Psychological Interview

Abstract

The psychological interview is a directed conversation used to elicit information for diagnosis, evaluation, and treatment planning. This article traces its development along the continuum from the unstructured clinical conversation to the fully structured diagnostic schedule, the instruments that standardize it, the Structured Clinical Interview for DSM, the Diagnostic Interview Schedule, the Composite International Diagnostic Interview, and the MINI, and the interrater reliability that standardization buys. It sets the interview against Meehl's finding that mechanical prediction outperforms unaided clinical judgment, treats the working alliance as the relational substrate of any interview, and examines two traditions in which the interview is itself the instrument or the intervention: the forensic cognitive interview and motivational interviewing. Three demonstrations model interrater agreement, clinical versus statistical prediction, and the motivational-interviewing readiness ruler.

Keywords: psychological interview, structured clinical interview, diagnostic reliability, clinical judgment, motivational interviewing

The psychological interview is the oldest and most widely used assessment method in clinical psychology and psychiatry, a directed conversation whose purpose is to elicit the information needed for diagnosis, formulation, and treatment. It is not ordinary conversation: the interviewer works to a purpose, samples specific domains of experience and history, and reasons continuously from what is said and how it is said toward a clinical judgment. For most of the twentieth century that judgment rested on the unstructured interview, in which the clinician chose what to ask and how to weigh the answers, and its diagnostic conclusions proved distressingly unreliable, with two competent clinicians often disagreeing about the same patient. The structured interview was the discipline's response, an attempt to make the interview a measuring instrument rather than an idiosyncratic art, and its introduction with the Structured Clinical Interview for DSM redefined how psychological assessment is done (Spitzer et al., 1992).

What the Psychological Interview Is

The psychological interview is defined, in the medical subject headings, as a directed conversation aimed at eliciting information for psychiatric diagnosis, evaluation, and treatment planning. Two features distinguish it from the surrounding categories of conversation. It is directed: the interviewer, not the respondent, sets the agenda and steers the exchange toward the domains that bear on the clinical question. And it is dual-channel: the interviewer attends both to the content of what is reported, the symptoms, history, and circumstances, and to the manner of its reporting, the affect, coherence, and behavior displayed in the room, which together constitute the mental status examination. The interview is therefore simultaneously a method of gathering self-report and a structured observation, and its unique standing in assessment rests on that combination.

The interview serves several distinct functions that are easy to conflate. As a diagnostic instrument it assigns a case to categories in a classification such as the DSM or ICD. As a formulation tool it builds an individualized account of how a particular person's difficulties arose and are maintained. As the opening of treatment it establishes the relationship, the working alliance, on which any later intervention depends. And in some traditions the interview is not a preliminary to intervention but the intervention itself. These functions impose competing demands, standardization serves diagnosis while flexibility serves formulation and rapport, and much of the history of the interview is the story of how the field has traded one against the other.

The Structured–Unstructured Continuum

Interviews are conventionally ranged on a continuum of structure. At one pole is the unstructured interview, in which the clinician decides in the moment what to ask, in what order, and how to interpret the replies; it is maximally flexible and maximally sensitive to the individual case, but its coverage is uneven and its conclusions vary with the interviewer. At the other pole is the fully structured interview, in which every question is scripted, asked verbatim and in fixed order, and the scoring of each answer is governed by explicit rules, so that the instrument can in principle be administered by a trained lay interviewer with no clinical background. Between them lies the semi-structured interview, which prescribes the domains to be covered and supplies standard probes but allows the clinician to follow up, rephrase, and use judgment, the form taken by the Structured Clinical Interview for DSM (Spitzer et al., 1992).

The great structured instruments were built for different purposes, and their design reflects it. The Diagnostic Interview Schedule was created for the Epidemiologic Catchment Area program so that lay interviewers could generate psychiatric diagnoses in large community surveys without clinicians (Robins et al., 1981). The Composite International Diagnostic Interview extended that logic across countries and diagnostic systems, becoming the instrument of the World Health Organization's World Mental Health surveys and yielding diagnoses comparable across cultures (Kessler & Üstün, 2004). The Mini-International Neuropsychiatric Interview was designed for speed, a short structured interview that produces reliable DSM and ICD diagnoses in a fraction of the time, for use in clinical trials and busy settings (Sheehan et al., 1998). Figure 1 places these instruments within the broader family of psychological interviews, and Table 1 sets out how they differ.

