
outs
(inWave)
(betwee
nWaves)
8Total
1119
710
1118
792
360.82
614
5849
361
6ALWA
4817
248
1540
990.85
155
416
261
6NEP
S129
074
2903
2450
0.84
432
213
10
NEP
S334
734
3469
2687
0.77
558
220
00
Notes:‘-’
does
nota
pply.
NEPS Survey Paper No. 34, 2018 Page 23
Zinn, Würbach, Steinhauer, & Hammon
Because of convenience, these cases were completely excluded from the panel.8 The column”Administered“ contains for the Waves 2 and 4 the gross sample sizes of the newly drawn indi-viduals of the subsamples NEPS1 and NEPS3.9 In total, 11649 individuals par cipated in Wave2 and gave their panel consent. This corresponds to 43.1% of the administered cases. In Wave2, 1927 members of the ALWA sample dropouts out temporarily. From these, 833 individualswere readministered in Wave 3 and 283 repar cipated. These cases (i.e., N=283), combinedwith the par cipants of Wave 2, cons tute the panel sample of SC6. In Wave 3, 76.4% of theadministered cases par cipated in the interview. In Wave 4, the ini al panel sample was aug-mented by a refreshment sample of 17111 persons. From the drawn gross sample, 30.4%par cipated in the panel study and gave panel consent. We see that the ALWA members aremore likely to par cipate in the survey than the individuals from the two other NEPS samples.In par cular, the NEPS3 subsample shows a strong decline in par cipa on rates: In the latestWave 8 only 77.5% of the administered persons agreed to par cipate, compared to 85.1% inthe ALWA sample. Figure 6 illustrates the SC6 panel progress. It is obvious that the tempo-rary dropouts decline more and more as me went by since at la er waves the panel consistsmainly of people who are willing to further par cipate.
3 Selec vity Analyses
Non-random a ri on across all of the panel waves is a common issue in non-mandatory panelsurveys. It does not pose a problem as long as it is accounted for in sta s cal inference. Oth-erwise, biased results might lead to erroneous research conclusions. In NEPS, selec vity (onthe level of the respondent) arises at two dis nct stages: in the ini al sample due to unit-nonresponse in the gross sample (yielding the panel samples at Wave 1) and due to wave non-response. Unit-nonresponse in the gross sample is usually handled by weighted analysis usingnon-response adjusted design weights or by including relevant design variables into the focalmodel of the substan ve research ques on. Non-response adjusted design weights are partof the SUFs (in theWeights file) and the design variables are described in detail in the sampledocumenta on. For further informa on see Würbach et al. (2016) for the SC1, Steinhauer etal. (2016) for the SC2, Steinhauer and Zinn (2016a) for the SC3, Steinhauer and Zinn (2016b) forthe SC4, Zinn et al. (2017) for the SC5, and Hammon et al. (2016) for the SC6. In a second step,a ri on along the panel waves has to be studied and individuals with higher dropout propen-si es to be revealed. This informa on can then be used to correct for non-random selec onprocesses in sta s cal analysis. Corresponding approaches are described in Sec on 4.
The main issue to start with is the examina on of the a ri on processes present in the NEPSStar ng Cohorts 1 to 6. Concretely, we explore how a ri on (final dropouts) distorts the NEPSpanel sampleswith respect to relevant design variables (such as stra fica on criteria) and panelmember characteris cs (like sex and birth year). For this purpose, we study the panel statusof each panel member–being part of the panel sample vs. final dropout–across all of the panelwaves with respect to star ng cohort and target popula on specific characteris cs. For consis-tency reasons, we consider some variables in each of themodels (corresponding to the dis nctstar ng cohorts). Each model comprises the region where a person is surveyed (Eastern Ger-many inclusively Berlin vs. Western Germany), her/his gender (female vs. male), the year of8These cases are not subsumed under the final dropouts.9For the remaining waves, this column reports all panel members who were asked for an interview and/or forpar cipa ng in competence tests.
NEPS Survey Paper No. 34, 2018 Page 24
Zinn, Würbach, Steinhauer, & Hammon
birth, the migra on background (target person and/or one of her/his parents are born abroadvs. otherwise)10, and the CASMIN of the father and/or the mother (elementary, secondary,and higher level of educa on according to length of educa onal experiences).11 If the percent-age of missing values in a variable exceeds 5%, we specify a missing category for this variable,otherwise missing values are imputed.12
We use discrete me event history models (see, e.g., Hougaard, 2000; Kalbfleisch & Pren ce,2002) to capture the dynamic nature of the a ri on process. Discrete me event history mod-els are perfectly suited to this kind of problem. At each panel wave, relevant variables areregressed on wether a ri on was observed for a panel member or not. Proceeding this way,the impact of me and individual characteris cs are considered simultaneously when model-ing propensi es for final dropouts. All models are specified as propor onal hazards model, socalled Cox models named according to their inventor (Cox, 1972). Hence, in our models theunique effect of a unit increase in a covariate is assumed to be mul plica ve with respect tothe a ri on propensity. To preserve the propor onal hazard property–as required by the Coxmodel–we specify our models as generalized linear models with a cloglog link func on.13 Allmodels across all star ng cohorts are es mated using the glm func on of the sta s cal so -ware R (R Core Team, 2017), see for example Broström (2012). Again, each of the star ngcohorts is analyzed and described in very detail separately.
