Psychological Techniques

Abstract

Psychological techniques are the methods by which psychology turns private mental events into public, measurable quantities. Because the mind cannot be observed directly, every finding rests on a technique that makes some hidden operation leave an observable trace: the time a response takes, the accuracy of a discrimination, the point at which a stimulus becomes detectable, the words of a think-aloud protocol, or the signal of a scanner. This article surveys the principal families of technique in cognitive psychology, from mental chronometry, signal detection, and psychophysics through verbal report, process tracing, and computational modelling, to the reproducible and open-science methods that now govern how all of them are applied. Three interactive demonstrations model signal detection, a psychometric threshold, and the inflation of false positives.

Keywords: psychological techniques, measurement, research methods

Psychological techniques are the procedures cognitive psychology uses to make the operations of an unobservable mind yield observable, quantifiable data. The discipline has no instrument that reads a thought directly, so its progress has always depended on ingenuity of method: on arranging a task so that a hidden process is forced to reveal itself in a number that can be recorded and compared (Donders, 1868/1969). The National Library of Medicine's Medical Subject Headings groups these methods under the descriptor Psychological Techniques, and this article follows the major families in turn, asking of each what mental quantity it estimates and by what logic it recovers it.

What Psychological Techniques Are

A psychological technique is a disciplined procedure for eliciting, recording, and quantifying behaviour under conditions controlled well enough that the record can be read as evidence about a mental process. The defining difficulty of the field is that its subject matter is private: perception, memory, decision, and thought produce no signal an experimenter can measure the way a physicist measures a current. Every technique is therefore a solution to the same problem — how to arrange a stimulus, a task, and a response so that an otherwise invisible operation is compelled to leave a trace in something observable, whether the latency of a key-press, the proportion of correct answers, a spoken report, or a physiological recording (Donders, 1868/1969).

The families of technique surveyed here differ in which trace they exploit and which quantity they recover, but they share a common structure. Each pairs a task, engineered so that the process of interest is engaged and, ideally, isolated, with a measurement model, an explicit account of how the observed data map onto the unobserved quantity. The task without the model yields uninterpreted numbers; the model without the task has nothing to fit. What has changed most over the discipline's history is not the tasks, many of which are more than a century old, but the sophistication of the measurement models and, most recently, the methodological standards that decide whether a result obtained with any of them can be believed.

Types of Psychological Techniques

The National Library of Medicine's Medical Subject Headings places Psychological Techniques as a formal descriptor and hangs a set of narrower techniques beneath it. This classification is an indexing tool for the biomedical literature rather than a theory of method, so its groupings answer the question of under what heading a study is filed rather than how a mental quantity is best estimated, and the two need not coincide. The categories are also orthogonal to the thematic families this article is organised around: a single descriptor such as Reaction Time belongs at once to the chronometric tradition and, as a measured latency, to the wider apparatus of behavioural measurement. Table 1 lists the direct children of the descriptor; only those that are treated at length in their own articles are linked.

Mental Chronometry: Timing the Hidden Stages

The oldest quantitative technique in psychology reads the organisation of thought off the time it takes. Franciscus Donders introduced mental chronometry in 1868 by timing reactions of graded complexity and subtracting one from another: a simple reaction with one stimulus and one response, a go/no-go reaction that adds a stage of discrimination, and a choice reaction that adds a stage of response selection on top. The difference between two such reactions estimates the duration of the stage that distinguishes them, so an unobservable mental operation is recovered as an interval of time (Donders, 1868/1969). This subtractive logic, and the family of paradigms built on it, is treated in full in the article on Reaction Time; here it stands as the founding example of the discipline's method — a task engineered so that a hidden process is forced to declare itself in a number.

Donders' subtraction assumes pure insertion, that a stage can be added to a task without disturbing the others, and that assumption is often false. Saul Sternberg answered the difficulty a century later with the additive-factors method, which manipulates the difficulty of processing rather than inserting or deleting stages: two factors that act on the same stage interact in their effect on reaction time, while factors acting on different stages simply add, so the pattern of additivity and interaction maps the stages without assuming any can be cleanly inserted (Sternberg, 1969). His memory-scanning task supplied the classic demonstration, reaction time rising linearly with the number of items held in memory at a slope near forty milliseconds per item, evidence of a serial comparison running through the set one item at a time (Sternberg, 1966). Chronometry established the template every later technique would follow: pair a task with an explicit model of how its output maps onto the hidden quantity.

