《BiasesinEnvironmentalDecisionAnalysis-MultipleCriteria在環(huán)境決策分析-多標準偏差》由會員分享，可在線閱讀，更多相關(guān)《BiasesinEnvironmentalDecisionAnalysis-MultipleCriteria在環(huán)境決策分析-多標準偏差（42頁珍藏版）》請在裝配圖網(wǎng)上搜索。

1、Can We Avoid Biases in Environmental Decision Analysis ?,Raimo P. Hmlinen Helsinki University of Technology Systems Analysis Laboratory raimohut.fi www.paijanne.hut.fi,Structure of the presentation,Background & decision analysis interviews Goals of the study Case: Regulation of Lake Pijnne Splitting

2、 bias & swapping of levels Description of the experiment Results of the experiment Conclusions ?,Environmental decision analysis,Parliamentary nuclear power decision (Hmlinen et. al) Decision analysis interviews (Marttunen & Hmlinen) Spontaneous decision conferencing in nuclear emergency management

3、(Hmlinen & Sinkko),Cognitive biases,Splitting bias attribute receives more weight if it is split origins: subjects give rank information only (Pyhnen & Hmlinen) Not observable in hierarchical weighting,Decision analysis interviews,Opinions of large groups of people traditionally collected through qu

4、estionnaires Decision analysis interviews may provide a more reliable way to collect these opinions Idea: one value tree for all = common terminology emphasis on finding the viewpoints of different stakeholder groups interactive, computer supported,Research interest,Existence of biases in a real cas

5、e Can biases can be avoided through training and proper instructing ? Identify what can go wrong in the Lake Pijnne case Compare the well trained university students and spontaneous stakeholders responses,The Lake Pijnne case,Regulation started 1964 Main aims were to improve hydroelectricity product

6、ion and to reduce damages caused by flooding Environmental values & increase in free time need for an improved regulation policy,,Splitting bias,When an attribute is split, the weight it receives increases,0.4,0.3,0.3,0.3,0.3,0.1,0.3,0.4,Swapping of levels,Does the order of the levels affect the res

7、ulting weights? Important question in environmental decision analysis: stakeholder groups may vary regionally Not studied before,Example of swapping of levels,,Lake Pijnne,,River Kymijoki,,Attribute 3,,Attribute 3,,Attribute 2,,Attribute 1,,Attribute 2,,Attribute 1,,,,,,,,,,,,,,,,,,,Attribute 3,,Att

8、ribute 2,,Attribute 1,,River Kymijoki,,Lake Pijnne,,River Kymijoki,,Lake Pijnne,,River Kymijoki,,Lake Pijnne,,,,,,,,,,,,,,,,,Earlier experiments on biases,Structure of the decision model affects the results Previous experiments typically: subjects: university students problems: artificial results: t

9、aken from group averages Lake Pijnne-case: a real problem with real stakeholders,Important new features,Realistic case Decision analysis interviews instead of passive decision support or survey Interactive computer support (resulting weights shown immediately) Instructions and training before the we

10、ighting,Subjects:,University students attending a course on decision analysis (N = 30) held during a tutorial session, not mandatory Habitants of Asikkala (N = 40) 3 groups of students 1 group of adults (volunteers) 3 experts from the Finnish Environment Institute & 2 summer residence owners,Experim

11、ental setting,Weighting done with the SWING method using a tailored Excel interface Subjects entered the numbers themselves, two assistants were present to help Resulting weights shown as bars Order of value trees partly randomized,Sessions,A short introduction to: Lake Pijnne case value trees & wei

12、ghting different structures of the value tree In HUT the avoidance of biases was emphasized more Duration: 60 - 90 minutes,SWING method,Easy to use Attribute ranges clearly presented Idea: choose the attribute you would first like to move to its best level assign it 100 points assign other attribute

13、s points less than 100 in respect to the first attribute,Flat-weighting,,,Muu talous ???,,Vesivoima,,Vesivoima,,Muu talous,,Ymprist,,Talous,,,,,,,,,,,,,,,,,,,Rantojen kytettvyys,,Virkistyskalastus,,Kalojen lisntyminen,,Rantakasvillisuus,,Lahtien,umpeenkasvu,,Virkistys,,Luonto,,,,,,,,,,,,,,,,Tulvat,

14、maatalous ja,teollisuus,,Tulvat, loma-asutus,,Vesiliikenne,,Ammattikalastus,Upper level weights:,,,Muu talous ???,,Vesivoima,,Vesivoima,,Muu talous,,Ymprist,,Talous,,,,,,,,,,,,,,,,,,,Rantojen kytettvyys,,Virkistyskalastus,,Kalojen lisntyminen,,Rantakasvillisuus,,Lahtien,umpeenkasvu,,Virkistys,,Luont

15、o,,,,,,,,,,,,,,,,Tulvat, maatalous ja,teollisuus,,Tulvat, loma-asutus,,Vesiliikenne,,Ammattikalastus,ENV5-tree:,,,,Luonto,,Virkistys,,Ymprist,,Talous,,,,,,,,Rantojen kytettvyys,,Virkistyskalastus,,Kalojen lisntyminen,,Rantakasvillisuus,,Lahtien,umpeenkasvu,,,,,,,,,,,,,,,ENV2-tree:,,,,Luonto,,Virkist

16、ys,,Ymprist,,Talous,,,,,,,EC5-tree:,,,Muu talous ???,,Vesivoima,,Vesivoima,,Muu talous,,Ymprist,,Talous,,,,,,,,,,,,,,,,Tulvat, maatalous ja,teollisuus,,Tulvat, loma-asutus,,Vesiliikenne,,Ammattikalastus,EC2-tree:,,,Muu talous ???,,Vesivoima,,Muu talous,,Talous,,,,Muu talous ???,,Vesivoima,,Muu talou

