Home

DEXi: A Program for Multi-Attribute Decision Making

1. mmm 50 4t Houndin down e no up Harmalizatia Sum 100 Mas 100 i i Apply Cancel In this editor you can set your required weights by moving the sliders of the corresponding attributes The numerical values of these settings are shown under Required In addition you can normalize these weights using the buttons Sum 100 The weights are proportionally scaled so that their sum is 100 Max 100 The weights are proportionally scaled so that the maximal weight is equal to 100 The blue bars and blue numbers show the actual Current weights These are determined from the utility function currently edited in Function Editor Rounding specifies how to calculate the values of non entered rules that fall just between two ordinal values For example let ordinal numbers of acc and good be 2 and 3 respectively For some non entered rule the linear approximation formula using weights can calculate the value 2 5 In this case Rounding specifies whether this value is rounded down to 2 meaning acc or up good The default setting of no does not favor any specific direction and leaves the rounding to the underlying arithmetic software 3 11 6 1 Remarks Absolute values of weights are not really important in DEXi only their relative proportions with each other do matter Both normalizations preserve this property It is important to understand that Current weights represent th
2. o improved handling of the ENTER key Function Editor o consistent implementation of the two non entered value handling strategies Use scale orders and Use weights o added pop up menu for improved data entry o added Enter values command Options Page o required confirmation before deleting an option Bug fixes Model Page Function transformations when moving attributes up and down Display of Evaluation Page when there are no options Displaying button glyphs and icons Avoiding initial crash hopefully when there are no printers installed Version 3 0 New program features Added Undo Redo functionality when editing models options and utility functions Added option analysis functionality o added Analysis Menu and toolbar buttons on the Evaluation Page o implemented three analysis methods Plus minus 1 analysis Selective explanation and Compare options 57 e Improved option data entry right click menus are now available also on the Options and Evaluation Pages e Added Shrink Expand commands to alter the display of attribute trees on the Model Page 5 4 2 Changed program features e Added Undo Redo selection to Settings Advanced e Added file name on report headings 5 4 3 Changed documentation e Updated English help e Updated DEXi Web Page e New URL of the DEXi Web page http kt ijs si MarkoBohanec dexi html 5 4 4 Bug fixes e Function Editor The right click menu now correctly enters
3. 35 use weights 36 value 32 weights 17 Function Editor 31 pop up menu 33 status bar 33 toolbar 30 32 G global weights 18 H Help Menu 26 hierarchy 14 hyperplane 17 36 if then rule 16 import function 24 import options 24 increasing scale 15 Internal browser 43 interval 16 L linked attribute 14 local weights 18 Main Toolbar 23 menu Analysis 40 Chart 41 Edit 26 File 23 Help 26 pop up Charts Page 41 Function Editor 33 Window 26 model decision 12 description 24 inputs 14 multi attribute 5 12 multi criteria 12 outputs 14 Model Page 27 Model Window 26 monotonicity 35 multi attribute model 5 12 multi critera model 12 N non entered value 32 handling 35 normalization 18 34 normalized weights 18 O option 20 analysis 20 40 Car Evaluation Model 49 data file 24 description 20 38 49 edit 37 evaluation 20 39 49 evaluation results 20 value 38 39 options compare 406 51 export 24 import 24 Options Page 37 ordered scale 15 P page Attributes 41 Charts 40 Evaluation 38 Model 27 Options 37 41 Settings Advanced 45 Import Export 45 Report 44 participants decision analysts 11 experts 11 in decision process 11 stakeholders 11 users 11 Plus minus 1 analysis 40 50 Preview 43 toolbar 43 Q qualitative decision model 12 quantitative decision model 13 R radar chart 41 53 redo 28 33 3
4. The overall evaluation of an option is finally obtained as the value of one or more root attributes of the model On this basis the decision maker can compare and rank the options and possibly identify and select the best one In the evaluation undefined values of basic attributes denoted are interpreted as sets of all possible values that can be assigned to corresponding attributes There DEXi evaluates options trying all these values and keeps track of the evaluation results in these cases Therefore an evaluation result is not necessarily represented by a single attribute value but can also be a set of values In DEXi results of option evaluation are shown on its Evaluation Page as well as in charts and reports 2 13 Analysis Analysis is one of the key concepts in Decision Analysis In contrast with evaluation which is merely a calculation directed from inputs option descriptions to outputs evaluation results analysis is 20 understood as an active involvement of participants who are trying to find answers to questions such as e Are option evaluations in accordance with expectations Are they correct If not why e How do the options compare with each other Which one is the best and why e Can we explain and justify the evaluations What are the most important strong and weak points of individual options e What if something changes What if we try a new option What if an option becomes unavailable What if
5. or options transposed e ndent specifies whether attribute names are indented or not Indentation adds whitespace to outline the tree structure of attributes Function data format These controls specify the format and contents of exported function data files e Attribute values specifies the format of attribute values which can be either text strings or ordinal numbers starting at 0 base 0 or 1 base 7 e Rules export all or only entered decision rules e Interval format specifies the format of function values which are in general intervals Possible formats are o from to o from and to separated by the TAB character or o Using interval symbols 2z z and 3 19 3 Advanced Page This page contains two controls Link equal attributes specifies whether DEXi performs automatic linking of attributes or not Undo Redo to activate or deactivate the Undo Redo functionality when editing models options and utility functions 45 4 Example Car Evaluation Model This is an example of a DEXi multi attribute model for the evaluation of cars This is a very small and simple model used to illustrate main concepts of multi attribute modeling and DEXi and is not meant to address the problem of car evaluation at any realistic level This model has been traditionally handed out together with all previous versions of the programs DEX and DEXi For this reason it is shown here without change in spite of its age 4 1
6. CAR Car 7 Cart medum kw mee 4 big high Car medium medium moe 4 big medium 4 4 1 Interpretation At the bottom the table shows two options Car and Car2 described by the qualitative values assigned to the six basic attributes of the tree These values then aggregated from bottom to the top of the tree of attributes according to the structure of the tree and defined utility functions In this way intermediate evaluation results are first obtained and assigned to the attributes PRICE COMFORT and TECH CHAR see the table at the top Finally the values of PRICE and TECH CHAR are aggregated into CAR giving the overall evaluation of both cars 4 5 Some Car Option Analyses These examples show some option analysis reports obtained from the Car Evaluation Model These reports can be created by commands of the Analysis Menu or corresponding buttons on the Evaluation Page 4 5 1 Plus minus 1 analysis The first example shows results of Plus minus 1 analysis for the option Car2 and the aggregate attribute CAR The column Car2 displays the current values of Car2 The column 1 displays the values of the attribute CAR when each corresponding lower level attribute s value changes by one step down independently of other attributes Similarly the column 1 shows the effects of increasing the value by one step up Empty fields denote no effect and the brackets and indicate that t
7. PRICE Frice af a car BUY PRICE Buying price MAINT PRICE Maintenance price TECH CHAR Technical characteristics COMF ORT Combo rt PERE Maximum number af passengers SDOORS Number of doors LUGGAGE Size of the luggage boot SAFETY Cars safety 4 2 Scales for Car Evaluation The scales of attributes in the Car Evaluation Model are defined as follows Attrib ute Scale CAR unacc acc good exe PRICE high medium low BUY PRICE high medium low MAINT PRICE high medium low TECH CHAR bad acc good exe COMF ORT small medium Aigh oor to 2 3 4 more SDOORS 2 3 4 more LUGGAGE small medium big SAFETY sm all medium high 4 3 Utility Functions for Car Evaluation Consider the root attribute of the Car Evaluation Model According to the tree of attributes the root attribute CAR depends on two lower level attributes PRICE and TECH CHAR Thus the utility function that corresponds to CAR maps all the combinations of values of PRICE and TECH CHAR into the values of CAR see the scales of these attributes The function is defined by the table shown below TECH CHAR CAR bad unacc acc unacc good unacc exc unacc bad undacc acc acc good good axe unacc good exc exc TECH CHAR 47 4 3 1 Interpretation The attributes PRICE and TECH CHAR have three and four values respectively so the number of rows in the table is 3x4 12 Each row provides a value of CAR for one combination of the values of PRICE and TECH CHAR
8. Tree of Attributes for Car Evaluation The Car Evaluation Model has the following tree structure of attributes CAR PRICE TECH CHAR BUY PRICE MAINT PRICE COMFORT SAFETY PERS EDOOKS5 LU AE 4 1 1 Interpretation This structure can be interpreted as follows 1 Decomposition In order to evaluate a CAR we consider its PRICE and TECHnical CHARacteristics PRICE is further decomposed into BUYing PRICE and MAINTenance PRICE Similarly TECH CHAR are decomposed into COMFORT and SAFETY and COMFORT is further decomposed into the number of PERSons passengers number of DOORS and size of the LUGGAGE boot 2 Dependency The attribute CAR depends on PRICE and TECH CHAR Similarly COMFORT depends of ZPERS ZDOORS and LUGGAGE Etc 3 Aggregation The values of ZPERS ZDOORS and LUGGAGE are aggregated into a value uf COMFORT Then in the following order BUY PRICE and MAINT PRICE are aggregated into PRICE COMFORT and SAFETY are aggregated into TECH CHAR and PRICE and TECH CHAR are aggregated into CAR 4 1 2 Attribute Types The attributes in this tree are of the following types e Basic attributes are BUY PRICE MAINT PRICE ZPERS DOORS LUGGAGE and SAFETY e Aggregate attributes are CAR PRICE TECH CHAR and COMFORT e The root attribute is CAR 46 4 1 3 Attribute Descriptions In DEXi s reports this tree is printed together with attributes descriptions and appears as follows Attribute Description CAR Quality of a car