Signal Detection: Separating Sensitivity from Bias

Raw accuracy is a treacherous measure, because how often an observer responds affirmatively depends not only on how well they can tell signal from noise but on how willing they are to respond at all. Signal detection theory, imported into psychology from radar engineering, dissolves this confound by modelling detection as a decision between two overlapping distributions of internal evidence — one generated by noise alone, one by signal added to noise — with the observer placing a criterion along the evidence axis and responding affirmatively whenever the evidence exceeds it (Swets et al., 1961). The distance between the two distributions is the observer's sensitivity, written d′, and the placement of the criterion is the response bias; the same hit rate can arise from keen sensitivity with a strict criterion or from poor sensitivity with a lax one, and only the model tells them apart.

The technique's power is that both quantities are recovered from two observed rates. The hit rate is the proportion of signal trials called present and the false-alarm rate the proportion of noise trials called present; sensitivity is the difference of their z-transforms, d′ equal to the z of the hit rate minus the z of the false-alarm rate, and the criterion is a symmetric function of the same two z-scores (Stanislaw & Todorov, 1999). Because it strips out the decision criterion, signal detection theory has spread far beyond perception into memory recognition, diagnostic testing, eyewitness identification, and any task where a discrimination is confounded with a willingness to respond. The detailed geometry of the receiver operating characteristic and its applications are developed in the article on Signal Detection Theory; the first demonstration lets a reader move the criterion and the sensitivity and watch the hit and false-alarm rates, and the recovered d′, follow.

Psychophysics: Measuring the Threshold

Before psychology could time a process or model a decision, it had to show that sensation could be measured at all, and that is what psychophysics accomplished. Gustav Fechner, founding the field in 1860, defined the techniques that relate the physical intensity of a stimulus to the sensation it produces: the absolute threshold, the intensity at which a stimulus becomes detectable, and the difference threshold, the smallest change in intensity that can be told apart. Because a threshold is not a sharp cut-off but a region over which detection rises gradually with intensity, it is estimated as a point on a smooth psychometric function — most often the stimulus value detected on half of trials. These constructs are treated at length in the articles on Sensory Thresholds and the Differential Threshold.

The classical methods of measuring a threshold — the method of constant stimuli, the method of limits, and the method of adjustment — differ in how they sample the stimulus range, but each ends by fitting a psychometric function whose location gives the threshold and whose slope gives the precision of discrimination. The regularity Fechner built on had been established by Ernst Weber: the difference threshold is not a fixed increment but a constant fraction of the baseline intensity, so a heavier standard requires a proportionally larger change before the change can be told apart (Algom, 2021). Fechner's own contribution went further: assuming that every just-noticeable difference is a subjectively equal step, he integrated Weber's law to derive a logarithmic law relating sensation to stimulus intensity, the first quantitative law of mind — though later analysis stresses that Weber's empirical regularity and Fechner's derived law are distinct claims that should not be conflated (Algom, 2021). Psychophysics thus gave psychology both a measurement procedure and a model of the quantity it measured, and its threshold-fitting logic remains the backbone of research on perception. The second demonstration fits a psychometric function and reads the threshold and the just-noticeable difference off its shape.

Verbal Reports and the Limits of Introspection

The most direct technique for studying the mind is to ask it to describe itself, and the systematic use of self-observation is as old as experimental psychology. Wilhelm Wundt, founding the first psychology laboratory at Leipzig in 1879, made trained introspection a laboratory method, having practised observers report the elementary sensations and feelings evoked under tightly controlled stimulation. The approach fell into disrepute when it proved unable to settle disputes between laboratories, but the underlying idea — that verbal report can be data — was rehabilitated in a disciplined form. K. Anders Ericsson and Herbert Simon showed that think-aloud protocols, in which participants verbalise the contents of working memory as they perform a task without interpreting or explaining them, yield reports that reliably reflect the information being attended to, provided the reporting does not itself alter the process (Ericsson & Simon, 1980).