3.1 Star ng Cohort 1
The SC1 panel sample consists of four waves with surveys in an interval of approximately oneyear covering the me period 2012 to 2015. Star ng from a gross sample of 8483 targets, 3481individuals responded in Wave 1. The corresponding logit model with the propensi es for par-cipa on is given in Würbach et al. (2016, Chap. 4.1). This model contains only a restricted set
of explaining variables owing to the fact that very limited informa on was available in advancefrom the registra on offices (asked for providing informa on on the target popula on). Theseare mainly characteris cs of the newborns used for sampling. Addi onal informa on from thehistory of contacts was included. That is, the number of contact a empts was used to controlfor accessibility. This model indicates onlymodest selec vity of the par cipants with respect tothe gross sample. Respondents with non-German ci zenship show a slightly lower propensityfor par cipa on than respondents with German ci zenship.
Table 9 documents the results of the selec vity analysis regarding the latest published SUF forthe SC1 (Waves 1 to 4). The figures are reported in reference to the panel sample of the SC1 atstart (N=3431). In the SC1 the target popula on are newborns but the respondents are theirlegal guardians. It is possible that the contact person changes between twowaves, for example,in the first two waves we got all informa on from the mother and in the last two waves thefather par cipated and gave informa on (both with panel consent). If there was no change ofthe contact person, all relevant child and parent data was carried over from previous CATI.
10This characteris c is quan fied by the genera on status variable provided by the NEPS, see Olczyk, Will, andKristen (2014).
11Further informa on on the CASMIN classifica on is given in, for example, Brauns and Steinmann (1997).12Imputa on was done by mul variate imputa on by chained equa on with one repe on step. We used the R
packagemice fot this to do, see van Buuren and Groothuis-Oudshoorn (2011).13It can be shown that there is a direct rela onship between the Cox model and a binary dependent variable
model with a cloglog link func on, see for example Beck (2008).
NEPS Survey Paper No. 34, 2018 Page 25
Zinn, Würbach, Steinhauer, & Hammon
Table 9: Selec vity Analysis for the SC1 Panel Sample along Waves 1-4.
Variable Reference category Hazard Ra o p-Value
Gender (P) femalemale 0.828 0.581
Year of Birth (P) <19761976-1980 1.044 0.8001981-1985 0.967 0.822>1985 0.891 0.447
Month of Birth (T) FebruaryMarch 0.986 0.924April 0.910 0.558May 0.682 0.018June / July 1.093 0.552
Region Eastern GermanyWestern Germany 0.800 0.089
BIK <50,000 inhabitants50,000 up to 500,000 1.079 0.609> 500,000 inhabitants 0.983 0.909
CASMIN (P) 1a, 1b, 2b1c, 2a 0.864 0.3312c 0.655 0.0063a, 3b 0.431 <0.001no informa on 0.488 0.253
Employment Status (P) employednot employed 3.859 <0.001no informa on 0.726 0.729
Migra on Background (P) noyes 1.571 <0.001Family Status married / life partnershipdivorced / widowed 0.893 0.705single 1.084 0.517no informa on 11.411 <0.001
Numbers of Children in Household 1 child2 children 0.812 0.0663 children 0.937 0.6914 children or more 0.643 0.120
N 3431Notes: Dependent variable is a ri on (yes or no). (P) parent informa on, (T) target informa on.
NEPS Survey Paper No. 34, 2018 Page 26
Zinn, Würbach, Steinhauer, & Hammon
In case of change, usually the parent data was obtained from the new respondent and thusbeing updated. This updated informa on is used for modeling. The remaining missing valuesare imputed as men oned above. We considered the residen al community size, the employ-ment status and the family status of the repor ng parent as well as the number of childrenin the household as relevant variables to model a ri on in SC1. All covariates included wereregarded as me invariant, because changes–if at all–are only modest.In detail, Table 9 reports the hazard ra os for a ri on across all of the four waves observed sofar. The results show a significant increase in the propensity to drop out from the panel sam-ple when the respondent is currently unemployed or has a migra on background (genera onstatus lower than three) compared to their reference categories. Moreover, respondents witha higher level of educa on have a remarkably lower propensity to be a final dropout. Opposedto respondents with school leaving cer ficate lower or equal to secondary educa on withoutvoca onal training (reference category), respondents of the groups higher educa on entrancequalifica on (with or without voca onal training) as well as respondents with university degreeor a technical college qualifica on are significantly more willing to par cipate. Regarding thehousehold and family structure two further outcomes emerge. Missing informa on for familystatus is strongly associated with a ri on. In addi on, we see a tendency for large families tobe more willing to par cipate. That is, having two or more children in the household increasesthe propensity to stay in the panel sample, though not being significant. The me effects werehighly significant, indica ng significant a ri on at all of the waves following Wave 1.
