The history of psychometrics is the history of learning how to justify measurements of psychological attributes and educational performance. Its central problem is not simply how to assign a number. It is how to show that responses to particular tasks support a particular interpretation, for particular people, under particular conditions. A good historical account follows the changing evidence behind those claims, including the limits of the instruments and institutions that made them.
This guide follows a selected route through early twentieth-century testing, correlation, comparative judgement, professional organisation, reliability, item-based models and validity standards. It connects historical sources with original numerical examples so that the reader can inspect an argument rather than memorise a procession of names. The examples are invented teaching cases, not reconstructions of historical datasets or instruments for assessing an individual. For current methods in greater technical depth, continue to The Gold Standard of Psychometrics.
The wider route begins at The Gold Standard master and The Gold Standard of Education. The How X Works hub connects measurement to the systems that collect information and act on it. Readers interested in the school setting can also use History of Educational Psychology.
Choose your psychometrics history route
Chapters 1–3: Measurement, correlation and early testing
Chapters 4–6: Scales, reference groups and institutions
Chapters 7–9: Consistency, items and constructs
Chapters 10–12: Standards, access and transfer
Chapter index
Measurement, correlation and early testing · Chapters 1–3
1. Before the timeline: what has to be measured?
2. Spearman in 1904: making relationships part of the problem
3. Binet’s 1905 text: the purpose comes before the conclusion
Scales, reference groups and institutions · Chapters 4–6
4. Thurstone in 1927: comparisons become material for a scale
5. Norms and reference groups: a score needs a comparison story
6. The Psychometric Society, 1935–1936: a field needs places to argue
Consistency, items and constructs · Chapters 7–9
7. Cronbach in 1951: consistency becomes a question about test structure
8. Lord’s 1951 report: the questions help shape the scores
9. Cronbach and Meehl in 1955: a construct must face evidence
Standards, access and transfer · Chapters 10–12
10. Testing standards: evidence becomes an explicit obligation
11. Reading the difficult parts: classification, access and missing voices
Learning projects and source routes · Chapters 13–14
CHAPTER 1 OF 14 · Measurement, correlation and early testing
1. Before the timeline: what has to be measured?
A ruler can be placed alongside a desk. There is no equivalent operation that places a ruler alongside a person’s understanding, confidence or reasoning. We observe answers, actions or reports and infer something beyond those observations. Psychometrics studies the quantitative problems in making that move. The Psychometric Society’s account of the field places it within measurement in psychology, education and the social sciences; it is broader than administering intelligence tests.
The related History of Measurement follows physical standards such as length, mass and time. Comparing the two histories is useful precisely because their routes from observation to a justified quantity differ. A psychological score needs an argument about the attribute represented by the tasks as well as careful calculation.
Start with four distinguishable things. A construct is the attribute under discussion, such as understanding written arguments. A task is an occasion for showing relevant performance. A response is what the person actually does. A score is produced by a rule applied to that response. Confusing any two creates a shortcut in the reasoning. Correctly adding points does not establish that the tasks represent the intended construct.
An original example: the same total, different evidence
Imagine a ten-question reading exercise. Five questions ask learners to locate a stated detail and five ask them to explain an inference. Learner A answers all five detail questions and none of the inference questions correctly. Learner B answers two detail questions and three inference questions correctly. Both receive 5/10 under equal weighting. Their totals match; their observed response patterns do not.
It would be premature to declare stable personal strengths from only five questions in each category. Nevertheless, the example proves that a total can discard distinctions. A historian should therefore ask what an old score combined, what it suppressed, and what its users thought it meant. The answer may be different for a classroom exercise, a research scale and a selection procedure.
This also explains why no single invention can supply the whole history. A mathematical technique, a practical test, a professional journal and a rule for interpreting results solve different problems. Their development overlaps. The chapters that follow use dated sources as points of entry, without treating every later date as an automatic improvement or every earlier method as an error.
CHAPTER 2 OF 14 · Measurement, correlation and early testing
2. Spearman in 1904: making relationships part of the problem
Charles Spearman’s 1904 paper on general intelligence argued for studying correlations between performances rather than assuming that a laboratory task already represented a familiar mental faculty. Its introduction explicitly confronted the gap between relatively simple experiments and complex activities outside the laboratory. Read historically, this is a proposal about how psychological claims should be supported, as well as a contribution to theories of intelligence.
