Changes in family socio-economic status as predictors of self-efficacy in 13-year-old Polish adolescents

Data were used from a prospective cohort study with the main objective of examining the impact of early childhood factors (mostly nutritional) on body weight disorders in adolescents. Figure 1 illustrates this study’s three stages. The first two stages were funded by a governmental health policy programme (Mikiel-Kostyra et al. 2002), whilst the third one was an independent project (Jodkowska et al. 2010). In the first phase, hospital staff carried out a survey in 427 maternity wards and included all children born in Poland between 1 and 10 January 1995. Three years later (January 1998), a subset of 9,612 healthy children was selected. The following inclusion criteria were applied: gestational age ≥37 week; birth weight ≥2,500 g; lack of neonatal problems, and discharge at 2–15 days after birth. Then, by systematic sampling of every fifth case (20 %), 1,923 children were selected randomly. A postal survey was conducted with a response rate of 65.0 %. In the third phase, 10 years later, 94.2 % of the families who took part in the second stage, were identified, and separate questionnaires for parents and children were issued. The response rate to the third stage was 51.4 %. Six hundred 13-year-old children formed the cohort of the current analysis.

The main dependent variable was the average self-efficacy score which was measured by using the GSE (General Self-Efficacy) questionnaire designed by Schwarzer and Jerusalem (1992); and validated in more than 30 languages (Luszczynska et al. 2005). The respondents assessed each of 10 statements related to coping in life. A high score, on the GSE scale, indicated high self-esteem; self-acceptance; and an optimistic attitude. In our sample, the GSE scale was unidimensional, and the Cronbach’s alpha reliability coefficient was equal to 0.842.

The second-dependent variable was the KIDSCREEN-10 health-related quality of life index. It was used, in the literature, as an abbreviated 10-item version of the full KIDSCREEN-52 questionnaire (Ravens-Sieberer et al. 2010). In our sample, KIDSCREEN-10’s reliability was equal to 0.774. The sum score was converted into a 0–100 scale, with a low score indicating a poor quality of life—feeling unhappy; unfit; and dissatisfied as regards family life; peers; and school life.

The main independent SES variables were the mother’s education and family affluence. Data, from early childhood (1995 or 1998) and adolescence (2008), were considered and the change between these two periods was determined. Education was recoded in three comparable categories: below secondary; secondary (including post-secondary courses); and higher (tertiary undergraduate or postgraduate) education. A dichotomous variable, measuring change, was created, and a value of “1” was assigned to mothers who had increased their education level. In the models, taking into account only data from year 2008, six original categories of education were maintained: primary; basic vocational; secondary; post-secondary; BA (bachelor’s) degree; and higher than BA degree.

The analysis, of changes in family affluence, was more difficult due to the lack of comparable items. In 1998, only one question was available concerning retrospective subjective assessment of family living conditions in the first year of the child’s life (1995). Family conditions were classified in the following four categories as: very bad (found only in one family) or bad; average; good; and very good. In 2008, similar categories of family conditions were distinguished based on the modified FAS (Family Affluence Scale). Originally, FAS was composed of four items related to the adolescent having his/her own room; the number of computers and cars in the family; and holiday trips with the family outside the place of residence (Currie et al. 2008). After the fifth item, measuring subjective family wealth, assessed from the parent’s perspective, was added, the reliability, of this scale, rose from 0.49 to 0.58. A modified standardized index of affluence was estimated using the principal component’s method; this improved its continuity. Families, surveyed in 2008, were divided into four groups, proportionate to the earlier stage of research. These groups were labelled arbitrarily in the same manner as in 1995. Rather than illustrate the absolute level of family poverty/affluence, this division showed their relative position in the sample under consideration. There were five categories of family, whereby affluence, in the studied period, was described as: fell significantly (a difference of −3 or −2); fell slightly (−1); stayed the same (0); increased slightly (+1); or increased significantly (+2 or more).

An attempt was made, also, to include other available confounding variables: mother’s age (2008); her marital status (1995); the father’s education (2008); and family structure (2008). Because no significant relationship was noted, the use of these factors was discontinued.

