Recency Effect in Appraisal

What the recency effect in appraisal is

The recency effect in appraisal is a performance rating built on the last few weeks of the appraisal period alone, so that the rating describes part of the period and is read as describing all of it.

Its mechanism is a narrowed evidence window. The period runs twelve months and memory supplies two of them. The appraiser does not set out to disregard what came earlier; they use what they can recall, and what they can recall is what is nearest.

That framing matters because it decides what can be done about it. An error of intention is answered by being told about it. An error in what is available to recall is not, and the only thing that works on it is what was written down while it was happening.

It moves the rating in both directions, and its third face is the worst

It is usually assumed to inflate ratings. It inflates and deflates them alike, according to what happened last. An employee whose performance was steady all year, whose final project ran into trouble a few weeks before the cycle, has their year read out of a single stumble. An employee whose performance was uneven, who delivered something visible just before the appraisal date, has months folded away behind it.

Then the team works out the rhythm, and it stops being an error of recall and becomes a behaviour. Visible work gets saved for the weeks before the cycle, and timing becomes part of performance. This is the damaging one, because the first two distort one person’s rating and this one corrupts what the instrument measures for everybody.

Measuring the evidence before measuring the rating

This error is measured on the record of observations before it is measured on the rating, because its seat is in the evidence rather than in the judgement. The figures below are assumed, and they set out a manager’s record about one employee over an appraisal period of twelve months.

• Months one to six: two recorded observations.
• Months seven to ten: three observations.
• Months eleven and twelve: nine observations.

So the total recorded is 2 + 3 + 9, which is 14 observations. The last two months hold 9 of those 14, which is about 64 percent of everything written down in the period. Their share of the period itself is 2 months of 12, or about 17 percent. Dividing the first share by the second gives close to four times, meaning the density of recording in the final sixth of the period runs at nearly four times what its length would call for. The two denominators are different things and have to be named as such: one counts the observations recorded, the other counts the months of the period.

At the other end the same record reads: the first half of the year holds 2 of the 14 observations, which is about 14 percent of what was written down, against 50 percent of the period. Six months of this employee’s work entered the cycle carrying two observations, and whatever judgement is formed about those months rests on memory rather than on a record.

This calculation is run before the form is opened and needs no knowledge of the rating, which is what makes it preventive rather than diagnostic. If the density comes out this way while the rating is still unwritten, it is still possible to go and retrieve what happened in the first half from its sources. Run after the rating is written, it explains a result nobody can now change, because what was not recorded at the time is rarely recovered as it was.

The question that decides every case

Every borderline situation in this area resolves to one question: was the evidence from the first half of the period there and then outweighed by something more recent, or was it never there at all?

The first is a fault in weighting, and it is treated by putting the older evidence back on the table. The second is a fault in recording, and no amount of discussion in the final meeting treats it. The two look identical in the finished rating and they have nothing in common behind it.

The question also separates this error from the one it is usually paired with. An early impression that persists takes the beginning of the period and does not drop the end of it; it reads the end through the beginning, so later evidence arrives and is filtered rather than lost. The recency effect drops the beginning outright, and what was dropped was not outweighed by anything. It was forgotten. That is why returning the old evidence to the table treats one of them and not the other: in the second case the later evidence was present all along and was read in a particular way.

Two other errors sit off this axis entirely. Central tendency error is a fault in how much of the scale gets used, and contrast error in appraisal is a fault in what the rating is measured against. Either of them can occur on the same form as this one, on the same rating, without any of the three explaining the others.

A declared weighting is not this error

An organisation may decide openly that the final quarter weighs more than the rest, because it sits closest to the next plan or because the principal project closed in it. That is an administrative choice, written into the description of the instrument, said to the employee at the start of the period rather than at the end, and discussed as policy.

What separates it from the recency effect are properties of the decision rather than of the outcome. A declared weighting is announced, known to both parties before the period begins, while the recency effect is not known to the appraiser about themselves. A declared weighting is uniform, applying to the whole team equally, while the recency effect varies with whatever happens to have stuck about each individual. And a declared weighting is revisable by changing a line in the description of the instrument, while the recency effect cannot be revised because it is not written anywhere that can be edited.

