From attrition to evidence: why we should rethink how we select rangers.

Selection is the first thing a new ranger learns about an organisation. It happens before any instructor speaks, before any uniform is issued, before any new skill is learned. It tells the candidate what the organisation values, how it evaluates, and what kind of behaviour it rewards. Long before the training begins, it has already started shaping the ranger they will become.

This makes selection the single most consequential investment an organisation can make, and the one most organisations in our sector have under built. We invest relatively heavy in curricula, instructors, and equipment. We treat selection as a formality (a physical test, or an informal nomination, or both) and then wonder why the training does not seem to produce the rangers we expected.

The cost of poor selection is not theoretical. It shows up downstream; in eighteen-month turnover, in rangers who cannot maintain discipline unsupervised, in deteriorating community relations, in integrity incidents, and, at the outer edge of the failure distribution, in candidates who do not survive the process itself.

I lead the ranger training practice at LEAD Conservation. We work with protected-area practitioners across sub-Saharan Africa and Southeast Asia on instructor and leadership development. A while ago, we went back to the literature on how selection actually works. Not how we think it should be done.

What the evidence tells us

The field is called personnel selection, and it is one of the most extensively studied areas of applied psychology. The numbers tell a story that does not flatter the model our sector, in large, still uses.

Start with the most uncomfortable finding. In 2023, Coombs and Hauenstein, writing in Military Psychology, published a study that tracked 1,321 candidates through the U.S. Army’s Ranger Assessment and Selection Program. They were looking for what actually predicted who made it through and who did not. The answer was counter-intuitive. Physical fitness at entry (the variable the Army had selected on) turned out to be the strongest single predictor of who would quit. Not the weakest. The strongest. The fittest candidates disproportionately walked away. The variable that best predicted success across every phase of the programme was something the Army had not been measuring at intake: conscientiousness. A stable personality trait, associated with reliability, discipline, and follow-through. Not with running speed.

If that is true of the U.S. Army’s special operations pipeline (a pipeline with more resources, more science, and more experience in selection than any conservation agency) it is almost certainly true of our sector too.

There is a century of research that points in the same direction. Schmidt and Hunter, in a 1998 meta-analysis in Psychological Bulletin, synthesised eighty-five years of personnel selection studies and ranked methods by how well they predict subsequent job performance. That work was updated in 2016, and again, most rigorously, by Sackett, Zhang, Berry, and Lievens in a 2022 paper in the Journal of Applied Psychology; the current reference point for anyone who wants to know what actually predicts who performs. Over time, the numbers have been refined.

A small number of selection methods reliably predict who will perform well, and most of them are not the ones we use.

Structured interviews, the kind where every candidate is asked the same questions in the same order and against the same rubric, are among the strongest predictors we have. In a foundational 1994 review, McDaniel and colleagues found that a well-structured interview accounts for about a fifth of the difference between strong and weak performers once they are on the job. An unstructured one accounts for about a tenth. Sackett’s 2022 update, the current reference point, keeps the structured figure in the same range. To put that in perspective, no method in the whole of personnel selection reliably explains more than about a quarter of that difference; structured interviews are close to the ceiling of what is achievable. The standard ranger interview (informal, unsystematic, conducted by whoever happens to be available) sits near the bottom of that range. Structuring the interview is the single biggest upgrade most recruiters could make, usually for the cost of a morning spent writing the questions down and agreeing on what a good answer looks like.

Work samples (asking candidates to do a scaled-down simulated version of the job) predict reliably too. Roth, Bobko, and McFarland’s 2005 meta-analysis put them at roughly the same level as an unstructured interview on their own; where they earn their place is in combination. A structured interview plus a work sample identifies the right candidates better than either method alone, because the two are measuring different things. This is the evidence base for asking a ranger candidate to carry a realistic load over a realistic distance on representative terrain, or to observe and report what they see in an unfamiliar landscape, or to solve a team problem without an appointed leader. Not to race. To do.

Personality predicts. Barrick and Mount’s 1991 meta-analysis, replicated by Salgado across European samples in 1997, established conscientiousness as the most consistent personality predictor of sustained job performance across occupations. It predicts who will still be doing the job well, properly, three years from now. In rangers (who spend most of their working life unsupervised, in difficult conditions, where the test is whether they do the job right when no one is watching) that is exactly the trait we need.

Learnability predicts too. In the contexts where we work, it arguably matters more than any other single attribute, because so much of what a ranger needs to be able to do has to be built during training rather than brought to it. Hunter’s 1986 meta-analysis, and Ree and Earles’ 1991 work on U.S. Air Force selection, established what the research community calls general mental ability as the single strongest predictor of how well someone learns during training; ahead of experience, education, and prior skill. Sackett’s 2022 update keeps it near the top of the hierarchy.