Before all analyses, missing values were imputed by conditional mean procedure (Barzi and Woodward 2004), if applicable. Changes, in the level of education, were tested using the McNemar–Bowker method, i.e. the Chi-squared test for dependent variables. Mean GSE scores were compared using the one-factor ANOVA and the post hoc Scheffe’s test. A series of linear regression models was estimated whereby GSE was the dependent variable. A model, specified on the basis of the automatic selection procedure, was compared with: (1) the full model including all the factors; (2) the model using only the 2008 data; and (3) the model including only variables related to the changes between early childhood and adolescence. The standardized regression coefficients and the R-squared coefficient of determination were shown.

In the final part of the study, a potential further use, of all the obtained results, was discussed. The structural equation (path) model was assessed to evaluate whether self-efficacy could be the mediator of the relationship between a change in the mother’s education/family affluence and general well-being (KIDSCREEN-10). The effect of mediation was checked using the Baron and Kenny criterion (1986) and Sobel’s test.

The calculations were carried out using the PASW Statistics v. 17.0 and AMOS v.20 software.

Table 1 gives the mean GSE scores with regard to gender and SES variables with comparative response categories at two points of time. On a basis of post hoc Scheffe’s test on the 2008 data, a significant difference was noted between the children of the mothers with lower and higher than secondary education (p = 0.029) and when secondary and higher than secondary education were compared (p = 0.046). When considering the impact of affluence in 2008, the highest difference occurred between families living in poor and good conditions (p = 0.019).

There was found, also, to be a significant impact of a change in the mother’s education level. When comparing groups, with and without a change in the mother’s education, the average GSE scores were 73.4 and 68.1, respectively, p = 0.003. Comparing groups, with the same mother education level in 2008, the higher GSE score was found if the mother started from a relatively lower level of education (Fig. 2).

The difference in the mean GSE values, in the groups of families with extreme changes, i.e. when affluence rose and fell sharply, exceeded eight points; however, this result was not statistically significant (p = 0.130). This could be due to the small size of the extreme groups and the observation of a great variability of results in these groups. In general, in the group of families where affluence had not improved, the mean GSE was 67.9 (SD = 14.66) as opposed to 70.9 (SD = 13.04) in the group where an increase was noted (p = 0.019).

When comparing the mean GSE indexes, calculated for boys and girls, a generally higher impact of the mother’s education could be observed in girls.

A series of alternative regression models, adjusted for gender, were specified and estimated (Table 2). When all potential explanatory variables were added to Model 1, only the change, in the mother’s education level, and the modified FAS index were significant. In Model 2, only the 2008 data were used and a significant impact of current affluence; however, nothing of the mother’s current education was noted. In Model 3, only the effects of changes, in mother’s education and in family wealth, were considered and only the first parameter was significant. In Model 4, suggested by the method of the automatic selection of variables, the GSE was explained by a change in the mother’s education and family’s current wealth. These factors explained the GSE variability in 3.0 %, slightly better than in Models 2 or 3. Standardized parameters were higher in Model 4 than in Model 1; this suggested interaction between mother’s education and family affluence (both measured in 2008) as predictors of GSE.

When considering that self-efficacy was an important determinant of the proper functioning of adolescents, and the obtained results could be used in devising more complex models featuring various outcome measures. A simple path model, whereby self-efficacy was the mediator of the relationship, between a change in family SES (both mother’s education and affluence) and well-being, could serve as an example. Although an increase, in the mother’s education and family affluence, resulted in a significant GSE increase, only a slight tendency for growth was present with regard to the KIDSCREEN-10 index. A strong correlation was found between the quality of life index and self-efficacy (r = 0.497; p < 0.001). The path model was estimated including direct and indirect relationship paths (Fig. 3). The Sobel’s test confirmed a significant mediation effect arising from a change in the mother’s education (p = 0.005), but not from a change in family affluence (p = 0.397). In addition, some goodness of fit statistics improved after the path of direct relationships was eliminated. For example, the RMSEA value fell from 0.084 to 0.035, and the upper limit, of the confidence interval, decreased to the recommended level of less than 0.08. The CMIN/DF value fell from 5.23 to 1.75. In the model, including direct paths, the RFI index was much worse (0.724 vs. 0.908), as was the TLI index (0.764 vs. 0.958).

