Data Collection and Cross-Sequential Twin Design

To date, there is neither a central population register nor a health or disease register in Germany. To establish a register-based longitudinal panel, the TwinLife project had to implement an innovative sampling strategy and obtain addresses directly from the many residents’ registration offices located at community level in Germany. Here, twin families were identified through a multistage process. In a first step, a sample of about 500 communities (with more than 5000 inhabitants) was drawn where residents’ registration offices were asked to identify potential twins in four target birth cohort groups: 2009−2010 (cohort 1), 2003−2004 (cohort 2), 1997−1998 (cohort 3), and 1990−1993 (cohort 4) (Mönkediek et al., Reference Mönkediek, Lang, Weigel, Baum, Eifler, Hahn, Hufer, Klatzka, Kottwitz, Krell, Nikstat, Diewald, Riemann and Spinath2019). The twins were then identified by locating children registered as currently living in the same household, having the same sex, and birth dates. By limiting the study to same-sex twins, the costs of reaching the required total sample size for fraternal twins were reduced, and it was easier to find a sufficient number of twin families with a balanced ratio of identical and fraternal twins. To facilitate the identification of a sufficient number of twin families in the targeted birth cohorts (about 1000; i.e., up to 5% of all twin births in Germany), TwinLife includes twins from subsequent birth years (subsamples A and B) per cohort. For cohort 4, a restriction to only children currently living in the same household was not possible, due to some of the twins potentially no longer living with their parents. Therefore, twins of cohort 4 were identified by utilizing registries of residents that lived under the same address dating several years back (Lang & Kottwitz, Reference Lang and Kottwitz2020). After the residents’ registration offices provided a gross sample of around 19,000 addresses, in a final step, 13,359 addresses (families) were sampled and contacted until about 1000 twin families per birth cohorts agreed to take part in the first data collection in 2014−15 (Brix et al., Reference Brix, Pupeter, Rysina, Steinacker, Schneekloth, Baier, Gottschling, Hahn, Hufer, Kaempfert, Kornadt, Krell, Lang, Lenau, Mattheus, Nikstat, Peters, Schulz, Schunck, Starr, Witt, Diewald, Riemann and Spinath2017; Lang & Kottwitz, Reference Lang and Kottwitz2020). The original sample consisted of 4096 twin families (36%), who were subsequently observed and surveyed annually.

The twin families were invited by mail to participate in the study before they were visited by an interviewer for a face-to-face interview (for a detailed description, see Brix et al., Reference Brix, Pupeter, Rysina, Steinacker, Schneekloth, Baier, Gottschling, Hahn, Hufer, Kaempfert, Kornadt, Krell, Lang, Lenau, Mattheus, Nikstat, Peters, Schulz, Schunck, Starr, Witt, Diewald, Riemann and Spinath2017, and Lang & Kottwitz, Reference Lang and Kottwitz2020). In addition to the twins, data were collected from the twins’ parents (biological and step-parents), a nontwin sibling (the one closest in age to the twins and at least 5 years old, if available), and partners of adult twins (if available). In the first wave, siblings and partners were only interviewed if they lived in the same household as one of the twins. From wave 2 on, participants who left the household (e.g., due to separation of parents, or due to siblings or twins moving out of the parental home) and other target people who moved into the household (e.g., partners of twins) were tracked, contacted and surveyed as part of their new households. While respondents were offered multiple opportunities to self-update their addresses in the address database (study’s hotline, email address, online address portal), address tracking was conducted via the ADDRESSFACTORY database of Deutsche Post AG. In addition, during the field phase, residents’ registration offices were contacted for individual inquiries about changed addresses (Lesaar et al., Reference Lesaar, Prussog-Wagner and Hess2020). These measures were conducted each month throughout the entire fieldwork phase and after. Starting with the fourth wave, twins were asked to provide information on their children.