The standardised tests that research used are a blunt instrument in our contexts; they are culturally loaded, literacy-dependent, and unfamiliar as a format. But the underlying capacity they try to measure (the ability to take in new information, make sense of it, and apply it) can be observed directly. We design our practical tasks so that part of each one is teachable on the spot: a brief instruction, a first attempt, a short correction, a second attempt. What we are watching for is not whether the candidate gets it right the first time. It is the distance between the first attempt and the third. Candidates who visibly learn during a few days of assessment are the candidates most likely to learn during the course that follows. That is what we are trying to see.

Combining these methods beats any one of them. The strongest predictions come from pairing a measure of learnability with a structured interview, a work sample, or an integrity assessment. Each pairing pushes validity well above what any single method achieves alone, because each method is measuring something the others miss.

What the evidence does not support

Some common practices are not supported by this evidence. Stress interviews (shouting, trick questions, deliberate hostility) do not predict how well a candidate handles stress on the job. What they reliably measure is anxiety about the interview itself. Sleep deprivation and food deprivation test tolerance of manufactured hardship, not job readiness. Timed competitive fitness tests reward short-term preparation and disadvantage smaller bodies and different physiologies without predicting sustained operational capacity. Unstructured community nomination, on its own, carries all the risks of nepotism and bias one would expect. None of these are meaningless rituals. They are simply not selection methods in the scientific sense.

Voice stress analysis deserves its own paragraph, because it has quietly become common in our sector as an integrity screen, for recruitment and for internal investigations. The evidence against it is unusually clear. A National Institute of Justice field study by Damphousse and colleagues in 2007 tested the two leading commercial systems on arrestees questioned about recent drug use, with laboratory results as ground truth: the systems identified around 15 per cent of the lies, and performed at roughly chance overall. Harnsberger and colleagues, in the Journal of Forensic Sciences in 2009, found the same chance-level performance with false-positive rates between 40 and 65 per cent. The problem is not the equipment. Stress in the voice is measurable; its relationship to deception is not established. The one real effect these studies did find is worth knowing: people who believe they are being tested admit more than people who do not. That is an interview effect, and it can be achieved without the machine. In a sector where a single flagged reading can end a ranger’s career, or exclude a candidate who has done nothing wrong, that is not a defensible basis for a decision. For integrity we rely on structured questions about past behaviour, verified references, and situational judgment responses scored against a rubric.

Two philosophies

There are, broadly, two philosophies of selection. One starts with the question: who can we break? The other starts with the question: who can we recognise?

Negative selection is designed to eliminate. It imposes escalating stress and retains whoever remains. Endurance equals suitability. The process is the test. High failure rates are read as proof of rigour.

This model has a particular risk profile in the places where we work. In much of rural sub-Saharan Africa, a ranger position is one of the few formal, paid jobs available within reach of a community. Selection exercises routinely draw hundreds of candidates for ten or twelve posts. People for whom failing is not a realistic option will push themselves well past a safe limit to stay in the process. In the last few years, reports from Uganda and Kenya have described candidates collapsing and dying during recruitment exercises for wildlife ranger and defence-force positions. They had been told that a long run, against the clock and against their peers, was part of the process. Some of them did not finish. Some of them never went home.

Nobody who designed those exercises wanted anyone to die. Many of the officers running them are themselves graduates of the same system; they are doing what they were taught, by the people who trained them, who were taught the same thing before. The model has a long lineage. It was built in military tradition, refined through decades of colonial and post-colonial policing, and inherited, more or less intact, by many of the organisations now protecting our parks. It persists because the people running it believe it works. The trouble is, it doesn’t; not even on its own terms, and not for the organisations running it.

Positive selection is designed to identify. It begins with a clear description of what a successful ranger looks like at the end of training, and works backwards. It asks: given the right training, who is most likely to reach that standard? The assessment is the test, not the process.

At LEAD we use the second philosophy. This is not about lowering the bar. A well-designed Basic Field Ranger course itself is demanding, physically and cognitively. What changes is where the difficulty sits. In the attrition model, the selection punishes and the training picks up whoever is left. In the latter model, the selection is rigorous but humane, and the training does the hard work of development.

How we run it

Our process runs over roughly three to four weeks. We publish an honest description of the role, including the parts people often leave out: the long days on foot, the weeks away from family, the strict rules about use of force and the treatment of communities. The announcement is delivered through community meetings in local language, through local radio or WhatsApp groups, through word of mouth from serving rangers. The goal is informed self-selection at the front end. If a candidate applies, they do so with open eyes.