9. Weights can also be normalized or not This is because some scales can have more values than the others Geometrically larger scales appear longer they have lower slopes and consequently smaller weights Normalization refers to the procedure in which all scales are adjusted to the same length unit interval before determining the weights Usually this is the better method of weight assessment and comparison of attributes 2 10 4 Combinatorial Explosion Consider a utility function f that maps the values of the attributes X X X into the value of the aggregate attribute Y In DEXi a utility function maps all the combinations of the lower level attribute values into the values of Y Suppose that each X has a scale consisting of s values Then the number of combinations and thus the size of fis equal to SEs X5 X X s 18 In other words when defining f you should define S decision rules 2 10 4 1 Example Let all the n lower level attributes have s valued scales In this case the size of f equals to S s The following table shows how fast S grows with the increasing n and s Number of lower level attributes x Easa al c 4 4 8l 256 4 8 m 64 256 NEE 243 END 2 10 4 2 Recommendations Experience shows that utility functions of size up to 25 are small and usually quite easy to define The difficulty grows towards the size of about 100 which is already quite difficult Everyth
10. affect the decision or its consequences for example competition response weather e Multiple and possibly conflicting goals e Group decision making involvement of different decision makers or groups that have different and possibly conflicting goals e Important consequences of the decision such as possible big financial losses or environmental impacts e Limited resources to conduct the decision process most often available time and expertise 2 3 Decision Process The ultimate goal of a decision process is to solve a decision problem that is to make a decision In Decision Analysis the decision process is understood as a process of careful and in depth analysis of the decision problem It involves a systematic acquisition and organization of knowledge about the decision problem which is done by participants of the decision process and typically includes e assessing the problem e collecting and verifying information e identifying options alternatives 10 anticipating consequences of decisions making the choice using sound and logical judgment based on available information justification and informing others of the decision and its rationale evaluating decisions and their consequences In general such a process should provide all the information needed for a sufficiently good decision reduce the chance of overlooking important information and making other errors improve the effectiveness and efficiency o
11. and with italic letters in reports On the Options Page however any chain of linked attributes appears only once so there is no need for duplicate data entry Linked attributes are also properly handled during the evaluation and on the Evaluation Page 2 8 3 Recommendations 1 Before editing a tree in DEXi sketch it on paper Before making a real tree create an unstructured list of attributes Brainstorm At this stage try not to overlook important attributes but do not bother about their structure or redundancy 3 When making a structure of attributes create meaningful subtrees that contain related attributes Try structuring your unstructured list in two directions o Bottom up Group similar attributes together into a single higher level attribute It is usually a good indication if you can find a meaningful name for it o Top down Decompose complex attributes into simpler ones 4 Avoid meaningless redundant duplicate inessential and unoperational attributes In other words check each basic attribute that o ithasa well defined meaning o It does not duplicate or overlap with some other attributes o it does affect the decision and you know how at least approximately o itcan be measured or assessed with sufficient accuracy 5 Avoid aggregate attributes that have more than three descendants Too many descendants cause a combinatorial explosion on the size of corresponding utility functions making them extremely difficult t
12. are only those rules whose argument values in the file are matched with the ones in DEXi Even though DEXi tries to recognize the data format used in the function file it is strongly recommended that function data is exported and imported using the same Import Export Settings Import Import options Open an option file and load its data into the current model Existing options that have same names as imported options are overwritten by imported data Otherwise imported options are inserted into the model Export Export tree Export the current tree of attributes to an external file Two data formats are supported e Tab delimited textual format with columns containing attribute names scales and descriptions e GML Portable Graph File Format for creating tree graphs by programs such as yEd Graph Editor Export Export function Save the utility function that is currently selected in the current window Data is written to a function file Before saving you are asked to define the name and format of this file Export Export options Extract option data from the current model and save it on an option file Only options that are selected checked on the Options sub page of the Charts Page are actually exported Before saving you are asked to specify the name and format of the file Model description View and edit a description of the current model The entered text is used only for documentation and can be displa
13. d Model Edit the current DEXi model its structure and components ph Options Edit option descriptions 26 Evaluation Evaluate and analyze options El Charts Compose and view charts 3 9 Model Page The Model Page of DEXi model window provides workspace and commands for editing a DEXi model its structure attributes scales and utility functions but excluding options and program settings 3 9 1 e Model sze Options Km Evaluation H Charte E m CAR PRICE s 4 BUY PRICE eq MAINT PRICE TECH CHAR m COMFORT Quality of a car JA 4 PERS Scale jx HDOORS pomi m 4 LUGGAGE unace acc good exc 4 SAFETY Attributes 10 6 basic O linked aggregate Scales 10 Functions Options Z Attribut Mame Car Description Utility Function Rules 12 12 100 00 determined 100 00 unace B ace 1 good 2 exc 3 Description 2 5 EN as ee E AA Workspace The workspace consists of four main areas that correspond to the following model components 1 4 3 9 2 Tree of attributes On the left there is an Explorer like Tree View that displays the structure of the edited model It has an associated vertical toolbar All toolbar buttons and most of the corresponding Edit Menu commands refer to the currently selected attribute in the tree Attribute There are two entry fields in which you can name or rename the currently s
14. function values e Corrected calculation of weights of underspecified utility functions 5 5 Version 3 01 5 5 1 New program features e Slightly changed File Menu added Import and Export submenus e Added File Export Export tree menu item for exporting trees of attributes using two data formats o Tab delimited textual format with columns containing attribute names scales and descriptions o GML Portable Graph File Format for creating tree graphs by programs such as yEd Graph Editor 5 5 2 Bug fixes e Generation of decision rules of incompletely specified utility functions correctly interpreting the setting entered rules only e Treatment of linked attributes in Plus minus 1 Analysis 5 6 Version 3 02 5 6 1 New program features e Added Model Description functionality o Added File Model Description File Menu item o Added Model description report item e Added functionality to search attribute names descriptions and scales for text strings o Added Edit Menu items Edit Find and Edit Find next e Added a drag and drop operation to join two father son attributes into a single attribute provided they have compatible scales e Improved information displayed in status bars 5 6 2 Bug fixes e Activation of Export function e Proper counting of attributes of different types in status messages 58 5 7 Version 3 03 5 7 1 New program features Improved handlin
15. in normal typeface and by default they are recalculated whenever the table changes The calculation is based on already entered values and other available information particularly weights DEXi uses two strategies for calculating non entered values which can be individually activated or deactivated using the two corresponding checkboxes on the Function Editor toolbar Use scale orders and Use weights These strategies are available only for functions whose attributes have increasingly ordered scales For other functions Function Editor does not show the corresponding checkboxes 3 11 7 1 Scale orders This strategy takes into account the ordering of scales Consider the function y f x where both y and x have increasingly ordered scales Then whenever x increases it is clear that f x should also increase or remain constant This function property is known as monotonicity SE Decision rules TECH CHAR B x l Use scale orders Use weights small 2 small medium bad 3 small high bad 4 medium small bad 5 medium medium acc B medium high 900 high small bad 8 high medium b ac n J high high exc Rules 4 9 44 4475 determined 85 1995 bad 5 acc 3 qgood 2 exc 3 It is easy to see that in general monotonicity narrows the intervals of values that can be assigned to non entered cells For example the rule 5 above assigns acc to the combination COMFORT medium and SAFETY medi
16. is as follows 13 e basic attributes represent inputs of the model e root attributes represent its main outputs and e other aggregate attributes represent intermediate results of option evaluation 2 9 2 Linked Attributes In principle DEXi s multi attribute models have a strict tree structure attributes are structured so that there is exactly one path from each aggregate attribute to the root of the tree This means that each attribute other than the root influences exactly one parent in the tree Sometimes this is not enough and we wish to introduce attributes that influence more than one parent In other words we wish to create attribute hierarchies directed acyclic graphs rather than simple trees For this purpose DEXi version 2 0 introduced the concept of inked attributes The idea is that whenever there are two attributes in the tree that have equal names and equal scales and at least one of them is basic they are declared linked and they logically represent a single attribute Attribute linking is done automatically by DEXi but only when explicitly enabled in Settings Advanced By default linking is disabled and equally named attributes are considered different This concept allows that DEXi s trees still retain their basic tree structure In tree displays linked attributes appear separately so the tree structure is preserved it is only that linked attributes are represented by special symbols Ul in Tree Views
