The history of learning research explains how experience came to be studied through observable change: acquiring a response to a signal, changing an action after its consequences, finding a route, or reproducing something seen in another person. Its central practical lesson is that better performance today is evidence to investigate, not a complete explanation of what has been learned. A useful account specifies the experience, the behaviour measured and the conditions under which the change appears.
This guide follows selected experimental milestones in the study of learning, rather than attempting a complete history of education or of every culture’s teaching traditions. Pavlov, Thorndike, Tolman and Bandura’s research team asked different questions with different methods. Understanding those differences helps us avoid turning their work into slogans such as “just repeat it”, “just reward it” or “just watch an expert”. The numerical examples below are original teaching cases, not reported historical results.
Read the historical chapters first, then use the practical comparisons and transfer exercises to test your understanding. The present-day interdisciplinary field is covered in The Gold Standard of Learning Sciences. This historical root connects with The Gold Standard collection and Education: learning that transfers into life.
Choose your learning history route
Chapters 1–3: Define learning, actions and consequences
Chapters 4–6: Study signals and learned relationships
Chapters 7–9: Test routes, models and decisions
Chapters 10–12: Interpret results and transfer
Chapter index
Define learning, actions and consequences · Chapters 1–3
1. Learning needs evidence of change and an account of experience
2. Thorndike brought repeated performance into the comparison
Study signals and learned relationships · Chapters 4–6
4. Pavlov made signals and responses experimentally visible
5. Prediction requires a comparison with what happens without the cue
6. Tolman questioned whether immediate performance shows all learning
Test routes, models and decisions · Chapters 7–9
7. A route can work without revealing a general map
Interpret results and transfer · Chapters 10–12
10. Measure speed and accuracy together
Study the history and sources · Chapters 13–14
CHAPTER 1 OF 14 · Define learning, actions and consequences
1. Learning needs evidence of change and an account of experience
A novice completes a digital form slowly on Monday and quickly on Friday. Something has changed, but several explanations are possible. The novice may have learned the procedure. The form may have become shorter. A colleague may now supply the answers. The Friday task may contain easier cases. Faster performance is the observation; learning is one possible explanation requiring a closer account.
Begin by recording what remained constant and what changed. Was the person using the same interface? Did the task require the same decisions? Was help available on both occasions? Were errors counted alongside completion time? Without this information, a comparison can reward an easier environment while attributing the improvement entirely to the learner.
A practical definition for this guide
Here, learning means a change in capability attributable to experience. Capability is inferred through suitable tasks; it is not directly visible as a single number. A successful response provides evidence, but we usually need more than one occasion and more than one arrangement to understand how dependable the change is. This is a working definition for evaluating ordinary learning claims, not a claim that every historical researcher used identical terminology.
Consider a trainee who completes a procedure correctly while reading a checklist. The performance may demonstrate effective use of the checklist, which can itself be valuable. It does not automatically demonstrate unaided recall of the procedure. If the real work permits the checklist, removing it from the assessment may create an irrelevant obstacle. If unaided action is essential, leaving it available may conceal a gap.
The first question is therefore not whether help is good or bad. It is what capability the task requires and what evidence would show that capability. Historical experiments become clearer when read in this way: each arrangement defines a particular opportunity to learn and a particular opportunity to demonstrate the result. The apparatus, instructions and scoring are part of the argument, not incidental background.
CHAPTER 2 OF 14 · Define learning, actions and consequences
2. Thorndike brought repeated performance into the comparison
Edward Thorndike’s animal-learning experiments examined changes across repeated encounters with problem situations, including escape tasks. His 1911 account of the experimental work challenged explanations based chiefly on impressive anecdotes. The historical contribution was to record performance over trials and ask how successful actions became established, rather than inferring a complex thought process from one striking success.
This does not mean that every later learning problem should be reduced to an escape task. The record was evidence about animals, apparatus and procedures used in that research. An educational application requires further reasoning and testing. Thorndike’s own account also contains qualifications that disappear when his position is compressed into a one-line rejection of all animal thought.
Read a performance series before naming the mechanism
Imagine an invented sequence of completion times: 90, 75, 72, 48, 51 and 35 seconds. The overall direction is improvement, but the sequence is not perfectly smooth. From the first to the last trial, time falls by 55 seconds. Relative to the initial 90 seconds, that is approximately a 61.1% reduction. Calling the person “61.1% more intelligent” would substitute a different construct for the measured quantity.
