eduKateSG · Judgement · The Gold Standard of Education
Combine additive and multiplicative rankings without hiding the trade-offs
Choose the explanation you need, or follow all twelve chapters in order.
Full chapter index · Independent practice · Judgement subject guide
WASPAS, the Weighted Aggregated Sum Product Assessment method, blends a weighted-sum score with a weighted-product score. The blend can be useful because the additive and multiplicative models react differently to weak criteria and scaling. It does not remove value judgements. Criteria, directions, normalisation, weights and the blending coefficient lambda still determine the result. This guide calculates a fictional example, tests lambda sensitivity and shows when the method should not be used.
A high-quality decision record keeps the raw matrix, data sources, missing-value treatment, weights, normalised matrix, component scores and lambda scenarios together. It also records thresholds applied before ranking and the smallest input change that reverses the leading pair. That bundle matters more than a polished final rank because it shows where judgement entered the calculation.
The method is most useful as a transparent comparison aid. It should prompt discussion about evidence and trade-offs, not replace responsibility for the final decision or conceal disagreement behind decimals.
Before calculation, build a data dictionary. It should state whether each value is measured, estimated or scored; the observation period; the responsible source; and the uncertainty or permitted range. Qualitative judgements converted to numbers need anchored rubrics so that two assessors can apply the same meaning. Missing values should trigger an explicit rule, not an unnoticed blank treated as zero. After calculation, invite challenge to the criterion set and thresholds as well as to the arithmetic. A stable rank built on an incomplete objective remains a poor decision. Record minority views about weights, and show at least one reasonable alternative scenario when stakeholders disagree. This lets readers distinguish a robust leader from an option that wins only under one narrow preference profile.
Finally, separate decision support from decision authority. The matrix can organise evidence and expose trade-offs, but the accountable decision maker must consider feasibility, rights, implementation capacity and information that does not fit a compensatory index. If the final choice differs from the calculated leader, record the lawful and evidence-based reason instead of altering weights retrospectively.
Verification should include both arithmetic and meaning. Recalculate the model with an independent implementation or a small hand-worked fixture, confirm that cost and benefit directions were not reversed, and inspect the normalised bounds. Check that the product component uses the intended exponents and that displayed rounding does not change the order. Then review semantic questions: does a one-unit change mean the same amount across the observed range, are any criteria proxies for the same underlying factor, and could an apparently beneficial criterion create harm beyond a threshold? A model can pass every numerical test while failing these questions.
Governance becomes especially important when rankings affect people. Preserve the version used for the decision, restrict access to sensitive source data, and publish an explanation at the level the audience can use. Avoid presenting individual-level scores as objective worth. Where affected people can correct inaccurate data or challenge a criterion, include that path in the process. Periodically test whether evidence sources have drifted and whether the criterion definitions still match the decision. A ranking is a time-stamped analysis, not a permanent fact.
If several options are effectively tied across reasonable scenarios, say so. The responsible output may be a shortlist, a request for better evidence or a staged pilot rather than a forced first place. Near-ties are often more informative than extra decimal places because they reveal where implementation learning could improve the later choice.
Continue from The Gold Standard of Judgement within The Gold Standard of Education: learning that transfers into life and the How X Works learning library.
Full chapter index
See the mechanism · Chapters 1–2
Build the worked model · Chapters 3–5
Test what changes · Chapters 6–8
Check evidence and limits · Chapters 9–10
Practise and continue · Chapters 11–12
Name the decision owner, alternatives, deadline and constraints. Separate eligibility rules from ranking criteria. An option that fails a legal, safety or budget threshold should not be rescued by high scores elsewhere unless the policy explicitly permits that compensation.
Define each criterion with a unit, evidence owner and preferred direction. Benefit criteria are better when larger; cost criteria are better when smaller. Avoid duplicate criteria that count the same advantage twice.
Evidence check: for Frame the decision before the matrix, preserve the observation before the interpretation. Record decision owner, alternatives and constraints. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
The weighted-sum model forms an additive score from normalised values. A strong criterion can compensate for a weak one in proportion to the weights. The weighted-product model multiplies powered normalised values. A value near zero can sharply reduce the product, so weak performance is penalised differently.
WASPAS calculates both components and blends them: Q equals lambda times the sum score plus one minus lambda times the product score. Lambda between zero and one selects the balance. Lambda of one is pure sum; zero is pure product.
Evidence check: for Understand the two component models, preserve the observation before the interpretation. Record sum and product model meanings. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 3 OF 12 · Build the worked model
3. Normalise benefit and cost criteria carefully
For a common ratio normalisation, a benefit value is divided by its column maximum. A positive cost value is transformed using the column minimum divided by the value. This creates dimensionless values where larger is better.
The formula assumes meaningful positive ratios. Zero or negative values make the product component undefined or misleading. Interval scales, arbitrary offsets and signed net values may require a different transformation or a different method. Document the chosen rule before calculation.
Evidence check: for Normalise benefit and cost criteria carefully, preserve the observation before the interpretation. Record normalisation and positivity assumptions. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
Consider three fictional options evaluated on learning access, reliability and weekly cost. Access and reliability are benefits; cost is a cost. The raw rows are A=(8,7,6), B=(6,9,5), C=(9,6,8), with weights 0.40, 0.35 and 0.25.