3.2 Star ng Cohort 2
The SC2 panel sample consists of six waves with one survey every year covering the me pe-riod 2011 to 2016. In Wave 1 the SC2 panel sample contains 3007 children from kindergarten.Compared to the gross sample (N=4515), the panel sample has a lower propor on of childrennot speaking German at home. Furthermore, the panel sample comprises a lower propor onof children raised by a single parent opposed to children being raised by both parents. The cor-responding logit model with the propensi es for par cipa on is given in Steinhauer, Aßmann,Zinn, Goßmann, and Rässler (2015, Chap. 3.1).The panel sample of the augmenta on subsample K1_AUG (N=6341) reveals only minor selec-vity of par cipa ng school children compared to the gross sample (N=16784). We found that
the propor on of children being earlier enrolled for school is slightly lower than in the grosssample, see Steinhauer et al. (2016, Chap. 3.2). Again, the set of variables used for analyzingselec vity between the gross and net sample is naturally restricted to the sampling informa-on (because no other informa on was available in advance). Please note, that no general
statements can be made regarding the selec vity apart from this.Table 10 documents the results of the selec vity analysis regarding the latest published SC2SUF (Waves 1 to 6), in which all subsamples (KIGA_IND, KIGA_PANEL, K1_AUG) were tested andsurveyed again. The figures are reported in reference to the SC2 panel samples at start (N=9336in total) but separately for each of the three subsamples. The number of explaining variablesdiffers between the subsamples. For the children of the augmenta on subsample (K1_AUG)a lot of informa on from the target as well as the school context is available. We consideredthe level of urbaniza on, the funding of the school, the me of enrollment for school as wellas the presence of special educa onal needs as relevant variables to model a ri on in the SC2subsample K1_AUG.
NEPS Survey Paper No. 34, 2018 Page 27
Zinn, Würbach, Steinhauer, & Hammon
Table10
:Selec
vity
Analysisforthe
SC2Pa
nelSam
plealon
gWav
es1-6(KIGA_
Pane
l/KIGA_
IND),a
ndWav
es3-6(K1_
AUG),respe
cvely.
KIGA
_PAN
ELKIGA
_IND
K1_A
UG
Varia
ble
Referencecatego
ryHa
zard
Rao
p-Va
lue
Hazard
Rao
p-Va
lue
Hazard
Rao
p-Va
lue
Gend
er(T)
female
male
1.64
90.14
11.15
80.27
01.07
80.45
7
Year
ofBirth(T)
2004
/05
2006
/07
1.10
10.78
91.02
30.88
41.17
60.16
5
Region
EasternGe
rman
yWestern
Germ
any
1.04
70.91
31.14
50.42
71.71
60.00
2
Urban
izaon
Level
rural
semi-u
rban
0.41
20.04
7-
-0.97
40.88
3urba
n0.72
30.47
1-
-1.12
60.49
9
Fund
ingof
Scho
olprivate
public
--
--
0.61
90.02
3
Scho
olen
rollm
ent
earlier
later
--
--
0.76
40.44
0regu
lar
--
--
0.80
40.30
7
Specialedu
caon
alne
eds
noyes
--
--
1.34
20.25
7CA
SMIN
(P)
1a,1
b,2b
1c,2
a0.94
50.94
00.82
80.34
11.64
50.05
12c
0.80
60.78
40.75
90.22
31.34
70.25
83a
,3b
0.39
70.27
20.64
20.06
80.80
40.43
3no
inform
aon
1.16
80.87
70.54
80.01
72.24
30.28
2
Migra
onba
ckgrou
nd(P)
noyes
--
--
0.74
10.03
3no
inform
aon
--
--
0.48
90.33
5
N57
624
1963
41No
tes:De
pend
entv
ariableisa
rion
(yes
orno
).(P)p
aren
tinforma
on,(T)
target
inform
aon
.
NEPS Survey Paper No. 34, 2018 Page 28
Zinn, Würbach, Steinhauer, & Hammon
Similarmanifold informa on is available for the school children from the subsample KIGA_PANEL.However, due to the small overall sample size (N=576) and the resul ng small case numbersin single cells, some variables were inten onally excluded when modeling a ri on for theKIGA_PANEL to increase efficiency. Concretely, this applies to the funding of the school, thelevel of urbaniza on, the school enrollment, the special educa onal needs as well as the mi-gra on background of the parent. When modelling a ri on propensi es in the KIGA_IND sub-sample, we added to the variables described in the introduc on of this sec on the urbaniza onlevel. All covariates included were regarded as me invariant, because changes–if at all–areonly modest.
Table 10 reports the hazard ra os for a ri on across all six waves observed so far (i.e., Waves3 to 6 for K1_AUG, respec vely) in detail. In all three subsamples targets whose parents havea higher level of educa on show a remarkably lower propensity to be a final dropout, though,not being significant. Opposed to targets of parents with school leaving cer ficate lower orequal to secondary educa on without voca onal training (reference category), having parentsof the groups higher educa on entrance qualifica on (with or without voca onal training) aswell as having parents with university degree or a technical college qualifica on significantlyincreases willingness of the target to par cipate.
In the KIGA_PANEL subsample the propensity to drop out from the panel sample is significantlydecreased for targets living in semi-urban areas opposed to those living in a rural area. Forthe KIGA_IND subsample only the missing informa on regarding the CASMIN of the parentsshows a significant effect on the panel a ri on. However, the effect is counterintui ve be-cause the presence of missingness in the CASMIN is related to a lower propensity for a ri onhere. The results show that in subsample K1_AUG respondents fromWestern Germany have asignificantly increased propensity to drop out from the panel compared to those from EasternGermany including Berlin. Regarding the funding of the school and the migra on backgroundof the parents we observe posi ve effects on panel willingness. Children from public schoolsas well as school children with parents having a genera on status lower than three are morewilling to par cipate.
The me effects were highly significant at all waves for the KIGA_PANEL and K1_AUG subsam-ples, indica ng a significant loss of panel members at all of the waves following Wave 1 forKIGA_PANEL, and a er Wave 3 for K1_AUG, respec vely. The me effects for KIGA_IND areinsignificant up to Wave 6. This is not surprising, because KIGA_IND was pending in the Waves3 to 5.