17. it deletes that attribute Thus to completely delete a sub tree of attributes you should delete it twice de Cut Equivalent to Copy followed by Delete subtree or item Copy Copies the currently selected sub tree into the clipboard for further use E Paste Inserts previously Cut or Copied sub tree into the model positioning it as a descendant child of the currently selected attribute Ei Duplicate Makes a copy of the currently selected sub tree and inserts it as a new top level tree in the model so that it can be easily moved around Find Opens a window in which you can define a text to be searched for in the model You can also select components in which to search attribute names attribute descriptions and or scales The search can be case sensitive or not FA Find next Continues searching from the current position in the model 1 Move up Moves the currently selected attribute one place up according to what is shown in the tree view Move down Moves the currently selected attribute one place down according to what is shown in the tree view El Shrink tree Shrinks the display of the tree by one level El Expand tree Extends the display of the tree by expanding the inner most collapsed sub trees Undo Undo the last tree editing operation Redo Redo the last undone tree editing operation I Scale Invokes the Scale Editor to create or edit the scale of the currently selected attribute Delete
18. scale Deletes the scale of the currently selected attribute F Utility function Invokes the Function Editor to create or edit the utility function of the currently selected aggregate attribute Delete function Deletes the utility function of the currently selected aggregate attribute 28 3 9 3 Remarks Changing the structure and scales of attributes can severely affect previously defined utility functions In some cases for example when adding or deleting an attribute value DEXi tries to adapt the affected function so that its meaning is preserved as much as possible Unfortunately such adaptation is impossible with more extensive changes particularly when adding or deleting function arguments In these cases the function must be deleted and defined anew Before deleting a function DEXi issues a warning and asks for your confirmation Consequently you might want to develop your tree structure as completely as possible before attempting to define utility functions 3 9 4 Tree View Tree View is an important part of the DEXi model window and its Model Page It shows the structure of the currently edited DEXi model i PRICE 4 BUY PRICE Ei TECH CHAR E m COMFORT m q HPERS 4 HDODAS In Tree View you can e Selecta single attribute by clicking it or using keyboard arrows e Invoke Model Page commands Most of these commands refer to the currently selected attribute e Move attributes and subtrees shown in th
19. three or more dimensions the maximum is six The next chart shows the evaluation of cars using the three attributes PRICE COMFORT and SAFETY Car1 Car2 co MFOR SAFETY 53 5 Further reading 5 1 Documentation in Slovene The DEXi User s Manual is available as Jereb E Bohanec M Rajkovi V DEXi Ra unalni ki program za ve parametrsko odlo anje Moderna organizacija Kranj 2003 Further information on decision analysis multi attribute modeling fundamental DEXi concepts and underlying methods is available in Bohanec M Odlo anje in modeli DMFA zalo ni tvo 2006 5 2 Selected Publications Bohanec M Rajkovi V Ve parametrski odlo itveni modeli Organizacija 28 7 427 438 1995 http kt ijs si MarkoBohanec org95 index html Bohanec M Rajkovi V Multi attribute decision modeling Industrial applications of DEX Informatica 23 487 491 1999 Bohanec M Zupan B Rajkovic V Applications of qualitative multi attribute decision models in health care nternational Journal of Medical Informatics 58 59 191 205 2000 Cestnik B Bohanec M Decision support in housing loan allocation A case study DDM 2001 ECML PKDD 2001 Workshop Integrating Aspects of Data Mining Decision Support and Meta Learning Positions Developments and Future Directions eds Giraud Carrier C Lavrac N Moyle S Kavsek B Freiburg 21 30 2001 Leben A Bohanec M Evaluation of life e
20. use the default dxi file format The formats dax and xml are obsolete and do not support all DEXi s features They are implemented only for compatibility with DEX and jDEX The dxi files are composed as XML documents so you may use other XML tools to process them In order to conform to the XML standard and fully support UTF 8 strings the format of dxi files has changed from DEXi 1 04 to 2 0 DEXi versions prior to 2 0 cannot process the new files and refuse to load them DEXi 2 0 and above support both formats as follows e Loading File format is recognized automatically e Saving By default files are saved using the new format For the old format select DEXi version 1 in the Save file dialogue For versions 2 0 and above DEXi files are in principle backward compatible for reading but not for writing This means that any previous version of DEXi can read a file that was created by a more recent version however attempting to save this file by the previous version may result in data loss Regular updating of DEXi to new versions is therefore strongly advised 3 3 Main Toolbar The main toolbar of DEXi provides four buttons for fast activation of top level DEXi model commands New Create a new DEXi model and show it in a new window ar Open Load an existing DEXi model from a DEXi file and show it in a new window ll Save Save the current DEXi model on a DEXi file File name is requested only when the model is saved for th
21. value setup B medium lee low Enter values Z low high high B low medium low gj low low low 31 3 11 1 Table The central part displays a table of decision rules Table rows contain all the possible combinations of values of function s arguments lower level attributes in the tree The rightmost column such as PRICE above represents the output function value This is the only editable column in the table It contains cells into each of which you can assign e asingle value taken from the corresponding scale for example high or e an unknown value denoted this is the initial value set to all rows of a newly created table The values entered by you appear in bold Other values are called non entered and are shown in normal typeface By default non entered values are handled by DEXi and recalculated each time the table has changed In general non entered values are intervals You can enter a cell value in one of the following ways e selecting an item from the Rule values field which appears above the table e right clicking and selecting an item from the pop up menu or e pressing one of the keys 1 2 on the keyboard which represent consecutive scale values or the key which represents an unknown value To delete the currently selected cell value press the Delete key or the button Deleting changes cell status to non entered Clicking on a table heading shows the description and scale values
22. 5 A A PEAT A OEE EES EAEE TEETE OEE E AT E T O RA 16 2102 88 9 8 a Sia 16 2103 A O banana 17 ZAQA gt GCombinatorialEXplOSIOT xi eite e detener etos be buena fucus esae ie ees Poste ostendere dne Ad 18 2 11 OPUS eat 20 212 EVA OF OPUS mmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmi 20 ah Ae 4 32 s SE PI o ia Ded rain atus 20 3 DEXI components and Gorman sia a ee i n 22 SN MES ee o A A EO PA MEJ Po a A O RE MI 22 22 DEA FE did iio 23 29 Mari OOM en a E A A T A a E E MN 23 SA FENE ee O A a E 23 3 4 1 Opiom Ee TE ee OMEJI 24 3 4 2 Function Data er e bre Cr b A E 25 39 e Mia Umm 26 JO VITO OW IL Us AAA 26 SW Aeb Men o o a 26 Jo Model WNdOW osere A tibias 26 39 MOde Pages ies m 27 3 9 1 OS DA aos 27 3 9 2 s aM a RR 27 3 9 3 B e II CR EST c O RE EN RATE 29 3 9 4 Tree VW cernenni na N N NES N R EANN ANN 29 310 Scale Edi mnt 30 SOS AO A e ds E 30 310 2 COMMANA Sms bois 30 A 2153 mmm mmmmmm mmmmmmmmmmm mmrmrmmmmmmm mmmmmrmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmimmac 31 3 11 FUNCION Editor oo den Latet os N 31 AIET Table RE 32 xs Fd PME Vellei EC E EDD 32 SUN TROP Uedu iN 33 Cr RR 33 S2 115 o cette eine ime ice enter d tO ebrei bal o dna nee ona brata 33 SVS eight oe pe ndn n Andi iesus O OE C td O b ern TO PM edad 34 3 11 7 Handling Non Entered Function Values eese eene nntne rnnt nnn rincon 35 SE Fro EN CUO EG NG ETE 36 CN NES O l
23. 8 39 45 Report 42 elements 42 required weights 34 root attribute 13 rounding 34 S scale 14 Car Evaluation Model 47 class 61 pad 15 good 15 decreasing 15 edit 27 30 increasing 15 order 35 ordered 15 ordering 15 size 15 value 15 Scale Editor 30 scatter chart 41 52 Selective explanation 40 51 set of values 20 Settings 44 Advanced 45 Import Export 45 Report 44 stakeholders 11 strategy scale orders 35 weights 36 T tab Attributes 41 Charts 40 Evaluation 38 Model 27 Options 37 41 table size 36 toolbar Evaluation Page 39 Function Editor 32 Main 23 Model Page 27 Options Page 37 Preview 43 Scale Editor 30 tree aggregation 13 decomposition 13 dependency 13 of attributes 13 Car Evaluation Model 46 edit 27 restructure 19 root 13 Tree View 29 U undefined value 20 undo 28 32 38 39 45 unordered scale 15 use scale orders 35 use weights 36 users 11 utility function See function V value defined 20 entered 32 function 32 interval 16 non entered 32 35 set 20 undefined 20 W Weight Editor 34 weights 17 36 approximation 17 current 34 edit 34 global 18 hyperplane 17 local 18 62 normalization 17 18 34 normalized 18 required 34 types 18 window Model 26 Window Menu 26 X XML 23
24. EXi makes a report in which it compares the primary option with each secondary one The primary option values are displayed in full whereas the secondary options values are displayed only when they differ from the primary option When selecting options on the Compare options dialogue you can right click on the list of options which gives you two additional commands IV Select all options Select all options other than the primary one l Unselect all options Exclude all options from comparison 3 15 Charts Page The Charts Page of DEXi model window provides workspace and commands to create graphical displays of evaluation results 40 Car1 PRICE Im PRICE 4 BUY FRICE 4 MAINT PRICE 3 Mi TECH CHAR E a COMFORT COMFORT Md HPERS M HOOORS Car Pd LUGGAGE k 4 SAFETY lAttribukes 10 6 basic 4 aggregate Seales 10 Functions 4 Opine 2 8 3 15 1 Workspace The workspace consists of a selection area on the left and chart display on the right The selection area consists of two sub pages tabs e Attributes This page shows the tree structure of the current DEXi model There you can check attributes that constitute chart s dimensions e Options This page shows the list of currentl
