Research Methods, Glossary and Standards

In short

Most research on social stories and video modelling uses single-case designs with one to a few children. They can show that a change happened for those children, but not how often it would happen for children in general.
Randomised trials are rare in this field, so we flag every one we find and say what it was compared with.
"Evidence-based" in national autism reviews means an approach has enough positive studies of acceptable quality, not that it works for every child or for children who are not autistic.
Our reviews are narrative evidence reviews, prepared with AI assistance, with every reference and number checked against its source, edited by a person and not peer reviewed.

How we label strength of evidence

The evidence map on the research hub uses five plain labels. They are editorial judgements that summarise our reviews, not formal GRADE ratings.

Strong Systematic reviews or meta-analyses and well-conducted randomised trials agree, with at least moderate certainty.
Moderate Consistent positive findings, but mostly from small or single-case studies, or from trials with risk of bias or low certainty.
Mixed A substantial body of studies whose results point in different directions.
Emerging Only a few small studies so far, mostly positive.
Very limited Case reports, feasibility or quality-rating studies, or a single small controlled study.
Untested No study has evaluated it.

Study designs

Systematic review: A review that searches the literature with a pre-specified, reproducible method, applies stated inclusion criteria and appraises each study. It sits near the top of most evidence hierarchies, but it is only as good as the studies it finds.
Meta-analysis: A statistical pooling of effect sizes from several studies into one average effect, usually with a confidence interval. Meta-analyses of single-case studies pool overlap indices or single-case effect sizes instead.
Scoping review and evidence and gap map: Reviews that map what has been studied (populations, interventions, outcomes, designs) without judging how well an intervention works. Useful for spotting gaps, such as how few studies involve girls or animated models.
Narrative evidence review: A structured but non-systematic summary of the evidence written by experts or editors. All Cala Luna reviews are narrative evidence reviews: we search widely and verify every source, but we do not follow a registered systematic-review protocol.
Randomised controlled trial (RCT): Participants are randomly allocated to an intervention or a comparison group, so that, apart from chance, differences in outcome can be attributed to the intervention rather than to differences between the groups. The strongest single design for testing whether something works.
Cluster randomised trial: Groups such as schools or classes, rather than individual children, are randomised. Used when an intervention is delivered to a whole setting; ASSSIST2 randomised 87 schools.
Waiting-list, usual-care and attention controls: A waiting-list control receives the intervention later; usual care continues normal support; an attention control receives a similar amount of contact without the active ingredient (for example a poem instead of a social story). Comparisons with waiting lists tend to show larger effects than comparisons with active controls.
Controlled (non-randomised) study: A study with a comparison group that was not created by random allocation. Groups may differ in ways that explain the results.
Single-case experimental design (SCED): One or a few participants are measured repeatedly, before and after an intervention is introduced. Experimental control comes from replicating the change at different times. Most research on social stories and video modelling uses these designs.
Multiple-baseline design: A single-case design in which the intervention starts at staggered times across participants, behaviours or settings. If behaviour changes only when the intervention starts, each time, that suggests a causal effect.
Withdrawal (ABAB) and alternating-treatments designs: In a withdrawal design the intervention is introduced, removed and reintroduced; in an alternating-treatments design two conditions (for example animated and human video models) are rapidly alternated to compare them within the same child.
AB design and case report: A single baseline phase followed by one intervention phase, or a description of one case. These cannot rule out other explanations and are treated as weak evidence.