3.3 Star ng Cohort 3
The SC3 panel sample covers seven waves, mostly in an interval of one year, ranging from2010 to 2016. During this me, 6112 students (subsample G5) have been surveyed and testedfrom Grade 5 to Grade 10. The 2205 students of subsample G7_AUG have been surveyed andtested from Grade 7 to Grade 10. The relevant design variable used for stra fica on in bothsubsamples is the school type in which a student had ini ally been sampled. The correspond-ing secondary school types (offering educa on to students in the Grades 5 to 10) are listed inTable 11.
Some students changed schools and possibly also school types over the course of the panel.Unfortunately, there is no consistent and complete informa on on the school type histories
NEPS Survey Paper No. 34, 2018 Page 29
Zinn, Würbach, Steinhauer, & Hammon
Table 11: School types in Germany
Abbrevia on German name Englisch name
FS Förderschulen schools offering schooling to students with spe-cial educa onal needs in the area of learning
FW Freie Waldorfschulen Rudolf Steiner schools
GS Grundschulen elementary schools
GY Gymnasien schools leading to upper secondary educa onand university entrance qualifica on
HS Hauptschulen schools for basic secondary educa on
IG IntegrierteGesamtschulen
comprehensive schools
MB Schulen mit mehrerenBildungsgängen
schools with several courses of educa on
OS SchulformunabhängigeOrien erungsstufen
schools only covering the orienta on stage
RS Realschulen intermediate secondary schools
of the SC3 panel members available. This is why we s ck to the sampling informa on whenmodelling a ri on propensi es. In addi on to the individual characteris cs described in theintroduc on of this sec on, we consider themathema cal competence of a student in Grade 5and Grade 7 (low, medium, high, and no informa on) as explanatory model variable. All of theconsidered covariates are me invariant. This also holds for the mathema cal competencies inGrade 5 and Grade 7, incorporated as cross-sec onal informa on into themodel because therewas no tes ng in Grade 6. Table 12 shows the results of the respec ve analysis for the two sub-samples of SC3. For the subsample G7_AUG there are no es mates displayed for mathema calcompetence in Grade 5, because this informa on is not available by design. Further, there areno es mates given for certain school types (special need schools FS, elementary schools GS, andorienta on stage schools OS), because either no students were sampled in the correspondingschool type (FS), or the school type does not host any students in Grade 7 (GS, OS). In the firstfour waves, G5 contains students with special needs sampled in school type FS. Since thesestudents were dismissed from the panel a erWave 4 (cp. Table 5), we excluded them from ouranalysis. The dominant effect of having no informa on on several variables on the a ri onpropensity is obvious, although only relevant for mathema cal competence among studentsof the G5 subsample. Besides that, students of the G5 subsample having good or mediummathema cal competence show a smaller propensity to drop out of the panel, compared tostudents with bad mathema cal competencies. The same holds for G5 students who have ini-ally been sampled in OS (school type independent orienta on stages). This is because these
students had to leave OS a er Grade 6, and thus, are individually surveyed. Finally, studentsfrom theG5 subsample living inWesternGermany have a higher a ri on propensity than thoseliving in Eastern Germany (incl. Berlin). Characteris cs like gender, age group or the migra onbackground do not affect the a ri on propensity in G5.
NEPS Survey Paper No. 34, 2018 Page 30
Zinn, Würbach, Steinhauer, & Hammon
Table 12: Selec vity Analysis for the SC3 Panel Sample along Waves 1-7 (G5), and Waves 3-7(G7_AUG), respec vely.
Subsample G5 Subsample G7_AUGVariable Reference Hazard Ra o p-Value Hazard Ra o p-Value
Gender femalemale 0.879 0.231 1.238 0.275
Year of Birth 1994-19992000-2003 1.018 0.870 1.116 0.572
Migra on background noyes 1.259 0.073 0.986 0.952no informa on 1.180 0.389 1.140 0.650
Region Eastern GermanyWestern Germany 1.951 0.010 2.914 0.010
Mathem. Competence badin Grade 5good 1.317 0.113 - -medium 1.242 0.141 - -no informa on 2.058 0.001 - -
Mathem. Competence badin Grade 7good 0.651 0.028 0.658 0.157medium 0.698 0.041 0.618 0.048no informa on 1.831 <0.001 0.894 0.830
CASMIN (P) 1a, 1b, 2b1c, 2a 1.344 0.181 0.647 0.2422c 1.001 0.998 0.586 0.2283a, 3b 1.042 0.873 0.138 0.004no informa on 1.103 0.658 0.477 0.037
School type GYFS - - - -GS 0.475 0.069 - -HS 0.816 0.280 1.192 0.622IG / FW 0.798 0.387 1.781 0.155MB 1.093 0.774 2.006 0.061OS 0.535 0.028 - -RS 1.170 0.273 1.351 0.306
N 5525 2205Notes: Dependent variable is a ri on (yes or no). (P) parent informa on. Abbrevia ons for school typesare given in Table 11.
NEPS Survey Paper No. 34, 2018 Page 31
Zinn, Würbach, Steinhauer, & Hammon
We find that students of the G7_AUG subsample living in Western Germany have a higherpropensity to drop out of the panel than students from Eastern Germany (incl. Berlin). Com-pared to G7_AUG students with bad mathema cal competencies, students with a mediummathema cal competence have a lower a ri on propensity. Students with parents havinga high educa onal background (measured by CASMIN), or no informa on on the educa onalbackground have a higher probability for remaining in the panel sample, compared to studentswhose parents have a lower educa onal background.