Because our study was an analysis of secondary data, it had some limitations. The first two stages of this three-wave study were designed for a different purpose than the last one. In addition, they were designed, even less so, to study the influence of changes in family socioeconomic status on selected internal resources in adolescence. Moreover, a limited number of variables describing a changing family situation was applied. Despite the attempts to extend the set of confounding variables, no predictor of deterioration, in GSE, was identified. Similar to the National Longitudinal Survey of Youth (Ryan and Claessens 2012) we could consider only additional data on mother marital status, at the time of the child’s birth, and on family structure in early adolescence. Having two points of time did not allow us to establish when the change, in family status, occurred. In particular, it was difficult to evaluate the effect of negative changes, such as divorce. If over time, single-parent and blended families become more common, parents and children might perceive changes, in these structures, as more normal, more predictable, and, therefore, less stressful. Finally, no information was collected about parent self-efficacy (in particular, the mother’s self-efficacy). It was known, from other cited studies, that mothers, with a high level of self-efficacy, continued to study more frequently. In adolescents, the determinants of self-efficacy including school and peer environment, were definitely more complex (Caprara et al. 2002). This kind of research ought be carried out on a larger sample; this would allow for a multi-factor analysis. In considering the large size of the primary sample at the first stage, only a fraction of the cases were included in the final analysis. This was a consequence of including only 20 % of the initial sample at the second stage.

Nevertheless, our study had a number of advantages which, we hoped, would counterbalance the presented shortcomings. Currently, in Central and Eastern Europe, studies, which are based mostly on a cross-sectional design, are being conducted on changes in social gradient of health and health behaviours (Mazur and Woynarowska 2011; Pitel et al. 2012). Cohort studies, such as presented herein, were recognized as the best sources of data in linking early childhood events with later outcomes (Robert 1997). One advantage of the study was its significant territorial diversity; usually, this was not achieved in studies of adolescents carried out in schools. Another advantage included the use of a coherent age group and, therefore, eliminating an additional age factor.

The use of the FAS Scale, as a measure of material conditions, could be questioned. This scale was used widely in other papers (e.g. Holstein et al. 2009), but some authors criticised its low internal consistency (Molcho et al. 2007). The modification of the original FAS, improved its reliability, but not sufficiently. However, the results of validation studies, conducted, also, in Poland in 2003 (Mazur 2010) and 2005 (Andersen et al. 2008), could serve as an argument in favour of using it. These studies showed high compliance of child–parent responses and strong correlation between the overall FAS index and external income measures. The original method of analysing a change in family affluence was devised; this increased data comparability. This method was the authors’ adaptation of an approach, used in research of households, based on income distribution (quantiles). To our knowledge, this study is one of few studies describing that kind of relationship on the basis of prospective design.

Our study belongs to a stream of research relating to social health inequalities from the life course perspective. It referred to practical measures, in particular those related to the strategy of lifelong learning and creating a home learning environment (HLE). In contrast to other studies, the focus was not on the negative impact of low social position but on the positive influence of raising it. Attention was brought to the fact that the transition from late childhood to adolescence occurred at a time when families underwent a series of economic transformations in their aspiration to improve living conditions and to achieve stable affluence. According to Delors’ report (Delors 1996), lifelong learning should encompass many activities enabling people to familiarize themselves with the phenomena of the changing world; social changes; and, above all, themselves. Raising parental education, apart from having direct benefits for themselves and for society development, was very important, also in addition, for their children. Knowledge and skills, acquired by the parents, might support a harmonious functioning of families and normal child development. It might be assumed, that parents, themselves interested in further education, would create a positive home learning environment. Currently, HLE creation was considered to be one of the strategies for reducing poverty and social inequalities in health.

The obtained results might serve as an inspiration to undertake research concerning the impact of changes in socioeconomic status on other developmental assets conducive to adolescent health.