Educational attainment emerged from preliminary analysis as an important variable with wide differentials across groupings and proved to be the most robust outcome measure for the respondents' age and stage of development. In this chapter, the relationship between educational attainment and other study factors is explored further. Bi-variate analysis is used initially to examine the relationship between educational attainment and other, potentially important, variables. For this examination participants were divided into three groups: respondents who had completed part of second level (n=33), those who had completed all of second level but did not go on to third level (n=26) and those who did go on to third level education (n=38)
23. In the second part of this chapter, log linear models are used in an attempt to develop a predictive model for educational attainment.
23 Where expected cell counts fall below five in more than 25% of cells, education level was recoded into two categories (those who had completed all or part of second level (n=59) and those who had gone on to third level (n=38).
6.2 Educational Attainment: Descriptive Variables
6.2.1 Personal factors
Somewhat unexpectedly, no significant interaction between educational attainment and gender was identified. However, as noted in the preceding chapter (Chapter 5) this lack of gender difference might be explained by differential gender contact rates. The contact rate between Phase One and Phase Two of the study was 60% for females and just 46% for males. There was a significant association between educational attainment and parenthood in that those respondents who had gone on to third level education were less likely to have children and those who only completed part of second level were more likely to have children ( = 11.496, df = 2, p < 0.01). There were no significant differences across the education groups in relation to social network. Another potential source of social support, religious belief, also proved to be non-significant in relation to educational level. The association between education and satisfaction with various aspects of the respondents' lives (work, finances, housing and leisure opportunities) was similarly non-significant. This was also true for self-esteem and locus of control.
The association between current employment status (and job satisfaction) and educational level did not prove to be statistically significant. There were however significant differences between the groups in relation to their current socio-economic grouping ( = 17.973, df = 2, p < 0.001
24). The relationship between education level and socio-economic status is in the expected direction in that those with third level education are likely to be in the professional and managerial socio-economic groups. This is illustrated in Table 6.1. Examination of the mean ranks confirms that a linear trend exists, with those leaving school after only part of the secondary cycle showing the lowest socio-economic status.
Table 6.1: Educational level and participants' socio-economic status
24 Ten respondents were not in employment because they were still in full-time education.
6.2.3 Psychological Health Status in Phase Two
None of the respondents had received a clinical research diagnosis (based on the SCID assessment) therefore this factor was examined at the level of likely diagnosis. There was no significant association between this categorisation of diagnosis and educational attainment. Neither was there a relationship between educational levels and respondents' reports of an emotional problem and/or receiving treatment for psychological problems in the intervening years.
6.2.4 Contact with the Law
Table 6.2: Educational level and contact with the law
Table 6.2 illustrates the results of an analysis of the relationship between participants' education level and whether they had been in trouble with the law. The significant relationship evident here ( = 12.151, df = 2, p < 0.01) reflects the association already detected in the preceding analysis (Chapter 5). It is clear that respondents who completed only part of second level education were more likely to have offended. A significant difference was also found between educational attainment and frequency of involvement with the law ( = 12.747, df = 2, p < 0.01). This is illustrated in Table 6.3. Inspection of the mean ranks suggested a linear trend with those completing less education being in trouble more frequently.
Table 6.3: Education level and frequency of contact with the law
6.3 Childhood Characteristics and Educational Outcome
In this section factors identified as important in Phase One of the study are examined in relation to educational outcome. These include respondents' and mothers' diagnosis, categorisation of behavioural deviance and respondents' IQ (assessed in Phase One of the study) and economic disadvantage during the respondents' childhoods.
6.3.1 Respondents' Psychological Health and Behavioural Status
As indicated in the previous chapter (Chaper 5) a diagnostic categorisation at age eleven proved to be an important indicator of educational outcome. This is further illustrated in Table 6.4. There was a significant association between educational attainment and this factor ( = 9.2, df = 2, p < 0.01).
Table 6.4: Diagnosis (Phase One) and present educational level.
Almost half of those without a diagnosis went on to third level, in contrast to 17% of those with a diagnosis. A similar trend was evident for those with a deviant categorisation (Table 6.5). A significant difference emerged in relation to educational attainment ( = 21.9, df = 2, p < 0.001) between the deviant and the non-deviant group (measured by the Rutter B2 instrument). Two-thirds of those with a behaviourally deviant categorisation on this instrument did not complete second level, compared with 19% of the non-deviant group.