The rehabilitation came with a sharp boundary. Richard Nisbett and Timothy Wilson assembled evidence that people have little direct access to the processes underlying their judgments and choices, and that when asked why they responded as they did they often confabulate a plausible cause rather than report the true one, drawing on lay theories about what should have influenced them (Nisbett & Wilson, 1977). The lesson that survives is a division of labour: verbal reports are trustworthy about the current contents of attention, as protocol analysis requires, but unreliable about the causal processes that generated a response, which is exactly why the field turned to indirect techniques that measure a process without asking the participant to describe it.

Process Tracing and Physiological Measures

Between the millisecond of a reaction time and the retrospective word of a report lies a broad class of techniques that record the unfolding of a process while it happens. Process-tracing methods capture the intermediate steps of cognition rather than only its outcome: eye-tracking follows where attention is deployed, information-board and mouse-tracking paradigms record which pieces of a problem are inspected and in what order, and the continuous trajectory of a hand reaching toward one of two responses reveals a decision still forming (Freeman & Ambady, 2010). A survey of the field documents how these methods, once confined to decision research, have become general-purpose tools for testing process models against the moment-to-moment record rather than the final choice alone (Schulte-Mecklenbeck et al., 2017).

Physiological techniques read the body's accompaniment to mental activity. The galvanic skin response indexes autonomic arousal through changes in the skin's conductance; pupillometry and heart-rate measures track effort and emotion; and functional neuroimaging localises the neural correlates of a task. Each buys access to a hidden process at the price of an inferential gap, and functional magnetic resonance imaging makes the gap explicit: it measures a slow haemodynamic signal that is only an indirect correlate of neural activity, so its spatial and temporal resolution, and the inferences it will support, are bounded in ways the technique's users must respect (Logothetis, 2008). The recurring caution across all these measures is reverse inference — the temptation to read a mental state off a physiological signal that is neither specific to it nor uniquely produced by it.

Computational Modelling as Measurement

The most consequential shift in psychological technique over the last half-century is the treatment of a formal model not as a summary of data but as a measuring instrument. Rather than compare mean reaction times or accuracies across conditions, a researcher fits a process model to the full pattern of behaviour and reads off its parameters, each a psychologically interpretable quantity that the raw data confound. The diffusion decision model is the paradigm case: from the distribution of reaction times for correct and error responses together with accuracy, it estimates the rate of evidence accumulation, the amount of evidence required, and the time consumed outside the decision, separating processing efficiency from caution in a way no summary statistic can (Ratcliff, 1978; Ratcliff & McKoon, 2008).

Model-based measurement has become routine because accessible software has lowered its cost and because its parameters answer questions that mean performance cannot (Voss et al., 2013). Two observers with identical mean reaction times may differ sharply in drift rate and caution; only the model distinguishes them, and only its parameters map cleanly onto constructs such as evidence quality or response conservatism. The approach generalises well beyond two-choice tasks, to memory, categorisation, and value-based choice, and it is developed further in the articles on Reaction Time and Decision Making. Its promise is also its hazard: a flexible model can fit data for the wrong reasons, so the technique is only as trustworthy as the tests that discipline it, a theme the reproducibility movement has since made central (Ratcliff et al., 2016).

Reproducible and Open-Science Methods

A technique is worthless if the findings it produces do not hold up, and the last decade has forced psychology to treat reproducibility itself as a methodological problem. A large collaborative effort that repeated one hundred published studies found that a substantial fraction of the effects failed to replicate, and that replicated effects were on average about half the original size, a result that turned diffuse worry into a measured estimate of the field's reliability (Open Science Collaboration, 2015). The diagnosis pointed less at fraud than at ordinary practice: the many undisclosed choices available when collecting and analysing data — when to stop sampling, which conditions to compare, which covariates to include — give a researcher so many chances to find a significant result that the nominal five-percent false-positive rate balloons far beyond it (Simmons et al., 2011).