3.4 Star ng Cohort 4
The SC4 panel sample covers nine waves, mostly in an interval of one year, ranging from 2010to 2016. During this me, 16425 students have been surveyed and tested from Grade 9 on-wards. Students get to choose their track of educa on a er Grade 10. Here students caneither stay in school, enter the academic track (ACA) and do their A-levels (Abitur) or they canleave secondary school. In the la er case, students start a voca onal training or enter the Ger-man transi on system. Both groups, voca onal training and transi on system, are summarizedin the voca onal track (VOC). The relevant design variable used for stra fica on is the schooltypewhich a student had been ini ally been sampled. Here, all secondary school types listed inTable 11 except elementary schools (GS) and orienta on stage schools (OS) apply. Comparedto the SC3, in the SC4 more students changed schools over the course of the panel and likelyalso the school type. Unfortunately, there is no consistent and complete informa on on theirschool type history available, which is why we s ck to the sampling informa on. Besides theindividual and design characteris cs men oned above, we consider the mathema cal compe-tence of a student in Grade 9 (low, medium, high, and no informa on) as explanatory modelvariable. Because students change their educa onal track a er Grade 9 we incorporated theeduca onal track as a me-varying covariate into the model. Table 13 shows the results of therespec ve analysis.
The dominant effect of having no informa on on several variables on the a ri on propensityis obvious, although only relevant for migra on background and parental CASMIN. Comparedto students in the academic track, students in the voca onal track have a higher probability todrop out of the panel sample . This is mostly due to the fact that students in VOC are surveyedand tested individually, so that the peer pressure of tes ng groups in schools is not presentanymore, making it easier to refuse. Apart from this, the VOC group of students is more mo-bile and thus harder to track. We find that the school type has a strong effect on panel a ri on.Compared to students who have been sampled in schools leading to upper secondary educa-on (GY), students in other school types are more likely to drop out. Commonly, students in GY
stay longer in school as students in other school types (who offer schooling mostly un l Grade10). Accordingly, students who have been sampled in schools dominantly passing their stu-dents over the voca onal track (i.e., schools for basic secondary educa on HS, comprehensiveschools IG, Rudolf Steiner schools FW, schools with several courses of educa on MB, interme-diate secondary schools RS) have a lower propensity to remain part of the panel, compared tostudents in schools of upper secondary educa on (GY).
Students in special need schools (FS) are, compared to students in schools of upper secondaryeduca on (GY), less likely to leave the panel sample. Thismight be due to the fact that these stu-dents do not switch or leave their schools. Moreover, male students have a higher propensity todrop out of the panel as compared to female students. Students belonging to the younger part
NEPS Survey Paper No. 34, 2018 Page 32
Zinn, Würbach, Steinhauer, & Hammon
of the cohort have a lower probability to drop out. Concerning the mathema cal competence,students with medium or high mathema cal competencies are more likely to remain part ofthe panel sample as compared to students with a lower achievement in themathema cal com-petence tests. Finally, the parents’ educa onal background (measured by CASMIN) influencespanel a ri on. Here, studentswhose parents have at least a secondary school qualifica on anda completed voca onal training (or higher degrees of educa on) are more likely to remain inthe panel as compared to students whose parents do not have at least a completed voca onaltraining.
Table 13: Selec vity Analysis for the SC4 Panel Sample along Waves 1-9.
Variable Reference Hazard Ra o p-Value
Gender femalemale 1.131 <0.001
Year of Birth 1991-19951996-1999 0.885 <0.001
Migra on background noyes 1.007 0.850no informa on 1.496 <0.001
Region Eastern GermanyWestern Germany 0.747 <0.001
Mathem. Competence badin Grade 9good 0.644 <0.001medium 0.865 <0.001no informa on 0.981 0.813
CASMIN (P) 1a, 1b, 2b1c, 2a 0.813 0.0042c 0.558 <0.0013a, 3b 0.483 <0.001no informa on 1.502 <0.001
Educa onal Track AcademicVoca onal 7.744 <0.001
School type GYFS 0.522 <0.001HS 1.389 <0.001IG / FW 1.301 0.001MB 1.383 <0.001RS 1.387 <0.001
N 16425Note: Dependent variable is a ri on (yes or no). (P) parent informa on. Abbrevia ons forschool types are given in Table 11.
NEPS Survey Paper No. 34, 2018 Page 33
Zinn, Würbach, Steinhauer, & Hammon
3.5 Star ng Cohort 5
The panel sample of SC5 consists of nine waves with one survey every six months ranging from2010 to 2015. The first wave sample comprises 17910 students. Relevant design variables arethe type of university at which a student started her/his studies (i.e., public or private univer-sity, and university or university of applied sciences), whether a student studied with the aimbecoming a teacher14 (i.e., yes vs. no), and whether a student has graduated with a degree al-lowing for tradi onal university admission15 (i.e., tradi onal university admission in Germany,tradi onal university admission abroad, and nontradi onal university admission). The field ofstudy is a further stra fica on criterion. However, over the course of the panel many studentschanged their study field (in parts or completely). There is strong evidence that many stu-dents who dropped out have changed their study field. Consequently, no current informa onon their study field is available. Including outdated informa on into our analysis would give awrong picture. Thus, we decided to omit it. Clearly, students have also changed universi es.However, here we could not find evidence for high incidence. Hence, we included this criterioninto our analysis. In addi on to the individual characteris cs described above, we consider themathema cal competence of a student in the winter semester 2010/11 (low, medium, high incomparison to peers) as explanatory model variable. All of the considered covariates are meinvariant.