An observer might propose several explanations. The learner may identify an effective action, become physically faster, stop attempting unhelpful actions or learn where a control is located. The times alone do not distinguish all of these. A more informative record could include the actions attempted and the errors made before success.
Now imagine that one trial succeeds very quickly by chance, followed by several slow trials. The single success is not strong evidence of dependable mastery. Conversely, one slow trial after several reliable successes does not erase the entire learning history. Examine the pattern and the conditions, rather than selecting whichever trial supports the preferred story.
For a practical exercise, record five attempts at a harmless new software task. Note time, errors and assistance separately. The purpose is not to rank yourself. It is to discover which aspect of the task changes with practice and which remains unresolved.
CHAPTER 3 OF 14 · Define learning, actions and consequences
3. Consequences and information are not the same thing
A response can be followed by a consequence that makes it more likely to recur. A learner can also receive information that helps them understand an error. These functions may occur together, but they should not be collapsed into one vague idea of feedback. A cheerful sound after submission may encourage another attempt without explaining which part of an answer is wrong.
Use an original example involving a sorting task. A learner places objects into two categories and sees only “correct” or “incorrect”. Over repeated trials, the learner might discover the rule. Alternatively, the learner might memorise the category of each displayed object. Both routes can improve performance on the same examples while producing very different results on new ones.
Ask what the consequence actually changes
Suppose a system awards points for answering quickly. The learner becomes faster but makes more mistakes. The system may be shaping the behaviour it rewards while failing to improve the capability its designer intended. If reliable classification is the goal, speed alone is an incomplete target. The scoring arrangement should make accuracy visible and preserve enough time for the relevant judgement.
Now suppose the system gives the correct category and explains the decisive feature. The experience has changed again. The learner receives explanatory information, not merely a consequence after an action. If later performance improves, the explanation may have contributed. It would be misleading to attribute the whole improvement to points or praise when several elements changed together.
This distinction matters in classrooms, training and product design. “We provided feedback” is not a sufficiently precise description of what happened. State whether the learner received a result, an explanation, a worked correction or an opportunity to revise. Then assess the effect on the intended performance.
The existing Gold Standard of Feedback develops the practical cycle of information, correction and a new attempt. In historical reading, keep the terminology attached to the actual procedure. Similar words can conceal different experimental operations, and different operations cannot be compared responsibly as though they were identical treatments.
CHAPTER 4 OF 14 · Study signals and learned relationships
4. Pavlov made signals and responses experimentally visible
In his 1927 lectures, Ivan Pavlov described investigations of conditioned reflexes. Lecture II includes a demonstration in which a metronome sound elicited salivation after the relevant experimental history. It also explains efforts to control unintended stimuli from the environment and experimenter. The historical point is more precise than the familiar story of a bell: an initially different event could acquire a signalling function under specified conditions.
The lecture is a historical physiological account, not a complete modern theory of learning. Its terminology and explanations should be read in their period. The demonstration concerns a measured response and a controlled arrangement; it does not establish that complex human understanding is simply the same response written on a larger scale.
Separate the cue from the event it predicts
Consider a harmless illustrative example. A display changes colour shortly before a delivery becomes available for collection. A worker begins checking the collection point when the colour changes. The cue, the delivery and the checking behaviour are three distinct elements. To investigate the role of the cue, record their timing rather than merely noticing that they often occur together.
If the colour changes at the same moment as a loud announcement, either event could be guiding the response. If the experimenter always turns towards the collection point just before the cue, that movement could also matter. Removing ambiguity requires varying the arrangement carefully and checking what information is actually available.
A useful source-reading exercise is to identify the stimulus, measured response, prior experience and competing cues in a historical demonstration. Do not begin by assigning the whole event to a named theory. Begin with the sequence that another reader could reconstruct.
The lesson for ordinary instruction is modest but valuable: learners may respond to signals the instructor did not intend to teach. A heading colour, an adult’s glance or the position of an answer on a page can become informative. Successful performance under those cues may not survive when the cues disappear.
CHAPTER 5 OF 14 · Study signals and learned relationships
5. Prediction requires a comparison with what happens without the cue
Two events occurring together does not tell us whether one provides useful information about the other. Suppose a training display shows a symbol twenty times, and a target event follows sixteen of those appearances. The target also occurs sixteen times during twenty comparable periods without the symbol. The target is frequent, but the symbol adds no difference in these observed rates.
