Research methods & statistics Glossary

Evidence-based practice joins three inputs

Evidence-based practice combines the best available research evidence with clinical expertise and the client's values and context. Slocum and colleagues describe this ABA framework as a decision process rather than a list of automatically approved procedures.

The BACB BCBA Test Content Outline, 6th edition includes evaluating evidence, contextual variables, client preferences, and data-based decisions. It is examination content. It does not authorize practice or endorse an intervention for a particular person.

When reading a claim, ask who was studied, what changed, how the outcome was defined, which supports were present, how missing observations were handled, and whether benefits and unwanted effects were measured.

Separate study quality from case fit

Internal validity asks whether the design and execution support the study's causal claim. External validity asks how findings may extend across people, settings, responses, procedures, and time. Measurement validity asks whether the recorded variable represents the intended construct. Social validity asks whether goals, procedures, and outcomes matter and are acceptable to the people affected.

A study can be strong on one dimension and limited on another. A carefully controlled experiment may answer a narrow question with little evidence about generalization. A large observational study may describe an important pattern while leaving causal explanations unresolved. Read the design for the question it can answer.

For clinical use, map the research participants, setting, procedures, implementer training, comparison, dose, supports, follow-up, and adverse-event monitoring to the current case. Record material differences. Adaptation may be appropriate, but it changes what can be inferred from the original study and requires its own measurement.

Document that reasoning before the clinical decision is made.

Effect estimates need context

An effect size expresses the magnitude of a contrast or change in a defined metric. Different designs and questions call for different effect-size methods. Raw data, graphs, baseline variability, overlap, trend, immediacy, consistency, clinical importance, and client experience still matter.

The What Works Clearinghouse single-case technical documentation discusses design standards, visual analysis, combining studies, and effect-size estimates. It is historical technical guidance, rather than one mandatory metric for all current ABA research.

A confidence interval is a range produced by a method designed to cover the target parameter at a stated long-run rate under its assumptions. The NIST engineering statistics handbook explains the frequentist interpretation. A 95% interval is not a 95% personal probability that the fixed parameter lies inside the observed interval.

Wide intervals signal limited precision. Narrow intervals can still surround a biased estimate when measurement, sampling, model, or missing-data assumptions are poor.

Reviews answer defined questions

A systematic review uses a preplanned, transparent method to search for, select, appraise, and synthesize evidence for a focused question. The National Academies' standards for systematic reviews cover protocol development, searching, screening, extraction, appraisal, synthesis, and reporting.

A meta-analysis is a statistical synthesis of results from studies judged sufficiently compatible for a defined analysis. A systematic review can include no meta-analysis, and a meta-analysis is only as interpretable as its question, included studies, effect definitions, dependence handling, and assumptions. The AHRQ methods guide discusses compatibility, heterogeneity, model choice, missing results, dependence, and uncertainty.

Publication bias arises when the available record differs systematically from the full set of conducted studies, often because results influence whether and how work is published. Search breadth, registries, gray literature, outcome switching, small-study patterns, and sensitivity analyses can inform an assessment. One funnel plot does not prove or remove bias.

Replication tests how far a finding travels

Replication repeats a study or demonstration to test whether a finding appears again under stated conditions. Exact duplication is rarely possible, so authors should identify what stayed constant and what changed.

Systematic replication deliberately varies selected participants, settings, responses, procedures, implementers, or other features while preserving the central question. It can show boundary conditions and generality. A different result may reveal a useful limit rather than a failed exercise.

For single-case work, replication can occur within a design, across participants, or across later studies. Count the number of opportunities, phases, effects, participants, and replications separately.

Read a finding in layers

Suppose a review finds eight studies of a teaching procedure. Five show clear improvement, two show mixed patterns, and one has an uninterpretable baseline. A pooled estimate from only the five clear studies would answer a selected question and hide part of the evidence base.

A better summary reports the study count, participant count, design quality, outcome definitions, effect estimates, uncertainty, missing results, and reasons for excluding any study from synthesis. It also asks whether client priorities, assent, access needs, adverse effects, maintenance, and generalization were measured. The evidence can inform a clinical decision while leaving meaningful uncertainty.

Explore clinical roles at Finni practices. During diligence, ask how clinicians access complete evidence, preserve raw data, discuss uncertainty, and involve clients in decisions.