The Intellectual History of HR and How the Past Informs the Future of Human Resources

“The most influential HR research” is a big article to write but as we thought about the trajectory of the past and what it means to our work on the future, it seemed a task necessary to tackle. So, we read fifty years of well cited peer-reviewed work on managing people the way an economist reads a market history; as a sequence of wrong assumptions, because each one changes beliefs thanks to a study that evaluated the practice of the time, established a durable finding, and built a new framework to make our collective work more meaningful.

For most of the past of human resource work, peer-reviewed research has studied two employment doors extensively. The hiring door (who should we let in, and what predicts whether they will perform) has produced some of the best replicated findings in all of social science. The exit door (who leaves, and why) produced decades of turnover models. Interestingly, the time between them, the months and years of daily behavior after the offer letter and before a layoff or resignation letter, has been theorized a lot and studied far less.

HR built a science of the front door and a science of the back door. Meaningful studies below are catalogued in the Behavioral Workforce Intelligence Research Library. We want to put together a history of HR and what we might learn from it to do better in the middle, providing intelligence about employees and the future of work.

What Caused Corrections

What we found we might call the Citation Fallacy; mistaking the volume of references a paper attracts (a popularity measure) for the magnitude of the error it corrected in practice (an impact measure). The root cause of this is that academic incentives pay for being cited. We wanted to wade through that to get a sense for what’s meaningful.

So, we asked better, whether a study did at least one of these things:

• Changed HR practice, meaning managers and HR teams now do something differently because of it.
• Established a empirical finding, one that survived replication and meta-analysis (a statistical method that pools many studies to estimate the true size of an effect).
• Created a major theoretical framework that later research organized itself around.
• Overturned a widely held assumption, including assumptions the field itself had spent decades building.

Run this way, it seems history organizes itself around problems rather than journals. Each era that asked a question, answered it better than the era before.

The 1970s Moved Motivation Out of the Person and Into the Job

Hackman and Oldham (1976) made job design an HR problem

Before J. Richard Hackman (then at Yale) and Greg Oldham (then at the University of Illinois) published “Motivation through the Design of Work: Test of a Theory” in Organizational Behavior and Human Performance, the default managerial assumption was that motivation lived inside the employee; some people had it, some didn’t, and HR’s job was to find the ones who did. Their Job Characteristics Model argued that five properties of the job itself (skill variety, task identity, task significance, autonomy, and feedback) produce the psychological states that drive motivation, satisfaction, and performance.

Blaming an employee for low motivation in a badly designed job is blaming a houseplant for wilting in a closet; you could swap it out for a hardier plant, but the smarter fix is opening the door and letting some light in.

The shift reframed HR’s question from “how do we motivate employees?” to “how do we design work so motivation can occur?” That reframe is the ancestor of everything we now call employee experience, job crafting, and meaningful work.

But the model also carried a detail managers routinely forget; Hackman and Oldham treated growth (how much a person wants to learn) as needing strength; the thinking, that the same job energizes one employee while overwhelming another because they can’t handle it. Job design and individual difference were never separable, and the field worked that way for about forty years before we really took people seriously.

The next year, William Mobley turned quitting into a sequence instead of an event

William Mobley’s “Intermediate Linkages in the Relationship Between Job Satisfaction and Employee Turnover” proposed that dissatisfaction does not convert directly into a resignation. It passes through intermediate steps; thinking about quitting, evaluating whether a search is worth the cost, searching, comparing alternatives, forming an intention, and only then leaving. Turnover became something HR could model rather than merely observe, and decades of research on retention, commitment, and replacement economics grew from that root.

Rodger Griffeth, Peter Hom, and Stefan Gaertner’s meta-analysis of turnover antecedents tested the predictors that grew out of Mobley’s work and found the steps closest to leaving (withdrawal cognitions and quit intentions) predicted turnover better than satisfaction did.

Mobley drew a map. Every step in his model happens between the two doors, and nearly every step is private and cognitive, which means the instruments HR actually owned (the application at one end and the exit interview at the other) could not see any of it. The field accepted a theory of the hallway between the doors but kept measuring the doorways, because the doorways were the only place where data was already being written down.