The important reading move is to separate a statistical pattern from its explanation. If people who perform well on one task tend to perform well on another, the association is evidence requiring interpretation. It does not, by itself, identify the cause, establish an immutable capacity or demonstrate that a task is suitable for every practical decision.
Worked calculation: perfect association can conceal a shift
Consider four fictional people with scores 1, 2, 3 and 4 on one occasion, and 3, 4, 5 and 6 on another. The means are 2.5 and 4.5. Subtracting each mean gives the same deviation sequence: −1.5, −0.5, 0.5 and 1.5. The sum of cross-products is 5, and each sum of squared deviations is also 5. The Pearson correlation is therefore 5/√(5 × 5) = 1.
Yet every score increased by two points. A perfect correlation records the preserved linear relationship; it does not show that the two occasions produced identical marks. Anyone reconstructing an old investigation from its correlation alone could miss the change in level. Conversely, a change in average marks need not mean that the ordering of people changed.
The repair is to ask for the means, spread, scoring rule and conditions alongside the association. If those details have not survived, report the limit rather than filling it with a convenient story. Our small calculation is not an estimate of a population relationship. It is a demonstration of what the statistic can and cannot logically establish, and why historical tables deserve to be read with their surrounding methods.
CHAPTER 3 OF 14 · Measurement, correlation and early testing
3. Binet’s 1905 text: the purpose comes before the conclusion
The 1905 text reproduced in Elizabeth S. Kite’s 1916 translation begins by delimiting its question. It addresses a child’s condition at the time of examination and explicitly sets aside explaining its origins or predicting future improvement. This is a valuable historical boundary: a test result about present performance is not automatically an account of cause or destiny. The text also contains terminology and classifications that are demeaning and should not be adopted as present-day descriptions of people.
Reading an early testing document responsibly requires both observations at once. A statement may show methodological restraint in one respect while reflecting assumptions and institutional arrangements that deserve criticism in another. Quoting only the restraint makes a falsely comforting history. Quoting only a shocking label can prevent the reader from understanding how the procedure actually worked.
A case of three different questions
Suppose a fictional learner cannot complete a set of instructions today. The observation is a failed performance under recorded conditions. Why did it happen? Perhaps the vocabulary was unfamiliar, the instructions were not heard, the learner was tired, or the relevant skill was not yet developed. Those are competing explanations requiring additional observations. What will the learner do after instruction next month? That is a further question requiring evidence about change.
A single present result cannot answer all three. A useful report would preserve the task, the response and the conditions, then distinguish what was observed from what remains to be investigated. It would not turn a descriptive score into a permanent identity.
For historical reading, make a two-column note: the author’s stated purpose and the claims actually made from the result. If a later institution used the instrument for a different purpose, investigate that later use separately. An inventor’s intention does not settle every consequence of adoption; nor does a later misuse establish that the original document made the same claim. This distinction helps connect the history of measurement with the educational settings discussed in History of Educational Psychology.
CHAPTER 4 OF 14 · Scales, reference groups and institutions
4. Thurstone in 1927: comparisons become material for a scale
Louis L. Thurstone’s A Law of Comparative Judgment, published in 1927, considered how comparative responses could support psychological scaling. His examples extended beyond physical intensities to judgements of specimens such as handwriting. The paper treated variability across judgements as something to model, rather than assuming that every comparison reveals a fixed, directly observable quantity. It also distinguished cases with different simplifying assumptions.
That broadens the historical question. Measurement need not begin with a person selecting a number from a printed scale. It can begin with repeated choices between alternatives. However, moving from proportions of choices to distances on a scale requires a model. Counting preferences and explaining their structure are different operations.
Worked case: a preference proportion is not a quality ratio
Imagine twenty independent judgements of two fictional instruction sheets, A and B. Fourteen favour A for clarity and six favour B. The observed proportion favouring A is 14/20 = 0.70. It is correct to say that A was selected in 70% of these judgements. It is not correct to say that A is 70% clearer, or that its clarity is fourteen-sixths of B’s.
Those ratios would need a defined quantity and a defensible scale with the relevant properties. A preference count supplies neither automatically. It also matters whether the twenty judgements came from twenty readers, one reader on twenty occasions, or a discussion group that reached twenty recorded decisions. The same numerical total can represent different dependencies.
