How to weight an interview scorecard: four ways to set the weights, and three tests that show whether they matter

Weight an interview scorecard at intake, before anyone is interviewed. First turn the true must-haves into pass-or-fail gates, then weight the competencies that trade off by one of four methods: each decision-maker allocating 100 points independently, weights derived from an agreed ranking, importance ratings from a job analysis, or equal weights. Convert the result to percentages that sum to 100, then test it: check how much scores actually spread on each competency, and how small a change in weight would flip your top two candidates.

The scorecard layout and the weighted-total arithmetic are in the weighted interview scorecard page. This one is about choosing the numbers and checking that they do what you think.

Should you weight an interview scorecard at all?

Combining scores with any fixed rule beats combining them by discussion. Kuncel, Klieger, Connelly and Ones (2013, Journal of Applied Psychology) compared mechanical and holistic data combination in selection and found a consistent loss of validity when data were combined holistically, even by experts; in predicting job performance, the difference translated into an improvement in prediction of more than 50% for the mechanical methods. Which rule you use matters less: Dawes (1979, American Psychologist) showed that even "improper" linear models, including equal weights, outperform intuition when predicting a numerical outcome from numerical predictors, and that unit weighting is quite robust. So the order of priorities is: a fixed rule first, gates second, careful weights third.

Step 1: separate gates from weights

A gate is a minimum that no strength elsewhere can buy back: a license, a core skill the role cannot run without. Write it as "must score 2 or higher on a 1–4 scale" and give it no weight at all; a candidate who fails it is out before the total is calculated. Weight only the competencies where a little less of one can genuinely be made up by more of another. If everything feels like a gate, the role is probably described too narrowly, and the intake meeting is the place to test that.

Method 1: independent points allocation

Each person who will decide (the hiring manager, a senior peer, the director) splits 100 points across the weighted competencies, alone and before any discussion. Average the allocations; the spread between people tells you what to talk about.

Invented numbers. Data analysis has the widest spread, 25 to 40 points: the peer expects the new hire to do the analysis, the director expects them to commission it. That is a disagreement about the job, and it is far cheaper to settle now than in a debrief.

Method 2: weights from an agreed ranking

When people can agree on an order but argue about numbers, derive the numbers from the order. Two common formulas, for n competencies ranked 1 (most important) to n:

Rank-sum weight for rank k        = (n − k + 1) ÷ (1 + 2 + ... + n)
Rank-order centroid for rank k    = (1/k + 1/(k+1) + ... + 1/n) ÷ n

Example, four competencies:
Rank-sum:            4/10, 3/10, 2/10, 1/10   = 40%, 30%, 20%, 10%
Rank-order centroid: 52.1%, 27.1%, 14.6%, 6.3%

The centroid weights are much steeper; the top competency gets more than half. Use rank-sum unless the first-ranked competency really does matter more than the other three combined.

Method 3: job-analysis importance ratings

For roles you hire often, ask several people who do or supervise the job to rate each competency's importance on a 1–5 scale, and separately how costly a mistake on it would be, then weight by the average of importance × cost, normalized to 100. As a cross-check, O*NET publishes importance ratings for the skills and abilities of each occupation on a 1–5 scale running from "Not Important" to "Extremely Important", shown on O*NET OnLine converted to 0–100 (O*NET OnLine, scales and ratings, as of November 2026). O*NET describes the occupation in general, not your seat, so treat it as a prompt for the discussion, not the answer.

Method 4: equal weights

Give every weighted competency the same share. It is the default to beat: it needs no meeting, it is easy to explain to a candidate or a committee, and, as Dawes found, it is robust. Move away from it only when the people deciding can say which trade-off they are willing to make and why.

Test 1: does the scheme change the ranking?

Score the same three invented finalists (1–4 scale, all past the gates) under each scheme:

Candidate A leads or ties under every scheme. Second place does not hold: B is last with equal weights, ties C under rank-sum and beats C under the centroid weights. When the order depends on which reasonable scheme you picked, the totals are not deciding anything; go back to the evidence for B's data analysis and C's consistency, as described in how to compare candidates after interviews.

Test 2: is the weight you set the weight you get?

A competency can only separate candidates if their scores on it differ. Its influence on the total is roughly its weight multiplied by the spread (standard deviation) of its scores:

Influence of competency i = weight_i × SD_i
Share of influence        = influence_i ÷ sum of all influences
(a simplification that ignores correlations between competencies)

Example, using the points-allocation weights and invented spreads across the last twelve finalists:

The top-weighted competency carries about a sixth of the movement in totals, because nearly everyone scores 3 on it. Either the screen already filters for it, which is fine, or the questions and anchors do not discriminate, which is a design problem; anchors that separate levels are covered in behaviorally anchored rating scales.

Test 3: how much would a weight have to move to flip the decision?

For two candidates, take the gap in weighted totals and ask how much weight would have to move from one competency to another to erase it:

Gap D = total(A) − total(B)
Moving weight Δ from competency j to competency k changes the gap by
Δ × [(A_k − B_k) − (A_j − B_j)]
Break-even Δ = D ÷ [(A_j − B_j) − (A_k − B_k)]

With the points weights, A leads B by 3.03 − 2.98 = 0.05. Moving weight from stakeholder management (A ahead by 1) to data analysis (B ahead by 2) changes the gap by Δ × (−2 − 1) = −3Δ, so the break-even is 0.05 ÷ 3 = 0.017, about 1.7 percentage points of weight. Nobody's weights are accurate to two points, so this is a tie in practice, and the decision should rest on the evidence and the open questions, not the total.

Rules that keep weights honest

• Set and lock them at intake, with a one-line reason per weight, and apply the same weights to every candidate.
• Keep them coarse, in multiples of 5 or 10. Test 3 shows why finer precision is false.
• Do not weight what was not assessed. Divide by the weight covered, as the interview scorecard template shows.
• Show the parts beside the total in any comparison, as the candidate comparison matrix does.
• Re-run Test 2 after each search and drop weights that are doing nothing.

Common mistakes