The collection mode alternated between face-to-face interviews (F2F) at the respondents’ homes and computer-assisted telephone interviews (CATI) across survey waves (see Figure 1). To limit the time interviewers spent in the household during the F2F interviews, computer-assisted self-interviews, and paper and pencil interviews were implemented in addition to the computer-assisted personal interviews (e.g., Brix et al. Reference Brix, Pupeter, Rysina, Steinacker, Schneekloth, Baier, Gottschling, Hahn, Hufer, Kaempfert, Kornadt, Krell, Lang, Lenau, Mattheus, Nikstat, Peters, Schulz, Schunck, Starr, Witt, Diewald, Riemann and Spinath2017). Alternatively, computer-assisted web-based interviews (CAWI) were also offered to respondents in the later waves. For example, the interviews of the fourth F2F wave, which were originally planned as household interviews, had to be substituted with CATI and CAWI interviews due to contact restrictions during the COVID-19 pandemic. In order to assess the impact of the pandemic on respondents, between 2020 to 2024, additional information was collected in four COVID-19 supplementary surveys (Cov 1–Cov 4 in Figure 1).

Starting with the fourth wave of the F2F survey, the cohorts overlap in terms of age, making it possible to combine the data. For example, in the fourth F2F wave, the youngest twins (cohort 1), aged 11 years old, were of the same age as the second cohort when they were first interviewed (see Figure 2). As of 2025, five F2F interviews and four CATI interviews have been conducted. While the originally planned observation period covers 10 years (between 2014 and 2024), the panel was extended in spring 2025 to include a further CAWI follow-up survey. Together, the established cohort-sequential design covers an age range of up to 28 years, from age 5 to age 33 (see Figure 2). This cross-sequential twin design allows to separate the effects of age, time, and cohort on genetic, environmental, and phenotypic differences in several characteristics that were measured repeatedly over time.

Figure 2.

Age overlaps between the TwinLife cohorts.

Note: Age refers to mean age during each survey. F2F, face-to-face interviews; CATI, computer-assisted telephone interviews; Cov, Corona-based inventories; CAWI, computer-assisted web-based interviews.

Source: TwinLife homepage (own depiction).

All participants were informed in writing prior to the surveys about the aims of the study, incentives, and that their participation was voluntary and could be withdrawn at any time. In all F2F surveys, participants received a two-page leaflet summarizing the procedures concerning data processing, protection and privacy, and informing them about their rights and how to exercise them in accordance with applicable data protection laws. At the beginning of each F2F interview, informed verbal consent was obtained from participants, or the participants’ legal guardians when the participants were underage, and documented. When the study started, written informed consent was not required, in line with German national legislation and the institutional requirements. Nevertheless, for all supplementary online surveys, written informed consent was obtained from all participants.

Panel Stability and Sample Weights

In total, 16,951 individuals from 4096 twin families participated in the first wave of TwinLife. The complete sample comprises data on 19,152 individuals, including the twins, their parents, partners, and a sibling of the twins (if available). Figure 3A shows the development of the sample sizes in the F2F and CATI surveys by cohorts. For simplicity, all sample sizes and participation rates are calculated based on at least one family member participating in the respective survey. As shown in Figure 3A, there was a substantive drop in the number of participating families from the first home interview (F2F 1) to later participation in the panel (CATI 1). Participation rates of the families were around 70% in all cohorts in CATI 1 and in the younger two cohorts in F2F 2. For the older two cohorts, the participation rates fell to around 60% in F2F 2 which corresponds to the third survey wave. This drop is not unusual in the context of panel studies with register-based sampling designs (e.g., in comparison to opt-in studies, where individuals proactively choose to participate), as it takes time for participants to feel committed to a study. Furthermore, Figure 3A shows that the number of participating families stabilized across panel waves, particularly in the younger two cohorts from the second wave of home interviews onward. The decline in the number of participating families across waves can be particularly observed in the oldest twin cohort due to the twins’ age and living situations (e.g., twins leaving the parental home), which made it harder to reach and interview them. The participation rates of the twins and their families are comparable to other large German panel studies with register-based sampling of families, such as the German Socio-Economic Panel (SOEP, Kroh et al., Reference Kroh, Kühne, Siegers and Belcheva2018) or the adult cohort of the National Educational Panel Study (NEPS; Zinn et al., Reference Zinn, Würbach, Steinhauer and Hammon2020).