The response has been a set of techniques aimed at the research process rather than the participant. Preregistration fixes the hypotheses and the analysis plan before the data are seen, converting an exploratory analysis into a confirmatory one and removing the flexibility that inflates false positives; registered reports extend the idea by peer-reviewing the plan before results exist. Alongside these run standards for open data, open materials, and analytic transparency, and, in neuroimaging especially, explicit guidance for making a complex pipeline reproducible (Poldrack et al., 2017). A widely endorsed manifesto gathered these measures into a coherent programme for reproducible science (Munafò et al., 2017), and a subsequent review distinguished the several things the term reproducibility can mean, from re-running the same analysis on the same data to obtaining the same conclusion with new data, so that the field can state precisely which kind it is claiming (Nosek et al., 2022). The third demonstration shows how quickly the false-positive rate climbs as undisclosed analytic choices accumulate.

Current Directions

The technical frontier is now as much about infrastructure as about individual paradigms. Experiments that once required a laboratory are increasingly run in the browser, and validated software for building and delivering behavioural tasks online has made large, diverse samples routine while raising new questions about timing precision and data quality (Anwyl-Irvine et al., 2020). The move online has been coupled with the open-science reforms: preregistered, high-powered studies delivered to thousands of participants are becoming a standard unit of evidence, and the resulting datasets are shared for reuse and reanalysis.

A second current concerns the reliability of the measures themselves. The many-analyst studies, in which independent teams analyse the same data, have shown that defensible analytic choices can move a conclusion, prompting a search for measurement procedures robust to who applies them and for reporting standards that make analytic flexibility visible (Nosek et al., 2022). In parallel, the model-based tradition continues to consolidate, with retrospectives cataloguing where sequential-sampling and related models are secure and where their flexibility still outruns the tests that constrain them (Ratcliff et al., 2016). The common thread is a discipline turning its measuring apparatus on itself, treating the reliability and validity of its own techniques as an empirical question.

Criticisms and Open Questions

Each family of technique carries a characteristic vulnerability. The subtractive method assumes stages can be inserted without interaction, an assumption that is frequently false and that motivated the additive-factors method and the process models that followed (Sternberg, 1969). Verbal reports are reliable about the contents of attention but not about the causes of behaviour, so a technique that asks participants why they acted risks recording a confabulation (Nisbett & Wilson, 1977). Neuroimaging measures an indirect signal and invites reverse inference, reading a mental state off a physiological correlate that is neither specific nor unique to it (Logothetis, 2008). And computational models can be so flexible that fitting one well fails to establish its mechanism, so that distinguishing them requires experiments designed for the purpose rather than post-hoc fits (Ratcliff et al., 2016).

Above these specific worries sits the general one the reproducibility movement exposed: a technique can be sound in principle yet yield unreliable results in practice when the flexibility of its application is not controlled (Simmons et al., 2011; Open Science Collaboration, 2015). The open question is not whether psychology's techniques can measure the mind — a century of cumulative findings shows that they can — but how to apply them so that what they measure is stable, interpretable, and shared. That is a question about method and incentive as much as about instrumentation, and it is the one the field is now working out (Munafò et al., 2017).

Worked Example

The first demonstration reproduces a signal detection analysis. Model the internal evidence as two unit-variance normal distributions, noise centred at zero and signal-plus-noise centred at the sensitivity d′, with the observer responding affirmatively whenever the evidence exceeds a criterion placed at position λ on the evidence axis. Take a sensitivity of d′ equal to 1.5 and a criterion at λ equal to 1.0. The hit rate is the probability that signal-plus-noise exceeds the criterion, Φ of d′ minus λ, that is Φ of 0.5, about 0.69. The false-alarm rate is the probability that noise alone exceeds it, Φ of minus λ, that is Φ of minus 1.0, about 0.16. Recovering sensitivity from these rates, the z of the hit rate is 0.5 and the z of the false-alarm rate is minus 1.0, so their difference is exactly 1.5, the d′ we set. The criterion measure, λ minus half of d′, is 1.0 minus 0.75, or 0.25, a conservative bias — the observer requires more evidence than an unbiased placement would, which is why hits and false alarms are both comparatively low. Those are the numbers the demonstration reads out as its sliders move.