Modern HR software inherited the same blind spots for the same reason, which is the argument behind Your HR Platform Is a Coroner; It Should Be a Doctor.

Selection Became a Science at the Hiring Door, and Then the Science Corrected Itself

Schmidt and Hunter (1977, 1998) proved some hiring methods beat others everywhere

What seems to be the most consequential body of work on employee selection starts with Frank Schmidt and John Hunter’s 1977 validity generalization paper. Before it, the prevailing belief was “situational specificity;” a test that predicted performance at one company supposedly told you nothing about the next, so every employer had to validate every test from scratch. Schmidt and Hunter showed that most of the variation between studies was statistical noise (small samples, unreliable performance ratings, and restricted ranges of scores) rather than real differences between workplaces.

Their 1998 review, “The Validity and Utility of Selection Methods in Personnel Psychology,” compared nineteen selection procedures side by side and ranked general mental ability, work samples, and structured interviews near the top.

For two decades that work was the closest thing organizational psychology had to an assessment of hiring.

Then in 1991, Murray Barrick and Michael Mount showed personality predicts performance, conditionally

Murray Barrick and Michael Mount’s “The Big Five Personality Dimensions and Job Performance: A Meta-Analysis” established the Big Five (openness, conscientiousness, extraversion, agreeableness, and emotional stability) as a usable framework for work. Conscientiousness predicted performance across every occupational group they examined, while the other traits mattered only for specific kinds of work (e.g. extraversion helps a salesperson far more than an accountant).

Who a person is matters at work, but only in relation to what you are asking them to do. That sounds obvious in 2026; in the personnel selection understanding that preceded it, it was not, and this study is a large part of why personality measures are now a major component of hiring.

In 2022, found a scale had been miscalibrated for decades

This work audited its own statistics and Paul Sackett, Charlene Zhang, Christopher Berry, and Filip Lievens, showed that a statistical correction applied across decades of work had inflated estimates.

You only ever observe job performance for people you hired, so the spread of scores among employees is narrower than among applicants, and they uncovered that hiring was extremely limited by this bias. The revised rankings put more structured interviews first and lowered cognitive ability tests.

Here’s the problem from another example: Height barely predicts scoring among NBA players because every one of them is tall; we naturally stop at concluding height is irrelevant to basketball because we know it to be true but judging based on scoring would field a team likely to fail. Sackett’s team showed that when a similar correction (knowing more from applicants) has been applied, which is like adjusting for the NBA height filter, it opens the field of applicants without simply saying scoring is poorer because players are shorter.

Selection science measures hiring with pretty remarkable precision, but the process (usually one supervisor rating, taken once) compresses the expectations of the employees into that single snapshot. This, “Range restriction” is a problem; the people you never hired are invisible, and the people you did hire are visible only at the moment their performance as an employee is recorded.

The case for screening on how people operate rather than where they trained is laid out in The 5 Levels of Agency; Why Startups Should Predict Them Before They Hire.

An Employee Staying Has Three Different Causes; Satisfaction Explains Less of Performance but We Favored It

Back to 1991, “commitment” was split into three things that a retention report cannot tell apart

John Meyer and Natalie Allen’s “A Three-Component Conceptualization of Organizational Commitment” separated attachment to an employer into three forms:

• Affective commitment, where the employee wants to stay.
• Continuance commitment, where the employee needs to stay because leaving costs too much (a vesting schedule, a mortgage, a non-transferable skill set).
• Normative commitment, where the employee feels obligated to stay.

Picture three guests still at a party at midnight; one is having the time of their life, one’s ride left without them, and one would feel rude leaving.

A mere headcount at the door records three satisfied guests. A retention rate is that headcount, and it treats a hostage and a fan as the same data point, because it records that someone is present and nothing about why.

Low turnover is not evidence of a healthy workforce, because continuance commitment produces exactly the same retention number as affective commitment while producing the opposite behavior in the workforce (the person who needs to stay does the minimum, and the person who wants to stay does the extra thing nobody asked for).