Now change the instruction from “clearer” to “more complete.” A longer sheet might win completeness comparisons while losing clarity comparisons. The intended attribute must therefore be stated before interpreting the result. This is a small practical route into the historical problem of scaling: comparisons require a criterion, observations vary, and a mathematical representation inherits assumptions. A good account explains those steps instead of treating the appearance of decimals as the moment subjective experience became unquestionable fact.
CHAPTER 5 OF 14 · Scales, reference groups and institutions
5. Norms and reference groups: a score needs a comparison story
Historical score reports often become difficult to interpret because the original reference group has disappeared from view. The report may preserve a rank or standardised value while omitting who was tested, how they were selected and which version they completed. This chapter is a reading exercise about that missing context, not a claim that one person invented every form of norm-referenced measurement.
A percentile locates a score relative to a distribution under a specified convention. It is not the percentage of questions answered correctly. A standardised score expresses a location relative to a chosen mean and spread. Neither can be interpreted responsibly when the reference has been silently replaced by a different population or occasion.
Worked calculation: one raw mark, two reference distributions
Suppose a fictional score is 70. In reference group A, the mean is 60 and the standard deviation is 10. Its standardised distance is z = (70 − 60)/10 = 1. In reference group B, the mean is 80 and the standard deviation is 5. The same raw score has z = (70 − 80)/5 = −2. These calculations require only the stated means and standard deviations; they do not assume normal distributions.
The person has not changed between our two calculations. The comparison has. Nor have we shown that either reference group is appropriate. If one consisted of beginners and the other of experienced participants, the numerical contrast would be unsurprising. If the groups took different tests, an additional comparability problem would arise before any interpretation of their raw marks.
For an archive, record the reference group, date, sampling method, test form, scoring rule and any exclusions. If the manual reports only a summary distribution, distinguish what is recoverable from what cannot be checked. Do not translate an old rank into a modern personal judgement merely because the report uses familiar-looking numbers. The historical evidence may support a claim about an institution’s classification practice more strongly than it supports a claim about an individual’s enduring attributes.
CHAPTER 6 OF 14 · Scales, reference groups and institutions
6. The Psychometric Society, 1935–1936: a field needs places to argue
The Psychometric Society’s official timeline dates its first organisational meeting to 4 September 1935 in Ann Arbor, Michigan, during an American Psychological Association meeting. It records the first issue of Psychometrika in March 1936. The account describes the journal project as preceding the society: its organisers wanted an outlet for quantitative work in education and psychology. These dates mark institutional organisation, not the beginning of every measurement idea.
A journal matters because methods need more than isolated success stories. Someone must be able to specify an assumption, publish a derivation, identify an error and offer an alternative that other readers can inspect. A professional community gives those exchanges a recurring home. Publication, however, is an opportunity for scrutiny rather than a guarantee that every application is sound.
How to read an institution’s own history
An official timeline is strong evidence for what the institution records about its organisation. It is not an exhaustive account of the people affected by testing, the experiences of examinees or every competing tradition. A historian should match the source to the question. Meeting dates can be checked against organisational records; the effects of a testing programme require evidence from its implementation and participants.
Consider a fictional collection containing a journal article, a publisher’s catalogue and a school memorandum. The article may explain a method. The catalogue may show how a product was advertised. The memorandum may show how staff were instructed to use it. Even if all three mention the same test, they answer different questions. Combining them carelessly can turn an advertisement into scientific evidence or a research proposal into a statement of actual practice.
The useful repair is to label each document’s role before extracting a conclusion. Ask who wrote it, for whom, at what stage of adoption, and with which access to information. This keeps institutional history connected to scientific work without reducing the field either to famous researchers or to a list of organisations.
CHAPTER 7 OF 14 · Consistency, items and constructs
7. Cronbach in 1951: consistency becomes a question about test structure
The ETS publication record identifies Lee J. Cronbach’s Coefficient Alpha and the Internal Structure of Tests as a September 1951 article in Psychometrika, volume 16, pages 297–334. Its title captures an important distinction for the reader: evidence about the relationships among parts of a test concerns its internal structure. It does not by itself settle the meaning or appropriateness of every use of the total.
Internal consistency, stability across occasions and agreement between scorers answer different questions. A collection of similarly worded questions might move together while sampling a narrow slice of the intended domain. Two scorers might agree perfectly while both apply an unsuitable rule. Repeating a test might preserve the order of people while practice shifts everyone’s marks, as Chapter 2 demonstrated.
