PRF cheat sheet

A study question should specify population, exposure or intervention, comparator, outcome, and timeframe (PICO); a vague question cannot be operationalized into measurable variables.

Hypothesis-driven designs test a priori predictions; exploratory designs generate hypotheses and cannot support confirmatory inference without pre-specification.

Internal validity concerns whether observed effects are causal within the study; external validity concerns generalizability to other populations and settings.

Randomized controlled trials minimize confounding by randomization but may have limited generalizability due to strict eligibility criteria.

Observational cohort designs estimate incidence and relative risk; case-control designs estimate odds ratios and are efficient for rare outcomes.

Cross-sectional designs measure exposure and outcome simultaneously and cannot establish temporal order, limiting causal inference.

Sampling strategy determines inferential scope: probability sampling supports population inference, while convenience sampling supports only descriptive inference about the sample.

A pilot study estimates feasibility, recruitment rates, and variance to inform sample size and protocol refinement, not to test the main hypothesis.

Pre-registration of the design, outcomes, and analysis plan reduces selective reporting and outcome switching.

Informed consent requires disclosure of purpose, procedures, risks, benefits, alternatives, and the voluntary right to withdraw without penalty.

Institutional review board or ethics committee approval must precede any data collection involving human participants, including pilot and minimal-risk studies.

The Declaration of Helsinki and the Belmont Report principles (respect for persons, beneficence, justice) underpin most research ethics frameworks.

Vulnerable populations (children, prisoners, pregnant persons, cognitively impaired adults) require additional safeguards and often assent in addition to consent.

Falsification manipulates or alters data; fabrication invents data; plagiarism appropriates others' work without attribution. All constitute research misconduct.

Conflict-of-interest disclosure covers financial ties, institutional relationships, and intellectual commitments that could bias design, analysis, or reporting.

Data protection regulations (e.g., GDPR, HIPAA) govern identifiability, storage duration, cross-border transfer, and the rights of data subjects.

Deviation from an approved protocol requires documentation and, when it affects participant safety or data integrity, amendment and re-approval.

A systematic review uses a pre-specified protocol, comprehensive search, dual screening, and risk-of-bias assessment; a narrative review does not.

PRISMA reporting covers identification, screening, eligibility, and inclusion, with counts at each stage and reasons for exclusion.

The Cochrane Risk of Bias tool assesses randomization, allocation concealment, blinding, incomplete outcome data, selective reporting, and other sources.

GRADE rates certainty of evidence as high, moderate, low, or very low based on risk of bias, inconsistency, indirectness, imprecision, and publication bias.

Publication bias is detected via funnel plot asymmetry, Egger's regression, and trim-and-fill; small-study effects inflate pooled estimates.

Heterogeneity is quantified by I-squared (proportion of variability due to between-study differences) and tau-squared (between-study variance).

A fixed-effect model assumes one true effect; a random-effects model assumes a distribution of true effects across studies.

Critical appraisal distinguishes internal validity (risk of bias) from applicability (whether findings transfer to the target population).

Reliability concerns consistency; validity concerns whether an instrument measures the intended construct.

Internal consistency is estimated by Cronbach's alpha or omega; values above 0.70 are commonly acceptable, though alpha assumes tau-equivalence.

Test-retest reliability uses intraclass correlation or Pearson correlation across administrations separated by an appropriate interval.

Inter-rater reliability is estimated by Cohen's kappa for two raters and Fleiss' kappa or ICC for more than two.

Content validity is established by expert review and alignment with a defined construct domain; construct validity by convergent and discriminant evidence.

Criterion validity compares scores against a gold standard (concurrent) or future outcome (predictive).

Measurement scales differ: nominal (labels), ordinal (rank), interval (equal spacing, arbitrary zero), and ratio (true zero, meaningful ratios).

Standardization of data collection procedures (training, manuals, calibration) reduces measurement error and systematic bias across sites.

Type I error is rejecting a true null hypothesis (alpha); Type II error is failing to reject a false null (beta); power equals 1 minus beta.

A p-value is the probability of data at least as extreme as observed, assuming the null is true; it is not the probability the null is true.

A 95% confidence interval excludes the null value when the two-sided p-value is below 0.05 for the same effect estimate.

Sample size for a two-group mean comparison depends on effect size, alpha, power, and variance; effect size is typically Cohen's d.

Parametric tests (t-test, ANOVA) assume approximate normality and homogeneity of variance; nonparametric alternatives include Mann-Whitney U and Kruskal-Wallis.

Regression coefficients represent the expected change in the outcome per one-unit change in the predictor, holding other covariates constant.

Confounding is addressed by randomization, restriction, matching, stratification, or multivariable adjustment; residual confounding may persist.

Missing data mechanisms are MCAR (missing completely at random), MAR (missing at random), and MNAR (missing not at random); multiple imputation is valid under MAR.

Purposive sampling selects information-rich cases; theoretical sampling iteratively selects cases to develop emerging theory.

Saturation is the point at which new data yield no new codes or themes; it is context- and question-specific.

Grounded theory builds theory inductively through open, axial, and selective coding with constant comparison.

Trustworthiness criteria include credibility, transferability, dependability, and confirmability, paralleling validity and reliability.

Member checking, triangulation, peer debriefing, and audit trails strengthen credibility and dependability.

Reflexivity documents the researcher's position, assumptions, and influence on data collection and interpretation.

Mixed-methods designs include convergent parallel, explanatory sequential, exploratory sequential, and embedded; integration occurs at design, methods, or interpretation levels.

Qualitative coding may be inductive (data-driven) or deductive (framework-driven); inter-coder agreement supports consistency but not interpretive validity.

A data management plan specifies collection, storage, access, retention, sharing, and destruction procedures before the study begins.

The FAIR principles require data to be findable, accessible, interoperable, and reusable, typically via metadata and persistent identifiers.

Version control (e.g., git) and documented code support computational reproducibility of analyses.

Data validation includes range checks, logic checks, missing-value flags, and double entry for critical variables.

A codebook defines each variable's name, type, permissible values, units, and missing-value codes.

De-identification reduces re-identification risk but does not eliminate it; quasi-identifiers can permit linkage attacks.

Reproducibility refers to obtaining consistent results from the same data and code; replicability refers to obtaining consistent results from new data.

An audit trail records data provenance, transformations, and analytic decisions so that results can be traced and verified.

Reporting guidelines are design-specific: CONSORT for trials, STROBE for observational studies, PRISMA for systematic reviews, COREQ for qualitative research.

The IMRaD structure (Introduction, Methods, Results, Discussion) organizes empirical reports; the Methods section should permit replication.

Absolute effects (risk difference, NNT) convey clinical meaning better than relative effects alone, which can inflate perceived benefit.

Selective outcome reporting occurs when published outcomes differ from pre-registered outcomes; trial registration enables detection.

Authorship requires substantial contribution, drafting or critical revision, final approval, and accountability, per ICMJE criteria.

Open access and preprints accelerate dissemination but may bypass peer review; versioning and peer-reviewed publication should follow.

Knowledge translation converts findings into practice through guidelines, decision aids, and implementation strategies with explicit target audiences.

Plain-language summaries and stakeholder engagement improve accessibility and uptake among non-specialist audiences.