25. IJS delovno poro ilo DP 11153 December 2012 Marko Bohanec DEXi Program for Multi Attribute Decision Making User s Manual Version 3 04 Institut Jo ef Stefan Ljubljana Slovenija Contents CF CES NUS pees NE AE RE TR O A OON ERES 2 1 Tah igejejbjoje ie a UR RO ei 5 ME i e PI Ne en T de Ne o ae ee ee 5 le FUNCIONAN A abs 5 Ro APPICAMONS ita 6 IR Developmentana NS ammmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmrmmmmmmmmmmmmmmmmmc 7 INSEL A A E TT 7 RO OS TT Tm 8 1 7 SACKAOWI OGMENI SE a a o dl deere 9 2 BASIC CONCE DIS usa EE 10 Dd gt DEGISIOMANAlNS SU a dr ii rl Matai li 10 Dd Decisi n GOW Mii T E E TNNT 10 2 35 DE SIMIO ro PE A A AA UL A A M M M E ME ME 10 2 9 1 Participants or the Decision PIOCeSS nee ao tea ota diea E 11 2 3 2 Decision Problem ldenuicalo ESTEE TE 11 2 4 Decision MOUOL ii oo deii aio 12 2 5 MULA DUO MOJEM sti ni idi deii qo rot aee cud E t es io c aste deme Bert EE 12 2 6 Qualitative Multi Attribute Model 12 PN MEN A AA 13 2 9 ANTE A o mmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm mmmmmmmmmmmmmmmmmmmmmmmmmmmmmimaHir 13 2 8 1 IOMA ge o y EE tener erent earn Seeker teen eer een este DLL DL 13 2 8 2 Ia e euge ENTRE 14 2 8 3 IsecormrmendaltlOREs dd sites elt eh el ED M MA AA tdo MERDA SAN AL IAM DAD take AN AA 14 2 07 Sale su IM MM E RC EuLG RE NE TE ED 14 2 9 1 Example Scales RA E E E E E 15 2 9 2 Recommendati NS ana aa iN a a a A a a a 15 TAS ee ENS TS 1
26. Interpreted as an elementary decision rule the fourth row for example means the following if PRICE medium and TECH CHAR bad then CAR unacc 4 3 2 Elementary decision rules The Car Evaluation Model has four aggregate attributes and consequently four utility functions These are defined by the following tables of elementary decision rules PERS DOORS LUGGAGE COMFORT PRICE TECH CHAR CAR 1 to 2 2 sm all sm all 1 high bad unace 2to2 2 medium small 2 high acc unacc 3to2 2 big sm all 3 high good unace 4to2 3 small small 4 high exc unacc 5to2 a medium sm all 5 medium bad unacc 6 to 2 4 big sm all 6 medium acc acc 7Tto2 4 small sm all 7 medium good good 8 to2 4 medium small 8 medium exc exc 9to2 4 big sm all 9 low bad unacc 10 to 2 more sm all sm all 10 low acc good 11 to 2 more medium small 11 low good exc 12 to 2 more big small 12 low exc exc 13 3 4 2 small sm all 14 3 4 2 medium sm all _ BUY PRICE MAINT PRICE PRICE 13 27 2 M sma sma ioe high high 4734 a medium medium 2 high medium high mu 18 34 J big medium 3 high low high k 19 3 4 4 sm all sm all 4 medium high high 20 3 4 4 di di 5 medium medium medium ao ee i 21 3 4 4 big high 6 medium low low 5 i 22 3 4 more sm all sm all 7 low high high ms i i 23 J more medium medium 8 low medium low 24 3 4 bi nmm H low low low mue Ig 19 25 more 2 small small 26 more 2 medium sm all COMFORT SAFETY TECH CHAR 27 more 2 big sm all 1 small sm all bad 28 more 3 small sm a
27. Multi attribute models used in DEXi are qualitative 2 6 Qualitative Multi Attribute Model DEXi s multi attribute models are called qualitative They are characterized by e using qualitative symbolic attributes whose scales are discrete and typically consist of words rather than numbers 12 e employing utility functions that are represented by tables of decision rules rather that numerical formulae Here the word qualitative is used for contrast with more traditional quantitative decision models which are characterized by e using continuous numerical attributes which typically represent the decision maker s preferences and e using numerical utility functions such as the weighted sum 2 Attribute Attributes are variables that occur in multi attribute models They are organized into a hierarchical structure called tree of attributes According to their position in the tree the attributes are either e basic attributes terminal nodes leaves of the tree or e aggregate attributes internal nodes in the tree Basic attributes represent inputs of the multi attribute model Options are described by the values of basic attributes Aggregate attributes represent option evaluations They include are the one or more roots of the tree which represent the overall evaluation of options In DEXi each attribute is defined by its e Name main identification of the attribute which is typically a short string used i
28. agro ivilstvu ur C Rozman S Kav i Pivola 18 19 3 2010 29 37 2010 Stubelj Ars M Bohanec M Towards the ecotourism A decision support model for the assessment of sustainability of mountain huts in the Alps Journal of Environmental Management 91 12 2554 2564 2010 nidar i M Bohanec M Handling uncertainty in DEX methodology URPDM 2010 Uncertainty and robustness in planning and decision making eds C H Antunes D R Insua L C Dias Proceedings of the 25 Mini EURO Conference University of Coimbra Portugal 15 17 April 2010 Pavlovi M Cerenak A Pavlovi V Rozman C Pa ek K Bohanec M Development of DEX HOP multi attribute decision model for preliminary hop hybrids assessment Computers and Electronics in Agriculture 75 181 189 2011 nidar i M Bohanec M Trdin N Qualitative assessment of data mining workflows Fusing decision support systems into the fabric of the context eds A Resp cio F Burstein 16 IFIP WG8 3 International Conference on Decision Support Systems 28 30 6 2012 An vissos Greece Amsterdam IOS Press 75 88 2012 Bohanec M Rajkovi V Bratko l Zupan B nidar i M DEX methodology Thirty three years of qualitative multi attribute modelling Proceedings of the 15 International Conference Information Society IS 2012 8 12 10 2012 Ljubljana 31 34 2012 55 Marini S Bohanec M Ve parametrsko vrednotenje varian
29. both basic and aggregate attributes Rows of the table correspond to all attributes and are displayed so as to indicate the tree structure of the current model You can move along the cells and select each of them When you select a cell that corresponds to a basic attribute a data entry field appears in the toolbar so that you can change the value of that cell in the same way as on the Options Page The effects of any change are immediately shown in the table 3 13 2 Commands To invoke a command you may either e press a button shown in the toolbar e select an item from the Edit Menu The following commands are available on this page Copy options Copies data of all the options into the clipboard for further use 3 Paste options Copies or inserts previously Copied option data back into the table The position where each option s data is pasted into the table is determined from option names Suppose an option named Opt has been previously copied into the clipboard If Opt exists in the current table then the paste command replaces all its data Otherwise a new option Opt is created Find Opens the same window as on the Model Page in which you can define a text to be searched for in the data table amp Find next Continues searching from the current position in the data table Undo Undo the last option editing operation Redo Redo the last undone option editing operation Analysis of options The followin
30. but wont like it good satishes my needs anc excellent i gt oOo p sa TO Deskscuelnnewatbues OK Cancel 3 10 1 Workspace Scale Editor consists of e Value list each value has a name usually a short string and an optional description used only for documentation e Scale order entry field at the top here you can define scale ordering unordered increasing recommended and decreasing e Toolbar at the right provides command buttons e The field Default scale for new attributes when checked this scale will be automatically assigned to all attributes created hereafter in the Model Page 3 10 2 Commands To invoke a command you may either e press a button shown in the toolbar or e select an item from the pop up menu that appears after right clicking the mouse button Scale Editor provides the following commands Add value Inserts a new value into the list The default name of this value is new value and you may change this name immediately after insertion Delete value Deletes the value that is currently selected in the value list 30 F2 keyboard Rename value Opens a small in line field in which you can quickly rename the currently selected value Edit value Opens a dialogue in which you can edit both the name and description of the currently selected value Set bad Set the class of the currently selected value to bad Set neutral Set the class of the currently s
31. cision maker s e combining these in order to arrive at optimal or at least sufficiently good decisions Decision Analysis and the DEXi program as well are aimed at supporting people in making decisions rather than making decisions themselves For this purpose they provide methods and tools for developing decision models and using them for the evaluation and analysis of options 2 2 Decision Problem In Decision Analysis a decision problem is understood primarily as a problem of choice which is defined as follows e Given a set of options also called alternatives which typically represent some object or actions either e choose an option that best satisfies the goals objectives of the decision maker or e rank the options according to these goals Making a choice usually occurs as part of a decision process Decision Analysis and DEXi are particularly interested in complex decision problems that is problems which are for some reason considered difficult by the decision maker and require careful elaboration and analysis Complex decision problems are usually characterized by e Novelty the decision maker is confronted with the problem for the first time and has insufficient knowledge or skills to address the problem e Unclearness unclear understanding of the problem and its goals unknown or incompletely defined options e Uncertainty existence of possible events that cannot be controlled by the decision maker but can