Effect sizes and statistics

Effect size (d, g, SMD): A standardised measure of how big a difference is. Cohen's d and Hedges' g express a difference between groups in standard deviations; by convention about 0.2 is small, 0.5 medium and 0.8 large, though what matters in practice depends on the outcome.
Odds ratio (OR): The odds of an outcome in one group divided by the odds in another. An odds ratio of 2 means the odds were twice as high.
Confidence interval (CI): A range of values compatible with the data, usually at 95%. A wide interval means the estimate is imprecise; an interval that includes no difference (0 for a mean difference, 1 for an odds ratio) means the result is not statistically significant.
p-value and statistical significance: The probability of seeing a result at least this extreme if there were really no effect. p < 0.05 is the usual threshold. A non-significant result in a small study does not show that an intervention has no effect; it shows only that the study did not detect one.
Percentage of non-overlapping data (PND): For single-case data: the share of intervention-phase points that exceed the best baseline point (or fall below the lowest one, for behaviours meant to decrease), introduced by Scruggs and colleagues. Reviews often treat about 70% or more as a sign of an effective intervention and 50–70% as questionable. Simple, but sensitive to a single unusual baseline point and blind to trends. (Scruggs et al., 1987)
Improvement rate difference (IRD): For single-case data: the difference between the proportion of improved data points in the intervention and baseline phases, from 0 to 1.
Tau-U: A non-overlap index for single-case data that can also correct for a trend already present in the baseline. Values usually fall between −1 and 1 (the trend-corrected version can stray outside that range). Authors use different conventions for what counts as small or large, so read the label each review gives alongside the number. (Parker et al., 2011)
Log response ratio (LRR): A parametric effect size for single-case data that expresses the proportional change in a behaviour between phases, allowing studies to be combined in a meta-analysis. (Pustejovsky, 2018)
Heterogeneity: How much results vary between studies or children. High heterogeneity means an average effect hides very different results and should be read with caution.

Quality and standards

What Works Clearinghouse (WWC) standards: Criteria from the US Institute of Education Sciences for judging whether a study, including a single-case study, provides credible evidence. For single-case designs they require, among other things, at least three demonstrations of an effect at different points in time. (Kratochwill et al., 2013)
GRADE certainty of evidence: A system used by Cochrane and guideline bodies to rate confidence in a body of evidence as high, moderate, low or very low, taking account of risk of bias, inconsistency, imprecision, indirectness and publication bias. (Guyatt et al., 2008)
Risk of bias: The chance that a study's design or conduct has distorted its results, for example when the people rating outcomes know which children received the intervention.
Publication bias: Studies with positive results are more likely to be published than studies that find nothing, which can make an intervention look more effective than it is. It is especially likely for small single-case studies.
Evidence-based practice classifications: National reviews such as the US National Clearinghouse on Autism Evidence and Practice (NCAEP, 2020) and the National Standards Project (2015) classify interventions for autistic children as evidence-based or established when enough studies of acceptable quality show positive effects. A classification says that an approach has support, not how large its effect is or for whom it works.
Fidelity: Whether an intervention was delivered as intended. Many social story studies do not report whether their stories followed Carol Gray's writing criteria.
Preregistration: Publishing a study's hypotheses and analysis plan before collecting data, which guards against selective reporting.

Outcomes

Generalisation: Whether a skill learned in one setting, with one person or one set of materials, appears in others. A key question for any story or video: does what the child sees on screen carry over to the real classroom door?
Maintenance: Whether an improvement lasts after the intervention stops, usually checked weeks or months later.
Social validity: Whether parents, teachers and children consider the goals, methods and results acceptable and worthwhile.
Goal Attainment Scaling (GAS): An outcome measure in which individual goals are set for each person and progress is rated on a common scale, allowing personalised goals to be compared across a group. Used, for example, in sensory integration trials. (Kiresuk & Sherman, 1968)
Proxy report: An outcome rated by a parent or teacher rather than the child or an independent observer. Common in this field, and open to bias when raters know which children received the intervention.
Video deficit: The finding that young children, especially under 3, learn less from a screen than from the same demonstration given live. It shrinks with age.

