Selection of 'lows'
Families were selected as 'lows' for this study if either or both twins were classified as 'low language' and/or 'low Parca', based on the 4 Year booklet data. Other selected families were designated as 'controls' if neither twin fell into these categories.
In the earliest stages of the study, in order to assist with staff training, a number of families assumed to be 'controls' were visited before any 'lows' were selected. When the first 'lows' were selected, the criteria for selection were based on the limited amount of 4 Year booklet data that had been entered so far. Throughout the study, due to time constraints, there were cases where the selection was made on the basis of a visual inspection of the 4 Year booklets, so that families could be visited without having to wait for data to be entered and incorporated into the dataset. All these factors led to some variability in the criteria used to identify 'lows' in the early stages.
The historical classification of twins as 'lows' has been retained in variables dlowlan1/2 (low-language) and dlowpar1/2 (low-Parca) in the in-home dataset. Derived from these historical classifications, 'control' families are labelled with variable econtrola in the dataset. For the reasons described above, these historical classifications were subject to early variations and to some apparent error when compared with strict classifications based on 4 Year booklet scores.
An updated classification, based on data currently in the 4 Year booklet dataset, has also been made in variables dllang1/2 (low-language) and dlparca1/2 (low-Parca). Derived from these updated classifications, 'control' families are labelled with variable econtrolb in the dataset. Hence, within the in-home dataset, it is possible to compare historical and updated classifications of lows and controls.
As the study progressed, the final criteria for selecting 'lows' were established. These are the criteria that have been used in the updated classification variables dllang1/2 and dlparca1/2, as mentioned above. These criteria relate to variables in the 4 Year booklet dataset, as follows:
- Low Parca Twins
(dlparca1/2 = 1 in the dataset).
The 4 Year child booklet included various measures of twin non-verbal cognitive ability, collectively called the 'Parca' measures. These included parent-administered tests and a parent-reported questionnaire (see the 4 Year study for further details of these measures). The 4 Year dataset contains a standardised overall Parca score, variable dparca1/2, derived from these measures and having a roughy normal distribution (see 4 Year derived variables). For the In Home study, a twin is classified as 'low Parca' if s/he scored in the lowest 5% of this non-verbal cognitive ability distribution. The cut-off for designation of 'low Parca' twins is made at values of dparca1/2 of -1.7 or less.
- Low Language Twins
(dllang1/2 = 1 in the dataset).
The 4 Year child booklet included various measures of language ability. These included parent-reported measures of vocabulary and general language ability, as well as a parent-administered pictorial vocabulary test (see the 4 Year study for further details of these measures). Twins are designated as 'low language' for the In Home study if they meet any of the following criteria on the basis of these measures:
- the parent-reported vocabulary score (dvocab1/2) was 21 or less
- the parent reported that the twin was not yet talking in sentences (dsay011/2 = 1,2,3 or 4)
- the parent-administered picture vocabulary test score (dpictot1/2) was 4 or less AND the parent reported concerns that the twin's language was developing slowly (dsayc1a1/2 = 1)
For the low-language classification of individual twins, the historical and updated categories agree for 417 lows and 1185 non-lows in the dataset. 56 twins were low-language according to the historical category but not the updated category; and 14 twins were not low-language historically but are low-language according to the updated category. 4 twins (1 low and 3 non-low according to the historical category) having missing language data in the 4 year booklet.
For the low-Parca classification of individual twins, the historical and updated categories agree for 358 lows and 1247 non-lows in the dataset. 28 twins were low-Parca according to the historical category but not the updated category; and 40 twins were not low-Parca historically but are low-Parca according to the updated category. 3 twins, not low according to the historical category, have missing Parca data in the 4 Year booklet.
For the control family classification, taking paired twins into consideration, the historical and updated categories agree for 310 control families and 473 low families. Only 1 family is classified as a control by the historical category but not the updated category. 54 families are classified as controls by the updated category but not by the historical category.
Each family selected for the In Home study was initially contacted by telephone, in order to seek verbal consent and to arrange a visit date. On the arranged date, the family would be visited in their home by two testers (researchers). Each tester would then conduct the full battery of tests on one twin. The two twins were tested simultaneously and independently by the two testers. The responses made by each twin were assigned numerical scores by the tester, using the rules specified in the test manuals, and the scores were recorded in the test score sheets (pdf).
In addition to the cognitive tests carried out on the twins, the battery included the Bayley behaviour ratings. These ratings were assessed by the testers, and recorded on the score sheets mentioned above. Guidance notes (pdf) for the behaviour ratings were provided for testers.
Occasionally, visits were abandoned (with little or no data collected) if either twin was discovered to have a severe medical condition such as autism or hearing difficulties that would affect the test outcomes. If a tester suspected that a twin had hearing difficulties, then a hearing test was administered on the spot (the test result is recorded in dataset variable ecatt1/2). Families excluded from analysis on medical grounds are flagged with the dataset variable emedexcl.
While the twins were being tested, the parent was asked to fill in the parent questionnaire (pdf), which included a consent form. In addition, each parent was asked to complete the 'Expressed Emotion' recording as described in a sheet of written instructions (pdf) (verbal instructions were also given). In outline, the parent was asked to talk about each twin as an individual, without making comparisons between the twins, spending five minutes talking about each twin. The recording was made by means of a cassette tape recorder armed with a blank audio cassette tape provided by the TEDS testers.
Immediately after the visit, each tester independently completed the post-visit ratings (pdf). In addition, miscellaneous notes relating to the twin tests or the visits generally would be recorded on the front of the twin score sheet.
For this study, all data were entered manually by TEDS staff in the TEDS office. The data were entered into an Access database that had been designed specifically for this purpose. Most of the data recorded in the twin test score sheets had already been scored by the testers, and these scores were entered directly. Verbatim text notes were also entered, although these do not form part of the analysis data.
The responses in the parent questionnaire and post-visit questionnaire are generally multiple-choice tick boxes; these responses were coded and entered as numerical data by the data-entry staff.
The tape recordings made by parents for the Expressed Emotion measure have not been transcribed or coded in any way, so these data do not form part of the dataset. The original recordings were made on audio cassette tapes. The surviving recordings were copied electronically to mp3 audio files in 2016, after which the cassette tapes were destroyed.
Other administrative data entered in the database included the visit date, the identities of the testers, whether each twin was categorised as 'low language' and 'low Parca' from the 4 Year booklets, whether the family was to be classified as a medical exclusion, and the results of hearing tests if carried out. Also entered were flag variables to record whether particular tests (or groups of tests) had been affected by events such as as twin refusal or testers' decisions to abandon testing.
The Access database into which the data were entered is still the primary source of electronic raw data for this study, although some modifications have been made to the database and the data have been cleaned. The database is described more fully in In Home raw data files.