Original diagnostic: identify the source of disagreement
Two fictional markers score the same four responses. Marker A gives 2, 4, 6 and 8. Marker B gives 4, 6, 8 and 10. Their ordering is identical, but B consistently gives two more points. The average difference is (2 + 2 + 2 + 2)/4 = 2. If a decision threshold is 7, the third response falls below it for A and above it for B.
The immediate diagnostic is a difference in scoring level. Adding more items that behave similarly would not directly resolve that difference. The scorers need to examine the interpretation of the criteria and examples around the decision boundary. A later check should use fresh work; agreement reached while discussing the original four responses is not an independent demonstration that agreement will persist.
This example is deliberately small and descriptive. It does not estimate a general reliability coefficient. Its purpose is to make the question visible before selecting a statistic. For a practical continuation, How Two Teachers Can Mark the Same Work Differently explains rubric interpretation and scoring drift. The modern Psychometrics guide develops the distinctions among reliability, internal consistency and measurement error.
CHAPTER 8 OF 14 · Consistency, items and constructs
8. Lord’s 1951 report: the questions help shape the scores
Frederic M. Lord’s 1951 ETS report, A Theory of Test Scores and Their Relation to the Trait Measured, examined how item properties affect score distributions and interpretations. Its abstract explicitly conditions the theory on assumptions and restrictions. The historical importance of this question is easy to miss when a score is treated as if it were produced by the person alone. A score also depends on the tasks and scoring procedure through which performance becomes visible.
This is one entry into the development of model-based test theory, not a claim that a single report invented all item response models. Nor should the report date be confused with the date of every later publication associated with Lord. Bibliographic precision prevents a familiar name from becoming a substitute for identifying the actual document being discussed.
Worked case: easier items can lift a total without demonstrating learning
Imagine two fictional ten-item forms. For a specified group under fixed conditions, suppose each item on Form A has a probability of a correct answer of 0.8, while each item on Form B has probability 0.5. With one point per correct answer, the expected totals are 10 × 0.8 = 8 and 10 × 0.5 = 5. Linearity of expectation gives these totals even without assuming that responses are independent.
If someone compares an average of 5 on B with an average of 8 on A and announces a three-point learning gain, the conclusion outruns the example. We constructed the difference through task probabilities. A claim about change in proficiency needs a design that addresses differences between forms, groups and administrations.
The repair is not to declare every comparison impossible. It is to establish a justified connection between the measurements. How Item Calibration Works explains how response data support estimated item parameters. How Anchor Item Selection Works explains why shared questions need stable functioning to support a connection between forms. These modern mechanisms make the historical concern concrete without pretending that a toy expected-value calculation performs an actual calibration.
CHAPTER 9 OF 14 · Consistency, items and constructs
9. Cronbach and Meehl in 1955: a construct must face evidence
In Construct Validity in Psychological Tests, published in 1955, Lee J. Cronbach and Paul E. Meehl developed an account of interpreting tests in relation to theoretical constructs. They placed their discussion in the context of earlier committee work and distinguished their own elaboration from an official committee position. The paper is useful for seeing validation as an argument connecting theory and observations, rather than a label conferred by one favourable result.
For a reader, the practical challenge is to state what should happen if an interpretation is useful and what finding would make it doubtful. Merely renaming a score after the construct does not meet that challenge. Calling a questionnaire “persistence” does not establish that every person with a high total persists in meaningful activities outside the questionnaire.
An original case: two interpretations compete
Suppose a fictional six-question form asks students how often they continue with difficult work. Interpretation A is that the total reflects persistence across learning situations. Interpretation B is that it mainly reflects willingness to describe oneself favourably in this setting. Both could produce plausible-looking responses, so the form’s wording alone cannot choose between them.
A useful investigation would specify additional observations before seeing their results. Does the score relate to sustained work when tasks become difficult? Does the relationship remain when reading demands are reduced? Does a change in anonymity alter responses? Each observation addresses part of the argument, with its own limitations. Behaviour on a single task could also reflect interest, prior knowledge or the time available.
If the predicted relationship fails, several explanations remain: the score interpretation may be wrong, the outside observation may be poor, or the prediction may have overlooked a condition. The response should be a more precise investigation, not an automatic rescue that makes the claim impossible to challenge. Historically, this way of reading a paper shifts attention from the prestige of a test name to the structure of the evidence. For current applications, it also prevents a correlation with a convenient outcome from silently becoming a claim about an entire personality.
