Improving STEM Outcomes for Young Children With Language Learning Disabilities by Intervening at the Intersection of Language and Scientific Thought: A Telehealth RCT

63 4- to 7-year-olds who have not yet begun 1st grade, who are monolingual speakers of English, and who have SLI will participate. Note that the investigators may recruit extra participants to allow for attrition. The investigators will adopt a Randomized Controlled Trial design, randomly assigning participants into three intervention conditions: science only (the control arm), science + vocabulary supports, and science + grammar supports. Pre- and post-measures will reveal the extent of learning in each condition and comparisons between conditions will reveal whether the grammar and vocabulary supports improved learning.

The hypothesis is that the language and learning of science are integrally related. Therefore, the investigators will use evidenced-based language interventions to improve the children's science-relevant language skills, with the prediction that this will cascade into changes in the acquisition of science concepts and practices:

1. Children in the science + language intervention conditions will show greater gains in taught science concepts across the 6-week intervention period than children in the control arm.
2. Children in the science + language intervention conditions will show greater gains from pretest to posttest on measures of generalized science concepts and practice than children in the control arm.
3. Children who demonstrate the greatest improvement in the use of the language targets will also demonstrate the greatest improvements in taught concepts, generalized concepts, and generalized practice knowledge.
4. Children will benefit from language supports directed at vocabulary as well as those directed at grammar, but these supports may differently benefit the science learning process.

First the investigators will document that the language supported interventions resulted in improved language abilities by comparing performance on probes of grammar and vocabulary at posttest to pretest performance. The investigators expect significant improvements in vocabulary knowledge for the vocabulary intervention condition as compared to the other two conditions, and significant improvements in use of complement clauses for the grammar intervention condition as compared to the other two conditions. Next, to be tested are the predictions associated with the specific aims via a series of logistic mixed models. Mixed models are appropriate for designs with unbalanced cell sizes due to missing data (due to non-response and dropout). There will be one model for targeted science concept outcomes with condition (control arm, science + vocabulary, science + grammar) and time as independent variables (Predictions 1 and 4). There will also be one model each for generalized concepts and generalized practice outcomes with condition (control arm, science + vocabulary, science + grammar) and time (pretest and posttest) as independent variables (Predictions 2 and 4). Within-subject correlation will be accounted for with random subject intercepts. Additional random effects (including random item intercepts or random condition slopes by item) will be determined by selecting the model with the best model fit (lowest AIC value). In each of the models, it is further expected that amount of improvement in grammar and vocabulary are mediators between the outcome and the other factors (Prediction 3). To assess this prediction, performance on the language probes will be considered as covariates. It is expected that performance on the language probes after instruction will be a significant predictor of science learning, and that including performance on the language probes as a covariate will reduce or eliminate the effect of condition because language performance will be the main factor predicting science performance.