Growing up in Scotland: change in early childhood and the impact of significant events

CHAPTER 2: MEASURES AND METHODS

In this chapter we examine in more detail the events and drivers of child outcomes, including a discussion of how these are defined and measured in the data, before moving on to outline the analytical techniques used in the project.

2.1 Key measures

Fundamental to this research is the construction of measures that allow us to explore the relationship between childhood events and the 'drivers'. Below we explain how each is measured using the GUS data.

2.1.1 Measuring events

GUS collects an assortment of information about children and their families. Main areas the study covers include childcare, education, social work, health and social inclusion.

At each annual interview mothers are asked whether a variety of events have happened in the last year. Table 2.1 shows the range of events and the percentage of children that experienced each event during their first five years. It shows quite a wide range of events that can happen during early childhood and seven in ten children experienced at least one of them. The most common event was the arrival of a new baby to the household, which happened to approximately two in five families.

Table 2.1 Events experienced in early childhood

Event	Per cent 1
Arrival of new baby in household	38
Death of grand-parent (or other close relative)	30
Parent has stopped living in household	12
New parent has entered the household	8
Parent has had a serious illness of accident	7
Either parent been away from child for three weeks 2 or more at a time	7
Sibling has had a serious illness or accident	4
Lived in temporary accommodation	3
Another child has stopped living in household	2
Another child has come to live in household	2
Death of parent or sibling	1
None	29
Bases 3
Weighted	3611
Unweighted	3610

Notes:

¹ Respondents were able to give multiple answers.

² In the second year the question was regarding a separation for three months or more.

³ Base sizes vary, smallest bases shown.

This research will look at four events that can occur during early childhood:

• parental separation;
• moving house;
• job loss; and
• the onset of maternal health problems.

Although Table 2.1 provides a guide to the incidence of these events, this is based on just one all-encompassing question asked to mothers. The GUS questionnaire actually asks more detailed information on each event and we used this information to construct more precise definitions of each event for use in further analysis throughout this report. Each definition is outlined below.

Parental separation

Family breakdown is a process that involves a number of risk and protective factors that interact in complex ways (Mooney et al., 2009). We investigate the event of parental separation that happened in the first five years of the study child's life. Separation is identified when families with two parents living together, whether married or cohabiting, are no longer living together when the interviewer returns a year later. Almost all (98%) of the GUS children whose parents separated went on to live with their mother in the year following separation and we focus only on those families where the mother did not re-partner during the period studied (7% of all families). The reason for this is to focus on the separation event by excluding the added complexity of any subsequent re-partnering event. We also only count separations after the first interview, which allows us to record the family's circumstances prior to separation. We compare separated families with families that remained intact during the period. These two categories of families with associated prevalence statistics are presented in Table 2.3.

Moving house

Again, for reasons stated above, we focus on house moves that occurred after the first interview. Given we do not have sufficient information on why families move house, we differentiate between those who did not move, moved house once and those who moved twice or more. See Table 2.3 for the number of children in each of these categories.

Job loss or significant decrease in working hours

To investigate changes in family employment levels we create a measure of Work Intensity Ratio ( WIR). This is based on the ratio of parents in employment in each family, taking into account the number of hours each parent works; either not in work (0-15 hours per week), in part-time work (16-34 hours per week) or in full-time work (35+ hours per week). Table 2.2 presents the categorisation given to single-parent and couple families according to the working hours of each parent.

Table 2.2 Work Intensity Ratio categorisation for single-parent and couple families

WIR	Single-parent family	Couple family
1	Parent working full-time	Both parents working full-time
0.75	-	One parent working full-time, the other part-time
0.5	Parent working part-time	One parent working full-time, the other not working OR both parents working part-time
0.25	-	One parent working part-time, the other not working
0	Parent not working	Both parents not working

To assess changes in a family's work intensity, we calculate differences in WIR from one year to the next. We exclude families where parents re-partnered or separated to avoid confounding employment changes with changes to parental composition (we look at parental separation as a separate event); 83% of all families who took part in all five sweeps were either stable couple or stable lone parent families throughout. We focus on families who experienced a year-on-year decrease in WIR of at least 0.5, which was not followed by a subsequent increase during the period studied. This is equivalent to a single parent losing a part-time job or, in the case of couple families, one parent losing their full-time job. The change is hence substantial and is likely to significantly affect the circumstances of the whole family, including children, particularly because the family's work intensity does not 'recover' during the period. To experience a decrease in WIR of a magnitude of 0.5 or more, a family needs to have a WIR of at least 0.5 in the earlier sweep, i.e. to be 'work-rich'. Therefore, the main comparison group is families who continuously had a high level of employment ( WIR of at least 0.5). See Table 2.3 for the number of families in each of these categories.