Figure 3.

Numbers of (A) families by twin cohort and (B) twin family members in the home and telephone interviews of TwinLife.

Note: Reported sample sizes and participation rates are defined by at least one family member participating in the respective survey. MZ, monozygotic; DZ, dizygotic; N = 6 cases of twins with unknown zygosity. F2F, face-to-face interviews; CATI, computer-assisted telephone interviews; CAWI, computer-assisted web-based interviews.

Figure 3B shows the development of sample sizes at the level of family members. As the figure illustrates, fewer family members were surveyed in CATI waves one to three. This was due to the following reasons: Due to their young age (<10 years old), the twins of the youngest cohort could not be interviewed in CATI 1 and CATI 2. Here, only parental reports about the children were available. In addition, in CATI 2 the twins in cohort 2 were not interviewed as the scope of the survey had to be reduced in this wave to reduce costs (see also Figure 4A and 4B). For the same reason, in CATI 2 and CATI 3 only one adult person was interviewed to collect important household information as well as information on the children.

Figure 4.

Number of (A) monozygotic and dizygotic twins as well as (B) male and female twins who participated in the home and telephone interviews of TwinLife by cohort.

Note: Twins in Cohort 1 were not included in the telephone interviews in CATI 1 and CATI 2 due to their young age. Twins in Cohort 2 were not interviewed in CATI 2. For cost reasons, the scope of the survey had to be reduced in this wave. F2F, face-to-face interviews; CATI, computer-assisted telephone interviews; CAWI: computer-assisted web-based interviews.

In TwinLife, the applied probabilistic sampling method (see section on data collection) has proven effective in counteracting the possible overrepresentation of monozygotic (MZ) twin pairs, which is a common problem in twin studies based on self-recruitment (for a discussion see: Lykken et al., Reference Lykken, McGue and Tellegen1987). Except for the oldest cohort, dizygotic (DZ) twins were slightly overrepresented in the different survey waves (compare Figure 4A): in cohorts 1 and 2, about 58%, and in cohort 3, about 52%. In the fourth cohort, however, depending on the survey wave, only 41% to 45% were DZ twin pairs. Furthermore, as described before, due to their young age, no twins were interviewed in the youngest cohort in CATI 1 and CATI 2, and twins in cohort 2 were not interviewed in CATI 2 to reduce the costs of the survey.

Similarly, the applied probability-based sampling design resulted in a relatively balanced sample with regard to the sex-ratio of the twins (compare Figure 4B). While TwinLife only includes same-sex DZ twins, across survey waves, about 51% to 53% of the twins in the two younger cohorts were female. In the two older cohorts, about 57% to 60% of the twins were female.

For the TwinLife data, several sample weights on the family level were calculated. To address unequal sampling probabilities of the twin families, design weights were calculated for F2F1 wave. To address nonresponse bias due to selective survey participation in F2F1, nonresponse weights were calculated. Finally, to address biased estimates due to selective panel drop out, panel weights were calculated for all subsequent waves (CATI1-F2F5). For more information on the survey weights, see Krell and Ruks (Reference Krell and Ruks2023).

Extended Twin Family Design

Using extended twin family data in TwinLife can provide more comprehensive insights into the sources of variance in various phenotypes. As classical twin models (CTM) rely on data from twins only, they rely on various assumptions (such as the absence of assortative mating) that do not always reflect the actual processes giving rise to individual differences (cf. Coventry & Keller, Reference Coventry and Keller2005). Through the inclusion of data from twins’ biological parents and, if available, also twins’ siblings in nuclear twin family models (NTFMs; Keller et al., Reference Keller, Medland, Duncan, Hatemi, Neale, Maes and Eaves2009), some of these assumptions can be relaxed. As a consequence, NTFMs provide more accurate estimates of heritability and environmental variance components under various conditions (Instinske & Kandler, Reference Instinske and Kandler2024). In particular, involving parental data allows to study mating patterns; for example, whether individuals select their spouses based on certain similarities in some characteristics (Horwitz et al., Reference Horwitz, Balbona, Paulich and Keller2023). Such assortative mating affects the genetic correlation between parents, which might increase the genetic relatedness within their offspring (except for already genetically identical MZ twins) and thereby affect estimates of genetic and environmental variance components in CTMs (e.g., Gonggrijp et al., Reference Gonggrijp, Silventoinen, Dolan, Boomsma, Kaprio and Willemsen2023).