32. created attributes 2 11 Options Options also called alternatives are basic entities studied in a decision problem Depending on the problem they can represent different objects solutions courses of action etc which are evaluated and analyzed by a multi attribute model In DEXi each option is represented by its name not necessarily unique and a set of values so that one value is assigned to each attribute in the tree of attributes These values are further distinguished into e option description vector of values assigned to basic attributes e intermediate evaluation results values assigned to aggregate attributes other than the roots of the tree e final or overall evaluation results values assigned to the root s of the tree Values assigned to basic attributes can be either defined or undefined e Defined means that a single qualitative value from the corresponding scale has been assigned to that attribute e Undefined means that the value is unknown or unspecified In DEXi this is denoted by an asterisk and is interpreted as a set of all values that can be assigned to the corresponding attribute 2 12 Evaluation of Options With multi attribute models options are evaluated in the following way 1 Each option is represented by a vector of basic attribute values The values of each option are aggregated in a bottom up way according to the defined structure of the model and corresponding utility functions 3
33. d functions may cause options to be evaluated by sets of values rather than single values Therefore you should check the status of your functions and ensure that they are 100 determined As long as your function is 100 determined the proportion of actually entered rules is not that important and can be less than 100 However even in this case you should be aware that non entered rules are more volatile than entered ones Entered values are never changed by DEXi but non entered values may be unintentionally changed later for example by changing weights in Weight Editor To protect your completed function from such changes you may want to run the Enter values command For final verification of your function you may also want to review its complex rules and weights 3 12 Options Page The Options Page of DEXi model window provides workspace and commands for editing option descriptions option names and option values assigned to basic attributes of the currently edited DEXi model ur dy Model TF Options x Evaluation E Chans can x Go GB Oa n Car meda m rhechum Attributes 10 6 basic 4 aggregate Scales 10 Functions 4 pan a 3 12 1 Workspace The workspace provides a table for editing option data Each column represents an option The first row displays options names and each of the remaining rows corresponds to one basic attribute Above the table there is a toolbar consisti
34. denotes the interval between and including the two values 2 10 2 Complex Rules DEXi uses complex rules in order to represent its utility functions in a more compact and possibly more comprehensible way than with elementary rules table rows Complex rules are obtained by joining several elementary rules which have the same function value In other words a complex rule represents one or more elementary rules In tables complex rules are characterized by the use of intervals Example This is the CAR utility function from the Car Evaluation Model represented with elementary rules PRICE TECH CHAR CAR 1 high bad unacc 2 high acc unacc 3 high good unacc 4 high exc unacc 5 medium bad unacc 6 medium acc acc T medium good good 8 medium exc exc 9 low bad unacc 10 low acc good 11 low good exc 12 low exc exc This is the same function represented with complex rules 16 PRICE TECH CHAR CAR 1 high i unacc EJ bad unacc J medium acc acc 4 medium good good 5 low acc good Bb medium exe exc T low gaond exc Notice the decreased number of rows from 12 to 7 and the use of symbols and gt For example the complex rule 1 says that if PRICE is high and regardless on the value of TECH CHAR the value of CAR is unacc This complex rule is a compact representation of the first four elementary rules 2 10 3 Weights Weights are commonly used in Decision Analysis to model the importance of attributes Weights are nu
35. ds At the top level DEXi provides a standard Windows MDI Multi Document User Interface In this case a document is a DEXi model You can work with one or more models at the same time using the following windows and commands e Main Toolbar e Menus o File Menu o Edit Menu o Analysis Menu o Window Menu o Help Menu e Model Window Gna Evan IE Chat ID PROGRAMME EEIeTITTTEPEIITEEITTUTTEUNT E EJ KNOWLEE 4 EXPEF OTEEOOK Model e Opel E Rouen Hd Chess i JACAR CAR dxi AA hem COMFORT Qusity of a ces 4 mPERS 4 BDOORS 4 LUGGAGE 4 SAFETY A NIKA F Flutes T2712 100 006 determined 100 005 10 6 basic 4 aggregate Scales 10 Functions 4 Options 2 3 1 DEXi Model The term DEXi Model refers to the document that is created and edited in DEXi s Model Window and stored externally on a DEXi File A DEXi model contains the data on e a qualitative multi attribute model e options and e program settings 22 3 2 DEXi File Each DEXi model which is created and or edited in DEXi can be stored on a DEXi File You can load and save files using the File menu commands DEXi supports three file formats dxi This is the native DEXi file format which stores a single model the corresponding option data and program settings dax File format used by the older program DEX xml File format used by yet another old program jDEX Normally you should only
36. e and PgDn Shows the next page Zoom Controls the magnification of display Print Prints the report Page setup Opens a dialog to define page size orientation and margins ll Save Saves the current report on a file The available formats are Text File txt Tab delimited File tab Html File html Enhanced Metafile emf and Bitmap bmp The former three save the entire report while the latter two save only the current page Copy Copies the current page to clipboard for its transfer to other applications The page is saved in two formats Windows Metafile and Enhanced Metafile amp View report in browser Converts the report to html and opens it in a browser The browser can be either internal or external depending on Settings Report v Close Closes the current preview and returns to DEXi 3 18 Internal browser Internal browser provides an alternative way to viewing DEXi reports when they are represented in the html format The internal browser is invoked only if Use default system browser is not selected in Settings Report otherwise the default system browser is invoked externally from DEXi The internal browser can be activated from 1 the preview already showing some report or 2 instead of preview when Reports in html is selected in Settings Report Once open the internal browser shows the current report and provides the following buttons 43 0 Zoom controls Control t
37. e Tree View using standard drag and drop mouse commands e Expand and collapse the display of sub trees by clicking the symbols Hand E respectively 3 9 4 1 Status While editing the model the status of its attributes is shown in Tree View using the following symbols lt Incomplete basic attribute an attribute having an undefined scale 4 Completely defined basic attribute W Aggregate attribute whose utility function cannot be constructed due to undefined scales of its own and or its descendants Aggregate attribute with an undefined utility function but everything is ready for its construction Aggregate attribute with a partially or incompletely defined utility function Aggregate attribute with a fully defined 100 determined utility function 29 Linked attribute 3 9 4 2 Remark Usually the goal is to completely define all DEXi model components In Tree View this is indicated by all green status symbols Occasionally and only with a good reason it is acceptable to leave a partially defined utility function and or to have a linked attribute UJ 3 10 Scale Editor Scale Editor is a window in which you can create and edit one attribute scale Basically a scale is just an ordered or unordered list of values you can add and delete these values give them names and optional descriptions change their order and define their bad or good class Value Description ra unacceplable car gt acc acceplable
38. e closest possible match with your Required weights and the rules already entered in the table In general thus Current weights differ from the Required weights and there are two main reasons for this e Utility function space in DEXi is discrete not continuous The functions are discrete too In general it is impossible to exactly match continuous Required weights to a discrete Current utility function e Ahyperplane constructed from Required weights depends on already entered function values The more the entered values the less the freedom for the hyperplane and the lower the chance to match the Required weights 34 For example there is no freedom left with a fully defined utility function In this case Required weights have no effect The default setting of Rounding is no for compatibility with DEXi versions prior to 3 03 However it is recommended to explicitly set Rounding to down or up for better stability of numerical calculations 3 11 7 Handling Non Entered Function Values In Function Editor utility function values are either entered by the user or non entered Entered values are shown in bold and are never changed by DEXi during the editing session This is however not the case with non entered values which are handled by DEXi with the purpose to aid and simplify the function editing process and maintain the consistency of function definitions Non entered values are shown
39. e first time Report Create and preview a report 3 4 File Menu The DEXi File Menu is available at all times for working with the currently active DEXi model window Mainly it provides commands for creating loading and saving DEXi files but there are also commands for importing and exporting other data making reports changing program settings and exiting DEXi L New Create a new DEXi model and open a corresponding window Initially the model is almost empty containing only one root attribute and no options ar Open Open and load an existing DEXi file through a standard Open a file dialogue 23 Close Close the currently active DEXi model window Before closing DEXi checks whether the model has been saved on a DEXi file and if not asks you whether to save the model Yes discard the model No or to Cancel closing the window Il Save Save the currently edited model on a DEXi file This is a quick save that normally rewrites the output file without confirmation However it does ask you to provide a file name for a newly created and yet unsaved model Save as Save the currently edited model on a DEXi file In this case a Save file dialogue appears before saving and always lets you to define or redefine the name and possibly format of the file Import Import function Open a function file and load its decision rules into the currently selected utility function The function is created if necessary Imported