The first conditional rate is 16/20, or 80%. The second is also 80%. Their difference is zero percentage points. A learner could encounter many cue–target pairings while the cue remains uninformative relative to the background arrangement. This is an illustrative probability comparison, not a claim about the exact design of Pavlov’s demonstration.
Change the background and the interpretation changes
Now imagine sixteen target events after twenty cue appearances, but only two target events in twenty comparable no-cue periods. The observed rates are 80% and 10%, a difference of seventy percentage points. The cue now distinguishes conditions in the recorded sample. That arithmetic does not prove a particular internal learning process, but it describes a meaningful feature of the information available.
The comparison depends on how opportunities are counted. A ten-second cue period cannot casually be compared with an hour without the cue. Longer periods provide more opportunities for an event. Define comparable windows or use an appropriate rate, and state the choice. Otherwise the denominator can manufacture an apparent association.
For a practical example, consider a notification that appears before nearly every important update but also appears constantly when nothing useful has changed. Its usefulness depends on both kinds of occasion. Counting only successful alerts makes the notification seem more informative than it is.
This chapter supplies a way to question claims about learning signals. Ask what happens when the proposed signal is absent. Ask whether other signals are present. Ask whether the observation windows are comparable. These questions do not replace an experiment, but they prevent a vivid pairing from being mistaken for a complete account of the learner’s experience.
CHAPTER 6 OF 14 · Study signals and learned relationships
6. Tolman questioned whether immediate performance shows all learning
Edward Tolman’s 1948 paper on cognitive maps discussed maze research and argued for a richer account of learned environmental relationships. In reviewing latent-learning work, he credited Blodgett’s earlier experiments and described related work with Honzik. A central issue was that experience could have effects that became clearer after the conditions for performance changed.
The paper’s map language was a theoretical interpretation of behavioural evidence, not a photograph of a map inside a rat’s head. Tolman also distinguished narrow route-like learning from broader representations that might support altered routes. The historical importance lies in asking what was acquired during experience and what conditions allowed it to be expressed.
A human illustration without pretending it is the same experiment
Imagine a visitor who walks through a public building several times while accompanying a friend. The visitor is not asked to find any particular room. Later, the visitor is asked to locate an exit and succeeds quickly. Earlier wandering may have supplied useful information, even though it was not followed by an explicit test of route knowledge.
That example does not prove that all unstructured exposure produces useful learning. The visitor may have followed signs at test, remembered one distinctive landmark or received unnoticed help. To investigate prior exposure, compare appropriate conditions and document what information is available during the final task.
It also does not imply that a poor performance always conceals excellent knowledge. Sometimes the required knowledge was not acquired. The distinction between learning and performance is a reason to investigate, not a licence to assume an invisible success whenever the visible result is disappointing.
A practical diagnostic can vary one condition at a time. Offer a clearer goal, remove an irrelevant time pressure or ask for a different response format, then observe what changes. State exactly what support was added. A better result under support demonstrates capability in that supported condition; whether it transfers elsewhere remains a separate question.
CHAPTER 7 OF 14 · Test routes, models and decisions
7. A route can work without revealing a general map
Suppose a learner knows a sequence of turns from a main entrance to a room: left, right, straight ahead. The learner reaches the room reliably from that entrance. When starting from a side entrance, the same sequence leads elsewhere. This is an original illustration of the difference between a learned route and knowledge of relationships that supports a changed starting point.
The failure is not evidence that the original route was never learned. It shows that the tested capability was narrower than someone assumed. A route can be useful for a stable task. Problems arise when a narrow success is advertised as broad flexibility without testing the changed conditions.
Construct a meaningful transfer test
Begin with a simple diagram containing four named locations and two connecting corridors. Ask the learner to reach a destination from one start. Then change the start while preserving the layout. Finally block one corridor and ask for an alternative route, if one exists. These are different demands, so report the results separately rather than combining them into a single unexplained score.
For invented results, suppose a group succeeds on eighteen of twenty familiar-start tasks and eleven of twenty changed-start tasks. Those are 90% and 55%, a thirty-five-percentage-point difference. The gap identifies a practical limitation in the tested setting. It does not identify a unique psychological mechanism; unfamiliar instructions or differences in task difficulty might also contribute.
