[AUDIO LOGO] We are the Association for Child and Adolescent Mental Health, or ACAMH for short. And this is ACAMH Learn. Welcome to Mind the Kids, a podcast series dedicated to exploring the latest advancements in child and adolescent mental health research and practise. I'm Clara, an Academic Clinical Fellow in Child Psychiatry. And I'm really passionate about understanding and addressing the diverse mental health challenges that are faced by young people globally. In this series, I will be joined by renowned researchers and clinicians from around the world to discuss their cutting edge research, their innovative interventions, and best practise in child and adolescent mental health. And today, I have the pleasure of receiving Dr. Emma Horton and Dr. Daniel Matthews from the University of Manchester and the University of Sheffield, respectively. We're going to be discussing their paper-- unequal educational outcomes for children with similar early childhood vocabulary capillary but different socioeconomic circumstances that was published at the GCPP. So thank you so much for agreeing to do the podcast, Emma. And if you could introduce yourselves. Hi, I'm Dr. Emma Thornton. I am currently a postdoctoral research fellow at the University of Manchester. I did my PhD at the University of Liverpool, which is where this paper came from. We also worked with Dr. Colin Bannard and Dr. Praveetha Patalay on this paper as well, so just to acknowledge them as well while we're here. I'm Daniel, I'm a Professor of Psychology at the University of Sheffield. It's one of Emma's supervisors, along with Colin and Praveetha. Thank you so much for agreeing to the podcast, Emma and Daniel. It's an honour to have you here today. And your paper approaches a fundamental topic in our society, which is basically socioeconomic inequalities and how those impact the educational attainment children are going to have later on. And I wanted to ask you, what inspired you to answer this specific research question? And we know that socioeconomic inequalities affect attainment. So what was the main gap? You were trying to address with your study. So the study was actually part of my wider PhD thesis, so that overall PhD examined the extent to which childhood vocabulary can be used to predict adolescent outcomes that have important implications for later in life. So we looked at mental health, and in this paper, academic attainment, and importantly, the role of socioeconomic status in those relationships. So as you've mentioned, we know that there are socioeconomic inequalities in educational attainment. And it's also well-known that there are socioeconomic inequalities in childhood vocabulary. And there have been efforts to develop interventions to promote vocabulary prior to school entry. So an overall aim of my thesis for this paper was to provide insight into whether those interventions that have been developed are likely to benefit children in the longer term. So the specific gaps for this paper included whether or not vocabulary predicts educational attainment at the end of secondary school, beyond those demographic inequalities, and whether or not there were long-lasting impacts of childhood inequalities in vocabulary, sort of beyond whether or not the individual differences have an impact later in life, and whether or not they are felt more by some groups than others. So the question that kind of guided this paper was, do children with the same level of cognitive ability, which we measured using vocabulary, achieved the same academic outcomes regardless of their socioeconomic position? Thank you so much for that overview, Emma. And I think, as you said, you were trying to see if children with the same cognitive ability would achieve the same outcomes. And in your introduction, you talk a bit about the discussion around that. And I love that you referenced the book of Irony of Merit by Michael Sandel. I'm reading another book of his, Justice. And this author, Michael Sandel, he challenges this notion of meritocracy that it is everybody is able to achieve the same things if given the same resources. And you do make the point in your introduction that even though it's widely recognised that we do not live in a meritocratic society because there are lots of inequalities, lots of available early childhood interventions are actually underpinned by this logic of if we give the same intervention to all children, they are going to be able to achieve the same thing. And I wanted to ask to you and also to Dr. Matthews, to Daniel, why do you think that is? Why do you think so many childhood interventions are based on that? Yeah, it's curious, probably because some people like the meritocracy story maybe. It's something that people feel it gives a sense of control. But perhaps a better way of looking at it is that early interventions seek to help children by providing them with skills that help them to thrive. And when studies like the one this one of Emma's come along and show that that's likely to be more successful for some groups than for them for others, then we want to do something to help, make sure that all children benefit from education. But of course, some children will struggle in school regardless, perhaps because of SEND, so Special Educational Needs and Disabilities, or other reasons. And then a different kind of support is needed there. So I think you can recognise the tyranny of merit at the same time as wanting to support early interventions and try and have a holistic programme of support for children. [INAUDIBLE] Absolutely. And regarding this notion of meritocracy, can you tell us a bit about how your paper approached this notion in your research questions? Yeah, so I'm not sure whether meritocracy is the right term, but we looked at the relationship between early language ability as a marker of cognitive ability generally, so early vocabulary as a proxy for general cognitive ability. And you would expect that children with strong cognitive ability arriving at school, so large vocabulary, you'd expect them