Only behavior, observed over time, separates them.

Finally, we looked the Satisfaction-Performance link

Timothy Judge, Carl Thoresen, Joyce Bono, and Gregory Patton’s “The Job Satisfaction-Job Performance Relationship: A Qualitative and Quantitative Review” in 2001, concluded, “The mean true correlation between overall job satisfaction and job performance was estimated to be .30.”

It refuted the management claim that happy employees are dramatically more productive.

A later Texas A&M dissertation by Allison Cook, testing whether the relationship is spurious, controlled for personality and job characteristics and found the correlation to only be about .16, about half the original. It could be concluded that satisfaction and performance look linked because of the person’s traits and the job’s design (Hackman and Oldham again, forty years later). The argument between “satisfaction drives performance” and “satisfaction is irrelevant” was always a dispute of the shared cause of what a person is doing every day.

HR Practices Work as a System or They Barely Work at All

The Bundle Fallacy is treating HR practices as a menu of independent items whose effects add up (a training program, an incentive plan there, a team structure somewhere else), when the evidence says the effects of HR multiply, which means they can also be counter-productive in that a weak practice drags down the strong ones next to it. Buying a transmission, a set of tires, and an engine from three different cars and then wondering why you can’t drive is the Bundle Fallacy.

We’re Using the Car Analogy Because of John Paul MacDuffie’s 1995 study

“Human Resource Bundles and Manufacturing Performance” studied 62 automobile assembly plants around the world. HR practices (teams, training, contingent pay, employee involvement) mattered most as internally consistent bundles, and those bundles performed best when they matched the plant’s production system. A lean production line paired with a command-and-control HR bundle underperformed, because the line depended on workers solving problems the HR system discouraged them from touching.

This is very similar to what we see in startups when the management style of a small, idea stage team, must be very different than what happens as the venture matures through the work required to get to market, commercialize, and fundraise. Beyond pay, in particular given the nature of the work, the communication styles, micro-managing or expecting high agency, being a self-sufficient or directed team, are all HR styles evident through the founders before they manifest as policies through HR.

In 1995, Mark Huselid tied HR systems to the financial statements

In “The Impact of Human Resource Management Practices on Turnover, Productivity, and Corporate Financial Performance,” we find that Huselid surveyed nearly a thousand firms and found that systems of high-performance work practices were associated with lower turnover, higher productivity, and stronger corporate financial performance.

This paper put the effect in dollars:

“a one-standard-deviation increase in High Performance Work Practices yields a $27,044 increase in sales and a $3,814 increase in profits”

That number moved HR from an administrative cost center toward a reasonable argument of competitive advantage, and it is the foundation under modern people analytics, workforce strategy, and the idea of the CHRO as a business leader rather than a compliance officer. The case that organizations still choose to treat talent as overhead, despite thirty years of this evidence, is one made in Treating Talent as a Cost Center Is a Choice, Not a Constraint.

Eileen Appelbaum explained why the bundle works

Thomas Bailey, Peter Berg, and Arne Kalleberg, Eileen Appelbaum’s work in Manufacturing Advantage in 2000, drew on visits to 44 plants and surveys of more than 4,000 employees in steel, apparel, and medical electronics. They popularized the Ability-Motivation-Opportunity (AMO) framework, which organizes most strategic HR research:

• Ability, whether the employee has the capability to do the work well.
• Motivation, whether the employee has the incentive and willingness to apply it.
• Opportunity, whether the organization actually lets the employee use that capability (decision rights, team structures, access to information).

AMO explains the Bundle Fallacy. Performance behaves like a product of the three, so hiring brilliant people (ability) into a job with no decision rights (opportunity near zero) produces something close to nothing.

And in 2006, Becker and Huselid added a condition that led to the next changes

Patrick Wright and Gary McMahan had laid the theoretical groundwork for HRM in “Theoretical Perspectives for Strategic Human Resource Management,” defining is as the planned pattern of HR deployments and activities intended to help an organization achieve its goals.

