Psychological Testing

Overview

A psychological test is a standardised procedure for sampling behaviour under controlled conditions, scored by fixed rules so that results can be compared across people and occasions. Standardisation has two meanings that must not be confused: uniform administration and scoring, so that differences in scores reflect differences in test-takers rather than in procedure, and the collection of normative data, so that an individual score can be located in a reference distribution. Tests differ from informal judgement because their items, instructions, time limits and scoring are fixed, and because their measurement quality is investigated rather than assumed. A test score is always an estimate: it contains true score plus measurement error, and the examiner's job is to keep error small, to report uncertainty, and to avoid over-interpreting small differences.

Testing serves selection, classification, diagnosis, counselling, research and programme evaluation. In each use the same cautions apply. Scores should supplement, not replace, interviews, records and observation. Cut scores should be justified by evidence and consequences, not chosen by convenience. Cultural and linguistic fairness must be examined, because an item that assumes unfamiliar knowledge measures background as well as ability. And feedback must be given responsibly, in language the test-taker can understand, with limits and next steps stated plainly.

Types of tests and test construction

Tests are classified in several overlapping ways. By function they include intelligence, aptitude, achievement, personality, interest, attitude and neuropsychological tests. By administration they are individual or group tests, and by scoring they are objective, with fixed keys, or subjective, requiring judgement, as in projective interpretation. Speed tests reward rapid responding under time pressure, while power tests allow time for difficulty to be the limiting factor. Verbal tests require language, non-verbal and performance tests reduce language load, and culture-fair tests attempt to reduce cultural bias, though no test is fully culture-free. Standardised tests have uniform procedures and norms; teacher-made or informal tests may be useful locally but lack that evidence base.

Construction follows a disciplined sequence. First, define the construct and purpose: what is to be measured, for whom, and for what decision. Second, write a large pool of items covering the construct domain, following rules of clarity, single idea per item, plausible distractors for multiple-choice items, and freedom from ambiguity, double negatives and cultural traps. Third, try the items out on a sample similar to the target group and analyse them: difficulty index, the proportion answering correctly, should span a range with many items near moderate difficulty for maximum discrimination; discrimination index, the difference between high and low scorers on the item, should be positive and substantial; and distractors should attract low scorers more than high scorers. Weak items are revised or dropped. Fourth, assemble the final form with balanced content, appropriate length and clear instructions. Fifth, standardise administration and collect norms. Item response theory refines this logic by modelling the probability of a correct response as a function of ability and item parameters such as difficulty, discrimination and guessing, allowing item banks and adaptive testing.

Test standardisation: reliability, validity, norms

Reliability is consistency of measurement. Test-retest reliability correlates scores on two occasions and reflects stability over time, but it is affected by memory, practice and real change. Parallel-forms reliability correlates two equivalent forms and avoids memory effects but requires truly parallel forms. Split-half reliability correlates two halves of one test and reflects internal consistency, corrected by the Spearman-Brown formula to estimate full-test reliability. Coefficient alpha extends this logic to all possible splits and is the common index of internal consistency for scales. Inter-rater reliability is essential when scoring involves judgement. Reliability sets a ceiling on validity: a test cannot correlate with a criterion more highly than its own consistency allows, but a highly reliable test may still be invalid if it consistently measures the wrong thing. The standard error of measurement translates reliability into score uncertainty: a band around an observed score, rather than a point, should guide decisions.

Validity is whether a test measures what it claims for its intended use. Content validity asks whether items represent the full domain, judged by experts and by a systematic table of specifications. Criterion validity asks whether scores predict an external criterion: concurrent validity uses a criterion available now, predictive validity uses a future criterion, and validity coefficients are correlations whose size must be judged against the decision and base rates. Construct validity is the broadest form: it asks whether scores behave as theory predicts, through convergent evidence, where related measures correlate, discriminant evidence, where unrelated measures do not, and factor structure, experimental effects and developmental patterns. Face validity, whether a test looks relevant to test-takers, affects motivation and acceptance but is not evidence of measurement quality. Norms locate an individual score in a reference group: percentiles state the percentage scoring below a given score, while standard scores such as z-scores and T-scores place scores on a common scale. Norms must match the test-taker's population in age, language, education and culture, or interpretation will mislead.

