Assessment Centre

What an assessment centre is

An assessment centre is a selection method in which a group of candidates works through a number of exercises drawn from the job, watched by a number of trained assessors, with every attribute being measured observed in more than one exercise, and the observations then brought together in a single session from which the scores are derived.

The name suggests a place, and it is not one. The centre here describes an arrangement for measuring rather than a room or an issuing body, and it can be run on an employer’s premises or remotely without anything in the description changing. What changes with the location is what can be observed, not the method.

Four conditions the description rests on

The description is compound, and dropping any one of the four produces a different arrangement that keeps the name.

• Multiple exercises drawn from the job. Not one question, and not one exercise repeated in two phrasings.
• Multiple trained assessors. They observe and record independently before any of them speaks.
• Attributes defined before administration, described in behaviour that can be seen rather than in general terms, with each attribute observed in at least two exercises.
• A session to bring the observations together, in which the evidence is presented before the scores, and from which each attribute’s score is derived.

Anyone who has run a day of three consecutive interviews and then pooled the interviewers’ impressions in a meeting has run three interviews. The difference is not the number of hours or the number of people who attended. It is that the thing being measured was defined beforehand and deliberately distributed across the exercises.

The matrix is the whole design

The design tool here is a plain grid: attributes down one side, exercises across the other, and in each cell a note of whether that attribute is observed in that exercise. The rule that makes the method work is that every attribute has at least two cells filled.

That rule has a numerical consequence, and it is computed before the first session. Take a design with 4 exercises, where one exercise cannot carry serious observation of more than 2 attributes.

• Cells available: 4 exercises multiplied by 2 attributes each, which is 8.
• Cells required for four attributes: 4 attributes multiplied by 2 exercises each, which is also 8.

So the design supports exactly four attributes with nothing spare. Add a fifth and the requirement becomes 10 cells against 8 available, which resolves only by adding a fifth exercise or by raising what any one exercise carries to three attributes. The second is what usually happens, because it costs no time, and its price is that an assessor is asked to observe three things in 45 minutes, guesses at two of them and records one.

An attribute observed in only one exercise keeps its row in the grid and keeps its column of numbers. Its score is nonetheless the score of that one exercise. Reading it as a judgement about the attribute is reading a judgement about a single situation, and it is the most common misreading of these grids.

The usual exercises, and what each one can show

The exercises are not interchangeable. Each has something it shows and something it cannot reach, and that is what the distribution of attributes across the cells is built on.

• The in tray. The candidate is given a set of competing messages and requests in a limited time, sorts them, and writes what they would do about each. It comes closest to showing prioritisation and handling of incomplete information, and it shows nothing of what happens between people, because it is done alone.
• The group discussion. Several candidates are seated around one problem with no appointed chair. It is the only exercise that shows how a position is built with people over whom nobody has authority, and it is the most contaminated by the composition of the group: a candidate placed between two talkers reads as quiet, and reads as forthcoming in a different group.
• The role play. The candidate meets a trained actor in a situation from the job, such as an unhappy customer or a subordinate who is behind. Its strength is that the situation is identical for every candidate, and its condition is that the actor holds to the same script, since otherwise it is the situation that varies rather than the performance.
• The analysis and presentation. Raw material is supplied, a conclusion is drawn from it, presented, and questioned. It shows how an argument is ordered, and it favours whoever is used to presenting to a room, so its score is read with that in mind unless presenting is part of the role.

What follows from this is that an attribute’s score moves with the exercise it was observed in and not only with the candidate. That is why the two exercise rule exists. An attribute that appeared in two different situations supports a statement about the attribute. An attribute that appeared in one supports a statement about that situation.

Who sits as an assessor

Assessors are usually a mix of managers from the departments concerned and people from the HR function, for a practical reason: the manager knows what the role needs, and HR holds the administration consistent across candidates. A condition usually stated in this area is that an assessor should not, where possible, also take the final decision on the same candidate, because anybody who has decided about a candidate before the exercise observes what supports their decision.

That condition is hard to satisfy in a small organisation, where the manager is both the assessor and the decision maker. The answer there is not to pretend it has been satisfied. It is to write the evidence down before the session and read the decision against it, so that the effect of observation survives even where the two roles have not been separated.

The wash up session: evidence before numbers

The session in which the observations are brought together is not an administrative step at the end of the day. It is where the measurement happens. Its established order is that each assessor presents what they saw and heard in their exercise as a described event, before any number is mentioned. Then the scores are put up and the disagreements are examined.

The reason for that order is that a number stated before its evidence turns everything after it into a search for a justification. The disagreement between two assessors on one attribute is the most useful thing to come out of the session, because it points to one of three things: the attribute’s description is incomplete, or the exercise does not show it, or the candidate genuinely performed differently in two situations. Those are three different answers, and the average of the two numbers does not say which one it is.