We verify eligibility and baseline functional fitness through a medical screen, not a ranking exercise. Once we are honing in on the final candidates, we ask two or three people who know each candidate well (neighbours, teachers, former or current employers) the same five questions, looking for specific observed behaviour over time. Community input is not the decision; it is texture no short assessment can provide.

The heart of the process is a structured assessment centre that runs for three to five days, depending on local context. Candidates do multiple practical tasks drawn from the job: a route march with load, completed rather than raced; an observation and oral reporting exercise in an unfamiliar landscape; a multi-step instruction-following task; and small-group practical problems with no leader appointed. Each task is scored by at least two assessors against a behavioural rubric. Every candidate then sits a structured interview, in their preferred language, and responds to three situational judgment scenarios: realistic dilemmas with no clean right answer. What we are looking for is not the correct response. It is the reasoning. Does the candidate consider consequences? Do they think about other people? Do they show honesty about uncertainty? A candidate who says “I would ask my team leader because I am not sure” scores higher than one who invents a confident but reckless answer.

A panel of at least three brings the evidence together. No single person decides.

A few things we do not do. No timed competitive fitness race. No stress interview. No sleep deprivation, no food deprivation. No public elimination rounds. No written examinations; literacy varies across our candidate pool, and we are not assessing literacy. None of this is softness. It is the difference between testing for the job and testing for a model of the job borrowed from elsewhere.

The ninety who go home

The part of this that gets the least attention, and matters most, is what happens to the people who are not selected. If a hundred people apply and ten are selected, ninety go home carrying a story about the organisation. Those ninety stories shape future recruitment for years. They shape community relations. They shape reputation. A selection process that produces ten rangers and ninety resentful former candidates is a less dramatic failure than one that produces dead candidates. But it is still a failure.

So we invest in the things that often get cut. Individual feedback for every candidate, not only for those selected. “Not this time” framing, rather than rejection as judgment of worth. A meal. A certificate of participation. Travel support where we can manage it. Calm, observatory selector behaviour (no screaming, punishment, or a gatekeeper posture), and warm connection during down times. Decisions communicated privately, one to one, by a senior selector, not the most junior person on the team. None of this is sentimentality. It is part of what the process does.

The way an organisation selects people is the first thing those people learn about the organisation. If we want rangers who treat communities with respect, we treat candidates with respect. If we want rangers who exercise authority with accountability, we exercise authority with accountability while we are assessing them. Selection is the first day of a ranger’s education in what the organisation believes, long before any instructor steps in front of a classroom. Candidates notice. And ninety of them go home.

Whether the sector is willing to update

I am not writing this to criticise any specific agency. People running selection today are doing what they were taught, often what they themselves went through. The officers conducting these exercises are, in many cases, good people inside a model they did not choose.

But the evidence has been in for a long time now. The question is whether the sector is willing to update.

A few things would change immediately if we did. We would stop running timed competitive fitness races as the centrepiece of selection. We would keep rigorous functional fitness screening; cheaper, safer, and more predictive of the actual job. We would invest in structuring the interview. We would build selection around multi-day observation of task-based performance, not hours of endurance. We would treat community reference as structured qualitative input, not a substitute for competency assessment. We would look at personality (particularly conscientiousness) as seriously as we look at physical capacity. And we would track our selection decisions against subsequent performance, so that every year’s process improves on the last.

None of this is theoretical. It is standard practice in sectors that recruit for other demanding, high-accountability roles. It is the direction the U.S. special operations community itself is moving in. And for those of us who work in conservation (where the cost of selection failures is measured in poor community relations, high turnover, integrity incidents, and, sometimes but still too often, in candidates who never come home) the case for change is already overwhelming.

Selection is not a test of who can suffer most. It is a question about who will do the job well, for years, in conditions where no one is watching.

We know how to answer that question. We should start.

A useful comment offline reminded me to say this: none of what I’ve described requires a bigger budget. It is a mindset change, not a call for more funding.

FURTHER READING

Sackett, Zhang, Berry & Lievens (2022), Journal of Applied Psychology; Coombs & Hauenstein (2023), Military Psychology; McDaniel, Whetzel, Schmidt & Maurer (1994), Journal of Applied Psychology; Barrick & Mount (1991), Personnel Psychology; Roth, Bobko & McFarland (2005), Personnel Psychology; Arthur, Day, McNelly & Edens (2003), Personnel Psychology; Hunter (1986), Journal of Vocational Behavior; Ree & Earles (1991), Personnel Psychology; Damphousse, Pointon, Upchurch & Moore (2007), Assessing the Validity of Voice Stress Analysis Tools in a Jail Setting, National Institute of Justice; Harnsberger, Hollien, Martin & Hollien (2009), Journal of Forensic Sciences.