Pre-registration of treatment effect Data exclusions - We intend to include students in the ITT analysis as long as they saw a page in the treatment/control, signifying that they were randomly assigned - For grades effects, we do not consider students who are on IEPs or who are in special education to be comparable to other students, when their grades are determined using different criteria. And so we expect to exclude them from analyses of grades. - We intend to include students regardless of their English language learning status. We will carry out additional treatment heterogeneity analyses (ATT) using the following criteria: - Excluding those who saw treatment content but wrote nothing (or nothing of substance, as coded by third-party coders blind to condition) - Excluding those who received session 1 but not session 2 - Excluding those who are above a very high threshold for saying they were distracted or could not pay attention, on the fidelity measures - Or other measures determined by theory Covariates We intend for all regression analyses of primary outcomes of interest to control for, at a minimum: - prior performance - a fixed effect for school If it turns out that a given variable was, by chance, significantly different at baseline, we expect to control for prior values of that as well. We will examine whether controlling for the following reduces standard errors on the treatment effect: - Prior mindset (composite) In specific analyses of math, we will examine whether the following covariates also reduce standard errors: - Math anxiety - Math expectancies - Math interest Primary outcomes of interest: - GPA in core classes (science, math, English, social studies) - GPA in STEM classes (math or science) - GPA in math classes - Poor performance rates in core classes, STEM classes, or math classes (1 = A, B, or C; 0 = D or lower). - A stress composite H1: We expect that there will be an ITT effect on GPA in core classes, STEM classes, and Math classes, and on poor performance rates in each class. We have the strongest hypothesis about STEM and Math, as compared to English and Social Studies. We also expect that stress will be lower in the treatment condition (although that is an exploratory hypothesis that has not previously been shown). Secondary outcomes of interest H2: We expect that growth mindset composite scores (manipulation checks) will be significantly more changed in the treatment group as compared to the control group, in an auto-regressive model predicting T2 values with T1 values. H3: We expect that sense of purpose composite scores (manipulation checks) will be significantly more changed in the treatment group as compared to the control group, in an auto-regressive model predicting T2 values with T1 values. In prior studies these did not change significantly and so we expect weaker, less reliable effects on these scores. Moderation by prior performance - Prior performance is defined as 8th grade grades, - or as a latent variable of 8th grade grades and test scores and self-reported prior performance (on the session 1 survey), estimated in SEM using FIML o Each of these prior performance variables could be defined in terms of the course-specific grade (e.g., prior math performance moderating treatment effects on freshman math outcomes), or in terms of all core courses (the average of prior grades) - “Lower achievers” can be defined as the bottom third of performers in the full sample, or as a prior GPA under 2.0. H4: We expect that the ITT effect on the outcomes above will be moderated by prior performance, with a significant simple effect among lower-achievers, and a smaller or even non-significant simple effect for higher achievers. Issues related to prior performance are likely to arise when attempting to distinguish between school/context effects vs. individual difference. We will conduct exploratory analyses of whether the within-school prior performance distribution (i.e., lower performers within the school) or the between-school prior performance distribution (i.e., lower performers in the sample as a whole) is more predictive as a moderator. Moderation by group membership We expect to test for moderation by race/ethnicity and gender. H5: We expect larger ITT effects on grades for negatively-stereotyped students (black/African-American or Hispanic/Latino or native american) as compared to less-stereotyped students (white or Asian). H6: We expect larger ITT effects for girls in quantitative classes (math and science) than for boys in those same classes, but do not expect gender differences in English/social studies, for stereotype threat reasons. We will carry out exploratory analyses of moderation by other individual difference characteristics.