Reliability and the Standardized Interview

The reason structure matters is reliability: the extent to which two interviewers, assessing the same person, reach the same conclusion. Because the diagnosis a patient receives should not depend on which clinician happened to see them, interrater agreement is the first psychometric property any diagnostic interview must establish. Raw percentage agreement is a poor index of it, because two clinicians can agree by chance alone, the more so when a disorder is common or rare. The standard correction is Cohen's kappa, which compares observed agreement with the agreement expected if the two interviewers were guessing independently at the same base rates, and expresses the excess on a scale where zero is chance and one is perfect. Reasoning about a diagnostic judgment as a decision with hits, misses, and false alarms connects the interview to signal detection theory, in which the same confusion between true agreement and lucky coincidence is made explicit.

The historical case for the structured interview is that it lifts kappa out of the range that made unstructured diagnosis untrustworthy. Contemporary validation studies of the SCID for DSM-5 report substantial agreement for most major disorders: a clinical-validity study of the Clinician Version found good-to-excellent interrater and test-retest reliability across diagnoses (Osório et al., 2019), and a separate psychometric evaluation reported kappa values in the substantial range for the majority of categories (Shabani et al., 2021). The gain is not free: standardization can miss the atypical presentation that a flexible clinician would have pursued, and a high kappa for a common disorder can coexist with poor agreement on a rare one. The first demonstration is a Cohen's kappa calculator, letting the reader set the cells of two interviewers' agreement table and see how base rate drives a wedge between raw agreement and chance-corrected kappa; the Worked Example computes one such case by hand.

Clinical Versus Statistical Prediction

Reliability is necessary but not sufficient: an interview can be reliable and still predict poorly. The deepest challenge to the interview as a basis for judgment is Meehl's, who reviewed the studies then available comparing two ways of combining assessment data, the clinician's head and a mechanical formula, and found that the formula did as well as or better than the clinician in nearly every case (Meehl, 1954). The finding was unwelcome and endlessly re-examined, and it has proved remarkably durable. A meta-analysis synthesizing fifty-six years of such comparisons across clinical domains confirmed that mechanical prediction outperforms clinical prediction on average, by a margin that is small but consistent and that widens when the clinician lacks access to information the formula uses (Ægisdóttir et al., 2006). The lesson is not that the interview is worthless but that the human interviewer is a better gatherer of information than combiner of it: the clinician's advantage lies in eliciting and observing, and the mechanical rule's advantage lies in weighting the resulting cues consistently.

Why the unaided clinician loses is a question about judgment under uncertainty. The interviewer is subject to the same heuristics and biases that affect any decision making under uncertainty: confirmation-seeking, overweighting vivid detail, and inconsistency in applying the same criterion from one case to the next. A statistical rule, however crude, applies the same weights every time, and that consistency alone often outperforms the clinician's intuition, which varies with fatigue, order, and mood. The point is not to remove the interview but to feed its output into a consistent decision rule, an accommodation the field has been slow to make. The second demonstration contrasts a mechanical rule with a noisy clinical judgment built on the same valid cue, and shows how the clinician's inconsistency lowers accuracy across many cases even when the underlying cue is equally available to both.

The Working Alliance

Whatever its structure, an interview is a relationship, and the quality of that relationship is not merely a courtesy but a measurable determinant of what the interview achieves. The working alliance was given its enduring, pantheoretical definition by Bordin, who recast the older psychoanalytic notion into three components applicable to any helping relationship: the affective bond between the parties, and their agreement on the goals and on the tasks of the work (Bordin, 1979). So defined, the alliance is among the most robust predictors of outcome in the helping professions. A comprehensive meta-analytic synthesis covering hundreds of studies and tens of thousands of patients found a moderate but highly consistent association between the strength of the alliance and the outcome of treatment, holding across disorders, measures, and therapeutic approaches (Flückiger et al., 2018). For the assessment interview the alliance matters instrumentally as well: a respondent who trusts the interviewer discloses more, and more accurately, so the relationship is also a condition of valid data. The interviewer's capacity to represent the respondent's mental states, to track what they are thinking and feeling and to respond to it, is the mentalization that a good interview depends on and that no script can supply.

The Interview as Intervention: Motivational Interviewing

In most of the traditions considered so far the interview precedes the intervention. Motivational interviewing collapses the distinction: the interview is the intervention. Developed by Miller and Rollnick, it is a collaborative, goal-oriented style of conversation designed to strengthen a person's own motivation and commitment to change by eliciting and exploring their reasons for it within an atmosphere of acceptance (Miller & Rollnick, 2013). Its central premise is that ambivalence is the normal condition of a person contemplating change, and that direct persuasion tends to entrench it, because a person arguing against their own change talks themselves out of it. The interviewer therefore avoids confrontation and instead uses open questions, reflective listening, and selective reinforcement to draw out the person's own change talk, the self-motivating statements that predict subsequent behavior.