Furthermore, utilizing data from additional relatives of twins, the simplifying assumption of CTMs that either nonadditive genetic effects or environmental influences shared by twins approach zero can be relaxed. Due to the additional information, additional sources that contribute to phenotypic variance can be considered (Keller et al., Reference Keller, Medland, Duncan, Hatemi, Neale, Maes and Eaves2009). For instance, the additional consideration of parents and nontwin siblings of twins allows to distinguish between different types of environmental sources shared by family members, such as those shared across generations, only by siblings of a descendant generation, or only by twins (Bleidorn et al., Reference Bleidorn, Hufer, Kandler, Hopwood and Riemann2018; Hufer et al., Reference Hufer, Kornadt, Kandler and Riemann2020; Kandler et al., Reference Kandler, Instinske and Bell2025).

Moreover, NTFMs are based on information from individuals from two generations. Therefore, they enable examining whether intergenerational transmission of certain characteristics occurs due to genetic or environmental factors (e.g., Wolfram et al., Reference Wolfram, Ruks and Spinath2024). NTFMs allow to consider that parents may provide rearing environments that are influenced by their own genetic predisposition and thus associated with their children’s genetic predisposition (Briley et al., Reference Briley, Livengood and Derringer2018; Plomin et al., Reference Plomin, DeFries and Loehlin1977). This form of passive gene-environment covariance allows to study the extent to which socio-environmental circumstances shared by siblings reared together correspond to genetic differences (Kandler et al., Reference Kandler, Kühn, Mönkediek, Forstner and Bleidorn2024).

Drawing on data from the partners of twins, which is across waves available for 1142 participants of the older TwinLife cohorts, the TwinLife data can be used to study how aspects related to partnership (e.g., partnership quality), life events experienced by partners (e.g., severe illness), and partners’ characteristics (e.g., partner’s personality) can shape developmental mechanisms associated with social inequalities and life outcomes. Even though we do not have information on the exact timeline of the partnership duration, the data allow for a detailed examination of the circumstances under which partnerships are formed or dissolved. Since the twins and their partners are key components of their respective social environments, combining interviews from both parties provides more detailed insights into the specific reasons for the similarities between the partners, and thus also into the sources of individual differences. Specifically, partner similarity can be either due to genetic factors (i.e., genetic homogamy), due to a shared social background (i.e., social homogamy) or due to partner-specific interactions that increase their similarity over time (i.e., convergence; e.g., Luo, Reference Luo2017). As all these different causes underlying partner similarity may have different consequences for the genetic and environmental variance in a certain construct. The use of spouses-of-twin models (e.g., Kandler et al., Reference Kandler, Lewis, Feldhaus and Riemann2015) constitutes one way to derive more detailed conclusions on the sources of individual differences in this regard.

Certainly, the NTFM or spouses-of-twin model outlined here rely on simplified assumptions as well. Some of these assumptions and the requirements of NTFM or spouses-of-twin models are discussed in Keller et al. (Reference Keller, Medland, Duncan, Hatemi, Neale, Maes and Eaves2009) and Keller et al. (Reference Keller, Medland and Duncan2010). Even more complex models can be used to account for these limitations. However, these models require substantial sample sizes that are rarely accomplished (e.g., Keller et al., Reference Keller, Medland and Duncan2010). This emphasizes the potential of combining data from different empirical studies utilizing an extended twin family design.