40. e lead MM 50 452 Selective do LANA ON assent teu o AE RN O a AA etu ei 51 4 5 3 Compare ODNOSE on ane a arc a tcl a ea tcl dace aac ce 51 4 6 Some Car Evaluation Chnarts esses nennen nnne nnn tnra tetra terae tiara nro 52 4 6 1 Bar Char T E EI A AE TO ONA 52 452 sleni a y AAN AE n Pare Teme cuu MEM D MEM DL M MEE 52 4 6 3 mici ei mc H 53 5 Sa A a a a a a a a enda 54 5 1 Documentation in Slovene kekec 54 52 selected IPUblIeallOFiS anje pa espe EURO MEI UNI RUN AN Aa UU a SEINE MEI en aaa 54 DEX I VCP SIONS m C TT ee of F3 Verhon O e rer 57 5 93 1 INOW COMI OMG IS ciate e e co e el a 57 5 3 2 New program Teal ures idas pp 57 Ooo Gh anged program te atures iso 57 93947 BHO A A C ND es NE 57 gA Versiono Dd o do e od do eo de 57 5 4 1 New program Te alUles nurse bala 57 542 Changed program Tedlufes at i 58 5 4 3 Changed documentation 1 0 0 eceesesssssesssssssecscscsssssacscsssesesacscsesesacacecsesesacsesesesacacsusesesavsesesesacaceusesesaraeeseesevarensseesarans 58 SAA A eed odds et cov a t bU i ea cent CD HE MM Db bs MC UTE 58 Sa NV Om sO P roads 58 5 5 1 New program features ima oe Ra vut A ri de A a at 58 5 5 2 ESUIGHTDXO Sesame HUC Sper rate ater O UN CUN A EUNDUM MULUS UNUM EUM tee USQUC eer treater tT 58 DO VESON dat I 58 5 6 1 New program te ALUN CS usse cedido haesit ote canted noni look te anj ea dn
41. ed together in a single utility function 2 10 Utility Function Utility functions are the components of multi attribute models that define the aggregation aspect of option evaluation For each aggregate attribute Y whose descendants in the tree of attributes are X X X the corresponding utility function f defines the mapping EXX XXX y 15 In DEXi a utility function maps all the combinations of the lower level attribute values into the values of Y The mapping is represented in a table where each row gives the value of f for one combination of the lower level attribute values Rows are also called decision rules because each row can be interpreted as an f then rule of the form if X value and X value and and X value then Y value or value interval In the context of representation such rules are also called elementary rules This name is used for contrast with complex rules 2 10 1 Intervals An interval is a subset of consecutive scale values DEXi often uses intervals in connection with utility functions particularly when they are edited or represented by complex rules An interval of values is denoted in one of the following ways e the asterisk denotes any value of the corresponding attribute e gt value stands for better than or equal to value alternative interpretation at least as good as value value worse than or equal to value e valuef value2
42. eed for a simple computer program to be used in decision modeling courses in secondary schools and universities Consequently DEXi s design is a trade off between these two requirements It has a convenient MS Windows user interface which has been kept as simple as possible Some advanced DEX features have been deliberately dropped such as probability and fuzzy distributions of values chain attributes advanced transformations of utility functions some settings and reports In order to deal with missing incomplete or uncertain data DEXi employs value sets instead of more complex value distributions of DEX On the other hand more emphasis has been put on graphical and reporting capabilities in DEXi Also DEXi adds a new functionality to the treatment of attribute weights The first version of DEXi was developed in 2000 as part of the project called Expert Systems in Education The projects acronym was ESi where the letter i indicated education in Slovenian izobra evanje The name DEXi pronounced DEXy was coined using the same pattern to mean DEX for Education In subsequent years DEXi was increasingly used not only in education but also in more and more advanced decision support projects Some extremely complex decision models were developed in international projects proving somewhat surprisingly that DEXi was suitable not only for education but also for supporting difficult real life decision problems In
43. elected attribute and optionally provide its description Scale of the currently selected attribute You can either select an existing scale from a pull down list or press to open the Scale Editor Whenever the attribute s scale has been already defined the list of scales contains only compatible scales the ones having the same number of values Utility function of the currently selected aggregate attribute There are three parts a read only field showing the current status of the function the button F to invoke the Function Editor and a field for entering an optional description of the function The utility function area is not available for basic attributes Commands To invoke a command you may either press a button shown in the window select an item from the Edit Menu or select an item from the pop up menu that appears after right clicking the mouse button The Model Page provides the following commands 21 Add attribute This command creates a new attribute and inserts it into the tree as a descendant child of the currently selected attribute The new attribute is automatically called New so you should consider giving it a more meaningful name X Delete subtree or item This command depends on the type of the currently selected attribute e Foran aggregate attribute it discards all its descendants and its utility function effectively converting it into a basic attribute e Fora basic attribute
44. elected value to neutral nor bad nor good V Set good Set the class of the currently selected value to good Move up Moves the currently selected value one place up in the list Move down Moves the currently selected value one place down along the list 3 10 3 Remarks Editing a scale especially adding and deleting values may affect already defined utility functions DEXi tries to adapt the affected functions so that their meaning is preserved as much as possible but you should be careful and verify all affected functions after making such changes Increasing scales are strongly recommended they improve the comprehensibility of models and simplify the definition of utility functions The concept of value class bad neutral good is meaningful only with ordered scales Thus the commands OF do not work with unordered scales For ordered scales classes must be ordered too Therefore classes can be assigned only so that zero or more bad values are followed by zero or more neutral values which are then followed by zero or more good values 3 11 Function Editor Function Editor is a window in which you can create and edit one utility function Its workspace consists of a table two toolbars pop up menu and status bar Inl x wv Lse scale orders neh d Use weights BU PRICE i PRICE s pe l pes 2 medium 2 high medium high pen 3 high low high A 4 medium high high El medium medium medium w Automatic
45. em 10 complex 10 identification 11 owners 11 decision process 10 participants 11 steps 11 decision rule 16 inconsistency 35 interval 16 decision rules 32 decision support 10 decision making 10 decomposition 13 decreasing scale 15 defined value 20 dependency 13 determined measure 33 DEX 6 7 DEXi 5 applications 6 development 7 document 22 documentation 54 download 5 file 23 functionality 5 history 7 model 22 create 23 edit 27 open 23 save 24 name origin 7 publications 54 user interface 22 versions 7 57 E Edit Menu 26 editor Function 31 60 Scale 30 Weight 34 elementary rule 16 entered rules ratio 33 entered value 32 evaluation Car Evaluation Model 49 of options 20 results final 20 intermediate 14 20 50 overall 20 50 Evaluation Page 38 examples Car Evaluation Model 46 experts 11 export function 24 options 24 F file format comma separated 24 CSV 24 dax 23 dxi 23 tab delimited 24 25 43 xml 23 File Menu 23 function 15 Car Evaluation Model 47 combinations 16 18 complex rule 16 48 data file 25 decision rule 16 complex 16 elementary 16 definition measure 33 determination 33 edit 27 31 editing 36 elementary rule 16 48 entered rules ratio 33 export 24 if then rule 16 import 24 mapping 15 monotonicity 35 rounding 34 Size 18 36 status 33 table 16 32 cells 32 undefined 33 use scale orders
46. en P ag Osa E PP PE PEO nn E E 37 eZ WOKS PACO OO ROO 37 S 12 o 1 0 Co E A uM Mid NE 38 9 12 9 id ido 38 Sele EVAUAUON Radiata T 38 SX PON MB Tod ioco ipods 39 So a ala ae A E Ue O e PE 39 SE ii SESS E Oy O a 40 9 14 1 X9BIOD ANA SOS eudinsiiorpitn o ae udi oon tad aE 40 lo Chal Rada 40 o Ol WOIKSDOCO sosa 41 2 19 2 CZOImmabids outostuso m tnb eoi tests o foie rk MN a er E 41 LIG a 2 pS PE deat OE ta etu Gm MO Mte Metu I Pe d D M eee 42 9 16 1 e AAA 42 OW dii err 43 9e Internal DIOWSOL uecouapac m pudet diia 43 PEI o tutt 44 MOM RCT UT uo EE 44 94192 IMPOR EXPO Fate aussi nth tio 45 3 19 9 Advanced P3965 didi he P huis bu s n dat vA bd iv han v b blc Ado Pa Ure DUE 45 4 Example Car Evaluation Model testen e tu kale enka 46 4 1 Tree of Attributes for Car Evaluation sse nnne nnn nnn nnn 46 4 1 1 METAN REIS 46 4 12 JAXUUIDUTO TVDOS deese odo dw DRM ret N 46 4 1 3 JA bute Descriptions RR E 47 42 cales Tor Gar EvaluallolT mide tin E nnd Md E e 47 4 3 Uiit Functions for Car EVALUATION xi eine ttt potrete iaa 47 4 3 1 ee AION a A TE ET 48 4 3 2 Elementary decision TUS aio 48 5 9 Gomplextrules and WENS it tdt 48 4 4 Description and Evaluation of CAS 49 4 4 1 e EN A A TA DE VE O Ce Rego ee en O O A AT are A 50 45 Some Car Option Analyses uer eee e a ee a eli re er ea 50 4 5 1 Plus MINUS ANA VS S adiumentum de c
47. evitably in order to address these requirements DEXi was being gradually upgraded with new advanced features such as new methods for option analysis import export features and undo redo functionality 1 5 Versions DEXi version 1 00 was released in 2000 with a Slovenian user interface only DEXi versions 1 01 through 1 04 were gradually released in the period 2001 2006 They addressed only bug fixes and the addition of Slovenian help There were no major functional additions or improvements DEXi version 1 02 with English user interface was released in 2001 DEXi version 2 00 was released in 2007 introducing a number of additions and improvements Slovenian and English user interface for the same functionality editing options editing utility functions importing exporting options exporting utility functions report and report settings improved DEXi file format English help installation package improved DEXi web page bug fixes DEXi version 3 00 was released in 2008 undo redo functionality option analysis functionality improved option data entry updated English help and DEXi web page bug fixes DEXi version 3 01 was released in May 2009 e added exporting of attribute tree two formats tab delimited and GML e bug fixes DEXi version 3 02 was released in October 2009 added model description editor added a new operation to join attributes added searching functionality improved status disp