The evidence base is substantial. An early narrative and meta-analytic review established that motivational interviewing produces meaningful effects across problem behaviors from substance use to health-related change (Hettema et al., 2005), and a large meta-analysis found significant, durable effects across a wide range of target behaviors and settings (Lundahl et al., 2010). A systematic review of reviews in health and social care confirmed benefit across many applications while noting that effects vary with fidelity and target (Frost et al., 2018). A recurring practical tool is the readiness ruler, on which a person rates the importance of a change and their confidence in achieving it, two components whose product tracks readiness better than either alone. The third demonstration implements the ruler, letting the reader set importance and confidence and see how readiness, and the direction of change talk, responds.

The Investigative Interview

A parallel tradition developed outside the clinic, where the interview's purpose is not to diagnose or to treat but to recall: to obtain from a cooperative witness the most complete and accurate account of an event. The problem the cognitive interview solves is that standard police questioning, with its rapid closed questions and frequent interruptions, systematically suppresses recall. Fisher and Geiselman built an alternative from the principles of memory retrieval: reinstating the mental and physical context of the event, encouraging the witness to report everything however trivial, and prompting retrieval in varied orders and from varied perspectives, each a route back into the stored episodic memory that a single line of questioning would leave untapped.

The technique works. A meta-analytic review of a quarter-century of studies found that the cognitive interview produces a large and reliable increase in the amount of correct information recalled, with only a small increase in errors, so that overall accuracy is preserved (Memon et al., 2010). The gain is largest when the interviewer is well trained and the witness cooperative, and it comes at the cost of a longer, more demanding interview. The cognitive interview is a demonstration in miniature of the article's larger theme: that how the interview is conducted, the structure imposed on the conversation, determines the quality of the information it yields, whether the goal is a diagnosis, a decision, or a memory.

Worked Example

Consider two clinicians who independently interview the same 100 patients and each decide, for every patient, whether major depression is present or absent. Their decisions can be cross-tabulated in a two-by-two table. Suppose they both say present for 30 patients and both say absent for 55; the first clinician says present where the second says absent for 8 patients, and the reverse happens for 7. Observed agreement is the proportion on which they concur: the 30 both-present plus the 55 both-absent, divided by 100, which is 85 divided by 100, or 0.85. Raw agreement of 85 percent looks impressive, but some of it is coincidence, because both clinicians diagnose depression fairly often and would sometimes agree by chance.

Cohen's kappa removes the chance component. The first clinician called depression present in 38 of 100 cases, a marginal rate of 0.38; the second in 37 of 100, a rate of 0.37. If the two were judging independently at those rates, they would both say present by chance with probability 0.38 times 0.37, which is 0.1406, and both say absent with probability 0.62 times 0.63, which is 0.3906. Expected chance agreement is the sum, 0.1406 plus 0.3906, or 0.5312. Kappa is the observed agreement in excess of chance, divided by the maximum possible excess: 0.85 minus 0.5312, which is 0.3188, divided by 1 minus 0.5312, which is 0.4688. The quotient is 0.3188 divided by 0.4688, or about 0.68. A kappa of 0.68 falls in the range conventionally called substantial, comfortably above the 0.4 that unstructured diagnosis of the same disorder often failed to reach, and it illustrates the payoff of standardization: the same 85 percent raw agreement would correspond to a far lower kappa if depression were rarer, because chance agreement on a common category is high. The reliability demonstration lets these numbers be varied and recomputes kappa from the same formula.

Discussion

The history of the psychological interview is a sustained effort to convert a clinical art into a measuring instrument without destroying what made the art valuable. The structured interview succeeded in the first aim: by scripting the questions and the scoring, instruments such as the SCID, DIS, CIDI, and MINI raised interrater reliability from the poor levels that discredited unstructured diagnosis to the substantial levels modern validation studies report (Spitzer et al., 1992; Osório et al., 2019). But reliability is not validity, and Meehl's challenge remains the deepest unresolved tension in the field: even a reliable interview, when its output is combined by the unaided clinician, tends to predict outcomes less accurately than a simple mechanical rule using the same information (Meehl, 1954; Ægisdóttir et al., 2006). The rational response, feeding the interview's superior information-gathering into a consistent decision rule, is understood but unevenly adopted, and the interviewer's confidence in their own judgment is among the reasons why.