A repair could include comparing routes from several starts and explaining the relationships between landmarks. After practice, use new route questions. Repeating the exact failed question until it is memorised may remove that particular error without developing the wider capability you intended to test.
This distinction reaches beyond navigation. A spreadsheet user can memorise where a button sits without understanding which operation it performs. A reader can recognise a familiar example without identifying its underlying rule. The historical debate invites a precise question: what changes can the learner handle, and which changes break the demonstrated success?
CHAPTER 8 OF 14 · Test routes, models and decisions
8. Observational learning changed the unit of experience
Bandura, Dorothea Ross and Sheila Ross’s 1961 study compared children’s behaviour after exposure to aggressive or nonaggressive adult models, with a control condition. The report describes later imitation resembling the aggressive model’s behaviour. It also describes a structured testing situation, including an intervening arousal procedure. Omitting those details turns a specific experiment into an oversimplified story.
The historical contribution was evidence relevant to learning through observing another person’s actions, rather than requiring the observer to perform and receive direct consequences for every action during exposure. The study does not, by itself, establish every long-term effect of media or demonstrate that all observed behaviour will be copied in daily life.
Use a constructive teaching example
Imagine an instructor assembling a harmless paper model while learners watch. Some learners see the final movements only. Others also hear the instructor explain which edges must align and how to detect an error. A later comparison would involve different demonstrations: visible action alone versus action with explanatory information. It would not isolate observation from all other influences unless the design addressed those differences.
A learner may copy the order of movements while missing the decision that makes the sequence work. For example, turning a sheet clockwise may succeed only because the instructor began with a particular orientation. If the learner’s sheet begins differently, faithful copying can produce an incorrect result. Demonstration should make the relevant starting conditions visible.
To check understanding, ask the learner to predict the next step and explain its purpose before carrying it out. Then change a harmless surface feature while preserving the underlying relation. This adds a diagnostic task; it is an application inspired by the broader question of observational learning, not a reproduction of the historical experiment.
Historical studies involving distress or aggression should be read as evidence to evaluate, not as classroom activities to imitate. The learning question can be explored through benign examples, clear consent and ordinary instruction without reproducing the original procedure.
CHAPTER 9 OF 14 · Test routes, models and decisions
9. Copying a result and learning a decision rule differ
An expert solves a problem on a board. A learner reproduces the same steps immediately afterwards. The reproduction may be an important beginning, but it leaves several questions unanswered. Can the learner identify when the method applies? Can the learner choose between it and a competing method? Can the learner explain what would make the method fail?
Consider an original example with two categories of documents. Category A requires one review; category B requires two. If every demonstration of category B uses a blue cover, a learner might adopt the accidental rule “blue means two reviews”. The intended rule may instead depend on the document’s content. Perfect performance on the training set would not distinguish those rules.
Design examples that separate the explanations
Present a blue category-A document and a plain category-B document. Ask for the decision and the reason. These examples are valuable because the intended rule and the accidental cue now predict different responses. A good diagnostic example is not merely harder; it distinguishes plausible accounts of what the learner has learned.
Suppose the learner gets both decisions right but explains them using another irrelevant feature. The result warrants another check. A correct answer can arise from a fragile rule, just as an incorrect answer can arise from a small execution error despite a sound decision. The explanation is additional evidence, not an infallible window into thought.
After diagnosis, revise the demonstration. Use varied covers while keeping the relevant content criterion explicit. Show a counterexample and explain why it receives a different treatment. Then test with new documents rather than asking the learner to recite the instructor’s wording.
This is one reason observation, practice and explanation can work together. Watching supplies a model; attempting supplies evidence of performance; feedback identifies a discrepancy; varied examples test the generality of the decision. No single element should receive credit for the whole sequence without a comparison that separates its contribution. The history of learning is useful when it sharpens those distinctions rather than replacing them with a favourite technique.
CHAPTER 10 OF 14 · Interpret results and transfer
10. Measure speed and accuracy together
A training programme reports that participants finish tasks faster after practice. That may be useful, but it is incomplete if errors increase. For some activities, a small speed gain with more mistakes is a poor exchange. For others, a measured reduction in hesitation may matter once accuracy is already reliable. The judgement depends on the task and its consequences.
Take an invented batch of ten routine classification tasks. Before practice, the learner takes twenty minutes and classifies nine correctly. Afterwards, the learner takes fifteen minutes and classifies seven correctly. Completion time falls by five minutes, or 25%, while accuracy falls from 90% to 70%. A claim of unqualified improvement would conceal the changed balance.