48. example Scales Similar to Attribute tree but displays scales See an example Scale descriptions A long printout of scales For each scale in your model it prints out the scale name corresponding attribute description and all scale values together with their own descriptions if any This report element is rarely needed 42 Function summary Displays the tree structure along with the status of utility functions Rule tables Prints out all utility functions defined in your model Display format is determined in Settings Report See some examples Weights Similar to Attribute tree but displays attribute weights You can explicitly select displayed weight types in Settings Report See an example Evaluation results Similar to Attribute tree but displays evaluation results Only options that are selected on the Charts Page are included into this report element Again its format can be determined through Settings Report Chart Prints out the chart that is currently displayed on the Charts Page 3 17 Preview Preview displays reports created by DEXi You can view these reports and optionally print them save on files or copy individual pages for transfer to other applications such as Microsoft Word The toolbar which is displayed at the top of Preview provides the following buttons and controls Page status Displays the current page number and total number of pages and PgUp Shows the previous pag
49. f the decision making and improve the quality of the decision itself Usually the decision process involves at least the following steps eS hy Problem identification Modeling developing a decision model Evaluation and analysis of options Choice making the decision Implementation of the decision The decision support tool DEXi is primarily used in the steps 2 and 3 2 3 1 Participants of the Decision Process In general a typical decision process involves up to four types of participants either individuals or groups 1 4 2 3 2 Stakeholders also called decision problem owners individuals or organizations that have a legitimate interest in the decision making problem Usually these are the ones that need to make the final decision and are also responsible for that decision Experts People knowledgeable in the field so that they can provide information and advice relevant for the decision They may contribute to the overall decision problem identification to the definition of options goals and criteria and to the decision model development Decision analyst s Methodologists with experience in Decision Analysis that is the underlying methodology and tools Often they take the role of moderators or mediators of the decision making team Users People affected by the decision Decision Problem Identification The identification of decision problem occurs at the beginning of a decision process At this
50. g and display of utility functions o Added ability to describe individual functions Model Page Description o Extended display of function characteristics added distributions of class values Model Page Function Editor and Function summary report o Added value rounding control for functions defined by weights Weight Editor Rounding o Added menu item File Import Import function for importing utility function from tab delimited files o Added copying and pasting of utility functions Function Editor Copy and Paste buttons Reports o Added new report type Function summary o Added ability to save reports on tab delimited files as additional means of exporting data Search functionality extended to the Evaluation Page o Added menu items Edit Find and Edit Find next Bug fixes Report preview does not require an installed printer any more Model Page Timely refreshing of Tree View Proper positioning of windows on systems with multiple monitors 5 8 Version 3 04 5 8 1 5 8 2 New program features Reports o Added html reports o Html report browsing using an internal or external default system browser Function Editor o Adding the display of attribute information o Issuing a warning when entering inconsistent values and turning off Use scale orders Bug fixes Linking of attributes o Detecting unstable cases with two aggregate attributes having the same name o Avoiding an infinite loop by disal
51. g options with their basic features Which are these features 2 4 Decision Model The Decision Analysis approach is characterized by the use of decision models In general a decision model encodes knowledge and information that is relevant for solving the decision problem at hand Decision models are usually developed by participants of the decision process using tools such as DEXi Typical models used in Decision Analysis are e decision trees e influence diagrams e multi attribute models Among these DEXi employs qualitative multi attribute models Once developed the decision model is used to e evaluate options and e perform various analyses such as what if or sensitivity analysis The obtained evaluation and analysis results provide the basis for decision maker s assessment of options and possible choice of the best one 2 5 Multi Attribute Model Multi attribute models also called multi criteria models represent a class of models used in Decision Analysis that evaluate options according to several possibly conflicting goals or objectives In principle a multi attribute model represents a decomposition of a decision problem into smaller and less complex sub problems A model consists of e attributes and e utility functions Attributes are organized hierarchically into a tree of attributes Each attribute takes values from the corresponding scale Multi attribute models are used for option evaluation and analysis
52. g three buttons provide shortcuts to option analysis commands available in the Analysis Menu 39 Plus minus 1 analysis Investigating the effects of changing basic option values by one step up and down 3 Selective explanation Identifying particular advantages and disadvantages of an option A Compare options Comparing an option with some other options 3 14 Analysis Menu The DEXi Analysis Menu is available only from the Evaluation Page This menu provides commands for the analysis of options Each command creates and previews an option analysis report 3 14 1 Option analyses 1 Plus minus 1 analysis This analysis investigates the effects of changing each basic attribute by one value down and up if possible independently of other attributes The analysis is carried out for the currently selected option and displays the effects of changes on the currently selected aggregate attribute Selective explanation Displays particularly strong and weak values of the currently selected option This method finds and displays all connected subtrees of attributes whose values are either all good for strong points or bad for weaknesses A Compare options This command creates a report that is similar to the common Evaluation results report but highlights differences between options First DEXi opens a dialogue in which you can choose a primary option and a list of secondary options Compare options After selection D
53. g your mouse It has two parts the first one lists values that can be assigned to the current cell and the second part contains two items Automatic value setup Controls table recalculation after each change When checked DEXi does recalculate all non entered function values Otherwise there is no recalculation and the two toolbar checkboxes are disabled Enter values Changes the status of table cells from non entered to entered Entered cells are never recalculated by DEXi and can be only changed by you After issuing this command you can specify whether to change the status of all non entered cells or only those having a single non interval value 3 11 4 Status bar Status bar displays the status of the edited function 3 11 5 Utility Function Status When a utility function is created in DEXi it is completely undefined at first all cells in Function Editor contain the undefined value When you edit the function you usually assign values to more and more cells and the function becomes more and more defined Usually the goal is to completely define the function that is to precisely specify all cell values DEXi uses two measures of function definition which are both displayed as status progress indicators on Model Page and in Function Editor The first measure is called entered rules ratio This is a ratio between the number of entered rules that is cell values defined by you and the total number of rules function si
54. hat displays one option Copy chart Copies the current chart into the clipboard so that it can be pasted into other documents Available formats for pasting are Windows Metafile and Enhanced Metafile 3 16 Report Report provides a formatted and mostly textual presentation of key components of your current DEXi model After invoking this command a small window appears in which you can select elements to be included in the report Create report X Model description Attribute tree Scales Scale descriptions Function summary Rule tables Weights Evaluation results Chart Font Settings Cancel In addition to element selection you can press one of the buttons Font Specify font to be used in the report the default is Arial 10pt Settings Open Settings Report in order to specify the format of report and its elements Report Create a report consisting of selected elements and preview it on the screen Cancel Close this window without creating a report Also you can right click on the list and select one of the menu items Iv Select all reports Select all report elements l Unselect all reports all current selections 3 16 1 Report elements Model description Textual description of the current model which is optional and used only for documentation Attribute tree Displays attribute names and descriptions Tree structure is indicated by indenting and connecting attribute names See an
55. he attribute value cannot be decreased or increased respectively JO Attribute 1 Carl 1 CAR good BUY PRICE wunace medium exc MAINT PRI E unace medium exc PERS more DOORS 4 LUGGAGE big SAFETY unacc medium exc The above display shows for example that BUY PRICE considerably affects the evaluation of Car2 When BUY PRICE decreases by one step from medium to high the latter value is not shown the overal value of CAR becomes unacc In the other direction from medium to low the overall evaluation improves to exc The two brackets indicate that the values of corresponding attributes ZPERS and LUGGAGE cannot be increased any more preventing the 1 part of the analysis 4 5 2 Selective explanation Selective explanation highlights particular advantages and disadvantages of an option The method finds and displays only those connected sub trees of attributes for which the option has been evaluated as particularly good or bad Strong points Attribute Carl CAR exc Mies low LMAINT PRICE low PERS more COMFORT high UGGAGE big L SAFETY high This example shows that Car2 has three particularly strong parts two sub trees and one single attribute 1 overall evaluation which is strongly influenced by low MAINT PRICE 2 COMFORT due to very good ZPERS and LUGGAGE and 3 high SAFETY 4 5 3 Compare options This analysis compares one primary option with some other selected seconda