Table 14 shows the results of the respec ve analysis. Wefind significant effects of the birth year,the region, the competence score, and the university type. Younger cohorts (i.e., students bornlater than 1989) are less likely leaving the panel sample than persons born before 1989. Alike,people studying/having studied in EasternGermany (incl. Berlin) remainmore surely part of thepanel sample than those inWestern Germany. The same applies to students performing well inthe mathema cal competence test and to students studying at universi es (in comparison tostudents studying at universi es of applied sciences). The la er may be explained by studentscon nuing their studies by a doctorate programme at university. Such programmes do usuallynot exist at universi es of applied sciences. Thus, here the chance is higher that students moveand are not any longer accessible. Apart from this we see that students with no informa on ontheir university admission are surely dropping out. Moreover, we find strong me effects at allwaves, mirroring the significant loss of panel members at all of the nine waves. The strongesteffect arises at Waves 8, where for the first me all persons who did not par cipate in NEPSfor a period longer than 2 years were not administered since they had been converted intofinal dropouts a er Wave 7. Furthermore, we find evidence that final dropouts occur moreo en in CATIs than in CAWIs. Overall, the general tendency of more and more students leavingthe panel becomes apparent. The obvious reason for this that in Wave 8 most students havefinished their studies and move. Thus, they are hard to access, may lose their interest in thestudy, and stop par cipa ng.
14This group has been oversampled.15When establishing the sample, all universi es were asked providing informa on on the admission of their stu-
dents. Those with nontradi onal admission were fully surveyed. Thus, university admission is a design crite-rion of the SC5 sample.
NEPS Survey Paper No. 34, 2018 Page 34
Zinn, Würbach, Steinhauer, & Hammon
Table 14: Selec vity Analysis for the SC5 Panel Sample along Waves 1-9.
Variable Reference Hazard Ra o p-Value
Gender femalemale 1.000 0.998
Year of Birth <19891989/1990 0.898 0.008>1990 0.881 0.010
Migra on background noyes 1.024 0.599
Region Eastern GermanyWestern Germany 1.162 <0.001
Mathem. Competence in 2010/11 badmedium 0.868 0.144good 0.627 <0.001no informa on 1.375 <0.001
CASMIN Mother 1a, 1b, 2b1c, 2a 1.025 0.7142c 0.945 0.4603a, 3b 0.937 0.543no informa on 0.970 0.692
CASMIN Father 1a, 1b, 2b1c, 2a 0.969 0.7072c 0.949 0.5743a, 3b 1.028 0.786no informa on 0.933 0.424
Studying for Teacher Degree noyes 0.907 0.024
Public Ins tu on noyes 0.971 0.789
Ins tu on Univ. of Applied ScienceUniversity 0.891 0.006
University admission non-tradi onaltradi onal in Germany 0.890 0.642tradi onal abroad 0.827 0.711no informa on 27.48 <0.001
N 17910Note: Dependent variable is a ri on (yes or no).
NEPS Survey Paper No. 34, 2018 Page 35
Zinn, Würbach, Steinhauer, & Hammon
3.6 Star ng Cohort 6
The SC6 panel sample covers in total seven waves with surveys in an interval of approximatelyone year, ranging from 2009 to 2016. The first wave sample comprises 11649 par cipants, ofthese 11932 persons gave their panel consent and thus form the panel cohort at Wave 1 (i.e.,ALWA/NEPS1). In Wave 3 the panel sample was augmented by a refreshment sample of 5208par cipants (i.e., NEPS3). To comply with the different star ng mes, the SC6 selec vity analy-sis is conducted separately for ALWA/NEPS1 and NEPS3. Relevant design variables consideredin the analysis as covariates are gender, birth cohort, migra on background, whether someonelives inWestern or Eastern Germany (incl. Berlin), the size of the residen al community, maritalstatus as well as highest educa onal qualifica on a ained (mapped by the CASMIN classifica-on). Furthermore, the household size, the employment status and the presence of children
in the household are taken into account.
TheALWA/NEPS1model addi onally considers the susamplemembership (i.e., ALWAorNEPS1).All covariates included were regarded as me invariant, because changes–if at all–are onlymodest (especially concerning the presence of children in the household).
Table 15 shows the results of the respec ve analyses separatedby the two samplesALWA/NEPS1and NEPS3. In the ALWA/NEPS1 subsample, the individuals from the oldest birth cohort leavethe panel with a higher probability than those of the younger cohorts. Respondents who livein Western Germany are more likely to drop out from the panel than those from Eastern Ger-many (incl. Berlin). Likewise, leaving the panel is more likely for single and married persons asfor widowed or divorced ones. Respondents who live in communi es with more than 500,000inhabitants possess a lower dropout rate than individuals who live at loca ons with less than50,000 inhabitants. With increasing educa onal level, the likelihood of leaving the panel studydecreases. Furthermore, children in the household lead to higher panel affinity and three ormore household members result in a higher dropout probability.
For the NEPS3 sample, we observe–just like for ALWA/NEPS1–a higher probability of leavingthe panel for people of the oldest birth cohort and for respondents living in large households.However, there are also some differences in the effects as compared to ALWA/NEPS1. Theeduca onal level and whether someone lives in Western or Eastern Germany does not haveany significant effect on the a ri on propensity in NEPS3. However, we find that individualswith migra on background are more likely to drop out from the NEPS3 panel.
