Information from several linked data sources was used in this study, and included the Norwegian National Education Database; and The Historical Event Database, FD-Trygd; administered by Statistics Norway. These data sources offer rich longitudinal population data on income and wealth, welfare benefits and education, as well as demographic information whereby parents and children are linked. The Medical Birth Registry of Norway (MBRN) is also linked to a questionnaire survey on health, The Oslo Health Study 2000–2001 (UNGHUBRO). In 2000–2001, the survey was administered to all grade 10 pupils in Oslo, most of whom were aged 15–16 years. The overall response rate for the survey was 88 %. We included respondents born in Norway (n = 5 335). Compared with the national statistics obtained from the MBRN , the UNGHUBRO sample seemed to be fairly representative with respect to both birthweight and gestational length. All parents received written information about the questionnaire, and the students completed a consent form before participation. For youth less than 15 years of age the parents were contacted and asked to provide a separate informed consent form. The Regional Ethics committee, South East C approved the study.
Enrolment in higher education was the outcome variable, which was coded “1” if the individual entered higher education, and “0” otherwise. Any enrolment counted provided that the respondent was registered in higher education in October of any year between 2000 and 2011.
Infant health was measured using birthweight and the breathing effort, heart rate, muscle tone, reflexes and skin color (Apgar) score. Birthweight and the five-minute Apgar score are obtained from the MBRN. Birthweight was entered into the models as a normal logarithm of birthweight. The five-minute Apgar score is an overall assessment of newborn well-being five minutes following delivery. A score of 7–10 is defined as normal . Gestational length ranged from 23 to 47 weeks.
Self-rated health in adolescence
Self-rated health in adolescence was taken from the UNGHUBRO survey. The participants were asked: “What is your present state of health?, and they rated the current status of their own health on a four-point scale ranging from “very good” to “poor”. The self-rated health of the adolescents captured both the physical and psychological dimensions of well-being . The survey also contained a battery of questions relating to psychological distress in adolescence which covered:
Fear (suddenly feeling panicky for no reason).
Suddenly feeling frightened or anxious.
Feeling faint or dizzy.
Feeling tense or harassed.
Being self-critical (easily finding fault with oneself).
Feeling depressed or dejected.
Feeling useless and of little worth.
Feeling that everything is a burden.
Feeling hopeless about the future.
Psychological distress was measured by combining these 10 items into the Symptom Checklist (SC)-10 scale score – a validated 10-item short version of the original Hopkins Symptom Checklist (SCL-90) . Low scores indicated low levels of psychological distress and high scores high levels thereof.
Parental socio-economic status
Parental SES was measured by separate variables for parental education, income and wealth, and was taken from the register data. Parental education constituted the education level of the parent with the highest education or of the only parent who was present. Parental education was divided into four levels: compulsory school or less, upper secondary school, Bachelor’s level and Master’s level or higher. The parental income and wealth variables were measured as both parents’ gross combined mean income during the years that the persons in the sample were aged 7–16 years. Income included salary, income from self-employment and state support benefits, e.g., unemployment, sickness and maternity. Wealth included taxable assets and financial capital. Parents’ income and parents’ wealth were originally recorded in the Norwegian currency (the Norwegian Krone). The logarithm of parental income and wealth was used in the analyses. These variables were also centred on their mean. It was determined whether or not participants lived with both parents by asking them the following question: “Who do you live with at present?” This variable was categorised into a “two-parent household”, and a “one-parent household” /“other household arrangement”.
Physical activity was measured by asking: “During school hours, how many times a week do you take part in sport or participate in physical exercise to the extent that you feel out of breath or sweat?” Smoking habits were categorised into “non-smokers”, “previous smokers or occasional smokers”, and “regular smokers”. Alcohol consumption was measured by asking: “Have you every drunk so much alcohol that you became drunk?” Answers were coded into two categories of “no, never or once” and “2–3 times or more”.
Relationship to family was measured by asking: “When you think about your family, would you say that you feel attached to your family?” Relationships with friends were determined by the statement: “I feel closely attached to my friends”. Answers were coded according to two categories of “completely agree” and “otherwise”.
Grade points were calculated by combining the grades taken in grade 10 into 11 main school subjects (divided by 10 and centred around their mean). Future educational expectations were determined by the question: “What is the highest education that you have considered undertaking?” Answers were coded into two categories of “upper secondary school and lower” and “higher education”. Days absent from school due to illness were estimated by asking if the respondent had experienced pain or illness regularly over the last 12 months. The respondents were then asked if pain or illness had resulted in them having to stay home from school. The variable ranged from 0 to 10 days or more. Persons who reported no pain were coded as “0”.
An immigrant background, gender and the study year were obtained from the register data and considered to be covariates, and used in all five models. Immigrant background was categorised as “native origin”, “non-Western second-generation immigrants” and “Western second-generation immigrants”.
The analyses of enrolment in higher education were performed using logistic regression, with odds ratio (OR) and 95 % confidence interval (CI). There were some missing values in the dataset (between six and 185 cases out of 5 354). Excluding missing cases or including the missing cases as a separate category could have led to a biased estimate. Therefore, multiple imputations were run using the mi impute chained command in Stata® 13. Five imputed datasets were used. The procedure replaced each missing value with a set of plausible values based on all other variables in the dataset. For further details, see White, Royston, Wood .