TEDS Data Dictionary

Representativeness of TEDS Study Participants

Contents of this page:

Introduction

The purpose of this page is to illustrate changes in the composition of the participant samples of TEDS studies over time. This is done with reference to the 1st Contact study sample, which had the highest returns of any TEDS study (as is therefore thought to be the most representative of the general population), and which provided some useful demographic variables.

It is relatively easy to compare the representativeness of any later TEDS study sample with that of the 1st Contact study sample. Generally speaking, the participants in later studies were subsets of the participants in the 1st Contact study: only very small numbers of families participated in later studies without having participated in the 1st Contact study.

However, no attempt is made here to compare the representativeness of the TEDS sample(s) with the wider population of England and Wales, from which the original TEDS sample was taken.

Measures used for Comparisons

The following measures all originate from the 1st Contact booklet study, although twin sexes have subsequently been maintained and updated in the TEDS admin database. Further details of the variables can be found in the 1st Contact pages.

VariableDescriptionValues Representativeness comparison
aethnicEthnic origin of twins 1=white, 0=other% white
sex1/2Twin sex 0=female, 1=male% female (% of individuals, not pairs)
asesComposite variable measuring family SES, derived from 5 separate measures. See ases derivation for full details. standardised with mean=0 and standard deviation=1mean value
amojobDid the mother of the twins have a job, at the time when the 1st Contact booklet was completed. See amojob derivation for full details. 0=no (not employed), 1=yes (employed), 2=full-time parent% employed

Comparison table

The comparison figures shown in the table below were obtained using all participants having data in the respective datasets and having 1st Contact data. No exclusions were made for the purposes of these comparisons. Smaller-scale studies with selected samples (e.g. the 4 year in home study, and the 16 year Leap-2 booklets) are not included in the table.

The N column shows the number of individual twins for whom we have data in each study. The N for each representativeness comparison is generally slighly lower, due to missing data, but for brevity detailed Ns are not shown here.

Note that, due to the large Ns in all the major TEDS studies, differences in the comparison figures are generally significant even where the magnitude of the differences are small. For brevity, analysis of significance using p-values is omitted.

Parent and twin booklets were returned as a complete set by most families, even at age 16. Entries shown as "parent/twin booklets" include all twins for whom the parent and/or the twin booklet were returned.

StudyDescription of sample (data in dataset)N (twins) % white% femalemean ases % mothers employed
1st ContactParent booklet27444 91.7%50.1%0.00043.1%
2 YearParent/twin booklets12572 93.3%50.3%0.02741.3%
3 YearParent/twin booklets12118 93.2%50.7%0.06343.4%
4 YearParent/twin booklets16302 93.3%51.1%0.13144.3%
7 YearParent booklet15668 93.7%51.3%0.12846.4%
Twin phone tests10887 93.6%51.3%0.07745.6%
Teacher questionnaires12649 94.1%51.3%0.14546.1%
8 YearParent questionnaire13524 93.7%50.8%0.19147.0%
9 YearParent/twin booklets 6879 93.6%52.7%0.12846.4%
Teacher booklet5848 94.3%52.3%0.11147.0%
10 YearTwin web tests6171 93.7%54.1%0.16646.8%
Teacher booklet6160 94.1%52.4%0.12946.6%
12 YearParent/twin booklets 11829 93.5%52.7%0.19347.0%
Twin web/phone tests12179 93.7%53.3%0.18247.7%
Teacher booklet9881 93.6%52.9%0.18647.9%
14 YearParent/twin booklets7240 93.9%54.3%0.24047.0%
Parent SLQ7114 94.3%53.3%0.30948.7%
Twin web tests6642 93.5%58.3%0.24746.7%
16 YearParent/twin booklets 10320 93.5%55.3%0.22947.2%
Twin GCSE results13506 93.2%52.6%0.22548.1%
Twin web tests5762 92.2%57.6%0.17446.0%
18 YearTwin A-level results14321 93.0%52.9%0.20047.4%
21 YearTEDS21 phase 1 parent/twin qnrs 12661 93.1%56.9%0.23447.3%
TEDS21 phase 2 twin questionnaire8718 93.5%63.4%0.25147.0%
G-game web tests4644 94.4%68.0%0.30046.3%
Covid phase 1 twin questionnaire5062 94.0%68.0%0.31746.9%

Conclusions

The results in the table above show the following patterns among respondents in the TEDS studies.

  • Ethic origin
    The % white increased from just under 92% at 1st Contact to just over 93% in the 2 Year study. However, since then it has remained more or less constant at 93-94%.
  • Twin sex
    The % female increased gradually from 1st Contact (50%) up to age 12 (53%), and has increased more rapidly from age 12 upwards. It has risen to over 60% in recent studies where twins have participated independently of parents.
  • SES
    Both variables (ases, amojob) show a gradual increase in SES with twin age. The mean of 'ases' has increased from 0 at 1st Contact to roughly 0.2 at age 12 then up to 0.3 at later ages. The % mothers employed has increased from 43% at 1st Contact to 46-48% from age 7 onwards.

Factors that may have affected the observed patterns are as follows:

  • Cohort effects
    It has previously been observed that mean SES (measured using variables like ases) is higher for the 1996 cohort than for the 1994 and 1995 cohorts. The studies at ages 2, 3, 7 (phone tests), 9 and 10 did not include the 1996 cohort, hence their participants might be expected to show a lower mean SES than those in other studies that did include the 1996 cohort at roughly the same age. This might help to explain, for example, why the mean SES is lower at age 7 for the phone tests (which excluded the 1996 cohort) than for the parent and teacher studies (which included the 1996 cohort).
  • Study sample selection
    In some studies, for example at ages 7, 8, 16 and 21, the sample included essentially the entire set of contactable TEDS families. In other studies, for example at ages 4, 10 and 12, the sample was selected so as to include only those families that had participated in previous recent studies. (See the study samples page for details.) Given that families who have remained active over time have shown an increase in mean SES, excluding the less active families may have exaggerated the trend towards higher mean SES in such studies.
  • Phoning and reminders
    Some studies, like the 16 year GCSE results, 18 year A-level results and the 7 year phone testing study, have made use of extensive phone calling to families in order to maximise returns. A possible effect of this approach is to make the participants in such studies more representative of the wider population than studies that involved little or no phoning, and fewer reminders. This may help to explain, for example, the relatively high mean SES for participants in the 14 Year study, which involved very little phoning and few reminders.
  • Twin independence
    The increasing % female with age seems to imply that male twins have been more reluctant to participate than female twins as they have grown older. This may also reflect a greater degree of twin independence, and a decline in parental influence, with increasing twin age. This seems most likely to have been a factor in web studies, where web tests may have been completed by twins with minimal parental supervision in many families. In booklet studies, in contrast, families were generally invited to return the entire set of booklets in a single postage-paid envelope, so it is possible that it was less easy for twins to act independently in these circumstances. This may help to explain why the % female is higher for web studies than for booklet studies at ages 12, 14 and 16. The % female has remained lowest in teacher studies (ages 10 and 12), where parents provided teacher contact details without necessarily involving the twins, and in parent-only studies (8 Year, 14 Year Parent SLQ). The % female is also relatively low in the 16 Year GCSE and 18 Year A-level studies where, particularly at the phone reminder stage, data were often gathered directly from parents. These factors may help to explain variations in % female between different samples at a given age, but it is less clear whether they explain variations in SES.