Approaches discussed in our reviews

Social story / social narrative: A short, literal, usually personalised story that describes a situation, why people act as they do and what the child might do. "Social Stories™" is Carol Gray's trademark for her specific method; we use "social stories" for the general approach.
Video modelling: Showing a child a video of someone (an adult, a peer or an animated character) performing a skill or getting through a situation, before the child tries it.
Video self-modelling and feedforward: The child watches edited footage of themselves succeeding. Positive self-review uses existing best moments; feedforward edits together things the child can already do so that the child appears to manage a new, harder step.
Coping model and mastery model: A coping model starts out unsure or worried and works through it; a mastery model performs confidently from the start. Coping models suit children who doubt themselves.
Graded exposure and stimulus fading: Facing a feared situation in small, planned steps, starting with the easiest, often with rewards. In stimulus fading for selective mutism, new people are gradually brought into a situation where the child already speaks.
Family accommodation: Changes parents make to reduce a child's anxiety in the moment (for example speaking for the child or avoiding a situation). Understandable, but it can keep anxiety going over time; some parent programmes target it directly.
Bibliotherapy: Using books or written materials to help with emotional or behavioural difficulties, either clinically (for example a CBT workbook for parents) or developmentally (stories that normalise and explain).
Co-viewing and active mediation: An adult watching with a child and talking about the content: pausing, asking questions and linking it to the child's life. One of the most consistent predictors of children learning from screen media.

Our editorial standards

What these reviews are. Narrative evidence reviews written for researchers, teachers, clinicians and parents. They are not systematic reviews with a registered protocol, and they have not been peer reviewed.
How we search. PubMed and Crossref, web searches including ERIC and Google Scholar-style queries, arXiv and the ACM Digital Library for AI research, and a cross-check for missed studies with Perplexity. We prioritise systematic reviews, meta-analyses and randomised trials, then landmark studies.
How we check. Every reference is matched to its DOI, Crossref or PubMed record. Every number (sample size, effect size, p-value) is taken from the abstract or full text of the source itself. When we could only see a secondary summary, the review says so. Studies whose findings we could not verify are left out.
Use of AI. Literature searching and drafting are assisted by AI tools (Anthropic's Claude and Perplexity). An editor, Oleksandr Prykhodko, reviews every article before publication. AI tools make mistakes; that is why every source is checked and why we welcome corrections.
How we write. We report null and negative findings, state limits, and never describe anything as "proven". We say "research-based" about our own materials because they draw on this research; they have not been tested in any study.
Conflict of interest. Cala Luna, by Little Lynx Studios, publishes free social stories and sells personalised social cartoons and songs. We have no funding from, or connection to, any researcher or organisation we cite.
Updates and corrections. Each review shows its last-reviewed date and will be re-checked at least every six months, or sooner when an important study appears. To report an error or suggest a study, email hello@calalunakids.com; corrections are made promptly and noted here.
Terminology. "Social Stories™" is a trademark of Carol Gray. We use "social stories" for the general approach of short explanatory stories.

Corrections log

No corrections yet. Version 1 of the research centre was published on 24 October 2026.

References

Scruggs, T. E., Mastropieri, M. A., & Casto, G. (1987). The quantitative synthesis of single-subject research. Remedial and Special Education, 8(2), 24–33. doi.org/10.1177/074193258700800206
Parker, R. I., Vannest, K. J., Davis, J. L., & Sauber, S. B. (2011). Combining nonoverlap and trend for single-case research: Tau-U. Behavior Therapy, 42(2), 284–299. doi.org/10.1016/j.beth.2010.08.006
Pustejovsky, J. E. (2018). Using response ratios for meta-analyzing single-case designs with behavioral outcomes. Journal of School Psychology, 68, 99–112. doi.org/10.1016/j.jsp.2018.02.003
Kratochwill, T. R., Hitchcock, J. H., Horner, R. H., Levin, J. R., Odom, S. L., Rindskopf, D. M., & Shadish, W. R. (2013). Single-case intervention research design standards. Remedial and Special Education, 34(1), 26–38. doi.org/10.1177/0741932512452794
Guyatt, G. H., Oxman, A. D., Vist, G. E., Kunz, R., Falck-Ytter, Y., Alonso-Coello, P., & Schünemann, H. J. (2008). GRADE: An emerging consensus on rating quality of evidence and strength of recommendations. BMJ, 336(7650), 924–926. doi.org/10.1136/bmj.39489.470347.AD
Kiresuk, T. J., & Sherman, R. E. (1968). Goal attainment scaling: A general method for evaluating comprehensive community mental health programs. Community Mental Health Journal, 4(6), 443–453. doi.org/10.1007/BF01530764

How to read the evidence: methods, glossary and our editorial standards