Table 6.5: Behavioural Deviance/Non-deviance (Phase One) & Education Level
An examination of the mean ranks for each group suggests that those who had completed part of second level education had a higher number of deviant symptoms than other participants. This finding reflects the outcome of the analysis of educational level and a categorisation of behavioural deviance in Phase One. Inspection of the mean ranks indicated that those who had achieved lower educational levels had higher deviant scores. This is illustrated in Table 6.6
Table 6.6: Educational level & Mean/Standard Deviation Scores (Rutter B2 Scale)
25 The educational level attained by the participants and their score on the Rutter B2 was examined using a non-parametric test as scores based on this instrument are not normally distributed within the sample.
6.3.2 Intellectual Capacity (IQ score)
Table 6.7 below illustrates the mean and standard deviation IQ scores across education levels. There was a significant difference across the groups on IQ assessed in Phase One of the study (Fobs (2,86) = 5.565, p < 0.01). Those who entered third level had higher IQ scores than those who had completed either all or part of the second level cycle only.
Table 6.7: Educational Level & Mean and Standard Deviation IQ scores
When IQ scores were re-categorised into above and below average (Table 6.8) a significant difference between IQ categorisation and educational outcome was apparent ( = 15.452, df = 2, p < 0.01). Those with above average IQ were very likely (over 57%) to go on to third level while those (over 51%) with below average IQ tend to leave the educational system without completing second level.
Table 6.8: Educational Level by IQ Categories.
6.3.3 Mother's Diagnosis
In Phase One of the study mothers' psychological health status proved to have an important association with psychiatric diagnosis in the child. The relationship between results on the Clinical Psychiatric Instrument (Goldberg et al., 1970) and educational attainment is illustrated in Table 6.9.
Table 6.9: Maternal diagnosis in Phase One and Respondents' Educational level at Phase Two
Despite the importance of the diagnostic variable in Phase One, there was no significant association in this phase of the study between the education level attained at follow-up and whether the mother received a diagnosis of psychiatric disorder in Phase One. However, the trend is in the expected direction in that almost 50% of those with non-diagnosed mothers (in Phase One) went on to third level education. The other assessment of mothers' psychiatric status in Phase One (The Malaise Inventory (Rutter et al 1970)) proved to be similarly non-significant. Educational attainment was also analysed in relation to mothers' treatment for psychological difficulties between the two phases of the study but there was no significant relationship between this and the young person's educational level. Paternal psychological health also proved to be non-significant.
It was not possible to examine the relationship between educational level and parental marital status as the majority of mothers were married in both phases of the study. There were no differences in relation to parental educational attainment. The quality of respondents' relationships with their parents was also examined but no major differences were evident across the groups.
6.3.4 Socio-economic Background
The socio-economic status of the participants' family of origin is compared across education levels in Table 6.10.
Table 6.10: Educational Level & Family's Socio-economic Status (SES)
Given the small sample size (n=80) and the large number (eight), of categories in the SES variable it is not surprising that there were no significant differences between the groups. But again, the trend is in the expected direction with 36% of those with third level education coming from families in the Professional categories. An analysis of paternal unemployment (during the respondents' childhoods) produced no significant association. However there was an important difference between the groups in relation to another indicator of economic disadvantage i.e. whether the family was in receipt of state benefits ( = 15.162, df = 2, p < 0.01). This data is contained in Table 6.11.
Table 6.11: Educational Level & Family in receipt of State Benefits
Respondents from families who received state benefits completed only part of second level while those from families not in receipt of state benefits were more likely to go onto third level. To assess the impact of disadvantage further, the Index of Poverty used in Phase One of the study was examined for each family and the results are illustrated in Table 6.12 below.
Table 6.12: Educational Level & Poverty Indicators (Phase One)
As Table 6.12 demonstrates, there was a significant difference across the groups on this measure of poverty ( = 10.950, df = 2, p < 0.01). Inspection of the mean ranks indicated that those who had completed more education had higher scores on this scale, indicating less poverty at Phase One. In developing the predictive model in the following section, receipt of state benefits is included as the most robust indicator of socio-economic status.
6.4 A Predictive Model of Educational Attainment.
The two-way analyses above identified four factors from Phase One of the study as statistically related to educational attainment. These factors were: The family being in receipt of state benefit, Respondents' IQ categorisation, Diagnosis and Categorisation of deviancy. In addition, three of the four factors (Diagnosis, Deviancy and Receipt of Benefits) proved to be interrelated in that respondents with a diagnosis tend to have a deviant categorisation (on the Rutter B2 Scale) and come from families who were in receipt of state benefits. The respondents' IQ categorisation was independent of the other three variables. A predictive model for educational attainment was developed, based on these key variables, using Log Linear and Logit Models
26 These models are used when independent and dependent variables are categorical. In a logit model the value of the dependent variable is based upon the 'log odds'. It is the variation of this measure which is explained by the independent variables. In the present analysis, the dependent variable is educational attainment and the independent variables are: Family in Receipt of State benefit, IQ categorisation, Behavioural Deviance and Presence of a Psychiatric Diagnosis. The sample size is reduced to 76 when the five variables are considered simultaneously.