It needs to be noted that a family may experience a substantial loss in WIR yet still remain 'work rich' - for example, a lone parent who changes from full-time to part-time work, or a couple family where one parent stops their full-time job but the other is still employed on a full-time basis. However, even in such cases the change is deemed to be significant enough to be likely to influence the circumstances of the family.

The onset of maternal health problems

GUS asks mothers a number of questions about their health. We identify mothers who face an onset of health problems by selecting those who answered yes to two questions; the first asking whether they have any health problems or disabilities that have lasted or are expected to last more than a year, and the second asking whether this health problem or disability limits their ability to carry out normal day-to-day activities. As we are interested in events which are likely to have a large impact on family life, we focus on mothers who developed a persistent limiting health problem (2%). This was defined as mothers reporting no health problem in the first year of the GUS child's life and then reported health problems in at least two consecutive later years. The comparison group is mothers who reported no health problems throughout the period. Table 2.3 shows the number of families in each of these categories.

Unfortunately, in these sweeps, the study did not inquire about fathers' health problems. So although paternal health problems may have a major impact on family life, our analysis can only focus on the health problems that mothers face.

Table 2.3 Definition and incidence of events

Events	Experienced event		Unweighted base (missing in brackets)
	%	n
Parental separation 1:
Parents separated, mother did not re-partner	7	235	3621 (10)
Residential house move 1:
Moved once	32	1091
Moved twice or more	9	250	3621 (6)
Job loss/decrease in hours 2:
Decrease in WIR of 0.5 or more	6	167	3139 (131)
Onset of maternal health problems
Developed persistent limiting health problem	2	82	3621 (63)

Note: ¹Separations that occurred before the birth or between the birth and the first interview are not counted. Likewise, families that moved house before the birth or between the birth and the first interview are counted as non-movers. This is for analytical purposes, allowing the event to occur after 'baseline' information collected in 2005/06 and before the most recent information collected in 2009/10. ² Base: All stable couple and stable lone parent families taking part in all five years.

Base: All families taking part in all five years for all events except Job loss.

In the majority of cases the selected significant events did not co-occur. The majority (56%) of families who could have experienced all events ² did not experience any of the events at all; some 41% experienced one event while just three per cent experienced two or more events. Thus although some events can occur together, this is beyond the scope of this report.

2.1.2 Measures of 'drivers' of child outcomes

GUS also contains information on a range of factors that other research has identified as 'drivers' of child outcomes (Barnes et al., 2010; Barnes et al., 2008; Marryat and Martin, 2010; Jones, 2010; Deater-Deckard et al., 2009). The four drivers that we examine in this research are:

• home chaos;
• low income;
• maternal mental health; and
• parent-child relationship

- each of which has well-established relationships to child outcomes. Below we outline each of these measures ³ :

i) Home chaos

GUS includes a subset of four questions from the 15-item Confusion, Hubbub, and Order Scale ( CHAOS). This instrument is specifically designed to be administered to parents for assessing turmoil in the child's home across four areas: disorganisation, noise, having a calm atmosphere, and having a regular routine at home (Matheny et al., 1995). US research has shown household chaos to be associated with behaviour problems, inattention and cognitive development problems in children (Deater-Deckard et al., 2009; Dumas et al., 2006). We have combined the four items and taken the top third of the mean scores as an indicator of high level of chaos in the home environment.

ii) Relative low income

There is a well established link between growing up in a low-income household and poor outcomes for children. We use the bottom 30% of the (equivalised) income distribution to identify families living on low income. This is the same proportion of the income distribution focused on by the Scottish Government's anti-poverty strategy (and in fact approximately the percentage of GUS families below the poverty line (Barnes et al., 2010)).

iii) Maternal mental health

Previous analysis of GUS has shown that children who had more prolonged exposure to a mother with mental health problems were more likely to have adverse developmental outcomes (Marryat and Martin, 2010). Maternal mental health is measured in GUS by the Medical Outcomes Study 12-item Short Form ( SF-12) mental health component. The SF-12 is a widely used self-reported generic measure of health status, and is tailored for use in large health surveys of general populations. Higher scores are indicative of better health-related quality of life. The scale does not have thresholds defining whether a score suggests the presence of a psychiatric disorder, so we have followed the approach taken in a previous GUS report (Marryat and Martin, 2010) and defined a relative threshold below which we classify mothers as having 'poor' mental health (16% of mothers were in this category in 2009/10), as opposed to 'average or good' mental health. The threshold score is one standard deviation below the mean score for our analysed population, calculated separately for sweep 1 and sweep 5.

iv) Parent-child relationship

Attachment theory states that an infant needs to develop a relationship with at least one primary caregiver for social and emotional development to occur normally, and that further relationships build on the patterns developed in the first relationships (Cassidy, 1999). The Pianta scale (Pianta, 1992) is used to measure the mother-child relationship at year 5. The scale is constructed using the responses on the extent to which the mother feels a series of statements apply to her relationship with her child (such as 'I share an affectionate, warm relationship with [my child]'). The GUS questionnaire contains a subset of the full 30 items included in the scale. We have constructed measures of 'warmth' and 'conflict' to use in this research, adopting methodology used by Hobcraft and Kiernan (2010) for analysing Pianta questions in the Millennium Cohort Study.

