Our results suggested that teachers’ expectation of their pupil’s achievement was generally accurate at two Key Stages of UK education (ages 11 and 14). We found evidence that teacher expectation accuracy was related to some socioeconomic or demographic factors, principally pupil’s maternal education and age in year at both Key Stages, and parental social class and household income at Key Stage 2 and 3 respectively. This patterning was consistently in the same direction, whereby pupil’s from more disadvantaged backgrounds underperformed compared to their advantaged peers. Our findings conform to those from previous studies which have found differential teacher expectation accuracy towards certain groups of pupils such as those from lower socioeconomic position backgrounds 3,6,10,16,27,28,36−38. For example, disparity was found between teacher assessed measures and Foundation Stage Profile assessment on socioeconomic and demographic factors including income, gender, special educational needs status and ethnicity 4. We found little evidence that teacher gender, teacher experience or class size were associated with the accuracy of teacher expectations. This contrasts to previous research that has observed strong associations between these factors and the accuracy of teacher expectations 16,25.
We found mixed evidence for associations between genetic factors and teacher expectation accuracy. Using a polygenic score from a large GWAS of educational attainment 32, we found strong evidence that pupils with higher polygenic scores were more likely to outperform their teachers’ expectations compared to children with lower polygenic scores. Using all genomewide data within a GREML-GCTA framework we found only weak evidence for SNP heritability at both ages, though these results should be interpreted with caution due to their imprecision. These results suggest that some of the variation in teacher reporting accuracy can be explained by genetic variation at the pupil level. Conversely, they suggest that the overwhelming majority of variation in teacher expectation accuracy can be explained by non-genetic (environmental and residual) factors. Genetic liability towards educational performance could operate through a range of mediating mechanisms, such as personality characteristics or attitudes to learning and schoolwork 16,36. In this way, ‘invisible’ genetic variation may become visible to teachers, influencing their expectations of a pupil’s future performance. Future studies with larger sample sizes are required to verify these findings, however our results build upon previous studies demonstrating robust associations between genetic factors and achievement throughout schooling 31,39,40.
Our analyses were unable to determine whether inaccuracies in teacher expectation were due to error or detrimental bias on the teachers part, i.e. whether teachers were prejudiced against specific groups of pupils 2,41. Prejudice on the teachers part however may have been more likely to result in pupils over performing their teachers (unfairly negative) expectations. Similarly, pupil behavioural change due to self-fulfilling prophecies from inaccurate teacher forecasts may be expected to result in accurate teacher expectations, even if they are prejudiced 42. Regardless, our findings highlight that some groups of students systematically underperform their teachers’ expectations. We were unable to reliably investigate how ethnicity related to teacher expectations due to the ethnic homogeneity of the ALSPAC cohort and the low numbers of ethnic minority participants. Previous studies conducted on more ethnically representative cohorts have shown that teacher expectations differ by pupil ethnicity 6,18,30.
Several limitations exist with this study. First, generalisability of these findings to the wider UK school population may be limited. Our sample was ethnically homogenous and restricted to those who were recruited from a single geographical area over a three school-year period. This tightly defined sampling frame means that there will likely be reduced environmental and genetic variation in our sample compared to the broader population of the UK. Furthermore, within ALSPAC there is greater attrition for pupils from families of lower socioeconomic position and poorer general health 43, meaning that this demographic are underrepresented and our complete case analyses may be biased. Results from our multiple imputation analyses were broadly consistent with the complete case analyses, suggesting that bias due to attrition may be limited.
Second, the age of the sample may also limit the generalisability of our findings. We examined the accuracy of teacher expectations at ages 11 and 14, but our results may not be transportable to earlier of later stages of education. Additionally, the participants were educated between 2001 and 2006 and teacher expectations may have change since this period.
Third, the accuracy of teacher expectations may have differed across Maths, English and Science subjects as many participants will have had subject specific teachers. Our decision to combine across these subjects was taken to provide a more accurate measure of the pupil’s overall academic performance and reflect any general teacher expectation bias. Furthermore, teacher expectations of pupil achievement were provided as categorical levels for each subject, meaning that there was reduced variation in this measure when compared to point scores used for assessment.
Fourth, many variables were subject to potential measurement error. For example, family socioeconomic position is a complex construct encompassing education, income, wealth and other factors, yet we were only able to proxy this using weekly household income, parental social class and the highest level of maternal education 44,45. To improve statistical power, we leveraged the larger sample of responses from study mother’s reports of their partners occupation and income, but it is likely that the mother reports will have contained greater measurement error than direct partner reports.
Finally, because of the ethnic homogeneity of the ALSPAC sample, the European-centric focus of genetic studies, and the need to exclude non-Europeans due to systematic ancestry differences arising from population stratification (which can induce spurious genotype-phenotype associations) 46–48, we were only able to perform the analyses amongst white participants of European ancestry. Previous studies have demonstrated that the direction of teacher expectation accuracy varies by ethnicity 4. However, previous work has demonstrated that trait-associated genetic markers do not perform well across ancestral groups 49. Larger genotyped samples of ethnic minorities are therefore required to explore this issue further.
In conclusion, this study investigated potential patterns of teacher expectation accuracy by socioeconomic, demographic and genetic factors. We found evidence of systematic socioeconomic and genetic patterning in teacher expectations. Pupils from more disadvantaged backgrounds underperformed their teachers’ expectations compared to their more advantaged peers, and those with higher genetic liability for educational attainment outperformed their teachers’ expectations relative to pupils with lower genetic liability for educational attainment.