to do well in school. And if that's not the case, and if it's patterned by socioeconomic circumstances, then that tells us that there's something going on that we might want to address. So we used early vocabulary size as a measure of cognitive ability. And then we controlled for caregiver vocabulary size. So that's kind of a unique feature of this study. And of this really, really large data set, tens of thousands of children. There was also a researcher collected measure of parent language ability. And that allows us to control for that as a sort of proxy for the early caregiving environment and genetic factors in the family. It's not a perfect control, but it's a pretty good control. What we were expecting to see is that as children's cognitive ability increases at age five, so their chance of gaining benchmark qualifications at age 16 should increase. And then we were testing whether that relation is moderated by socioeconomic circumstances. Yeah, and I was going to say that, in your paper found this relationship right between language abilities at age five, and later, educational attainment at 16, even after adjusting for socioeconomic circumstances. And one of the things I found really wonderful as well is that in your second analysis, you tested if this predictive value was equal across different socioeconomic groups. So this is a bit what we were talking about before, where does it matter which socioeconomic group you belong to in terms of the predictive value of your language abilities at an early age in your educational attainment. And I was just wondering if you could tell us a little bit more about the results you found when you tested if different socioeconomic groups made a difference. Yeah, so-- the short answer is yes, we did find that they made a difference. So we used the benchmark qualifications, which were GCSEs. So we looked at core subjects of English, science, and maths, and whether or not they achieved grade 4 or C and above as sort of a benchmark. For those groups in the middle socioeconomic bands, yeah, we found that the larger the early vocabulary were for them, the more likely they were to attain those qualifications. For the lowest socioeconomic groups to the most disadvantaged children, we found that the relationship was actually weaker. So even children with the highest language skills, age five, they only had a 50/50 chance of attaining those benchmark qualifications, even though they had some of the strongest vocabulary capillary scales. And on the flip side of that, again, amongst the most advantaged children, the relationship was weaker again. Those children were likely to reach the benchmark qualifications regardless of what their early vocabulary was. So even those with very low language scores at age five were still quite likely to obtain GCSEs. The procedure fascinating result honestly, especially because-- and I was trying to make sense of it in terms of like, I'm a clinician, so in terms of what we see in the clinic. And I suppose when we see children with special educational needs, for example, I know that was not the focus of the paper, but I was thinking in terms of oh, a child that would have lower vocabulary, for example, sometimes as a child that also might have special educational needs. And I suppose, if you are from a higher socioeconomic strata and your child, your parents are much more likely to add-- because they have more resources, so they are more likely to be able to go to the right sources of support. They are more likely also to be able to pay out of pocket for things like speech and language therapy. I know the paper was not focused on children with SEND. But I was just thinking in terms of trying to make reasoning with myself why that would be. And I suppose we see as well in the clinic with children that have lower vocabulary at early ages. But, yeah, I thought that was such a fascinating result. Yeah, that actually, when you look at the graph in Emma's paper, what you see is that when you plot early vocabulary ability against likelihood of benchmark qualifications at age 16, you see a very steep gradient for children in the middle SEC band. So we split children into quintiles, so five different groups-- most disadvantaged at the bottom, most advantage at the top, and then three middle bands. And the three middle bands show the strongest relationship. The larger your vocabulary at age five, the higher the chances that you're going to get these benchmark qualifications. But that relationship is dampened both in the bottom group. So if you're experiencing socioeconomic disadvantage, even if with very strong early vocabulary, you only have a 50/50 chance of obtaining these benchmarked qualifications. But equally, that relationship is dampened for the most advantaged group. So there, even with very low vocabulary at the outset, you're still quite likely to get those GCSEs, which speaks to what you're saying about parents with the most resource being able to presumably put in the support to-- and possibly also valuing those academic qualifications. So we see that the relationship is attenuated at both extremes. Absolutely. And yeah, I love Dr. Matthews that you mentioned the 50/50 result for children in the most disadvantaged groups because-- just as you said, like I think it's important to acknowledge the reality while at the same time trying to use these results to confirm public policy and early intervention. But it is just really sad and really stark to realise that if you are a child from a lower socioeconomic background, even if you have good vocabulary skills at age five, you still only have a 50/50 chance to achieve good GCSE qualifications at age 16 no matter what. Can you guys help me make sense of these results? I wonder what happens along the way that this vocabulary doesn't make a difference in the same way that it makes for children in the middle group, for example? Yeah, it's a really striking finding. And I think we were all surprised by it, even though obviously we had preregistered this study, we were interested