Evidence-based practice in social work: a working guide

Evidence-based practice is a way of making decisions, not a list of approved programs. The original formulation, borrowed from medicine and adapted for social work in the late 1990s, describes three things that have to be brought together for any single decision: the best available external evidence, the practitioner’s own expertise, and the values, preferences, and circumstances of the person being served.

That third element does real work. A program with strong trial results that a client will not attend, cannot reach, or actively objects to is not the evidence-based choice for that client. Practitioner expertise does real work too: it is what tells you whether the trial population resembles the person in front of you, and what to do when it does not.

The phrase is often used loosely to mean empirically supported treatments — the manualized interventions that have cleared some evidentiary bar. Those lists are useful, but they are an output, not the process. Conflating the two is what produces the familiar situation where an agency adopts a branded program, delivers it to a population it was never tested on, and calls the result evidence-based.

Most descriptions of the process use five steps. They are worth knowing by name, because the ones people skip are predictable.

1. Ask an answerable question. Convert a vague practice concern into something a search can actually address.
2. Acquire the evidence. Search systematically rather than reaching for the first familiar article, and write down what you searched.
3. Appraise what you find. Judge the study on its design, sample, comparison, and measurement, not on its abstract’s confidence.
4. Apply it. Integrate the finding with your own expertise and the client’s values, resources, and context.
5. Assess the outcome. Track whether it worked for this person, and whether your process was any good.

Step five is the one that disappears. It is also the only step that tells you anything about your own practice rather than about the literature, so it is worth protecting even in a rough form: a simple repeated measure, a scaling question at each session, a documented check-in at week six.

A practice question and a search query are different objects. The usual bridge is PICO: population, intervention, comparison, outcome. Written out, a question looks like this:

The structured version is not what you type into a search box. Search engines match terms, not sentences, so pull out the concepts and add the synonyms an author might have used: foster care, out-of-home care, looked-after children; adolescent, youth, teen. Intervention names are usually the highest-yield terms, because authors rarely paraphrase them.

On Practora, a full question triggers a SocialGPT overview alongside the results, which is useful for orienting yourself to unfamiliar vocabulary. Keyword-style input skips the overview and gives you the ranked list directly. Once you have results, the practice filters for population, setting, intervention, and study design narrow them without your having to guess at the right phrasing. Try trauma-focused cognitive behavioral therapy foster care adolescents and then reach for the filters rather than rewriting the query.

Study design tells you what a piece of research is capable of ruling out. It does not tell you whether the study was done well, whether the outcome was measured sensibly, or whether the finding survives outside the conditions that produced it. Design hierarchies are a starting filter, not a verdict.

Questions worth asking of any study

• Who was actually enrolled? Age, setting, severity, referral route, and who was excluded. Exclusion criteria often remove exactly the co-occurring problems that define your caseload.
• Compared with what? Waitlist and no-treatment comparisons answer a much weaker question than an active alternative or treatment as usual. “Better than nothing” is a low bar that gets reported like a high one.
• How large was the effect? Look for effect sizes and confidence intervals, not just whether a p-value cleared .05. A statistically significant change of two points on a sixty-point scale may not be visible to anyone involved.
• Who left, and when? High or uneven attrition can manufacture an effect on its own, and the people who drop out are rarely a random sample.
• How was the outcome measured? Self-report at the end of a session, an independent rater, or an administrative record are three very different claims. Ask how long after the intervention the measurement happened.
• Who ran and funded it? When the developer of an intervention is also its evaluator, effects tend to run larger. That is not fraud; it is a known and measurable pattern.
• Has anyone else replicated it? One trial is a finding. Independent replication is evidence.

Start with systematic reviews and meta-analyses when they exist. They save time and, more importantly, show you the spread of results rather than one draw from it. Then read one or two primary studies in full so you know what the intervention actually involved.

Trial results are averages across groups. They describe what happened to a population, and they do not predict what will happen to one person. Transferring a finding is a judgment, and it is worth making explicitly rather than by default.

• How close is this client to the study sample on the things that plausibly matter — severity, co-occurring conditions, language, culture, housing stability?
• Can it be delivered here? Trained staff, session count, materials, transport, and cost are part of the evidence question, not separate from it.
• What does the client want? Preference and expectation affect engagement, and engagement affects everything downstream.
• What are the plausible harms and burdens, including opportunity cost and the effect of another failed referral?
• What is the plan if it does not work, and how will you know within a useful timeframe?

Documenting the reasoning takes a few sentences and gives your supervisor something to respond to. It also gives you a record when the case is reviewed later and the decision has to be explained.

Practora is built for the acquire step, and for part of the appraise step. It searches OpenAlex, normalizes the metadata, checks DOIs against Crossref, and formats APA 7 references so you are not transcribing fields by hand. The practice filters and the source pages are there to speed up screening.

What it does not do is appraise. It cannot tell you whether a comparison condition was adequate, whether an outcome measure was sensible, or whether a finding transfers to your setting. It also has coverage limits: books, chapters, dissertations, agency reports, and practice literature are indexed unevenly in open catalogs, so a serious review still needs your institution’s databases and a direct look at the organizations publishing gray literature.

When you have a working set of sources, the exports page will turn them into an APA reference list or an annotated bibliography starter. The FAQ covers what the verification labels mean.