Experimental Design Details
The study design aims to conform to CONSORT guidelines for reporting of RCTs. A two arm trial with randomisation at the level of the setting (cluster randomised) will provide a generalizable account of the efficacy of the intervention for all children in the final year of nursery (ages 3 to 4) in at risk populations in comparison to standard teaching practices (BAU). We will work in two areas in the country (Greater London and Teeside) with schools who postcode falls into the most deprived quintile based on the IDACI. The locations have been identified to test applicability and aid generalisability, which will enhance the robustness of the findings. All baseline measures (children, staff and setting) will be collected prior to randomisation and again following the completion of the training. There will be full reporting of selection, allocation and attrition of participants and settings.
Identification of classes and sampling: Within the two target areas, nursery schools and primary schools with nursery classes with the highest numbers of children at risk, as defined by the Income Deprivation Affecting Children Index (IDACI), will be identified; over two-thirds of disadvantaged 3 and 4 year-olds are in the maintained sector. Schools will be approached to participate, and the rationale and study design outlined. For each setting, we will collect setting details (including proportion of children eligible for pupil premium, previous OFSTED reports), local area initiatives, provision of school based speech and language therapy and child data (including EAL and SEN status).
Background information about individual staff will be collected using validated questionnaires including factors that may act as moderators or mediators of change (e.g. years of experience, training self-efficacy, openness to change, knowledge about language-supporting strategies) to explore their impact, (Ottley et al., 2015). Schools will be asked to distribute project packs to parents of all the children in the target age range (i.e. three to four years).
All data from quantitative measures will be initially screened for missing values, univariate outliers, skewness and kurtosis. Missing values analyses will be conducted to determine departures from randomness. Conventional linear transformations will be used to normalize distributions. Where outliers remain, we shall consider modelling outcomes with different link functions or replacing their values by less extreme scores.
Each pupil measure administered at initial and final assessment is likely to vary to some extent with the chronological age of the child and with time in the school year when the child was assessed. For this reason factor variables will be standardized residuals obtained by regressing each separately against chronological age and time in school year.
Maximising retention will involve ensuring that we continue to communicate with all schools throughout the trial period to ensure they do not lose interest or contact with us. We will provide a named contact to all schools should they have any queries at any point throughout the evaluation. We will minimise the burden on schools, especially those allocated to the BAU group. We will clearly communicate intervention schedules and data collection schedules, while at the same time being as flexible as possible to accommodate the schools’ needs. Testing at the end of the trial period will be necessary, but we will minimise the work that this involves for schools. Outcome data will still be collected from participants who discontinue or deviate from the intervention using intention-to-treat (ITT) analysis.
Child level data will be anonymised with a random ID number given to participants and parent questionnaires. Participant name and ID will be kept separately in a locked cabinet, in order to allow for data linkage at the post-test. Teachers will also have their own unique identifier for completing the online questionnaire (hosted on REDCap) which will allow the data to be linked pre- and post-test. All researchers on the project, including temporary RAs will complete Information Governance Assurance training. RAs will first complete scoring of child level assessments by hand and then enter it into excel spreadsheet at each site (with only the random ID number as identifier). Child level data will be entered weekly, with data validation checks in place (e.g. range checks, look-up table for acceptable data, presence check). Data from the Teesside sites will be sent to UCL in encrypted emails and merged with data from the Greater London sites. The merged data will be stored securely on the UCL Data Safe Haven, which complies with the ISO27001 information security standard and NHS Digital's Data Security and Protection Toolkit. An initial verification check of the baseline data will be completed by December 2021 by one of the research team.
1. Schools will not be considered fully recruited for the purposes of the impact evaluation until all required pre-trial data have been collected from schools, consent procedures have been carried out, and pre-testing has been carried out.
2. The primary outcome will be specified as an intention to treat (ITT) analysis of the baseline language measures. We will also explore the impact on compliers using a complier average causal effect (CACE) analysis (Angrist et al., 1996) based on an agreed definition of compliance to be included in the trial protocol.
3. Other available school- and pupil-level characteristics between the treatment and control groups; pupil-level covariates will also be used to improve precision in the analysis model, while, given the clustered nature of the trial, we will also include school-level fixed effects in the model and calculate cluster-robust standard errors at the school level.
4.The analysis model will be fully specified in the statistical analysis plan and changes will not be made once outcomes data have been collected.
5. Planned pupil-level data would include, for example, non-verbal ability, gender, whether English is an Additional Language. This data will be used for exploratory sub-group analyses to explore whether there is evidence of child (language level, SEN status, gender, EAL) or class factors (such as staff qualifications), which limit response to the intervention.
6. Fidelity checks for the intervention will occur in the spring term 2022 (i.e. during the intervention phase of the study) for each of the activities.
7.Missing data will be investigated to determine any non- random nature.
All data will be anonymised and the key kept in a locked storage base at the university separate to all data files. A list of children scheduled for assessment will be left at the school reception at the end of each assessment session, therefore no identifiable information will leave the premises. A data safe haven will be used to store the electronic raw data and conduct analyses securely. Data will be merged from the two sites using encrypted digital files within password protected folders. Only the necessary individuals will have access to the demographic data containing names and dates of birth (JD, JL, SM, JC, CF, KD). No names of children, teachers or schools will be published in the research findings.
Reports to the funder will be provided and all researchers acknowledged in the report. Summary reports will be sent to all schools in the summer term following the intervention. Publications in peer review journals will occur with the lead author taking the major responsibility for creating the first draft, including all PIs as authors. A conference for all participating schools will be held in September 2022, in conjunction with the UCL Centre for Inclusive Education with a parallel meeting in the north east. This will provide an opportunity to share research findings with the schools and for schools to share their experiences of the intervention, helping to shape best practice in future studies.