to look at this, like when you see the plot, the extent of the differences is quite striking, and all the more so given that it's a very large and nationally representative sample of UK children. So I think we've all seen SEC effect, Socioeconomic Circumstance effects, before. But when you have such a large data set of tens of thousands of children very carefully put together to be representative of the country, then it makes it especially robust test of this question. And then the answer is all the more striking. Of course, we don't really know what explains this moderating effect of socioeconomic circumstances. So several pathways exist in terms of ability to invest in education and different experiences of different education pathways. And that's really the burning question. We don't really know what drives that. And probably a good way of finding that out would be to go back and talk to families themselves and to try and do some qualitative work around that to better understand. It's not likely to be any single factor. And indeed, some of Emma's earlier work suggests that there's a kind of additive factors of different types of socioeconomic disadvantage, like caregiver education, family income, neighbourhood, and so on. So it's probably many, many things all adding up rather than one single explanation. You've mentioned that you guys used a very robust sample. You used the Millennium Cohort Study, which, yeah, I think is fantastic because for our listeners who don't know, the Millennium Cohort Study is a nationally representative UK sample. So it includes all four nations. You did a lot of subgroup analysis in your paper, including comparisons among the UK four nations. And I just wanted to ask if any of the results when you're doing this subgroup analysis surprise you. I mean, so-- like you say, we did quite a lot of different subgroup and sensitivity analyses. And I think the striking thing across those was that the findings were generally consistent. So no matter how we looked at the question, how we considered educational outcomes, the results were the same patterns of findings. In terms of the analysis that we did on the four nations of the UK, it's difficult to make any firm conclusions about Wales, Scotland, and Northern Ireland, purely because the sample size of those countries is so much smaller. The vast majority of the analytical sample and the Millennium Cohort Study is England just because England's bigger anyway. So that analysis likely was underpowered. But we saw a similar pattern emerging. I think a more interesting finding that came out of our sensitivity subgroup type analyses is that, and I think Danielle's touched on this a little bit, but we found that when we looked at the individual indicators of socioeconomic status, so our main analysis was a sort of combination, a combined score across parent education, household income, occupational status, where wealth and relative neighbourhood deprivation. But when we looked at those indicators separately, we found that it was caregiver education, household income, and occupational status that were driving the associations. We didn't see anything for wealth or neighbourhood deprivation, which I think kind of again, goes to what you were just discussing in terms of those possible pathways of how things might affect the relationship. I guess there's more proximal socioeconomic factors that are directly impacting within the caregiving learning environment, rather than the nondisposable wealth and what's going on in the wider neighbourhood. So that was interesting to us as well. And in your discussion section, if that's OK, just because you mentioned the subgroup analysis, I was wondering-- yeah, I think it's a study that has so many strengths, like you did so many subgroup analysis. You used very sophisticated imputation techniques. I read you did like 27 imputation models to be able to account for missing data. I think it's such a huge, robust piece of work. And the one thing I found interesting is that for educational outcomes-- I mean, lots of things are interesting, but one thing I found interesting is that for educational outcomes, it's self-reported. But you also did some comparisons with the National Pupil Database in terms of how the self-report correlates with the agreement in the National Pupil Database. And it's good agreement, isn't it? So we didn't do direct comparisons with our findings and the National Pupil Database. And original plan had been to actually use National Pupil Database data generally because that was available for the Millennium Cohort. But the timing of planning this paper and doing the bulk of the analysis was during the pandemic. So that threw up a lot of complications in terms of being able to access that data. And so that's why we went with the self-report data instead. The reported percentages of agreement between self-report and official statistics that we've quoted in the discussion are based on other papers that have done those comparisons, but agreement is fairly high. I can't remember off the top of my head, but I'm pretty sure those papers found 97% or so agreement. And I think although we would have liked to have used the objective statistics in this analysis, that would have meant a smaller sample, and we wouldn't have been able to glimpse an insight into the other UK nations, because that data is only available for children in England. But yeah. So I would like to think that if we did have the National Pupil Database data available to us for the study, then we would have found very similar patterns because in. And it's in multiple different studies as well that the agreement has been found to be quite high between self-report and national statistics. And yeah. And as you said, I think you would have missed this element of the four nations, which is so valuable in your study because I think a lot of the study, like large epidemiological studies, we have, including linkages because of the limitations of one of the databases. Usually, it's just focusing on England. So yeah, I