Psychological Experiments: From Classic Studies to the Digital Age

The experiments that built the field

Experimental psychology is less than 150 years old, and a surprising share of what is taught in introductory courses comes from a handful of studies. They are worth revisiting not only for their findings but for what they reveal about method: what was controlled, what was measured, and what later turned out to be wrong or unacceptable.

• Ebbinghaus (1885) memorised lists of nonsense syllables and tested himself at increasing delays, producing the forgetting curve and the spacing effect. A single-participant design, but with rigorous counting of repetitions and savings, and results that have replicated for over a century.
• Pavlov (1890s–1920s) showed that a neutral stimulus paired with food comes to elicit salivation on its own. Classical conditioning gave psychology a precise, measurable model of associative learning and the vocabulary of acquisition, extinction and generalisation.
• Stroop (1935) found that naming the ink colour of a colour word is slow when the word and colour disagree. The task is still one of the most widely used measures of cognitive control, and one of the easiest to run online.
• Asch (1951) asked participants to judge line lengths after a group of confederates gave obviously wrong answers. About a third of responses conformed to the majority, demonstrating normative social influence.
• Sperling (1960) flashed arrays of letters for 50 ms and used a tone to cue which row to report. Partial report showed that a brief iconic store holds far more than the four or five items people can report in full.
• Milgram (1963) instructed participants to deliver what they believed were increasingly painful shocks to a learner. Around two thirds continued to the maximum level, a result that reshaped thinking about obedience and about research ethics.
• Loftus & Palmer (1974) showed participants a filmed car accident and asked how fast the cars were going when they "smashed" or "hit" each other. The verb changed speed estimates and, a week later, false memories of broken glass, establishing the misinformation effect in eyewitness memory.

Ethical reckoning

Milgram's participants were deceived and visibly distressed; Zimbardo's Stanford prison study (1971) was stopped early after guards abused prisoners; and the Tuskegee syphilis study, though medical rather than psychological, was still running in 1972. These cases led directly to the Belmont Report (1979) and to the institutional review boards and ethics committees that now review every study with human participants. The principles are informed consent, minimisation of harm, the right to withdraw, and debriefing when any deception is used. Several classic studies could not be approved in their original form today, and replications of Milgram (Burger, 2009) had to cap the shock level and screen participants carefully.

The replication crisis

In 2015 the Open Science Collaboration reported the results of replicating 100 studies published in three leading psychology journals in 2008. Only about 36 % of replications produced a statistically significant result in the same direction, and effect sizes were on average about half of the originals. Social-psychology findings fared worse than cognitive ones. The causes were diagnosed as small samples, flexible analysis ("p-hacking"), selective publication of positive results and, in some cases, hypotheses formed after the data were seen.

Two developments have shaped the response. Preregistration, including Registered Reports where a journal accepts a study on the basis of its design before data collection, separates confirmatory from exploratory analysis and removes the incentive to adjust the hypothesis to fit the result. Online data collection attacks the sample-size problem: a study that once required a semester of undergraduate sign-ups can recruit several hundred participants in a day, and large, diverse samples make effect estimates more stable. Online experiments also make materials and code trivially shareable, which is what a replication needs.

What moves online well, and what does not

Not every classic translates. The table of winners and losers is fairly clear after a decade of web-based research.

• Transfers well: reaction-time and accuracy tasks with keyboard or mouse responses (Stroop, flanker, lexical decision, go/no-go, visual search); memory paradigms including the misinformation effect and most recognition and recall designs; judgement and decision-making tasks; questionnaires; and learning tasks with trial-by-trial feedback.
• Transfers with caveats: anything requiring brief displays under about 100 ms, such as Sperling's partial-report procedure, where display refresh and unknown monitor characteristics add noise; audio tasks, which depend on participants actually wearing headphones; and studies of very small effects between participants, where hardware variability adds unwanted variance.
• Does not transfer: studies that depend on live social pressure from confederates (Asch, Milgram), physiological recording, physical apparatus, or close experimenter observation. Some of these have web analogues, for instance conformity measured through displayed "previous responses", but they test a different question.

Running modern experiments on Cognition

Cognition is a hosting platform for jsPsych experiments, the open-source JavaScript library used for most browser-based behavioural research. You write or paste your experiment in a web editor, preview it, and share a link with participants; every trial is saved as it happens. Three features map directly onto the lessons above:

• Informed consent is configured per task in markdown and shown before the experiment starts, with a redirect URL for participants who decline.
• Balanced between-subject conditions are assigned on the server and exposed to your code as window.CONDITION, so a two-condition misinformation study ends up with equal groups without any extra logic.
• Data download in CSV or JSON, one file or a ZIP per participant, ready to deposit alongside a preregistration on OSF so that others can rerun your analysis.

If you want to see a classic in the browser, the Stroop demo and the lexical decision example are good starting points. Create a free account to host your own, or read the documentation.