4 Summary and Recommenda ons for Sta s cal Analyses
Our selec vity analyses have shown that–over the course of the panel–specific groups of indi-viduals have a higher tendency to drop out from the panel sample than others. All in all, highlymobile target persons (such as students leaving their parental home for university or voca onaltraining), people with migra on background, and persons with (or parents with) elementaryor lower secondary educa on have higher dropout propensi es than their counterparts. Like-wise, people living in theWestern part of Germany show a higher probability to leave the panelas compared to those living in the Eastern part inclusively Berlin. Furthermore, persons withlow mathema cal competence scores and those with missing values have a lower tendency toremain part of NEPS. Further findings of our analyses are ambivalent and differ between thestar ng cohorts.
NEPS Survey Paper No. 34, 2018 Page 36
Zinn, Würbach, Steinhauer, & Hammon
Table 15: Selec vity Analysis for the SC6 Panel Sample along NEPS Waves 1-7 (ALWA/NEPS1),and NEPS Waves 3-7 (NEPS3), respec vely.
ALWA/NEPS1 NEPS3Variable Reference Hazard Ra o p-Value Hazard Ra o p-Value
Gender femalemale 0.965 0.328 0.913 0.118
Birth Cohort 1944-19551956-1969 0.737 <0.001 0.891 0.1421970-1979 0.721 <0.001 0.757 0.0041980-1986 0.735 <0.001 0.656 <0.001
Migra on background noyes 1.050 0.295 1.168 0.026
Region Eastern GermanyWestern Germany 1.152 0.004 0.982 0.798
BIK <50,000 inhabitants50,000 up to 100,000 0.976 0.706 0.989 0.912100,000 up to 500,000 0.953 0.313 1.032 0.669> 500,000 inhabitants 0.898 0.027 0.888 0.133
Family Status divorced / widowedsingle 1.243 0.006 1.180 0.152married 1.188 0.015 1.011 0.919
CASMIN 1a, 1b, 2b1c, 2a 0.957 0.494 1.033 0.7682c 0.769 <0.001 0.858 0.2283a, 3b 0.664 <0.001 0.814 0.084
Subsample ALWANEPS W1 1.088 0.101 - -
Children in Household noyes 0.812 0.004 0.861 0.190
Employment Status not employedemployed 0.964 0.432 0.925 0.267
Household size 1 person2 persons 1.087 0.186 1.351 0.0043 persons and more 1.479 <0.001 1.776 <0.001
N 11932 5208Notes: Dependent variable is a ri on (yes or no).
NEPS Survey Paper No. 34, 2018 Page 37
Zinn, Würbach, Steinhauer, & Hammon
We see that the composi on of the NEPS cohort samples changes over me. Neglec ng thisfeature in sta s cal analysis likely yields biased results. As a guideline, we recommend apply-ing non-response adjusted designweights when conduc ng descrip ve sta s cs. Suchweightsare provided in the Weights file of the NEPS SUF. However, all of the weights provided referto the group of people who par cipated in a wave, not to a subgroup which may be of specialinterest to answer a par cular research ques on. For coping with a special subsample of acohort, further non-response weigh ng might be necessary. For this purpose, a non-responsemodel has to be specified, fi ed and adjustment factors have to be derived. For the NEPS, theaccordant processing in described in very detail in Steinhauer et al. (2015) as well as in Stein-hauer (2014). Concerning regression, we advise to include the stratum informa on–to accountfor the unequal selec on probabili es in the dis nct strata–into the focal model. Furthermore,all variables that have been found to have a significant effect on the a ri on probability of theconsidered sample should be included as explanatory variables. Missing values may be im-puted using mul variate equa on by chained equa on (van Buuren & Groothuis-Oudshoorn,2011) or modelled using the full informa on maximum likelihood approach (Enders, 2010).Both approaches work fine under missing at random (MAR) mechanisms. However, the situa-on complicates if a missing not at random (MNAR) process must be assumed and the missing
probability depends on the missing values themselves. Then, sensi vity analyses have to beperformed opposing different MNAR models such as selec on and pa ern mixture models.For the NEPS data, an accordant study with recommenda ons for the data users has been con-ducted by Zinn and Gnambs (2018).
Besides the recommenda ons listed here, users of the NEPS data are invited to use the NEPSfo-rum (h ps://forum.neps-data.de/) to ask ques ons answered by either other NEPS data usersor the data providers at the Leibniz Ins tute for Educa onal Trajectories.
NEPS Survey Paper No. 34, 2018 Page 38
Zinn, Würbach, Steinhauer, & Hammon
References
Beck, N. (2008). Time-series-cross-sec on methods. In J. M. Box-Steffensmeier, H. E. Brady,& D. Collier (Eds.), The Oxford Handbook of Poli cal Methodology (p. 475-493). OxfordUniversity Press. doi: 10.1093/oxfordhb/9780199286546.001.0001
Blossfeld, H.-P., Roßbach, H. G., & von Maurice, J. (Eds.). (2011). Educa on as a lifelong pro-cess: The German Na onal Educa onal Panel Study (NEPS) [Special Issue]: Zeitschri fürErziehungswissenscha (Vol. 14). Wiesbaden: VS Verlag für Sozialwissenscha en.