CHAPTER 10 OF 14 · Standards, access and transfer
10. Testing standards: evidence becomes an explicit obligation
The official Testing Standards site states that the American Educational Research Association, American Psychological Association and National Council on Measurement in Education have published the standards collaboratively since 1966. The 2014 edition, available through the official site, explicitly requires clarity about intended score interpretations, uses and populations in Standards 1.0–1.2. This is a dated edition used here as a historical and methodological source, not a claim that standards never change.
The change in viewpoint matters. “This is a respected test” is a statement about reputation. “This score supports this decision in this setting” is an evidence claim. A historian studying a testing programme should look for the documents and practices connecting those two statements, including revisions, challenges and restrictions on use.
One instrument, two proposed uses
Consider a fictional vocabulary exercise developed to identify words that need further classroom practice. A teacher uses the responses to plan a lesson. Later, an administrator proposes using the same total to decide access to an advanced programme. The second decision is more than a larger version of the first. It makes a different claim about what the score can support and carries different consequences.
The original exercise might contain repeated examples from a recently taught list, be sensitive to that lesson’s wording, or be too short to distinguish closely spaced results. Those features could be acceptable for a quick instructional check while making the new use difficult to defend. The appropriate repair is to investigate the proposed use, its evidence requirements and alternatives before treating the inherited score as sufficient.
A historical account should therefore trace uses, not merely instruments. When a test travels from a research setting to a school system, or from one language to another, ask what changed in the supporting argument. Standards can make obligations explicit, but their existence alone does not show that every organisation fulfilled them. Evidence about implementation remains necessary.
CHAPTER 11 OF 14 · Standards, access and transfer
11. Reading the difficult parts: classification, access and missing voices
Psychometric history concerns people who were measured as well as methods that were developed. The early Binet text discussed in Chapter 3 provides a direct example of why source language needs scrutiny. Its classifications should be described as historical categories, with attention to their function, rather than repeated as neutral truths about the people classified. Historical vocabulary can reveal an institution’s assumptions without making those assumptions acceptable today.
Avoid two shortcuts. One treats a numerical method as automatically fair because everyone received a score. The other treats every observed difference as proof of one particular bias. Neither explains how the result arose. A careful investigation needs the task, access conditions, sampling, scoring and decision process, together with evidence about the affected people.
Original case: identical instructions, different access
Imagine a fictional reasoning task presented in a language that some participants have only recently begun learning. A uniform script means the words were the same. It does not establish that the task placed the same language burden on everyone. If the intended attribute is reasoning with a diagram, difficult verbal instructions may obstruct the intended observation. If the intended attribute is comprehension in that language, the language demand has a different role.
The first repair is to define the intended interpretation. Next, examine whether the task elicits the relevant performance and whether a change in presentation preserves it. Simply deleting every challenging word could alter what a language assessment measures. Simply retaining the original wording could preserve an irrelevant barrier in another kind of assessment. The justification depends on the purpose.
For a historical archive, an absent participant account is an evidence gap, not permission to invent a reaction. Administrative records can show that a classification was assigned while saying little about how the person understood or contested it. Record the gap and seek other documents where available. This keeps the history critical without substituting speculation for testimony, and reminds readers that a technically sophisticated measurement system still requires accountable decisions about its use.
CHAPTER 12 OF 14 · Standards, access and transfer
12. Changed-context check: moving a paper test to a screen
Use the preceding history to solve a fresh problem. A fictional organisation has a paper-based reasoning exercise and moves the same questions to a tablet. The new version adds a visible countdown, permits scrolling and changes how answers are selected. Its designers want to reuse the old score interpretation without further investigation because the words and answer key are unchanged.
The relevant historical lesson is that measurement consists of more than text plus arithmetic. Spearman’s relationship problem, Thurstone’s attention to judgement conditions, Lord’s focus on item properties and the validity discussions all direct attention to the evidence connecting performance with interpretation. This is an application of those questions, not a claim that the historical authors studied tablet testing.
Work the case before reading the answer
Suppose sixty fictional participants complete both versions in a deliberately simplified comparison. Thirty take paper first and thirty take tablet first. In the paper-first group, mean scores are 20 on paper and 23 on tablet. In the tablet-first group, means are 20 on tablet and 23 on paper. All versions use the same maximum score and scoring rule. What do the averages show, and what do they leave unresolved?
