Classical (Crisp) Data

A classical, or crisp, data structure records an alternative's standing on a criterion as a single figure: a quoted price of 12,500 TL, a delivery time of 14 days, an examination mark of 78. No band surrounds the figure, no probability, no membership grade, no cushion for doubt. The number is what it is, and the decision matrix is an ordinary table: alternatives in rows, criteria in columns, one figure per cell.

Most methods devised for decision analysis assume exactly this structure. Every other data type exists to preserve information a bare number cannot hold: approximation, several plausible values at once, an admitted gap in knowledge, only a boundary, or a distribution. Classical data is therefore both the commonest starting point and the benchmark against which the others are understood.

Classical data belongs wherever a value has been measured, counted, invoiced, or given as one unhesitating figure by a single decision-maker. Objective criteria such as price, duration, quantity, ratio, capacity or test result fit this structure naturally, and so do subjective criteria for which an expert supplies one undisputed mark.

Classical data is the right point of departure for any decision problem. Moving to another data type is warranted only once the single figure can be shown to discard something the decision needs; otherwise, staying classical brings a wider choice of methods and a more readable result.

Classical data suffices wherever the cell's figure is unique and defensible. It falls short precisely where something the decision needs stands behind that figure and a single number erases it. Each kind of missing information points to its own data structure:

• •If the figure is a judgement or a forecast and is inherently "approximate" → Fuzzy
• •If several plausible figures exist for the same assessment (scenarios, experts, periods) → Hesitant
• •If the assessment rests on knowledge that is incomplete or conflicting and that gap needs reporting in its own right → Neutrosophic
• •If only a lower and an upper bound are known → Grey or Interval
• •If the value is a probability distribution (measurement error, a random process, simulation) → Stochastic
• •If the assessment was made in words and those words are to be kept rather than converted into numbers → Linguistic
• •If the figure comes with a separate statement of how much it is to be trusted → Z-Number

The test is a single question: "What information do I lose by writing one number into this cell, and would that loss change the decision?" If the answer is "none" or "it would not," classical data is enough.

Classical data asserts that the figure is unique, not that it is right. A number carrying a measurement error is still classical data, and so is a value simply measured wrongly. The classical structure does not represent uncertainty; it leaves it out.

For that reason, the claim

"The result is trustworthy because the number is exact"

is better replaced by

"Because the number is single, uncertainty never entered the calculation; how trustworthy the result is depends on where the number came from"

If uncertainty matters to the decision, move to a data type built to carry it; if it does not, staying classical is an honest choice.

Classical data's chief virtue is simplicity. Anyone can read the table, every step can be traced, and the outcome is interpreted directly, with no further decision rule needed: no defuzzification, no scoring function, no comparison of sets.

Because most decision-making methods were built for this structure, it offers the widest choice of methods. Asking an expert for a single number keeps the assessment burden light and lets different experts' scores be compared directly.

What classical data cannot do is carry uncertainty. An expert's hesitation ("7, though it might be 8"), the width of a forecast, a gap in the data and a measurement's margin of error are all erased on entering the table. The result appears as though none of this existed, and a reader may take it as more certain than it is.

There is a further cost when several experts' marks are folded into one classical table: an average is usually taken, and disagreement among the experts disappears inside it. Where that disagreement itself matters, classical data cannot show it.

The most frequent mistake is dismissing classical data as "too simple to be adequate." Moving a measured value into another structure merely to look more sophisticated adds no information; it manufactures uncertainty that was never there.

The opposite mistake is just as damaging: taking a criterion where the expert is plainly unsure, or the data plainly incomplete, and collapsing it to one figure labelled "classical data," which quietly hides information the decision needed.

A third mistake is mixing figures of different pedigrees in one table: a measurement in one column, a word converted into "good = 3" in the next, an averaged expert score in the third. Each is a single number, so the table looks classical throughout, yet the columns differ in reliability, and the report must say so.

The governing principle is this:

A classical figure is the right choice whenever the cell holds one defensible value; only once a genuine, decision-changing uncertainty stands behind that figure should you move to the structure built to carry it, and not before.