Brauns, H., & Steinmann, S. (1997). Educa onal reform in france, west-germany, theunited kingdom and hungary: Upda ng the casmin educa onal classifica on (ZUMA-Nachrichten No. 23). MZES. Retrieved from http://www.zuma-mannheim.de/publications/periodicals/zuma-nachrichten/zuna-44.htm
Broström, G. (2012). Event history analysis with R. London: CRC Press.Cox, D. (1972). Regressionmodels and life-tables. Journal of the Royal Sta s cal Society. Series
B (Methodological), 34(2), 187-220.Enders, C. (2010). Applied missing data analysis. Guilford.Hammon, A., Zinn, S., Aßmann, C., &Würbach, A. (2016). Samples, weights, and nonresponse:
the adult cohort of the na onal educa onal panel study (wave 2 to 6) (NEPS Survey Pa-per No. 7). Bamberg, Germany: Leibniz Ins tute for Educa onal Trajectories, Na onalEduca onal Panel Study.
Hougaard, P. (2000). Analysis of mul variate survival data. Berlin: Springer.Kalbfleisch, J., & Pren ce, R. (2002). The sta s cal analysis of failure me data. Hoboken, N.J.:
Wiley.Olczyk, M., Will, G., & Kristen, C. (2014). Immigrants in the neps: Iden fying genera on status
and group of origin (NEPSWorking Paper No. 41a). Bamberg, Germany: Leibniz Ins tutefor Educa onal Trajectories, Na onal Educa onal Panel Study.
R Core Team. (2017). R: A language and environment for sta s cal compu ng [Computerso ware manual]. Vienna, Austria. Retrieved from https://www.R-project.org/
Steinhauer, H. W. (2014). Sampling techniques and weigh ng procedures for complex sur-vey designs: The school cohorts of the na onal educa onal panel study (neps) (Doc-toral disserta on, University of Bamberg, Bamberg). Retrieved 07.11.2014, from http://d-nb.info/1061022536/34
Steinhauer, H. W., Aßmann, C., Zinn, S., Goßmann, S., & Rässler, S. (2015). Sampling andweigh ng cohort samples in ins tu onal contexts. AStA Wirtscha s- und Sozialsta s s-ches Archiv, 9(2), 131-157. doi: 10.1007/s11943-015-0162-0
Steinhauer, H. W., & Zinn, S. (2016a). Neps technical report for weigh ng: Weigh ng thesample of star ng cohort 3 of the na onal educa onal panel study (waves 1 to 3) (NEPSWorking PaperNo. 63). Bamberg, Germany: Leibniz Ins tute for Educa onal Trajectories,Na onal Educa onal Panel Study.
Steinhauer, H. W., & Zinn, S. (2016b). Neps technical report for weigh ng: Weigh ng thesample of star ng cohort 4 of the na onal educa onal panel study (wave 1 to 6) (NEPSSurvey Paper No. 2). Bamberg, Germany: Leibniz Ins tute for Educa onal Trajectories,Na onal Educa onal Panel Study.
Steinhauer, H. W., Zinn, S., Gaasch, C., & Goßmann, S. (2016). Neps technical report for weight-ing: Weigh ng the sample of kindergarten children and grade 1 students of the na onaleduca onal panel study (wave 1 to 3) (NEPSWorking Paper No. 66). Bamberg, Germany:Leibniz Ins tute for Educa onal Trajectories, Na onal Educa onal Panel Study.
NEPS Survey Paper No. 34, 2018 Page 39
Zinn, Würbach, Steinhauer, & Hammon
van Buuren, S., & Groothuis-Oudshoorn, K. (2011). mice: Mul variate imputa on by chainedequa ons in R. Journal of Sta s cal So ware, 45(3), 1–67. Retrieved from http://www.jstatsoft.org/v45/i03/
Würbach, A., Zinn, S., & Aßmann, C. (2016). Samples, weights, and nonresponse: the earlychildhood cohort of the na onal educa onal panel study (wave 1 to 3 (NEPS Survey Pa-per No. 8). Bamberg, Germany: Leibniz Ins tute for Educa onal Trajectories, Na onalEduca onal Panel Study.
Zinn, S., & Gnambs, T. (2018). Modeling competence development in the presence of selec onbias. Behavior Research Methods. (forthcoming) doi: 10.3758/s13428-018-1021-z
Zinn, S., Steinhauer, H. W., & Aßmann, C. (2017). Samples, weights, and nonresponse: thestudent sample of the na onal educa onal panel study (wave 1 to 8) (NEPS Survey Pa-per No. 18). Bamberg, Germany: Leibniz Ins tute for Educa onal Trajectories, Na onalEduca onal Panel Study.
Acknowledgements This paper uses data from the Na onal Educa onal Panel Study (NEPS):
• Star ng Cohort Newborns (DOI:10.5157/NEPS:SC1:4.0.0),
• Star ng Cohort Kindergarten (DOI:10.5157/NEPS:SC2:6.0.0),
• Star ng Cohort Grade 5 (DOI:10.5157/NEPS:SC3:7.0.0),
• Star ng Cohort Grade 9 (DOI:10.5157/NEPS:SC4:9.1.0),
• Star ng Cohort First-Year Students (DOI:10.5157/NEPS:SC5:9.0.0), and
• Star ng Cohort Adults (DOI:10.5157/NEPS:SC6:8.0.0).
From 2008 to 2013, NEPS data was collected as part of the Framework Program for the Pro-mo on of Empirical Educa onal Research funded by the German Federal Ministry of Educa onand Research (BMBF). As of 2014, NEPS is carried out by the Leibniz Ins tute for Educa onalTrajectories (LIfBi) at the University of Bamberg in coopera on with a na onwide network.
NEPS Survey Paper No. 34, 2018 Page 40