Project overview
Previous research has found a large decline in cognitive scores for high-ability low- socio-economic status (SES) children and a large increase for low-ability high-SES children, such that their trajectories traverse over time, the low-ability high-SES group performing better at older ages than the high-ability low-SES group. A recent paper by Jerrim and Vignoles (2011) argues that the patterns of change observed in previous research are an artefact of regression to the mean, due to both measurement error and non-comparability of tests over time, and illustrate how trajectories can change dramatically when attempts are made to adjust for regression to the mean.
The first source of regression to the mean that Jerrim and Vignoles identify is selection, which is caused by measurement error. In classifying children as high or low ability based on a single test score, by chance some children will have performed abnormally high and others will have performed abnormally low. As such, their performance in this test cannot be considered a measure of their true ability. The second source of regression to the mean is non-comparability in tests of ability over time. If the tests are assessing different skills, this will weaken the correlation between test scores over time, because those children who perform well at verbal tasks (for example) will not necessarily be the same children who perform well at mathematical tasks. Jerrim and Vignoles carried out their own analysis using both the Millennium Cohort Study and the Avon Longitudinal Study of Parents and Children. They adjusted for regression to the mean by classifying children into high and low ability groups based on their scores in a number of auxiliary tests and by using a different measure (of the same ability) to assess development over time. They show that the same pattern of change in high-ability low-SES children and low-ability high-SES children in earlier research is not evident once these adjustments have been made.
This project will look at this same issue, but will use a different methodological approach. Rather than assigning children to high and low ability groups according to a baseline measure, we account for regression to the mean by using all available measurements of language skills in the Millennium Cohort Study to define latent classes of developmental trajectory – a growth mixture modelling approach. We then examine whether the probability of a child belonging to a particular trajectory group depends on SES by regression of the trajectory classes on social class indicators and covariates. The approach also allows us to include children with missing responses on the language ability measures, under a missing at random assumption, while previous research has been based on children with a full set of measurements.
Following Jerrim and Vignoles we provisionally intend to measure children’s cognitive ability using the British Ability Scale regarding language skills conducted at the age of 3, 5 and 7 in the Millennium Cohort Study. However, we will also undertake a comprehensive review of other potentially suitable data sets during the initial phase of the project.
The first source of regression to the mean that Jerrim and Vignoles identify is selection, which is caused by measurement error. In classifying children as high or low ability based on a single test score, by chance some children will have performed abnormally high and others will have performed abnormally low. As such, their performance in this test cannot be considered a measure of their true ability. The second source of regression to the mean is non-comparability in tests of ability over time. If the tests are assessing different skills, this will weaken the correlation between test scores over time, because those children who perform well at verbal tasks (for example) will not necessarily be the same children who perform well at mathematical tasks. Jerrim and Vignoles carried out their own analysis using both the Millennium Cohort Study and the Avon Longitudinal Study of Parents and Children. They adjusted for regression to the mean by classifying children into high and low ability groups based on their scores in a number of auxiliary tests and by using a different measure (of the same ability) to assess development over time. They show that the same pattern of change in high-ability low-SES children and low-ability high-SES children in earlier research is not evident once these adjustments have been made.
This project will look at this same issue, but will use a different methodological approach. Rather than assigning children to high and low ability groups according to a baseline measure, we account for regression to the mean by using all available measurements of language skills in the Millennium Cohort Study to define latent classes of developmental trajectory – a growth mixture modelling approach. We then examine whether the probability of a child belonging to a particular trajectory group depends on SES by regression of the trajectory classes on social class indicators and covariates. The approach also allows us to include children with missing responses on the language ability measures, under a missing at random assumption, while previous research has been based on children with a full set of measurements.
Following Jerrim and Vignoles we provisionally intend to measure children’s cognitive ability using the British Ability Scale regarding language skills conducted at the age of 3, 5 and 7 in the Millennium Cohort Study. However, we will also undertake a comprehensive review of other potentially suitable data sets during the initial phase of the project.
Staff
Lead researchers
Collaborating research institutes, centres and groups
Research outputs
Katy Sindall, Patrick Sturgis, Fiona Steele, George Leckie & Robert French,
2019, Longitudinal and Life Course Studies, 10(3), 283-305
Type: article