All methods

SD

Available

Standard deviation weighting. Diakoulaki et al. (1995)

The idea

Wider spread, more weight
Spread σWeight = σ / 1.0076Profitability (%)0.28120.2791Market share (%)0.34350.3409Productivity (million $)0.38290.3800

Each dot is one alternative on the 0 to 1 scale; the ruler under the dots is σ.

Show as table
CriterionSpread σWeight
Profitability (%)0.28120.2791
Market share (%)0.34350.3409
Productivity (million $)0.38290.3800

Data: Greek pharmaceutical firms (Diakoulaki et al. 1995).

SD rescales every criterion to 0-1 (min-max) and gives more weight to criteria whose values spread more across the alternatives.

When to use it

  • No expert weights, and you want weights from the data alone.
  • A transparent baseline next to CRITIC and Entropy.
  • Data of any sign: negatives and zeros are fine.

Algorithm

  1. 1Min-max normalization

    Same as step 1 of CRITIC.

  2. 2Standard deviation

    Diakoulaki et al. (1995) print the population value (1/m); the weights are the same.

  3. 3Weights

    SD is CRITIC with the conflict term set to 1.

Used together with

Why these pairs
  • SD only weighs; a ranking method ranks. Both published examples feed SD weights into a weighted sum (Diakoulaki et al. 1995, Eq. 9; Mukhametzyanov 2021, Table 2).
  • CRITIC is SD times a conflict term, which reduces the double counting of correlated criteria (Diakoulaki et al. 1995, Tables 3-5).
  • SD, Entropy and CRITIC compared on four related data sets: SD weights varied least (Mukhametzyanov 2021, Table 2).
  • SD then MULTIMOORA or MOOSRA, SD then EDAS, COPRAS, TOPSIS or ARAS, SD then GRA, TOPSIS or ORESTE, as reviewed in Keshavarz-Ghorabaee et al. (2021).

Reference example

Diakoulaki, D.; Mavrotas, G.; Papayannakis, L. (1995)Matches the printed digits
Details

Diakoulaki, D.; Mavrotas, G.; Papayannakis, L. (1995). Determining objective weights in multiple criteria problems: The CRITIC method. Computers & Operations Research 22(7), 763-770. doi:10.1016/0305-0548(94)00059-H

Table 1 (input), Section 5 and Table 3 (SD). Eight firms, three criteria, negative values included. Our weights match the printed 3 decimals (largest difference 8.3e-5); the printed standard deviations are the population values and match once converted (3.5e-4). The four matrices of Mukhametzyanov (2021), Tables 1, 2 and 4, with cost criteria, match to 3 decimals as well (4.9e-4).

Checked by this site's test suite.

Sources

Show 5 sources
  1. Diakoulaki et al. (1995). Determining objective weights in multiple criteria problems: The CRITIC method. Computers & Operations Research 22(7), 763-770. doi:10.1016/0305-0548(94)00059-HSection 4 defines SD weights (Eq. 7) as the baseline for CRITIC, with a worked example; the paper credits the idea of contrast by standard deviation to Zeleny (1982).
  2. Diakoulaki, Mavrotas & Papayannakis (1995), Computers & Operations Research 22(7), 763-770 doi:10.1016/0305-0548(94)00059-H
  3. Mukhametzyanov (2021), DMAME 4(2), 76-105 doi:10.31181/dmame210402076i
  4. Zeleny (1982), Multiple Criteria Decision Making, McGraw-Hill
  5. Keshavarz-Ghorabaee et al. (2021), Symmetry 13(4), 525 doi:10.3390/sym13040525