The Molecular Genetic Extension TwinSNPs

An important extension of the core project and a further strength of TwinLife is the collection of saliva samples, from which molecular genetic data is obtained as part of the accompanying satellite project TwinSNPs. Written informed consent was obtained from participants who provided saliva samples. Here, each respondent received an information booklet about the goals of the extension projects, the processing of the saliva samples, the genotyping profiling, the future use of the data, data protection and their rights, and how to exercise them in accordance with applicable data protection laws. All materials were also made available on the TwinLife website (www.twin-life.de) so that they could be accessed at any time. If a participant was underage, a legal guardian gave written informed consent.

Based on saliva samples collected from TwinLife participants via self-collection kits (Oragene), DNA was extracted and respondents were genotyped (Illumina Global Screening Arrays, GSA+MD-24v3.0-Psych-24v1.1, Illumina, San Diego, CA, USA). Details on the processing of genotype data in TwinSNPs can be found in the cohort profile (Frach et al., Reference Frach, Disselkamp, Schowe, Andreas, Deppe, Maj, Rohm, Ruks, Wiesmann, Kandler, Mönkediek, Spinath, Nöthen, Binder, Czamara and Forstner2026). The TwinSNPs project enables the combination of the twin family design with genome-wide genetic variations based on single nucleotide polymorphisms (SNPs), which makes it possible, for example, to calculate and take into account polygenic scores (PGS). Besides the general incorporation of PGS in analytic approaches (for an overview, see Pingault et al., Reference Pingault, Allegrini, Odigie, Frach, Baldwin, Rijsdijk and Dudbridge2022), TwinLife offers the opportunity to combine genetically informative analyses based on PGS with twin-based analyses. PGS can be conceptualized as summary measures of (additive) genetic effects on a certain trait or disorder (e.g., Dudbridge, Reference Dudbridge2013; Harden & Koellinger, Reference Harden and Koellinger2020). This information can be considered beyond the covariance structure of MZ and DZ twins to estimate the extent to which phenotypic differences are attributable to genetic differences (e.g., Dolan et al., Reference Dolan, Huijskens, Minică, Neale and Boomsma2021). As such, the entire genetic variance can be decomposed into variance due to the common genetic variants incorporated in the PGS and the remaining genetic variance derived from the ratios of MZ and DZ twin similarity (e.g., Bruins et al., Reference Bruins, Dolan and Boomsma2020).

An overview of available PGS in TwinLife can be found in the download section on the TwinLife documentation website (www.twin-life.de/documentation/downloads) and in the cohort profile (Frach et al., Reference Frach, Disselkamp, Schowe, Andreas, Deppe, Maj, Rohm, Ruks, Wiesmann, Kandler, Mönkediek, Spinath, Nöthen, Binder, Czamara and Forstner2026). First publications based on the molecular genetic data from TwinLife addressed topics such as the PGS prediction of education (Starr et al., Reference Starr, Ruks, Giannelis, Willoughby, Pettersson, Pahnke, Maj, Andreas, Ahlskog, Arseneault, Fisher, Forstner, Kandler, McGue, Nöthen, Oskarsson, Spinath, Vrieze, Wertz and von Stumm2026), the associations of subjective well-being with personality traits (Deppe et al., Reference Deppe, Disselkamp, Maj, Aldisi, Nöthen, Diewald, Spinath, Forstner and Kandler2026), and neighborhood deprivation (Harerimana et al., Reference Harerimana, Liu and Ruks2025).

