Chapter 3: Comparing Vocabulary at Age 3 Across the Two Cohorts
This chapter considers the difference in language ability between children aged 3 in 2007/08 and children aged 3 in 2013.  It will also explore differences amongst children in different social groups and whether differences between children from more and less advantaged backgrounds have changed over time.
3.2 Key findings
- Children who were aged 3 in 2013 had slightly higher vocabulary scores than children aged 3 in 2007/08. This difference remained even when controlling for known differences between the cohorts such as parental level of education.
- Scores increased by a similar margin amongst children in all sub-groups.
- The difference in vocabulary between the most and least advantaged children (as measured by parental level of education) appears to have reduced slightly between the two cohorts.
3.3 Vocabulary ability at age 3 across the two cohorts
As outlined in section 2.2, language ability was measured via the naming vocabulary subtests of the British Ability Scales, with standardised t-scores assigned to enable comparison across the cohorts. The mean t-scores for each cohort are shown in Table 3.1. As the data show, children in BC2 had a slightly higher vocabulary score than children in BC1.
Table 3.1 Mean vocabulary t-scores, by cohort
|Mean vocabulary t-score||47.6||50.0|
Differences by cohort: p < .001
Children whose parents had higher educational qualifications tended to have higher average ability scores than those whose parents had lower qualifications (Table 3.2). This pattern is evident in both cohorts. Differences by parental level of education and within each cohort are statistically significant.
With the exception of those in the lower Standard Grade group, there was an increased vocabulary score between BC1 and BC2 among children from all parental educational backgrounds. The level of increase was broadly similar in most sub-groups but the difference between children with degree educated parents and those whose parents have no qualifications is slightly reduced for BC2 compared with BC1 (8.3 for BC1 compared with 8.1 for BC2). This tentative trend towards a weakening of the relationship between lower education and poorer vocabulary is evidenced by further results discussed below.
Patterns were similar for children in different income groups and those living in areas with different deprivation levels. On both measures and in both cohorts, children from more disadvantaged circumstances had lower average scores than those in more advantaged circumstances. Furthermore, ability increased between BC1 and BC2 amongst children in all sub-groups of each measure and by a similar margin.
Table 3.2 Mean standardised vocabulary t-scores, by parental level of education and cohort
|Parental level of education|
|No qualifications||Lower Standard Grades or VQs or Other||Upper level SGs or Intermediate VQs||Higher grades and upper level VQs||Degree level academic and vocational qualifications|
|Unweighted bases - BC1||174||179||745||1309||1516|
|Unweighted bases - BC2||124||168||652||1322||2204|
Differences by education level - p < .001; differences by cohort - p < .001; cohort*income p = NS
3.3.1 Controlling for known differences between the cohorts
As noted elsewhere ( e.g. Bradshaw, Knudsen and Mabelis, 2015), there are some notable differences in the socio-economic and demographic characteristics of parents and families in the two cohorts.  One particular difference is that a higher proportion of parents of BC2 children than of BC1 children were educated to degree level or equivalent (42% compared with 34%). In previous analysis of GUS language data (Bradshaw, 2011), as well as in a wide range of other analyses, parental level of education has been shown to be an important predictor of early language ability. As such, it is possible that the increased language ability seen in BC2 children is a function of the higher education level of parents in that cohort.
Further analysis was undertaken to determine whether children in BC2 still had higher average vocabulary ability than children in BC1 after controlling for differences in parental education levels between the cohorts  . The results indicate that, after controlling for differences in the education levels of parents in both cohorts, as well as other cohort differences, children in BC2 are still more likely than those in BC1 to have a higher vocabulary t-score. The remaining results are as may be expected: Lower vocabulary ability is associated with lower parental education levels, being a boy, having older siblings/not being first born, living in a household where other languages are spoken and being younger at the time of assessment.
In addition to controlling for known differences between the cohorts, the multivariate regression analysis also allowed consideration of whether the relationship between these factors ( e.g. parental level of education) and early language development was different for children in each cohort through adding interaction effects to the model. The results indicate that the relationship between language ability and each of the following factors were different in each cohort: parental level of education, whether the child is first born, and languages spoken at home. To explore these apparent differences further, separate multivariable models were run for each cohort.  These indicated that the difference in vocabulary ability between children whose parents had no qualifications and those whose parents were degree-educated was lower in BC2 than in BC1. However, the difference between children whose parents were degree-educated and those with other qualifications had increased. There was also a stronger relationship between having older siblings/not being first born and poorer language ability in BC2 than in BC1. Similarly, being in a household where no English was spoken or English and another language were spoken was more strongly associated with poorer language ability in BC2 than in BC1.
This seems to suggest that the difference in language between the most and least advantaged children (as measured by parental level of education) has reduced between cohorts.