think it's so wonderful that you were able to include all the UK four nations, even if Northern Ireland, Wales, and Scotland have smaller sample sizes. It's really valuable for in terms of public policy and implications. And speaking of public policy and implications, so your study was the first to test these moderating effects of socioeconomic circumstances in the relationships between language ability and educational attainment. And you did that, as we said in this nationally representative UK sample. And the other thing that I don't know if it's obvious to the audience, but in my study-- so the Millennium Cohort Study is a population-based sample. It's incredible really, because we have an electronic healthcare records databases that are quite large, but that's a clinical population. So it's really wonderful that you were able to do that in a general population sample, which is again has many more implications for public policy. And thinking about real-world implications of your work, yeah, what do you consider? Are your paper's main implications for schools and also public policy-wise? Yeah, I think the main implication is that despite the pupil premium, so that's about 1,000 pounds per disadvantaged pupil per year given to state schools in England to improve educational outcomes for disadvantaged children, despite that pupil premium, we still have a really long way to go to help children realise their cognitive potential. And perhaps even that funding isn't reaching its intended beneficiaries. We need to see better support in schools and better support for families. Obviously, that's easier said than done. But I know some schools that I work in really struggle at the moment, often don't have a teaching assistant, lots of children with lots of different needs. And it's very, very hard. It's a very difficult environment to teach in, so probably that would be our number one goal. We also, in the future, have the potential to harness technology to do better. I have a colleague, Sabrina Burr, who works in maths and how you can individualise children's learning environment according to the particular misconceptions they have, for example, when they're learning about a fraction or something. So there is huge potential in technology. It doesn't all have to be expensive, but probably some of this is going to cost us some money to fix this problem because we just need to better support children at school and try and help children at home as well. Obviously, different families are in different positions. And we need some evidence base on what to best do to promote school learning and to promote family support. And that has to be done in concert with families themselves. So figuring out what families want, what they would find helpful, what they would use, what they wouldn't use, and the same for teachers. So working really closely with them to figure out how we can best make a difference. Absolutely. And another thing that you just said in terms of the pupil premium and in terms of you're very consistent findings, just thinking that actually, and you did a lot of work to account for missing data. But actually, it might even be that your effect, like your estimates are an underestimate because we know that people who have lower vocabulary and people from the most disadvantaged sectors of society, they tend to drop out of longitudinal studies. So it might even be that the reality is even more-- like it's worse than what you found. So yeah, absolutely. And I think to wrap up, Daniel, thank you so much for being here and for sharing about your fantastic work. I wanted to ask you, looking ahead, if you could highlight one or more areas of research or maybe a research question that you would really like to see develop further in this field. You've mentioned working with families to find out what works best for them. Do you have any thoughts on that? I think a big area of research is sort of testing those causal pathways between what is it about the vocabulary, predicting the attainment, and then the role of the socioeconomic status coming in there. How is socioeconomic status having that impact on those different groups to fully unpack and understand why that association happens? And where is it going wrong? Because we've got these interventions in early childhood. And I think some of them have been shown to be promising for language ability at the time of the intervention. But then where is it dropping off between childhood and adolescence? And what's going wrong in there so that we can pinpoint tangible action points that we can target? And I think, as Daniel said, that needs to be done in tandem with children and their families and schools. It can't just be policymakers making those decisions because I mean, what works for one group of people won't work for everybody. I think that's the next step. I think that sums it up really well, Emma. And we're kind of excited that there are new causal inference methods available to us now that help us make better sense of these data sets, and that in concert with careful qualitative work and evaluations of interventions, which are getting better and better all the time, I think this is kind of in the field of education, a little bit behind, methodologically, the field of medicine in terms of how we evaluate interventions. But we're making progress all the time. I think it's those three strands, much better understanding of causal inference in these large data sets, randomised control trials, and all the implementation work that leads up to evaluating interventions and all of that always in tandem with working with families, teachers to interpret these findings for us and figure out what would work best for different families is really the way forward. Absolutely. And thank you so much for sharing your time today and for discussing your paper, and also for this amazing piece of work that you led Emma. And yeah, thank you so much for sharing this with us today. And thank you so much for the listen. Thanks for having us. Thank you for having us. [AUDIO LOGO]