This is the same discipline that the marking key in a work sample test depends on: the ratings for each criterion are stored separately, because the total says that there is a disagreement and only the components say where. It is also the point at which this method meets test reliability, since agreement between independent observers is the form of consistency that matters most in anything judged by people.

The cost arithmetic runs before the decision, not after

This is the most expensive selection method there is in assessor time, and the figure is computed rather than estimated. Take a round with 12 candidates, 4 exercises and 6 assessors, where each candidate is watched by 2 assessors in each exercise.

• Observations: 12 candidates multiplied by 4 exercises multiplied by 2 assessors, which is 96 observations.
• Time per observation: 45 minutes for the exercise and 30 minutes to write up the evidence, which is 75 minutes. So 96 multiplied by 75 is 7,200 minutes, which is 120 hours.
• The wash up session: 6 assessors for 4 hours, which is 24 hours.
• Assessor training before the round: 6 assessors for 8 hours, which is 48 hours.

The total is 192 hours of assessor time. Divided across the 12 candidates in the round, that is 16 hours per candidate. If 3 of the 12 are hired, the 192 hours fall on 3 appointments, which is 64 hours per appointment. At a loaded cost of SAR 250 an hour, which is assumed for this example and is not a fixed rate, those 192 hours cost SAR 48,000, or SAR 16,000 per appointment.

The figure is read alongside one further point. The 48 hours of training is spent once for the round and not once per candidate, so the larger the cohort the further it spreads. That is why this method suits a comparable group measured at one time and does not suit a single appointment opened and closed. Those assessor hours are internal recruiting time and belong in cost per hire, along with the basis on which the loaded hourly figure was built.

What spoils it

• Exercises that do not resemble the work. The argument here is the same as a work sample’s argument, drawn from the content of the exercise. A generic exercise measures something and keeps the name.
• Assessors untrained on the attributes. An untrained assessor observes what catches their attention rather than what they were asked to observe, and the halo effect and first impressions enter through that gap.
• Talking before recording. An assessor who has heard a colleague’s view before writing their own is not adding a second independent observation.
• Ranking candidates on the total alone. A total conceals that one low attribute may be the one the role requires.
• Running it for a role where one person will be appointed. The cost falls on a single candidate, and the comparison the method rests on has nothing to compare.

Telling it apart from the arrangements it resembles

What separates it from its neighbours is the material being judged. A day of several interviews judges conversation, and this method judges performance that is watched; how an interview itself is disciplined belongs to the structured interview, which may perfectly well be one of the exercises in a round while remaining a narrower thing than the round. An instrument marked against a single key, such as a cognitive ability test, judges an answer with no observer in the arrangement at all, which is why it sits inside a round without friction and why it cannot stand in for one.

The development centre is the harder case, because there the material and the form are identical and only the output differs: feedback to an employee for their development plan rather than a selection decision. That is why telling the participant which one they are in before it starts is a working condition rather than a courtesy. Somebody who believes they are being developed while they are being measured performs differently from somebody who knows, so a round run without that sentence has changed what it is measuring.

Where the attributes come from

The attributes in the grid are not invented for the round. They are taken from what the role requires, described in behaviour that can be seen, which is the same requirement competencies are held to: an observable behaviour, a defined level, and relevance to the role in question. An attribute written in general terms cannot be given a two exercise cell allocation, because nobody can say what would count as observing it, and that failure shows up at the wash up as an argument nobody can settle.

What it produces as data

The written evidence about each candidate, their scores and the record of the wash up session are personal data relating to an identified person, and they stand where the output of the other selection instruments stands. The duties that follow are set out on the cognitive ability test page, and this page states none of them. What is particular to this method, and is a point about the method rather than about those duties, is that what gets written down here is wider than what any other selection instrument produces: paragraphs describing how a named person behaved, rather than a bare score. A record of that shape is worth deciding about deliberately, and the page above is where to read what governs it.

What this page does not establish

We did not find, in our sources, anything establishing how far this method predicts performance, nor a ranking for it among selection instruments, nor a recommended number of exercises, assessors or attributes, nor a score at which it becomes acceptable. The figures in the example above are design figures assumed in order to show how the calculation runs, and they are not standards to copy.

Nor did we find a provision setting out a rule on an employer running selection instruments in the Saudi private sector, or on payment for a candidate’s time in them, so this page sets out nothing on either. That is a statement about our sourcing and not about the statute book.

Related terms

• Work Sample Test
• Test Reliability
• Test Validity
• Cognitive Ability Test
• Situational Judgement Test