6.4.1 Loglinear Saturated Model
Saturated models contain all possible effect parameters, that is, all individual effects of the independent variables and all interaction effects. A fully saturated model is therefore a perfect representation of the data. Thereafter effects with small estimated values can be deleted. This process was completed for all study variables and the findings are illustrated in Table 6.13.
Table 6.13: Main Parameter Estimates for Saturated Logit Model
The direct effect of three independent variables, categorisation of deviance, family in receipt of State benefits and IQ level emerged as important. Psychiatric diagnosis did not provide any additional explanatory power and was therefore eliminated from the analysis.
6.4.2 Loglinear Unsaturated Model
Table 6.14: Parameter Estimates for Specified Logit Model
At this stage various unsaturated models were tested. Independent variables and interaction effects, which have large estimated values (Coefficient and Z-Value) form the basis of the unsaturated model presented below (Table 6.14). Inspection of the Coefficients and the Z-values indicate that the categorisation of behavioural deviance has a slightly stronger influence than family in receipt of State benefits or IQ categorisation. The results of the analysis are summarised in Table 6.15.
Table 6.15: Log Linear Model: Categorisation of Deviance, Family in receipt of State Benefit, and IQ categorisation, by Level of Education achieved
As Table 6.15 illustrates, respondent's categorisation of deviance, family in receipt of State benefits and IQ all have a significant, independent, direct effect on educational achievement and these differences are clearly evident at all three levels of education. Eighty-five percent of respondents who were classified as behaviourally deviant, whose family were in receipt of state benefits during their childhood, and who were below average in IQ, did not complete second level education. Conversely, only 8% of those outside these categories failed to complete second level schooling. Overall, all non-deviant sub-categories are less likely to leave school without completing second level education. The differences in relation to third level education are stark. Eighty-five percent of the deviant group who were in receipt of benefits and with a below average IQ did not complete second level in comparison to 43% of non-deviant respondents who had similar attributes (i.e. below average IQ and family in receipt of state benefits). A categorisation of deviancy therefore, assessed at age eleven, is a very sensitive predictor of educational attainment.
Two other factors, family in receipt of state benefit and IQ categorisation, also have significant predictive powers, but are somewhat less important than a deviant classification. Respondents with below average IQ are less likely than those with above average IQ to pursue third level education. Those groups in receipt of state benefit are more likely to leave the educational system without completing second level. However, those in receipt of state benefit with above average IQ present a similar pattern of educational attainment as those with a below average IQ but with families not in receipt of state benefit. This finding indicates that socio-economic advantage can benefit individuals educationally. A similar pattern is evident for both the deviant and non-deviant groups but the deviant group is less likely to complete second level and go on to third level education.
This section explored the relationship between educational attainment and factors which emerged as important from preceding analyses. Neither gender nor current mental health status proved to be important in relation to educational attainment. However socio-economic status, based on current employment, was strongly associated with educational outcome. So also was involvement with the law.
Despite the importance of maternal diagnosis when the children were aged eleven, there was no significant relationship between this variable and educational attainment. There was an important association however between educational level and a diagnosis of psychiatric disorder and/or categorisation of behavioural deviance for the child. A significant relationship was also evident between respondents' IQ and educational outcome. The socio-economic categorisation of the family of origin did not prove to be important but another indicator of economic disadvantage, receipt of state benefits, was significantly related to educational outcome.
Four factors assessed in Phase One of the study (diagnosis, behavioural deviancy, IQ and family in receipt of state benefits) were identified as key to understanding educational outcome. When subjected to multivariate statistical techniques categorisation of deviance, IQ and family in receipt of state benefits all proved to have a significant, independent and direct effect on educational achievement. In particular, a categorisation of deviance, assessed at age eleven, proved to be an extremely sensitive predictor of educational attainment. IQ proved to be important in terms of educational attainment but this was associated with economic factors. Specifically, socio-economic advantage clearly benefits individuals in accessing third level education.
Structured Clinical Interview for DSM-IV Axis I Diagnoses (SCID)
Beck Scale for Suicide Ideation (BSSI)
Rosenberg's Self-esteem Scale
Arizona Social Support Interview Schedule (ASSIS)
Locus of Control
Frequency of SCID Diagnostic Categories