17、s,,Ymprist,,Talous,,,,,,,Swapping of levels:,Muu talous ???,,Kymijoki ja muut,,Pijnne,,Rantakasvillisuus,,Tulvavahingot,,,,,,,,Kymijoki ja muut,,Pijnne,,,,Muu talous ???,,Rantakasvillisuus,,Tulvavahingot,,Kymijoki ja muut,,Pijnne,,,,,,,,Rantakasvillisuus,,Tulvavahingot,,,,Flat weights vs. upper leve

18、l weights,Both in group averages and in results of individuals the total weights for the environment and economy were similar with both methods One explanation: symmetric value tree,Splitting bias,A typical resident in Asikkala,,,ENVIRONMENT,ECONOMY,5 1 5 2 1 1,5 1 1 1 5 2,Example from HUT(one of th

19、e best ones),ENVIRONMENT,ECONOMY,5 1 5 2 1 1,5 1 1 1 5 2,Why even weights ?,Some students: none of the attributes seemed to be important Asikkala: all of the attributes were important,,even weights for all attributes,What caused the bias ?,Similar points for all attributes in one branch regardless

20、 of the structure of the value tree,Effect of instructions,Students had good instructions only some had bias in their results In the spontaneous stakeholders sessions the information load was too high and thus the instructions were not adopted as well nearly all had systematically consistent bias,,

21、,,,,,,,,,STUDENTS,STAKEHOLDERS,Adjusted / not adjusted weights,Examples,STUDENTS,STAKEHOLDERS,Observation,The students and the experts from FEI could nearly avoid the splitting bias good background education + instructions did reduce the bias What did the students think? - Arithmetics or real avoida

22、nce of biases,Avoiding the splitting bias ?,Good instruction can eliminate it When the economical attributes were split, the magnitude of the bias was slightly larger Graphical feedback did not eliminate Hierarchical weighting,Swapping of attribute levels,If the order of the levels would not affect

23、the weigts, the pairs of weights should be equal (as in the first picture),Conclusions about swapping of levels ?,Only a few had clearly differing weights with the two trees No systematic pattern was found Less differences residents of Asikkala and students than with the splitting bias A simple scal

24、e lead to similar weights with both trees (100, 70 for example) Neither tree gained clear support,Solutions to reduce biases ?,Hierarchical weighting Models should be tested on real decision makers Interactiveness of weighting (= possibility to return to change the points given earlier ) Well balanc

25、ed trees,Other observations in Asikkala,Concept of weight seemed to be difficult for most subjects in Asikkala Information load was high Facilitators role becomes important when the DMs are uncertain,Problems related to the Lake Pijnne case,Current regulation policy cannot be improved very significa

26、ntly no big differences between the alternatives unrealistic hopes and false information are probably larger problems than the regulation itself money is not money strong feelings against the power companies and regulation (shape of value function ?),Suggestions for future research,Hierarchical weig

27、hting Encouragement to reconsider and readjust the statements iterate Decision Analyst must supervise!,R.P. Hmlinen, E. Kettunen, M. Marttunen and H. Ehtamo: Evaluating a framework for multi-stakeholder decision support in water resources management, Group Decision and Negotiation, 2001. (to appear)

28、 M. Pyhnen, Hans C.J. Vrolijk and R.P. Hmlinen: Behavioral and procedural consequences of structural variation in value trees. European Journal of Operational Research, 2001. (to appear) M. Pyhnen and R.P. Hmlinen: There is hope in attribute weighting, Journal of Information Systems and Operational

29、Research (INFOR), vol. 38, no. 3, Aug. 2000, pp. 272-282. Abstract R.P. Hmlinen, M. Lindstedt and K. Sinkko: Multi-attribute risk analysis in nuclear emergency management, Risk Analysis, Vol. 20, No 4, 2000, pp. 455-467. M. Pyhnen and R.P. Hmlinen: Notes on the weighting biases in value trees, Journ

30、al of Behavioral Decision Making, Vol. 11, 1998, pp. 139-150. Susanna Alaja: Structuring effects in environmental decision models, Helsinki University of Technology, Systems Analysis Laboratory, Theses, 1998.,References,M. Pyhnen, R.P. Hmlinen and A. A. Salo: An experiment on the numerical modeling

31、of verbal ratio statements, Journal of Multi-Criteria Decision Analysis, Vol. 6, 1997, pp. 1-10. R.P. Hmlinen and M. Pyhnen: On-line group decision support by preference programming in traffic planning, Group Decision and Negotiation, Vol. 5, 1996, pp.485-50. M. Marttunen and R.P. Hmlinen: Decision

32、analysis interviews in environmental impact assessment, European Journal of Operational Research, Vol. 87, No. 3, 1995, pp. 551-563. R.P. Hmlinen, A.A. Salo and K. Pysti: Observations about consensus seeking in a multiple criteria environment, in: Proceedings of the Twenty-Fifth Hawaii International

33、 Conference on System Sciences, Vol. IV, 1991, IEEE Computer Society Press, Hawaii, pp. 190-198. R.P. Hmlinen: Computer assisted energy policy analysis in the parliament of Finland, Interfaces, Vol. 18, No. 4, 1988, pp. 12-23. Also in: Case and Readings in Management Science, 2nd edition, M. Render, R.M. Stair Jr. and I. Greenberg (eds.), Allyn & Bacon, Massachusetts 1990 pp. 278-288.,