Two developments complicate any simple verdict that structure is better. First, the relational dimension of the interview, the working alliance, is itself a robust determinant of both disclosure and outcome, and it is precisely what a fully scripted schedule cannot cultivate (Flückiger et al., 2018). Second, whole traditions treat the interview not as a flawed measuring device but as an active method in its own right, motivational interviewing using the conversation to move a person toward change (Miller & Rollnick, 2013) and the cognitive interview using it to recover memory (Memon et al., 2010). The interview, in the end, is not one thing to be optimized on a single axis but a family of methods whose right degree of structure depends on the purpose: maximal for a survey diagnosis, minimal for building a formulation or an alliance, and specifically engineered for the extraction of accurate memory or the resolution of ambivalence.

Current Directions

The most consequential recent change is the migration of the interview to remote and digital delivery. The rapid, forced adoption of telemental health during the COVID-19 crisis showed that structured and clinical interviews can be conducted by videoconference at scale, and reports from large services documented the feasibility of moving whole clinics to home-based, remote assessment without abandoning the interview's core functions (Sharma et al., 2020). The change raises open questions the field is still working through: whether the mental status observations that depend on the physical presence of the respondent survive the video channel intact, whether the working alliance forms as readily at a distance, and how structured instruments should be adapted for self-administration on a screen. In parallel, the continued psychometric validation of the DSM-5 generation of structured interviews, establishing the reliability and validity of the SCID-5 across languages and settings, is extending the evidence base for standardized diagnosis into new populations (Osório et al., 2019; Shabani et al., 2021). The longer-term prospect, still largely unrealized, is the computer-administered adaptive interview that selects its next question from the answers already given, a direct descendant of the fully structured schedule freed from a fixed script.

Common Misconceptions

A structured interview removes the need for clinical skill.

Only the fully structured survey schedules approach that, and they trade coverage of the atypical case for the ability to use lay interviewers. The semi-structured instruments used in clinical practice, such as the SCID, require the interviewer to apply judgment in following up and scoring, and their reliability depends on that trained judgment (Spitzer et al., 1992).

High agreement between interviewers proves an interview is accurate.

It does not. Two interviewers can agree strongly by chance when a diagnosis is common, which is why agreement is corrected with Cohen's kappa; and even a reliable interview may combine its information less accurately than a mechanical rule, so reliability and predictive validity are distinct properties (Ægisdóttir et al., 2006).

The experienced clinician's judgment beats a formula.

Across more than half a century of head-to-head comparisons, mechanical combination of the same data predicts at least as well as, and usually better than, the unaided clinician, and experience does not reliably close the gap. The clinician's real advantage is in gathering and observing information, not in weighting it consistently (Meehl, 1954).

Glossary

Change talk.

In motivational interviewing, a person's own statements favoring change; eliciting and reinforcing it, rather than arguing for change directly, is the method's core mechanism.

Clinical prediction.

The combination of assessment data into a judgment or forecast by the clinician's own reasoning, as opposed to a mechanical formula.

Cognitive interview.

A forensic interviewing protocol built on memory-retrieval principles, using context reinstatement and varied retrieval to increase the accurate recall of a cooperative witness.

Cohen's kappa.

A statistic of interrater agreement that corrects observed agreement for the agreement expected by chance, scaled so that zero is chance and one is perfect concordance.

Composite International Diagnostic Interview.

A fully structured interview for lay administration, developed by the WHO to generate cross-nationally comparable psychiatric diagnoses in population surveys.

Diagnostic Interview Schedule.

The first fully structured psychiatric interview designed for trained lay interviewers, created for large community epidemiological surveys.

Interrater reliability.

The degree to which two independent interviewers assessing the same person reach the same conclusion; the first psychometric property a diagnostic interview must establish.

Mental status examination.

The structured observation, made during the interview, of a person's appearance, affect, thought, and cognition, complementing the content of what is reported.

MINI.

The Mini-International Neuropsychiatric Interview, a brief fully structured interview producing reliable DSM and ICD diagnoses quickly, widely used in clinical trials.

Motivational interviewing.

A collaborative, goal-oriented interviewing style that strengthens a person's own motivation for change by eliciting and resolving ambivalence rather than by direct persuasion.

Readiness ruler.

A motivational-interviewing tool on which a person rates the importance of a change and their confidence in making it, the two components whose combination tracks readiness.

Semi-structured interview.

An interview that prescribes the domains and standard probes but allows the clinician to follow up and use judgment; the form of the SCID.