Due to the extended family design, TwinLife also allows testing for direct and indirect genetic effects in a within-sibship (e.g., Howe et al., Reference Howe, Nivard, Morris, Hansen, Rasheed, Cho, Chittoor, Ahlskog, Lind, Palviainen, van der Zee, Cheesman, Mangino, Wang, Li, Klaric, Ratliff, Bielak, Nygaard and Davies2022) or trio-design (e.g., Balbona et al., Reference Balbona, Kim and Keller2022). This structure makes it well suited for applying methodological innovations in family-based genomewide association studies (GWAS) and other genetically informed designs; for example, those proposed by Bates et al. (Reference Bates, Sesia, Sabatti and Candès2020) and Balbona et al. (Reference Balbona, Kim and Keller2021) based on PGS. Utilizing the existing data on the twins and their parents, it is possible to contribute to research on indirect genetic effects, such as genetic nurture, that is, the indirect effects of parental genotypes on offspring outcomes via environmental pathways rather than direct genetic transmission (Kong et al., Reference Kong, Thorleifsson, Frigge, Vilhjalmsson, Young, Thorgeirsson, Benonisdottir, Oddsson, Halldorsson, Masson, Gudbjartsson, Helgason, Bjornsdottir, Thorsteinsdottir and Stefansson2018), on a wide range of outcomes. However, indirect genetic effects can also arise due to assortative mating and population stratification (e.g., Young et al., Reference Young, Benonisdottir, Przeworski and Kong2019). From a methodological perspective, genetic nurture, also known as passive rGE, as well as other sources of indirect genetic effects, can confound direct genetic effect estimates. Thus, incorporating genetic data from related individuals reduces bias in estimates of direct genetic effects (e.g., Young et al., Reference Young, Benonisdottir, Przeworski and Kong2019).

Based on its family design, TwinLife currently offers genotype data (including PGS) for different combinations of parent-offspring trios, where valid PGS information for either the biological mother, the biological father and at least one biological child is available (see Table 1 for further information on available person types and cases per cohort).

Table 1.

Cases of individuals per TwinLife cohort with valid trio PGS data, by person type

Note: Full biological siblings: both parents are the same as twins’ parents.

The Epigenetic Extension TECS

The aim of the TwinLife Epigenetic Change Satellite (TECS) project was to assess epigenetic changes in relation to the COVID-19 pandemic. For this purpose, DNA methylation (DNAm) data was obtained from the first two saliva samples (see Saliva 1a to 2b in Figure 1) for a selected group of twin participants using the Infinium MethylationEPIC BeadChip array (v1.0, Illumina, San Diego, CA, USA). Specifically, complete twin pairs with saliva samples collected at both timepoints, before and during the pandemic (see Figure 1), a valid consent form (signed by children and parents for those under 18 years of age), and participation in both the Cov 2 and Cov 3 surveys (see Figure 1) were prioritized for the assessment (n = 505 complete twin pairs). In addition, remaining slots on the arrays (n = 2128, two measures per participant) were used for individual twins (n = 54) that fulfilled these criteria.

For all twins with valid DNAm data after quality control (see Table 2), several epigenetic aging estimators were calculated, such as the Horvath clock (Horvath, Reference Horvath2013), the PedBE clock (McEwen et al., Reference McEwen, O’Donnell, McGill, Edgar, Jones, MacIsaac, Lin, Ramadori, Morin, Gladish, Garg, Unternaehrer, Pokhvisneva, Karnani, Kee, Klengel, Adler, Barr, Letourneau and Kobor2020), GrimAge2 (Lu et al., Reference Lu, Binder, Zhang, Yan, Reiner, Cox, Corley, Harris, Kuo, Moore, Bandinelli, Stewart, Wang, Hamlat, Epel, Schwartz, Whitsel, Correa, Ferrucci and Horvath2022), and DunedinPACE (Belsky et al., Reference Belsky, Caspi, Corcoran, Sugden, Poulton, Arseneault, Baccarelli, Chamarti, Gao, Hannon, Harrington, Houts, Kothari, Kwon, Mill, Schwartz, Vokonas, Wang, Williams and Moffitt2022). In addition, new epigenetic scores are continuously computed as they become available. To date, these include, for example, the epigenetic score for CRP (C-reactive protein, Barker et al., Reference Barker, Cecil, Walton, Houtepen, O’Connor, Danese, Jaffee, Jensen, Pariante, McArdle, Gaunt, Relton and Roberts2018) and BMI (Do et al., Reference Do, Sun, Meeks, Dugué, Demerath, Guan, Li, Chen, Milne, Adeyemo, Agyemang, Nassir, Manson, Shadyab, Hou, Horvath, Assimes, Bhatti, Jordahl and Conneely2023). First studies utilizing epigenetic clocks examined the genetic and environmental contributions to differences in epigenetic aging and how these vary across age and time (Kuznetsov et al., Reference Kuznetsov, Liu, Schowe, Czamara, Instinske, Pahnke, Nöthen, Spinath, Binder, Diewald, Forstner, Kandler and Mönkediek2025) and linked to personality differences (Instinske et al., Reference Instinske, Schowe, Czamara, Kuznetsov, Mönkediek and Kandler2026). An overview of all currently available epigenetic aging estimators and other derived scores from the DNAm data can be found on the TwinLife documentation website (www.twin-life.de/documentation/downloads). Detailed descriptions of the DNAm data generation, preprocessing, and epigenetic age score computations are provided in Frach et al. (Reference Frach, Disselkamp, Schowe, Andreas, Deppe, Maj, Rohm, Ruks, Wiesmann, Kandler, Mönkediek, Spinath, Nöthen, Binder, Czamara and Forstner2026).