The consequence is worth stating, because the mistake is tempting. Answering the recency effect by declaring a weighting for the final quarter does not treat it. It converts it into a rule. The treatment is to widen the record, not to narrow the period that counts.

When the end of the period is the right answer

Sometimes a rating that matches the final weeks is simply correct, and the question above sorts these cases quickly.

A genuine and continuous decline is one: an employee whose performance fell steadily across the quarters, where the appraiser holds recorded evidence from each quarter establishing the direction. The low rating rests on the whole period, and the fact that it agrees with the end of it does not make it a recency effect. A substantial change in the role is another: somebody who moved to different work halfway through the period has a first half that describes other work, so what is required is not a balance between the halves but a statement of which role the rating describes. A period that was short to begin with, such as an appraisal at the end of a brief probation period, has no distant first half to balance against, so the standard there is the density of the evidence rather than its distribution. And an observation window that was interrupted, by long leave, by a secondment or by a change of manager inside the period, genuinely shortened what could be observed; that gap is declared in the appraisal rather than treated by weighting whatever remains.

What treats the cause, and what only looks like treatment

A record of observations written as things happen, rather than assembled at the end of the period, is the only measure that treats the cause instead of the symptom. Reviews held during the period, each stage closed with a written note, mean the final cycle arrives with a record instead of a recollection. A regular one to one meeting is the nearest occasion for capturing what never appears in a report, provided what is said in it is written down at the time. Asking the appraiser for evidence distributed across the period, rather than for evidence in general, exposes the gap immediately, since a request for an example from the first half is answered or it is not. And recording the evidence has to come before the form is opened, rather than being searched for once the rating has been chosen.

What does not work is extending the discussion in the final meeting. It cannot return what was never written down. A performance calibration session works on what the appraiser recorded during the year, so it settles what a rating means and creates no evidence at all; this error sits in the evidence before it reaches the ruler. A self assessment does not treat it alone either, since it is subject to the same narrowing of the window. It does become a source of evidence about the early part of the period that the manager does not hold, but only where it asks for specific events with their dates.

Bringing the first half back to the table

Where the narrow window is discovered before the rating is written, what is needed are sources about the early part of the period that do not depend on the appraiser’s memory. Most organisations already hold them, even though they were not gathered for an appraisal: the minutes of regular meetings and the assignments and dates recorded in them, closure reports for stages and projects that finished in the first half, handover messages between the team and those it serves inside the organisation, logs of requests and tickets with their closing dates, and whatever the employee themselves recorded about the early months where specific dated events were asked for.

Those sources have a boundary worth naming. They describe what was done and describe far less well how it was done. They repair the countable part of the first half and repair nothing that leaves no trace on paper: handling a struggling colleague, or a remark made at the right moment that prevented a mistake. That is why they are a remedy for one cycle rather than a standing substitute for recording things as they happen. The same distinction explains why a well kept employee file helps here and does not solve it, since a file is built to hold documents and dates rather than observations of conduct.

When they are used, the appraisal should say that they were used. A rating built partly on records reassembled at the end of the period is not as strong as one built on observations written as they occurred, and saying so in the appraisal lets the following cycle start from a known position rather than from an assumption.

The limits of what is claimed here

We do not lay down here how many observations ought to be recorded in a period, nor how many reviews to hold within it, nor a distribution at which a record counts as sound. We did not find, in our sources, anything establishing such figures for the private sector in the Kingdom, and the numbers in the worked example are there to show the method rather than to serve as thresholds any organisation should be measured against.

What the page does say is that the distribution of evidence across a period can be measured before the rating is written, and that reading what that measurement returns remains an administrative judgement made inside the organisation.

Related terms

• Central Tendency Error
• Contrast Error in Appraisal
• Performance Calibration
• Employee File
• Peer Appraisal