The third demonstration quantifies researcher degrees of freedom. Suppose a researcher can run a test in several defensible ways — comparing two dependent measures, adding participants and re-testing, dropping or keeping a condition — and reports the analysis that reaches significance. If each of m such analyses is an roughly independent test at the conventional five-percent level, the probability that at least one reaches significance by chance alone is one minus 0.95 raised to the power m. With a single analysis the false-positive rate is the nominal five percent. With four analyses it is one minus 0.95 to the fourth, about 18.5 percent. With ten it is one minus 0.95 to the tenth, about 40.1 percent, and with twenty about 64.2 percent. The nominal five-percent guarantee, in other words, evaporates the moment the choice among analyses is made after seeing the data, which is precisely why preregistration fixes that choice in advance.

Discussion

The techniques surveyed here are best understood not as competitors but as a division of labour across the chain from stimulus to response, each recovering a different quantity by a different logic. Chronometry reads the timing of stages off differences in latency; signal detection separates sensitivity from bias in a discrimination; psychophysics locates the threshold that relates stimulus to sensation; verbal report exposes the contents, though not the causes, of thought; process-tracing and physiological measures record the intermediate course of a process; and computational modelling turns a formal account into a set of measured parameters. Table 2 sets the principal families side by side. What unifies them is the founding move of the field: engineer a situation in which an unobservable operation is compelled to leave an observable trace, then supply a model that maps the trace back onto the operation.

Read as a whole, the history of psychological technique is a movement from timing and counting toward modelling, and, most recently, toward a reflexive concern with the reliability of the methods themselves. What has not changed is the founding conviction that the mind, though it cannot be observed directly, can be measured indirectly by anyone willing to engineer the right trace and reason carefully from it. The open problems are now problems of application — how to constrain flexible models, how to build measures that survive the analyst, and how to run studies whose results replicate — rather than doubts about whether the mind can be studied by technique at all.

Glossary

Absolute threshold.

The minimum intensity of a stimulus at which it can be detected, estimated in psychophysics as the stimulus value detected on a set proportion, conventionally half, of trials.

Additive-factors method.

Sternberg's chronometric technique in which factors affecting the same processing stage interact in their effect on reaction time while factors affecting different stages add, revealing stage structure without assuming pure insertion.

Criterion.

In signal detection theory, the point on the internal evidence axis above which an observer responds affirmatively; its placement is the observer's response bias, independent of sensitivity.

Difference threshold.

The smallest change in a stimulus that can be reliably detected, also called the just-noticeable difference; the basic unit of Fechner's psychophysical scaling.

Mental chronometry.

The use of reaction time to infer the timing and organisation of the mental operations intervening between a stimulus and a response.

Method of constant stimuli.

A psychophysical procedure in which stimuli of preset intensities are presented in random order many times, and the resulting proportions detected are fitted with a psychometric function to estimate the threshold.

Preregistration.

The practice of specifying a study's hypotheses and analysis plan in a time-stamped public record before the data are collected or seen, converting an exploratory analysis into a confirmatory test.

Process tracing.

A family of techniques — eye-tracking, information-board, mouse-tracking — that record the intermediate steps of a cognitive process as it unfolds, rather than only its final outcome.

Protocol analysis.

Ericsson and Simon's method of treating think-aloud verbalisations as data about the contents of working memory, valid when the reporting does not alter the task being performed.

Psychometric function.

The curve relating the probability of a given response to the physical intensity of a stimulus, whose location gives the threshold and whose slope gives the precision of discrimination.

Psychophysics.

The branch of psychology, founded by Fechner, that measures the quantitative relation between the physical properties of stimuli and the sensations they produce.

Reproducibility.

The degree to which a research finding can be obtained again, ranging from re-running the same analysis on the same data to reaching the same conclusion with newly collected data.

Researcher degrees of freedom.

The many undisclosed choices available in collecting and analysing data that, exploited after seeing the results, inflate the false-positive rate far above its nominal level.

Reverse inference.

The fallible practice of inferring the engagement of a mental process from the presence of a physiological or neural signal that is not specific to it.

Sensitivity (d′).

In signal detection theory, the distance between the noise and signal-plus-noise evidence distributions, measuring how well an observer can tell signal from noise independently of response bias.

Signal detection theory.

A framework modelling detection as a decision between overlapping evidence distributions, separating an observer's sensitivity from the criterion that sets response bias.

Subtraction method.