Table 2.

Cases per TwinLife cohort with valid epigenetic data, by DNA-validated zygosity

Further Satellite Projects

There are several other satellite projects that have already begun or have even been completed. These projects are based on collaborations with colleagues from different research institutions that collected more specific data beyond the actual TwinLife survey and that tackle more specific research questions.

Zygosity Determination and Genetic Similarity

Zygosity was assessed at baseline using physical similarity questionnaires (see Hill Goldsmith, Reference Hill Goldsmith1991; Oniszczenko et al., Reference Oniszczenko, Angleitner, Strelau and Angert1993), which were administered during the first household interview. For the two youngest age groups, questionnaires were completed by the twins’ parents, whereas the two oldest age groups provided self-reports. In an early genotyped validation sample, collected during F2F1 including 287 pairs of twins, analyses indicated high validity of the questionnaire-based zygosity assessments and a low misclassification rate of 4% on average (see Table 3 and Lenau & Hahn, Reference Lenau and Hahn2017).

Table 3.

Frequencies of monozygotic and dizygotic twin pairs depending on classification method and misclassification rates

Note: F2F, face-to-face interviews.

Based on molecular genetic data collected as part of the TwinSNPs satellite project, the zygosity of a substantial share of the twin pairs was further validated. Out of the initial 4096 twin pairs, 1323 complete twin pairs were genotyped based on saliva samples collected in F2F3 and subsequent waves (Table 3). Comparing the results from the larger genotyped sample with the zygosity variable based solely on questionnaire data indicated a misclassification rate of 5.6%, further strengthening the validity of the questionnaire-based method of zygosity determination. Misclassifications of twins zygosity were generally higher for DZ twin pairs than for MZ twins. A variable combining all information sources, prioritizing DNA-based zygosity when available, is available in the dataset (zyg0112).

The available molecular genetic data allowed the calculation of the actual genetic similarity between siblings (i.e., their degree of relatedness or kinship). For DZ twins, this enables the use of empirically derived relatedness (or kinship) coefficients in twin models, rather than relying on the standard assumption of a genetic correlation in DZ pairs of r _G = .50. While DZ twins share on average 50% of their genetic makeup, their actual genetic similarity may vary, and can be precisely determined with the use of the kinship coefficients. Their use can help to alleviate the potential over- or underestimation of genetic variance components (see Therneau, Reference Therneau2024, for an implementation in R, and Yang et al., Reference Yang, Lee, Goddard and Visscher2011, for an application) as, for example, in the case of assortative mating of the twins’ parents, which would increase the average genetic correlation of DZ twins in the presence of positive assortative mating (Instinske & Kandler, Reference Instinske and Kandler2024). Figure 5 shows the kinship distribution in TwinLife‘s DZ twin pairs with available molecular genetic data (N _pairs = 665). The average genetic relatedness between DZ twins of a pair in TwinLife was 0.4997 with a standard deviation of 0.039. Thus, at least for the TwinLife data, the use of r _G = .50, which reflects the genetic similarity of fraternal twins, can be considered valid.

Figure 5.

Distribution of the genetic relatedness score.