Statistical prediction.

The combination of assessment data into a judgment by a fixed, mechanical rule that weights each cue the same way every time, independent of the clinician.

Structured Clinical Interview for DSM (SCID).

The standard semi-structured diagnostic interview, administered by a clinician, that operationalizes DSM criteria into a guided sequence of questions and scoring rules.

Unstructured interview.

An interview in which the clinician chooses in the moment what to ask and how to weigh the answers; maximally flexible but historically of poor interrater reliability.

Working alliance.

The collaborative bond between interviewer and respondent and their agreement on goals and tasks; a robust predictor of both disclosure and treatment outcome.

Key Researchers

Michael B. First (b. 1956). Professor of Clinical Psychiatry at Columbia University; co-author of the SCID and its DSM-5 successors and a central figure in DSM-IV and DSM-5 revision, he has kept the structured diagnostic interview aligned with successive editions of the diagnostic manual. Faculty Page - Wikipedia - ORCID

Ronald P. Fisher. Professor of Psychology at Florida International University; with R. Edward Geiselman he developed the cognitive interview, a memory-based investigative protocol that reliably increases the accurate recall of cooperative witnesses. Faculty Page - Wikipedia - Wikidata

Christoph Flückiger (Fluckiger). Professor of Clinical Psychology at the University of Kassel; he led the definitive meta-analytic syntheses of the therapeutic alliance, establishing the robust alliance–outcome association that grounds the relational dimension of the clinical interview. Faculty Page - ORCID - Wikidata

Paul E. Meehl (1920-2003). Regents' Professor of Psychology at the University of Minnesota; his 1954 monograph established that mechanical combination of interview and test data generally outpredicts the unaided clinical judgment of the interviewer, defining the limits of the interview as a measurement instrument. Wikipedia - Wikidata

William R. Miller (b. 1947). Distinguished Professor Emeritus of Psychology and Psychiatry at the University of New Mexico; he originated motivational interviewing, the client-centered directive style for eliciting behavior change by resolving ambivalence, and co-authored its defining text. Faculty Page - Wikipedia - Wikidata

Stephen Rollnick (b. 1952). Honorary Distinguished Professor at Cardiff University's School of Medicine; with William R. Miller he co-developed motivational interviewing and extended it into healthcare consultations where brief, ambivalence-focused interviewing supports behavior change. Wikipedia - Wikidata

Robert L. Spitzer (1932-2015). Professor of Psychiatry at Columbia University; he led the DSM-III revision that replaced impressionistic diagnosis with explicit criteria and co-developed the Structured Clinical Interview for DSM that operationalized those criteria into a standardized instrument. Wikipedia - Wikidata

Frequently Asked Questions

What is a psychological interview?
It is a directed conversation aimed at eliciting the information needed for psychiatric diagnosis, evaluation, and treatment planning, attending both to the content of what a person reports and to how they report it (Spitzer et al., 1992).

What is the difference between a structured and an unstructured interview?
An unstructured interview lets the clinician choose what to ask and how to weigh the answers, while a structured interview scripts the questions and scoring; structuring the interview is what raises interrater reliability from poor to substantial (Robins et al., 1981).

What is the SCID?
The Structured Clinical Interview for DSM is a semi-structured diagnostic interview, administered by a trained clinician, that operationalizes the DSM criteria into a guided sequence of questions and scoring rules (Spitzer et al., 1992).

How is the reliability of a diagnostic interview measured?
By interrater agreement corrected for chance, most often Cohen's kappa, which compares the agreement two interviewers actually reach with the agreement expected if they were guessing at the same base rates (Shabani et al., 2021).

Does a formula really predict better than a clinician?
On average, yes: across more than half a century of comparisons, mechanical combination of the same interview and test data predicts outcomes at least as well as, and usually better than, the clinician's unaided judgment (Ægisdóttir et al., 2006).

What is motivational interviewing?
It is a collaborative, goal-oriented interviewing style that strengthens a person's own motivation for change by eliciting and resolving their ambivalence rather than by arguing for change directly (Miller & Rollnick, 2013).

What is the cognitive interview?
It is a forensic interviewing method that applies memory-retrieval principles, such as reinstating context and prompting varied retrieval, to increase the amount of accurate information a cooperative witness recalls (Memon et al., 2010).

Why does the relationship in an interview matter?
The working alliance between interviewer and respondent is a robust predictor of outcome and of disclosure, so a trusting relationship is both a therapeutic factor and a condition of obtaining valid information (Flückiger et al., 2018).

References