Across the two equal-sized groups, the paper mean is (20 + 23)/2 = 21.5 and the tablet mean is also 21.5. In both orders, the second occasion is three points higher. The descriptive pattern is compatible with an order or practice effect. It does not show an average mode difference in this balanced example. It also does not establish equivalence for every participant or every item.
The correct next move is to inspect individual differences, response patterns, timing and relevant access conditions, and to distinguish practice from other changes between occasions. A formal study needs an appropriate design, adequate sample and uncertainty analysis beyond these four means. If one subgroup struggled with scrolling while another benefited from it, equal overall averages could conceal the difference.
The transfer check is passed when the reader resists both convenient conclusions: “same words, therefore same measurement” and “new technology, therefore invalid.” The supported conclusion is narrower: the proposed interpretation must survive the changes in the response process. The Learning Analytics guide offers a related route into interpreting digital traces without confusing recorded activity with learning itself.
CHAPTER 13 OF 14 · Learning projects and source routes
13. A practical route for learning the history
Begin with one primary document and one question. Spearman’s 1904 introduction is useful for asking what connects laboratory performances with broader claims. Binet’s 1905 text, in the identified 1916 translation, is useful for separating present observation from explanations and predictions. Thurstone’s 1927 paper is useful for asking what permits a set of comparative judgements to become a scale. Reading all three as if they were chapters of a single modern textbook would erase their different purposes.
For each document, make a short record of the publication date, the edition or translation consulted, the intended problem, the observations used and the inference attempted. Keep a separate line for your own interpretation. This simple separation prevents a modern explanatory gloss from being presented as an author’s explicit position.
A small project with a clear deliverable
Choose a claim from this article and produce a one-page evidence note. For example: “A correlation of one does not imply equal scores.” Include the four pairs from Chapter 2, calculate both means, show the centred deviations and state the exact conclusion. Then explain why an archive preserving only the correlation would be incomplete for a question about score change. This combines calculation with source criticism without requiring access to confidential records.
A second project can compare the roles of three documents: a research article, an official institutional timeline and a set of professional standards. Identify one question each can answer well and one it cannot answer alone. The deliverable is a map of evidence, not a ranking of which kind of document is universally best.
For present-day mechanisms, move from this history to Psychometrics, then to the specialist articles on item calibration, anchor selection and scoring agreement. Return to the historical document after learning the mechanism. The rewarding question is what the author actually established with the available evidence, rather than whether the document contains all the vocabulary used today.
CHAPTER 14 OF 14 · Learning projects and source routes
14. Frequently asked questions and source routes
Is psychometrics just the history of IQ tests?
No. Intelligence testing is an important part of the history, but psychological and educational measurement also includes judgements, attitudes, achievement, score structure and the relationships between observations and theoretical attributes. This guide’s route is selective. It is not a complete international history of every testing tradition, instrument or application.
Did the field begin when the Psychometric Society was founded?
No. The 1935 founding is an organisational milestone. The earlier papers discussed here demonstrate that measurement problems and methods preceded it. The society and its journal helped provide a continuing setting for quantitative work; they did not create every antecedent question.
Does a reliable score tell the truth about a person?
Reliability evidence concerns specified forms of consistency or precision. It does not settle every interpretation of the score. Our marker example showed how preserved ordering can coexist with a level difference that affects a threshold decision. The meaning and use of a result require their own justification.
Can an old score be compared directly with a new score?
Only when the relevant connection is defensible. Identical-looking numbers may come from different tasks, reference groups, scales or conditions. Look for the actual comparison design and its limits. A date on the report and a familiar test name are not enough.
Why use historical texts with objectionable terminology?
They are evidence of how authors and institutions framed their questions. Read them critically, identify their language as historical and avoid adopting it to describe people today. A source can be examined for a particular methodological argument without endorsing its classifications or social assumptions.
Where should I read next?
The primary-paper route runs through Spearman, 1904, Binet, 1905, in Kite’s 1916 translation, Thurstone, 1927 and Cronbach and Meehl, 1955. For the two 1951 milestones, the ETS records identify Cronbach’s alpha article and Lord’s test-score report. Those records are bibliographic and abstract-level sources here; they are not substitutes for reading the full technical works.
For institutional context, use the Psychometric Society timeline. For the standards discussed in Chapter 10, use the official Testing Standards site and its linked editions. Follow the Education parent to connect measurement with learning that can be used beyond the original task.