Sample Representativeness

The TwinLife data is based on a national probability sample drawn from population registers. To assess the representativeness of the TwinLife sample with regard to the social stratification in Germany, Lang and Kottwitz (Reference Lang and Kottwitz2020) compared the distributions of several indicators in TwinLife with those of a proxy-twin sample based on the German Microcensus. The German Microcensus is a representative household register sample of 1% of the German households, carried out annually since 1957 (Destatis, 2026). Proxy twin pairs were identified as two children with the same year of birth living in the same household with at least one of their parents (Lang & Kottwitz, Reference Lang and Kottwitz2020). The analysis showed that the share of households with tertiary-educated parents was approximately 15% higher in TwinLife compared to the German Microcensus. Additionally, TwinLife households had a slightly higher median household income and about 10% fewer households without German citizenship. These results indicated that participation in TwinLife was, to some degree, selective, especially with respect to parental education. However, the analysis confirmed that the TwinLife sample covers the full range of key socioeconomic indicators, including both the lower and upper bounds. Capturing also the bounds of the resources’ distribution is essential for stratification research.

Saliva for genotyping the respondents was collected from the third household wave (F2F 3) onward including all participants who agreed to provide a sample (about 70%). Since additional selection criteria apply for the TECS sample (see above), several indicators across the genotyped subsamples were compared with the full TwinLife sample at F2F1 to evaluate their representativeness (see Tables 4 to 6). The three twin subsamples — Valid Saliva Sample, TwinSNPs, and TECS (with valid DNAm data) — showed similar distributions to the full twin sample at F2F1 (Table 4) in terms of zygosity, sex, and community size. All included a higher proportion of participants without a migration background and a greater share of younger birth cohorts (Cohorts 1 and 2). Parental education and household income were slightly higher across subsamples. The two parent subsamples — Valid Saliva Sample and TwinSNPs — showed similar distributions to the full parent sample at F2F1 in terms of sex and community size (Table 5). Both had a lower mean age than the full parent sample and a greater share of younger birth cohorts. Parents’ own education and household income were slightly higher. The two sibling subsamples — Valid Saliva Sample and TwinSNPs — showed similar distributions to the full sibling sample at F2F1 in terms of sex and community size (Table 6). Both had a lower mean age and a greater share of younger birth cohorts. Parental education and household income were slightly higher. For more details on the genotyping and differences in the characteristics of the participants compared to the initial core sample see Frach et al. (Reference Frach, Disselkamp, Schowe, Andreas, Deppe, Maj, Rohm, Ruks, Wiesmann, Kandler, Mönkediek, Spinath, Nöthen, Binder, Czamara and Forstner2026).

Table 4.

Comparison of sample characteristics between genotyped twin subsamples and all twins in the full TwinLife sample

Note: For continuous variables: Mean (standard deviation); For categorical variables: Absolute frequencies (relative frequencies in %); ISCED, International Standard Classification of Education; OECD HH, net equivalent household income based on the modified OECD scale; F2F, face-to-face interviews.

¹ Following the definition of the German federal statistical office, a person has a migration background if they or at least one parent were not born with German citizenship. Thus, migration background is only calculated for persons with valid information on citizenship for both themselves and their parents.

Table 5.

Comparison of sample characteristics between genotyped parent subsamples and all parents in the full TwinLife sample

Note: For continuous variables: Mean (standard deviation); For categorical variables: Absolute frequencies (relative frequencies in %); ISCED, International Standard Classification of Education; OECD HH, net equivalent household income based on the modified OECD scale; F2F, face-to-face interviews.

Table 6.

Comparison of sample characteristics between genotyped sibling subsamples and all siblings in the full TwinLife sample

Note: For continuous variables: Mean (Standard Deviation); For categorical variables: Absolute frequencies (relative frequencies in %); ISCED, International Standard Classification of Education; OECD HH, net equivalent household income based on the modified OECD scale; F2F, face-to-face interviews.

¹ Following the definition of the German federal statistical office, a person has a migration background if they or at least one parent was not born with German citizenship. Thus, migration background is only calculated for persons with valid information on citizenship for both themselves and their parents.