Donders' original chronometric technique, which estimates the duration of a mental stage as the difference between the reaction times of two tasks differing only by that stage.

Think-aloud protocol.

A verbal-report technique in which a participant continuously verbalises the contents of attention while performing a task, without explaining or interpreting them, for use as process data.

Weber's law.

The psychophysical regularity that the difference threshold is a constant proportion of the baseline stimulus intensity, so the just-noticeable difference grows in fixed ratio to the standard against which it is judged; the empirical premise Fechner integrated to obtain his logarithmic law.

Key Researchers

Franciscus Cornelis Donders (1818-1889). Dutch physiologist at Utrecht University; founded mental chronometry, using the subtraction of reaction times to estimate the duration of hidden mental stages. Wikipedia

K. Anders Ericsson (1947-2020). Swedish-American psychologist at Florida State University; with Herbert Simon, established protocol analysis, the disciplined use of think-aloud verbal reports as psychological data. ORCID - Google Scholar - Wikipedia

Gustav Theodor Fechner (1801-1887). German physicist and philosopher at Leipzig; founded psychophysics, defining the methods for measuring sensory thresholds and deriving the first quantitative law relating sensation to stimulus intensity. Wikipedia

Richard E. Nisbett (b. 1941). Professor of Psychology at the University of Michigan; with Timothy Wilson, documented the limits of introspective verbal report about the causes of behaviour. Faculty Page - Google Scholar - Wikipedia

Brian A. Nosek (b. 1973). Professor of Psychology at the University of Virginia and co-founder of the Center for Open Science; led the empirical study of reproducibility and the development of open-science methods. ORCID - Faculty Page - Google Scholar - Wikipedia

Roger Ratcliff. Professor of Psychology at The Ohio State University; originated the diffusion decision model, the technique that treats a fitted process model as a measuring instrument for the components of a decision. ORCID - Faculty Page

Saul Sternberg (b. 1933). Emeritus Professor of Psychology at the University of Pennsylvania; developed the additive-factors method and the memory-scanning paradigm, extending Donders' chronometric logic. ORCID - Faculty Page - Google Scholar - Wikipedia

John A. Swets (1928-2016). American psychologist; imported signal detection theory into psychology, giving the field a technique for separating sensitivity from response bias. Wikipedia

Wilhelm Maximilian Wundt (1832-1920). German physiologist and philosopher; founded the first experimental psychology laboratory at Leipzig in 1879 and made controlled introspection a laboratory method. Wikipedia

Frequently Asked Questions

What are psychological techniques?
They are the methods psychology uses to turn unobservable mental processes into measurable data, pairing a task that engages a process with a model that maps the observed behaviour onto the hidden quantity (Donders, 1868/1969).

Why does psychology need special techniques at all?
Because the mind cannot be observed directly, every finding depends on a technique that forces a hidden operation to leave a trace in something observable, such as a response time, an accuracy, or a physiological signal (Donders, 1868/1969).

What is mental chronometry?
It is the technique of reading the timing and organisation of mental stages off reaction times, originally by subtracting the time of a simpler task from that of a task containing an extra stage (Sternberg, 1969).

What does signal detection theory add to a measure of accuracy?
It separates an observer's sensitivity, how well they can tell signal from noise, from their response bias, how willing they are to respond affirmatively, two things that raw accuracy confounds (Swets et al., 1961).

What does psychophysics measure?
It measures the thresholds and scales relating the physical intensity of a stimulus to the sensation it produces, including the absolute threshold for detection and the difference threshold for discrimination.

Are verbal reports reliable as data?
They are trustworthy about the current contents of attention, as in think-aloud protocol analysis, but unreliable about the causes of behaviour, which people often confabulate when asked (Nisbett & Wilson, 1977).

How is computational modelling a measurement technique?
Fitting a process model to behaviour yields parameters, such as the rate of evidence accumulation or the amount of evidence required, that are interpretable psychological quantities the raw data cannot separate (Ratcliff & McKoon, 2008).

Why did psychology adopt open-science methods?
Because many published effects failed to replicate and the flexibility of ordinary analysis inflates false positives, prompting techniques such as preregistration and open data to make results reliable (Open Science Collaboration, 2015; Simmons et al., 2011).

References