The percentage of children living in families at risk of negative child outcomes is presented in Table 2.4.

Table 2.4 Percentage of families with 'drivers' of negative child outcomes

Driver of child outcome	%	n	Base (missing in brackets)
High level home chaos	35	1,205	3621 (0)
Low income	31	829	3415 (216)
Poor maternal mental health	16	503	3621 (0)
Low parent-child warmth	23	763	3514 (107)
High parent-child conflict	17	574	3548 (73)

Base: All families taking part in all five years.

Note: Measures taken in the fifth year (2009/10).

2.1.3 Measures of family background

GUS contains a wealth of information on the background circumstances of children and their families. This research uses a range of measures to explore which children are most likely to experience each event. These measures are also used as control variables when exploring the association between an event and the drivers of child outcomes. The measures include:

Characteristics of the child:
- Gender
- Ethnicity
- Health
- Low birth weight

Characteristics of the child's parent/s:
- Age
- Education level
- Marital status
- Poor maternal mental health
- Mother's attachment with child ⁴
- Pregnancy planned or unplanned
- Duration of breastfeeding

Characteristics of the child's household:
- Tenure
- Social class
- Low income
- Home chaos
- Rurality
- Local area deprivation
- Number of dependent children in the household
- Family owns or has access to a motor vehicle

2.2 Analytical techniques

Multivariate analysis is used to help identify which families are most likely to experience each event and whether the event is associated with drivers of child outcomes (research questions 2 and 3 in section 1.3 above). Multivariate analysis is used to explore complex associations between an outcome variable and more than one explanatory variable.

Research question 2 involves investigating which families are most likely to experience each event and we use multiple regression analysis to identify which background characteristics are associated with experiencing an event, when accounting for other, potentially confounding, characteristics. For the separation, job loss and maternal health problem events, binary logistic regression is used (as the dependent variable is whether the event happens or not) while ordinal logistic regression is used to model the house moves event (as here we have three categories; no moves, one move and two or more moves). An explanation of these techniques and the relevant statistical output is included in section B of the technical appendix. Interpretation of the analyses is included in the relevant chapters of the main report.

Research question 3 involves investigating how each event is associated with acknowledged drivers of poor child outcomes. Again, we use multiple regression analysis (binary logistic regression) to identify whether there is an association between experiencing the event and the driver of child outcomes, when accounting for a family's background characteristics. Our approach makes the most recent measure of the driver (from 2009/10) the dependent variable in the model. This means the outcome variable always occurs later in time than the predictor variables, which can help with interpretation of the direction of any relationship. Where available, the model also includes an earlier measure of the driver, along with the measures of family background (all measured in 2005/06). This is important because of the possibility of an association between the same, or similar, drivers measured at two different time points ⁵ and because an association between an event and a driver may be different according to the level of the driver before the event. For example, the finding that parental separation is associated with later relative low income takes into account the fact that low income couples are more likely to separate in the first place. Thus, irrespective of prior income level separated families are more likely than stable families to experience income poverty.

It is important to note that the analysis presents significant relationships between the independent and dependent variables - the analysis does not unravel any cause and effect in the relationship. For example, if there is a relationship between moving house and a decline in mother's mental health, the analysis cannot definitively show whether moving house is a cause of declining mental health. However, as we allude above, because we measure mental health before and after moving house, we have more weight to such assertions than is possible in cross-sectional (static) analysis. But we should reiterate that these relationships are inherently complex, and ascertaining the direction of causality is difficult.

Separate regression models are constructed for each event, and for each event separate models are constructed for each of the drivers we explore. It is also important to point out that some events are relevant only to certain sub-groups of children. For example, the separation event is only relevant to children whose parents are initially partnered, whereas a house move can happen to all children, and the analysis groups are constructed appropriately.

GUS was not designed to focus specifically on these events however, the events are part of the life of young children and hence are captured in the study. However this does mean that relatively rare events will affect only a few families in the study. In addition, as noted above some events are only relevant to a subgroup of families or have for analytical purposes been defined in a way that further constrains the number of families in the sample for whom an event can be analysed (for example the job loss event analysis is limited to stable couple and stable lone parent families). While such simplifications make the findings clearer to interpret this is naturally at the expense of some of the immense complexity of the real world.

2.3 Technical Appendix

Readers interested in the details of the analyses should consult the Technical Appendix published alongside this report.