Choose the outcome before choosing the story
If the purpose is reliable classification, report accuracy first and speed alongside it. If the purpose is fluent performance after mastery, define a minimum acceptable accuracy and examine speed within that condition. Avoid selecting whichever measure makes the intervention look best after the results are known.
Even a composite score needs explanation. Assigning points to correct answers and subtracting points for time embeds a value judgement about the trade-off. Two scoring rules can rank the same performances differently. That does not make composite scores useless, but it means the weights should reflect the real task rather than a convenient headline.
For a personal learning record, keep separate columns for correct decisions, execution errors, completion time and assistance. You do not need a complex dashboard. A small, consistent record often reveals whether the bottleneck is understanding, selection or fluent execution.
Historical performance curves deserve the same care. A falling time measure may capture several changes at once. Interpreting it responsibly requires attention to what the apparatus permits and what the scoring omits. Measurement is not a neutral decoration added after the theory; it shapes which changes become visible and which remain outside the record.
CHAPTER 11 OF 14 · Interpret results and transfer
11. Diagnose the experience that is missing
When practice fails, the next step is not always more of the same practice. A learner may lack a clear model, an understandable rule, informative consequences, sufficient opportunities to act or a test that reveals the relevant capability. These possibilities call for different repairs. Treating them all as insufficient effort wastes time and can obscure an otherwise solvable problem.
Start with a concrete task. Ask the learner to describe the goal and identify the information needed. If that description is wrong, clarify the task before increasing repetitions. Next ask for an attempted solution with the usual tools available. Observe where the process breaks rather than judging only the final answer.
Four possible repairs
If the learner cannot see what a successful action looks like, provide a clear demonstration. If the learner copies the action but cannot choose when to use it, compare examples and counterexamples. If the learner understands the decision but repeatedly omits a step, practise the sequence with an appropriate check. If the learner performs only with one familiar cue, vary the cue while preserving the relevant condition.
These are practical design choices, not diagnostic labels for the person. Several may be needed together. Record what you changed so that later improvement can be interpreted honestly. If you change the explanation, examples, practice time and assessment simultaneously, you can evaluate the combined programme, but you cannot confidently attribute the result to just one component.
A useful repair also includes a new opportunity to perform. Explaining an error without allowing another attempt leaves the effect of the explanation untested. The learner may agree with the correction while still being unable to carry it out when the task returns.
The neighbouring History of Memory examines how retention, recall and relearning became measurable. Use that route when the difficulty appears after a delay. The present chapter concerns choosing an informative next experience. Together, they encourage a precise question: what can the learner do now, under what conditions, and what evidence would show that the repair lasts?
CHAPTER 12 OF 14 · Interpret results and transfer
12. Changed-context case: the training screen moves
A workplace trainee learns to process a routine request in a demonstration interface. The instructor always opens the same panel, selects the second option and clicks a green button. The trainee later completes twenty familiar requests with nineteen correct outcomes. The interface then changes: options appear in a different order, but their meanings remain the same. On twenty new requests, only twelve outcomes are correct.
The familiar-task accuracy is 19/20, or 95%. The changed-interface accuracy is 12/20, or 60%. The difference is thirty-five percentage points. This invented result suggests that performance depended partly on features of the familiar arrangement. It does not prove that the trainee learned nothing, or that all interface changes are equally difficult.
Separate the competing explanations
One explanation is that the trainee memorised positions rather than meanings. Another is that the new labels are unclear. A third is that the new requests contain harder cases. Before redesigning training, compare the content and inspect the errors. If wrong choices cluster at the old button positions, that observation supports the position-based explanation, though it may not be the only influence.
A practical repair begins by naming the decision each control represents. Demonstrate the same decision in two layouts. Ask the trainee to explain the chosen option before clicking. Include a request that cannot be completed because required information is missing. That last example checks whether the trainee understands when to stop, not just how to continue.
Then test new cases in a third arrangement. If success depends on seeing the instructor’s exact example again, the repair has not yet demonstrated the intended flexibility. If performance improves, report the particular changed conditions tested rather than promising unrestricted transfer.
