SAW
AvailableSimple Additive Weighting (Weighted Sum Model). Churchman and Ackoff (1954)
The idea
Each segment is one criterion's weight times its normalized value; the bar ends at S.
- GDP growth (%)
- Production growth (%)
- Average annual salary (EUR)
- Unemployment rate (%)
- Export / import ratio
Show as table
| Alternative | GDP growth (%) | Production growth (%) | Average annual salary (EUR) | Unemployment rate (%) | Export / import ratio | Score S |
|---|---|---|---|---|---|---|
| Lithuania | 0.3500 | 0.1000 | 0.1038 | 0.2004 | 0.1201 | 0.8744 |
| Estonia | 0.1840 | 0.0609 | 0.1459 | 0.2500 | 0.1152 | 0.7560 |
| Latvia | 0.2706 | 0.0404 | 0.1011 | 0.2257 | 0.0905 | 0.7283 |
| Poland | 0.1371 | 0.0522 | 0.1700 | 0.1205 | 0.1300 | 0.6098 |
Data: Baltic economies (Podvezko 2011).
SAW rescales every criterion so that the best value gets 1, multiplies each value by the criterion weight and adds the results.
When to use it
- All values are positive measurements: prices, rates, counts.
- You want a score anyone can recompute by hand.
- A transparent baseline next to TOPSIS, VIKOR or EDAS.
Algorithm
1Linear max normalization
Default; the sum option uses x / Σx, and 1 / x in place of x for a cost criterion.
2Contributions
Each criterion adds its weight times its normalized value.
3Weighted sum
Rank by decreasing S; equal scores share a rank.
Used together with
Why these pairs
- SAW only ranks; published studies take the weights from experts, Entropy or CRITIC (Podvezko 2011; Vujičić et al. 2017).
- SAW is WASPAS with λ = 1, and COPRAS with at most one cost criterion is SAW with sum normalization.
- The usual yardstick when a new ranking method is tested, as in the EDAS paper (2015).
- Run side by side with MOORA under six normalizations in an air-conditioner selection (Vujičić et al. 2017).
Reference example
Details
Podvezko, V. (2011). The Comparative Analysis of MCDA Methods SAW and COPRAS. Inžinerinė ekonomika - Engineering Economics 22(2), 134-146. doi:10.5755/j01.ee.22.2.310
Tables 1-3 and 5-6. Four countries, five criteria, weights from the paper. Sum normalization matches Table 5 and the scores of Table 6 to 3 decimals (largest difference 4.1e-4). Linear max matches Table 3 for three countries: Table 2 prints Lithuania's salary ratio as 0.599 instead of 306/501 = 0.611, so the paper has S(Lithuania) = 0.872 and we have 0.8744, with the same ranking; with the misprint our code gives Table 3. Ranks also match Keshavarz Ghorabaee et al. (2015) in 7 of 7 weight sets and Vujičić et al. (2017) for both weightings.
Checked by this site's test suite.
Sources
Show 7 sources
- Churchman and Ackoff (1954). An Approximate Measure of Value. Journal of the Operations Research Society of America 2(2), 172-187. doi:10.1287/opre.2.2.172A classical method with no single origin paper: first use is credited to Churchman & Ackoff (1954), the textbook statement is Hwang & Yoon (1981), and the formulas and the example here follow Podvezko (2011).
- Churchman & Ackoff (1954), Journal of the Operations Research Society of America 2(2), 172-187 doi:10.1287/opre.2.2.172
- Hwang & Yoon (1981), Multiple Attribute Decision Making, Springer doi:10.1007/978-3-642-48318-9
- Podvezko (2011), Engineering Economics 22(2), 134-146 doi:10.5755/j01.ee.22.2.310
- Keshavarz Ghorabaee, Zavadskas, Olfat & Turskis (2015), Informatica 26(3), 435-451 doi:10.15388/Informatica.2015.57
- Zavadskas, Turskis, Antucheviciene & Zakarevicius (2012), Elektronika ir Elektrotechnika 122(6), 3-6 doi:10.5755/j01.eee.122.6.1810
- Vujičić, Papić & Blagojević (2017), Tehnika 72(3), 422-429 doi:10.5937/tehnika1703422V