Areas of testing

Intelligence testing assesses reasoning, problem solving, memory and processing speed, historically through individually administered scales with verbal and performance components and now through group and computerised batteries. The score is useful for identifying strengths, weaknesses and support needs, but it is not a complete measure of a person, and it is sensitive to schooling, language and opportunity. Aptitude testing predicts future learning or performance in a domain, such as clerical, mechanical or musical aptitude, and is used in guidance and selection where a specific capacity matters more than general achievement. Achievement testing measures what has been learned, through school examinations, standardised achievement batteries and diagnostic tests that pinpoint gaps for remediation. The distinction matters: an achievement test that is used as if it measured innate aptitude will punish learners who were never taught the content.

Personality testing divides into self-report inventories, where respondents endorse statements about themselves, and projective techniques, where ambiguous stimuli invite responses that are interpreted for motives, conflicts and perception. Inventories are efficient and scorable but vulnerable to response styles such as social desirability, acquiescence and faking. Projective methods can reveal material that direct questions miss, but their reliability and validity are contested and interpretation requires extensive training. Interest inventories map preferences for activities and occupations and guide educational and career counselling without claiming ability. Neuropsychological testing assesses memory, attention, language, executive function and visuospatial skill after brain injury or disease, linking performance patterns to functional impairment and rehabilitation plans. In every area, the examiner must match the instrument to the question, check its norms and validity for the specific group, and combine scores with other evidence.

Attitude scales

An attitude is a learned tendency to evaluate an object, person, issue or group favourably or unfavourably, with cognitive, affective and behavioural components. Attitude scales attempt to place people on a continuum of favourability, and their construction must handle the gap between what people endorse on a questionnaire and how they actually behave. The Likert method is the most widely used: respondents rate agreement with a set of statements on a graded scale, typically from strongly disagree to strongly agree, with negatively worded items reverse-scored, and the total or average score represents attitude strength. Its strengths are ease of construction, good reliability and intuitive scoring; its risks are acquiescence, the tendency to agree regardless of content, central tendency, the avoidance of extremes, and social desirability on sensitive topics.

Other methods solve different problems. Thurstone scaling uses judges to place statements on an attitude continuum and then selects statements spanning the continuum, so that endorsing a statement locates the respondent, but judge labour is heavy. The semantic differential measures connotative meaning by rating a concept on bipolar adjective pairs such as good-bad or strong-weak, revealing evaluative, potency and activity dimensions. Guttman scaling seeks a cumulative order in which endorsing a stronger item implies endorsing weaker ones, which is elegant when it holds but restrictive in practice. Bogardus social distance scales measure willingness to accept varying degrees of closeness to a group. Whatever the method, scale quality depends on clear item wording, balanced positive and negative items, pilot testing, reliability evidence and validation against behaviour or related attitudes, not merely against other questionnaires. Attitude scores predict behaviour best when attitude and behaviour match in target, action, context and time, and when social pressure and opportunity are taken into account.

Computer-based testing and applications

Computer-based testing delivers items, records responses and scores automatically, bringing speed, precise timing, multimedia items, immediate feedback and reduced scoring error. Computerised adaptive testing goes further: after each response, an algorithm estimates ability and selects the next item at an appropriate difficulty, so each test-taker receives a tailored, shorter test with comparable precision. This efficiency depends on a large, calibrated item bank built with item response theory, on exposure controls so items are not overused and compromised, and on secure delivery that prevents cheating and item harvesting. Tests can also be administered remotely, expanding access for distance learners and rural candidates, but remote testing raises problems of identity verification, environment control, technical failure and unequal access to devices and connectivity.

Applications extend beyond delivery. Computers support test construction through item banking and analysis, scoring and reporting through profiles and norms, and interpretation through decision aids, though automated interpretation must be checked by a qualified professional and never presented as a diagnosis by itself. Ethical issues intensify online: data security and privacy, informed consent for data storage, accessibility for test-takers with disabilities, and transparency about algorithms that adapt items or flag behaviour. The future of testing is likely to combine adaptive efficiency with performance tasks, simulations and continuous assessment, but the old standards do not change: reliability, validity, fairness and appropriate use must be demonstrated for the computerised form itself, not borrowed from the paper form without evidence. A test that is fast and attractive but unvalidated is a worse instrument than a slow, validated one.

Reliability methods compared

Validity evidence compared