The historical connections are now concrete. Signals can guide behaviour unintentionally; repeated actions can become efficient; observation can supply a model; knowledge may be expressed differently when conditions change. These are useful questions to bring to the case. They are not four competing slogans from which one must select a single winner before looking at the evidence.
CHAPTER 13 OF 14 · Study the history and sources
13. Build a historical explanation that can be checked
A strong account of learning history should distinguish an investigator’s observations, theoretical interpretation and later applications. The observations describe what participants or animals did under particular conditions. The interpretation explains why the investigator thought it happened. An application proposes what someone else might do with that insight in another setting. These layers can support each other without becoming identical.
Create an evidence card for one source. Record the original publication date, the version consulted, the task, the comparison, the outcome and one limitation. Then add a separate sentence beginning “A practical question this raises is…” That sentence makes the transition from historical evidence to present application visible and prevents an attractive modern recommendation from being falsely attributed to the original author.
Compare questions rather than ranking famous names
Ask what each method made visible. Repeated task performance made changes across experience observable. Controlled signals made the timing and identity of stimuli important. Maze comparisons raised questions about knowledge beyond an immediately rewarded response. Model-observation studies examined what could be acquired from another person’s actions. The methods overlap, but their measurements are not interchangeable.
Also identify what a source leaves out. A short laboratory task may not measure long-term everyday behaviour. An animal experiment does not directly establish the best way to teach a human subject. A human demonstration study does not automatically explain every kind of learning. These limitations delimit the conclusion; they do not make the evidence worthless.
For an independent study route, reconstruct one procedure without looking, then reopen the source to correct it. Draw a simple sequence of experience, test and interpretation. Change one feature of the proposed design and explain how that would alter what could be concluded. This tests understanding of the method rather than recognition of a scientist’s name.
The How X Works hub provides routes into related mechanisms and subjects. Choose a next article because it resolves a specific uncertainty: the role of feedback, the persistence of memory or the breadth of transfer. A useful reading path is organised around questions whose answers change what you can explain or do.
CHAPTER 14 OF 14 · Study the history and sources
14. Frequently asked questions and primary sources
Is learning the same as doing better during practice?
Not necessarily. Better performance can reflect learning, but it can also reflect easier tasks, extra help or changed incentives. Describe the experience and test the intended capability under appropriate conditions. A delayed or changed-context check can reveal limitations that immediate repetition does not show.
Did Pavlov use only a bell?
The primary lecture linked here includes a metronome demonstration and discusses different stimuli. The familiar bell shorthand hides the wider methodological issue: which event functions as a signal, what response is measured and which other cues must be controlled? Read the actual procedure before repeating the shorthand.
Does a reward prove that learning occurred?
No. A consequence can change observed behaviour, but the resulting capability still needs to be specified and checked. A learner may become quicker, more persistent or more likely to repeat one response without gaining the understanding required for a different task. State the intended outcome and measure it directly where possible.
Does latent learning mean that teaching is unnecessary?
No. The concept raises questions about knowledge acquired before it is clearly expressed. It does not establish that every exposure produces understanding, or that poor performance always conceals mastery. Instruction, motivation, cues and assessment conditions need to be examined rather than assumed away.
Is watching an expert enough?
It may provide useful information, but successful independent performance should be checked. A learner can copy a visible movement while missing the decision behind it. Demonstration, explanation, practice and feedback can be combined, with new examples used to test whether the learner can apply the relevant rule.
What should I be able to explain now?
Distinguish a signal from a consequence, and a consequence from explanatory feedback. Explain why a familiar route does not demonstrate unrestricted navigation. Calculate a change in accuracy without confusing percentage points with a relative percentage. Design a test where an intended rule and an accidental cue predict different answers.
Primary reading
Thorndike, Animal Intelligence, Chapter II, 1911 collection: examine the experimental approach, repeated-performance evidence and qualifications in the account.
Pavlov, Conditioned Reflexes, Lecture II, 1927: examine the metronome demonstration and the attention given to experimental control.
Tolman, Cognitive Maps in Rats and Men, 1948: distinguish the theoretical map interpretation from the maze studies discussed, including the work credited to Blodgett and Honzik.
Bandura, Ross and Ross, Transmission of Aggression Through Imitation of Aggressive Models, 1961: inspect the comparison conditions, testing procedure and limits of generalising the result.
Continue to Learning Sciences for the wider study of learning in real settings. Carry forward one concrete standard: describe the experience, check the capability and test what happens when the conditions change.