That definition made fit the central question here.

Brian Becker and Mark Huselid’s “Strategic Human Resources Management” then pointed out that the practice is not inherently high performance; its value depends on fit with the organization’s strategy, workforce, capabilities, and competitive environment. The same equity allocated, high autonomy system that makes a startup hum would wreck a hospital’s medication dispensing unit, because the two businesses reward opposite behaviors.

Notice though that every study so far measured practices on one side (from HR policy surveys) and financial or operational results on the other (from annual reports and plant metrics), with almost nothing on the daily behavior connecting them.

The 1990s proved what’s in between mattered by measuring in and out of the workforce while inferring what happened inside, because the only data that existed then was written at policy reviews and quarterly closes, never during the workday.

Psychological Safety Gave “Culture” a Measurable Part, and Google Rediscovered It Fifteen Years Later

Amy Edmondson’s “Psychological Safety and Learning Behavior in Work Teams” did something HR had failed to do for decades; it took “culture,” a word executives use when they mean “the stuff we can’t explain,” and pulled out one component that could be defined, surveyed, and tested. The work defines team psychological safety as

“a shared belief held by members of a team that the team is safe for interpersonal risk taking”

In a study of 51 work teams at a manufacturing company, psychological safety was associated with learning behavior (asking for help, admitting errors, seeking feedback, discussing problems), and learning behavior translates to team performance.

Team efficacy, the team’s confidence in its own ability, did not predict learning once safety was controlled.

Confidence without safety produces teams that are sure of themselves and wrong in private, because the errors that would correct them never get said out loud.

This of a team without psychological safety as a room where everyone smells smoke and nobody pulls the alarm, because the last person who pulled it got asked to explain the inconvenience to management; the fire still happens, it just happens later and costs more.

Google’s Project Aristotle later studied its own teams and reached the same conclusion, ranking psychological safety first among the considerations that separate effective teams from the rest. The finding started rearranging how hiring should work, since it says that how a specific set of people work together matters more than who any one of them is individually, and the argument for applying that to your team is in The Best HR Advisors Already Know How to Build a Team. Predictive Intelligence Tells Them Who Fits It.

Psychological safety is in the workforce definition, rather than hiring or output; it exists only in daily interaction, it shifts when a manager changes or a layoff rumor circulates, and an annual survey samples terribly because it is only a snapshot of an ever persistent, primary consideration. At this point in history, HR has identified one of the most important conditions in the workplace and measured it once a year.

Every Era Stopped at the Black Box

By 2010 strategic HR had a credible claim that HR systems correlate with firm performance and no satisfying answer to the obvious follow-up; why? Researchers called the gap the “black box,” the seemingly unobserved stretch between an HR policy and a financial result.

Kaifeng Jiang, David Lepak, Jia Hu, and Judith Baer mapped the inside of the box

Their work in “How Does Human Resource Management Influence Organizational Outcomes?” pooled the evidence and sorted HR practices into the three AMO families as skill-enhancing (ability), motivation-enhancing, and opportunity-enhancing. Their model ran in stages, from HR practices to employee human capital and motivation, then to operational outcomes such as voluntary turnover and productivity, and only then to financial performance.

Sticking with our car analogies, the way a drive shaft carries power from an engine to the wheels, a “mediator” is a middle link that carries an effect from cause to outcome; you can measure the depression of the gas pedal and we can gauge engine output and wheel speed all day, but if the drive shaft is cracked you will never know why the car shudders. Jiang’s team showed the drive shaft exists. What they could not do, with the data available, was watch it turn.

Thirty Years of Evidence with the Same Blind Spot

Then we found a 2026 review in Human Resource Management, “Looking Back and Looking Forward,” which mapped 3,503 articles across 156 journals. It sorts the field into performance-focused, configurational (which bundles fit which contexts), process-based (how the effects travel through people), and newer crisis-oriented perspectives.