Using maximum normalisation for benefits and minimum-over-value for cost, A becomes approximately (0.889,0.778,0.833), B=(0.667,1.000,1.000), and C=(1.000,0.667,0.625). Preserve more digits in software and round only for display.
| Option | Access benefit | Reliability benefit | Weekly cost |
|---|---|---|---|
| A | 8 | 7 | 6 |
| B | 6 | 9 | 5 |
| C | 9 | 6 | 8 |
| Weight | 0.40 | 0.35 | 0.25 |
Evidence check: for Worked example: three study options, preserve the observation before the interpretation. Record every raw and normalised entry. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
For A, the weighted sum is 0.40×0.889 + 0.35×0.778 + 0.25×0.833, about 0.836. B is about 0.867. C is about 0.790. Under the additive model, B leads because its reliability and lowest cost compensate for weaker access.
Recompute every row independently and verify that weights sum to one. A spreadsheet fixture with known inputs and outputs is more reliable than visual inspection alone.
Evidence check: for Calculate the weighted-sum component, preserve the observation before the interpretation. Record weighted-sum arithmetic. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
For A, multiply 0.889^0.40, 0.778^0.35 and 0.833^0.25, giving about 0.832. B is about 0.850. C is about 0.776. The product remains positive and punishes B’s lower access more strongly than the sum, but B still leads in this example.
Use logarithms in software for numerical stability: log P is the sum of each weight times the log of the normalised value. Any non-positive value must trigger a validation error rather than a silent result.
Evidence check: for Calculate the weighted-product component, preserve the observation before the interpretation. Record weighted-product arithmetic. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
At lambda 0.5, A’s WASPAS score is about 0.834, B’s 0.858 and C’s 0.783. B ranks first, A second and C third. The score is a constructed utility index, not a probability, percentage of quality or predicted outcome.
The published method explores combining the two models to improve ranking accuracy under its assumptions. That does not make 0.5 universally correct. Lambda should be prespecified, justified or varied transparently.
Evidence check: for Blend with lambda, preserve the observation before the interpretation. Record lambda blend and score meaning. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
Calculate rankings at lambda 0, 0.25, 0.5, 0.75 and 1. If the order never changes, the decision is insensitive to the blend over that range. If two options cross, solve or approximate the crossing lambda and report the interval in which each leads.
Also vary plausible weights. Lambda sensitivity tests model blending; weight sensitivity tests preferences. They answer different questions and should not be combined into one vague robustness claim.
Evidence check: for Run lambda sensitivity, preserve the observation before the interpretation. Record lambda and weight sensitivity. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 9 OF 12 · Check evidence and limits
9. Check option-set and measurement effects
Ratio normalisation depends on observed column minima and maxima. Adding an extreme option can change every normalised value and potentially the ranking of existing options. Run dominated-option and deletion tests to reveal this dependence.
Propagate measurement ranges when evidence is uncertain. If cost could be 5 to 7 rather than exactly 6, rerun the model at defensible bounds or sample from a declared distribution. A precise rank from imprecise inputs is not robust merely because the arithmetic is exact.
| Test | Change | Fixed | Signal |
|---|---|---|---|
| Lambda sweep | Sum-product blend | Weights and data | Model sensitivity |
| Weight scenario | Trade-offs | Raw data | Preference sensitivity |
| Data interval | Measurements | Formula | Evidence sensitivity |
| Dominated addition | Option set | Old raw rows | Rank-reversal risk |
Evidence check: for Check option-set and measurement effects, preserve the observation before the interpretation. Record option-set and measurement changes. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 10 OF 12 · Check evidence and limits
10. Limits, misconceptions and non-compensation
WASPAS is compensatory. A sufficiently strong score can offset weakness elsewhere, especially in the sum component. It does not automatically model interactions, vetoes, uncertainty or causal outcomes. Weights are trade-off parameters, not simple statements of importance unless the elicitation method supports that interpretation.
Do not use a product model with zero, negative or arbitrary-origin scales without a defensible transformation. Do not describe a small score difference as meaningful without examining input and weight uncertainty.
Evidence check: for Limits, misconceptions and non-compensation, preserve the observation before the interpretation. Record compensation, scale and method limits. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 11 OF 12 · Practise and continue
11. Transfer practice: force a lambda crossing
Lower B’s access value while keeping all else fixed. Recalculate both components and look for a case in which A leads under the product model but B leads under the sum model. Find the approximate lambda where their blended scores are equal. Explain what that crossing means: the recommendation depends on how strongly weak criteria are penalised.
Then impose a non-compensable minimum reliability of 8. Remove ineligible options before ranking. Compare this constrained result with the unconstrained WASPAS order and explain why a threshold is a policy rule, not another criterion weight.
Evidence check: for Transfer practice: force a lambda crossing, preserve the observation before the interpretation. Record the crossing and veto contexts. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 12 OF 12 · Practise and continue
12. FAQs, reading routes and authoritative sources
Is lambda 0.5 mandatory? No. It is a common equal blend, not a universal truth. Are WASPAS scores probabilities? No. Can weights be changed after seeing the preferred winner? That would turn sensitivity analysis into result shopping. Must all criteria be positive for the product component? Under standard ratio-and-power treatment, yes.
Read the original Zavadskas, Turskis, Antucheviciene and Zakarevicius article, Optimization of Weighted Aggregated Sum Product Assessment. Continue with the verified COPRAS guide, the TOPSIS guide and the VIKOR guide. Each method encodes a different idea of acceptable trade-off.
Evidence check: for FAQs, reading routes and authoritative sources, preserve the observation before the interpretation. Record the narrowest defensible WASPAS recommendation. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.