CS School-Wide Curriculum Program Evaluation
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Fully de-identified dataset for the longitudinal program evaluation project of CharacterStrong SEL Curriculum. Please see to the file "Introduction and terms of CS de-identified data repository" Notes on De-identification of Data. The following rules provide an operational checklist for the de-identified school mental health program evaluation data before repository release. Institutions may adapt thresholds to local policy, legal guidance, and data context, but the governing principle remains the same: the released file should not permit a recipient investigator, or a reasonable person in the relevant school community, to identify an individual with reasonable certainty. A. Direct identifiers: remove completely • Remove all names of students, parents, guardians, siblings, teachers, clinicians, administrators, and other individuals. • Remove institutional identification numbers and usernames, including student IDs, employee IDs, login names, email addresses, and analogous identifiers. • Remove all direct contact information, including street addresses, mailing addresses, phone numbers, fax numbers, and social media handles. • Remove medical record numbers, health plan numbers, insurance numbers, account numbers, license numbers, and similar unique personal identifiers. • Remove biometric identifiers and images that could support automated or human recognition, including facial images, fingerprints, retina/iris scans, voiceprints, and identifiable signatures unless a separate review justifies retention in a non-human-subjects context. B. Dates and ages: generalize or transform • Remove exact dates directly related to an individual when those dates materially increase re-identification risk. Replace with broader units such as month, quarter, semester, school year, or relative time from baseline as analytically appropriate. • Convert exact birth dates to age bands or year of birth only when necessary and when year-level retention does not create small-cell or uniqueness risk. • Top-code very old ages or other extreme values that are rare enough to support recognition. • Review event sequences across time. Even if each date field is generalized, an uncommon longitudinal pattern may still be identifying and may require additional masking. C. Geography and institutions: reduce specificity • Remove or generalize street-level location, building room numbers, bus routes, exact clinic sites, and precise geocodes. • Replace school names, classroom identifiers, provider identifiers, and district identifiers with broader categories or randomly assigned repository labels if retention is essential for analysis and does not create contextual identifiability. • Avoid retaining combinations such as school + grade + race/ethnicity + service type when those combinations are rare in the local setting. D. Indirect or quasi-identifiers: assess in combination, not one variable at a time • Review demographic and programmatic variables jointly, including age, grade, race/ethnicity, gender, disability status, language status, housing status, service tier, attendance pattern, disciplinary involvement, and referral source. • Collapse sparse categories when rare combinations could permit recognition. The appropriate collapsing rule should be driven by context, not by a single universal threshold. • Treat contextual facts as quasi-identifiers when they are unusual in a local school community, such as a rare diagnosis-service combination, a unique extracurricular role, or a highly distinctive service history. E. Free text and narratives: remove or manually redact • Do not release open-ended comments, counselor notes, teacher narratives, incident descriptions, referral narratives, or other free-text fields unless every record has been manually reviewed and redacted. • Redact not only names, but also locations, dates, uncommon events, family relationships, program titles, and narrative details that could permit recognition even without a name. F. Small cells, rare categories, and uniqueness: apply disclosure controls • Suppress or collapse small cells in released outputs and, where necessary, in released analytic files. Small-cell policy should be based on institutional rules and contextual risk rather than a purely mechanical threshold. • Use complementary suppression or category coarsening where a suppressed cell could otherwise be reconstructed by subtraction. • Review outliers and unique patterns, not only cell counts. A single unusual multivariable pattern may remain identifying even when no individual table cell is small. G. Longitudinal linkage: use repository-specific random IDs only • If within-file longitudinal linkage is necessary, generate a new random study ID with no relationship to the original source identifier. • Do not release the original coded identifier or any crosswalk that would reconnect the released file to source records. H. Linked data and merger risk: assess proactively • Before release, ask whether the file could be linked with reasonably available public or institutional information in a way that would reveal identity. • Pay particular attention to local news events, public award lists, school rosters, publicly posted team memberships, and other information that may make a record unique in context. I. Residual-risk review before release • Conduct a final review of the complete released file, not just individual variables, using the question: could a recipient investigator, or a reasonable person in the relevant school community, identify an individual with reasonable certainty using this file alone or in combination with reasonably available information? • If the answer is yes or possibly yes, perform additional recoding, suppression, redaction, or aggregation before release. J. Output rules for downstream users • Require downstream users to suppress small cells and avoid disclosive descriptive examples, quotations, or case summaries in presentations, manuscripts, and reports. • Prohibit release of model diagnostics, influence statistics, predicted probabilities, or subgroup estimates when they could expose an individual or tiny subgroup. • Require a final disclosure review of outputs before public dissemination when the dataset contains sensitive student mental health information. Implementation note These rules are intended to support a de-identified repository model in which downstream investigators receive only de-identified analytic files, do not receive a code key, and are prohibited from subject contact or re-identification. Institutions should align the final rule set with their own privacy, FERPA, HIPAA, information security, records management, and human-subjects review requirements.
Fields of Study & Subjects
Early Childhood Education and Development
Evaluation and Performance Assessment
Values and Moral Education
Publication details
- Publication date
- 1 January 2026
- Language
- EN
- Document type
- Preprint