Read our clusters as a timeline of the questions explored in HR:

1. 1970s; how do we motivate people, and why do they leave?
2. 1980s; how do we select and retain the right people?
3. 1990s; do HR practices affect organizational performance?
4. 2000s; how do HR systems create competitive advantage, and under what fit conditions?
5. 2010s; what psychological and behavioral processes connect HR to performance?
6. 2020s; can we measure the workforce characteristics and behaviors that explain what happens after hiring, continuously and at the level of the individual?

The sixth question matters more than another paper showing that satisfied employees perform somewhat better, because it is the first one that benefits from instrumenting instead of inferring it.

The black box stayed black for a structural reason rather than an intellectual one. Researchers had theories of the active workforce as early as Mobley; what they lacked was longitudinal behavioral data from inside real employers, because HR systems were designed to record events (hire date, pay change, termination) and nothing that happens between them.

That gap between what platforms capture and when the behavioral sequence actually begins is the subject of Your Workforce Is Telling You Something. Your Systems Aren’t Listening.

Some Other Perspective from Outside HR Which Explains More Than Most HR Journals Have

Organizational psychology, labor economics, sociology, and behavioral science, we in:

• Goal setting; “Building a Practically Useful Theory of Goal Setting and Task Motivation” in American Psychologist summarized decades of evidence that specific, difficult goals outperform “do your best” instructions, which is why every serious performance-management system starts with goals.
• Intrinsic motivation; Richard Ryan and Edward Deci’s self-determination theory paper identified autonomy, competence, and relatedness as the psychological needs that sustain motivation without external pressure, and it explains why a bonus can actually crowd out the drive it was meant to reward.
• Judgment and bias; “Judgment under Uncertainty” documented the shortcuts human judgment takes under uncertainty, which the reason an unstructured interview behaves like a vibes-based lottery while a structured one helps.
• Networks; Look at the “The Strength of Weak Ties” which showed that acquaintances, more than close friends, carry new information across social circles, including information about jobs; referral hiring is a weak tie market whether HR appreciates it or not.
• Technology and skills; The 2003 Quarterly Journal of Economics paper on computerization showed computers substitute for routine tasks and complement non-routine problem solving and interaction, which is the template for what we all know too well now.

Every one of these findings describes behavior that unfolds over time inside the organization; goals are pursued, motivation rises and decays, judgments are made in sequence, networks route information, tasks get reallocated, skills accumulate. What we might find when we look outside of what is clearly work-related research is what might be in the hallway between doors, in the workforce, which might be why so little of it ever got wired into the systems HR actually operates.

The HR Reading List Organized by the Question Each Era Answered

Read the history and it becomes a list of retired assumptions, too many of them still running inside organizations.

The Future of Human Resources Must Be Built in the Hallway

Hackman and Oldham put motivation in the design of the job, Mobley drew the sequence that precedes a resignation, Schmidt, Hunter, and Sackett turned the hiring door into a science and then had the integrity to recalibrate it, Meyer and Allen showed that staying can mean three different things, Huselid and MacDuffie proved the system pays, and Edmondson and Jiang pointed at the conditions and processes in between.

Every one of them, sooner or later, arrived at the workforce between the hiring door and the exit door and had to infer what happened there, because nothing was recording it.

We should stop treating hiring and retention as separate departments with separate data, since the research says they are two ends of one behavioral sequence. We should measure the conditions already identified (job design, fit, commitment type, psychological safety) continuously instead of annually, and connect them to the outcomes they predict. We should judge every selection method by how its hires actually behave over their first year, not by a single rating, because that is the only way to see invisible talent the filter would otherwise discard. And we should build that measurement to improve jobs, teams, and managers rather than to police individuals, or we will destroy the very safety the measurement depends on.

Anyone working this problem, whether you run people operations, build HR technology, or fund the companies that do, is working the same open question the field has circled since 1977, and the most useful thing we can do is compare what we are each seeing inside the hallway rather than keep publishing about the doors.

If your organization had to explain its last ten resignations using only the behavior it observed before each person decided to leave, how much of the story would you actually be able to tell?