56. he magnification of display Print preview Prepares the report for printing and displays it on the screen Il Save Saves the current report on a html file EX Copy Copies the current html text to clipboard for its transfer to other applications v Close Closes the internal browser and returns to DEXi 3 19 Settings Settings is a dialogue in which you can specify how DEXi handles reports external files and other advanced features For this purpose it provides three pages Report Import Export and Advanced The selected settings apply to the current DEXi model and are saved on the corresponding DEXi File 3 19 1 Report Page Use this page to specify various aspects of report formatting Each section on new page When checked each report element is strictly started on a new page Otherwise multiple elements may appear on a single page Bad values and Good values Use these controls to specify the appearance of bad and good scale values Functions These controls specify the display of utility functions e weights specifies whether weights are displayed or not e rule numbers include or exclude row numbers e complex rules when checked represent the function with complex rules otherwise use elementary rules e entered rules only when checked display only rules entered by the user otherwise also show non entered rules Weights These controls specify the display of weights Use these controls to specify the n
57. ights see the definition below 17 DEXi uses weights for two purposes e asan approximate representation of utility functions used primarily for verification and overview see examples from the Car Evaluation model and e for defining utility functions or their parts see the weight based strategy of handling non entered function values 2 10 3 2 Weight Types Actually DEXi uses four types of weights as illustrated with the following weights from the Car Evaluation model Attribute Local Global Loc norm Glob norm CAR PRICE BO RE 53 53 BUY PRICE 5 30 50 26 MAINT PRICE 50 30 50 26 TECH CHAR 40 al a a COMFORT 50 20 5l 24 PERS JJ T 3B g HOCOORS 22 4 a B LUGGAGE 3g T J6 g SAFETY 5l 20 5l 24 The difference between ocal and global is due to the tree of attributes Local weights always refer to a single aggregate attribute and a single corresponding utility function so that the sum of weights of the attribute s immediate descendants function arguments is 100 Global weights on the other hand take into account the structure of the tree and relative importance of its sub trees A global weight of an attribute is calculated as a product of the local weight and the global weight of the attribute that lies one level above A global weight of the root attribute is 100 For example the global normalized weight of BUY PRICE is 50 its local normalized weight x 53 global normalized weight of PRICE which gives 26
58. ing above 100 is very difficult and everything above 500 is extremely hard if not impossible to define Also it is not only the size that matters The more the attributes the more difficult the function to define even if the size of the functions is comparable Combining four attributes together usually appears quite a hard task for human brain For these reasons the DEXi method strongly advises to limit the number of aggregate attributes descendants to three and to restructure the tree of attributes whenever this condition has not been met To prevent the combinatorial explosion the DEXi program issues a warning before creating a function of size 200 or more and disallows the creation of functions larger than 1000 2 10 4 3 Restructure Tree of Attributes In order to avoid combinatorial explosion in DEXi models it is strongly advised to structure the tree of attributes so that each aggregate attribute has only two or three immediate descendants Whenever you encounter an aggregate attribute with four descendants you may want to consider restructuring the tree below that attribute Usually there are several ways to do this 19 In all cases you should regroup the lower level attributes and introduce one or two new aggregate attributes denoted by O Usually the right structure is the one that appears the most logical so that e it groups together similar or related attributes and e itis easy to give names to the newly
59. iting commands depend on the currently edited model component that is on the page that is currently open in the model window sh Model s Options Evaluation There is no Edit Menu associated with the HEE Charts page 3 6 Window Menu The DEXi Window Menu is available at all times and provides commands for working with DEXi model windows These commands are particularly useful when there are multiple model windows loaded simultaneously and shown in the main window of DEXi TE Cascade Stack currently open windows so that each window title bar is visible This facilitates an easy selection of each window T Tile Horizontally Resize and move all currently open windows so that they are shown above each other occupying the full width of the main window H Tile Vertically Resize and move all currently open windows so that they are shown behind each other occupying the full height of the main window Minimize All Minimize all currently open windows so that they are shown only as small bars icons in the main window Arrange icons Neatly arrange all currently minimized window icons 3 Help Menu The Help Menu commands provide help on DEXi amp Help Show this help About Show DEXi version and copyright information For further information and latest news see the DEXi Web Page 3 8 Model Window A DEXi Model Window provides workspace for editing and using one DEXi model It consists of four pages tabs
60. lays bug fixes DEXi version 3 03 was released in March 2011 added descriptions summary displays and rounding control of utility functions added import and copy paste functionality for utility functions new report Function summary exporting reports to tab delimited files extended searching functionality bug fixes DEXi version 3 04 was released in December 2012 e added html reports e improved Function Editor added attribute information display and inconsistency warning e bug fixes Section 5 provides a more detailed description of different versions of DEXi 1 6 Credits Marko Bohanec Design and programming English help Web pages Vladislav Rajkovic Initial design and management educational aspects Eva Jereb Slovenian help Uros Rajkovic English translation of version 1 02 Zarja Vintar DEXi logo 1 7 Acknowledgments David Berneda http en wikipedia org wiki Teechart TeeChart Pro V4 Anders Melander http melander dk delphi gifimage TGIFImage for Delphi Jordan Russell http www jrsoftware org Inno Setup 5 1 11 2 Basic concepts 2 1 Decision Analysis Decision Analysis is a discipline popularly known as Applied Decision Theory It provides a framework for analyzing decision problems by e structuring and breaking them down into more manageable parts e explicitly considering the possible options alternatives available information involved uncertainties and relevant preferences of the de
61. ll 2 small medium bad 29 more 3 medium medium 3 sm all high bad 30 more 3 big high 4 medium sm all bad 31 more 4 small small 5 medium medium acc 32 more 4 medium high 6 medium high good 33 more 4 big high 7 high small bad 34 more more small small 8 high medium good 35 more more medium high 9 high high exc 36 more more big high 4 3 8 Complex rules and weights This example shows all the utility functions defined in the Car Evaluation Model They are represented in terms of local weights and complex rules 48 PRICE TECH CHAR CAR BU a ANS 1 high unacc as bad unacc 3 medium acc acc 4 medium good good 5 low acc good h medium exc exc i low good exc BUY PRICE MAINT PRICE PRICE 20 50 1 high P high 2m high high a medium medium medium 4 gt medium low low 5 low zzmediumi low COMFORT SAFETY TECH CHAR 5 9n SOS 1 sm all k bad Ze small bad a medium medium acc 4 medium high qood 5 high medium good B high high exc SPERS ZDOORS LUGGAGE COMFORT odo 2 JHY to2 i small FE 2 jj sm all a T ki sm all sm all 4 3 4 a medium medium 5 3 4 zag medium medium H gt 34 d medium medium T s a j 2 24 big high H more 4 big high Q more gt d medium high 4 4 Description and Evaluation of Cars Options evaluated by the Car Evaluation models are obviously cars This example illustrates two basic concepts description of options and evaluation of options 49 Option PRICE COMFORT TECH CHAR
62. lowing links to ancestor attributes Model editing o DeleteFunctionValue bug resulted in List out of bounds error after deleting a scale value o Redesign of utility function modification methods for a better control over the values of non entered function values thanks to Gabriele Fortino for pointing out this problem Reports Proper selection of the chart element 59 INDEX A aggregate attribute 13 aggregation 13 15 alternative See option analysis Car Evaluation Model 50 compare options 40 51 of options 20 40 Plus minus 1 40 50 selective explanation 40 51 what if 38 Analysis Menu 40 applications 6 attribute 13 aggregate 13 basic 13 description 13 47 edit 27 importance 17 linked 14 name 13 root 13 scale 14 47 status 29 tree 13 types 46 weights 48 B bar chart 41 52 basic attribute 13 C Car Evaluation Model 46 analyses 50 charts 52 evaluation 49 functions 47 options 49 scales 47 tree of attributes 46 cell 32 edit 32 value 32 38 39 chart 41 bar 41 52 Car Evaluation Model 52 radar 41 53 scatter 41 52 type 41 Chart Menu 41 Charts Page 40 combinations 16 18 combinatorial explosion 18 Compare options 40 51 complex decision problem 10 complex rule 16 current weights 34 D data file function 25 option 24 Decision Analysis 10 decision analysts 11 decision model 12 qualitative 12 quantitative 13 decision probl
63. m Proc Information Society IS 2006 Ljubljana 89 92 2006 Bohanec M Cortet J Griffiths B nidar i M Debeljak M Caul S Thompson J Krogh P H A qualitative multi attribute model for assessing the impact of cropping systems on soil quality Pedobiologia 51 3 239 250 2007 Ta kova K Stojanova D Bohanec M D eroski S A qualitative decision support model for evaluating researchers Informatica 31 4 479 486 2007 nidar i M Bohanec M Zupan B Modelling impacts of cropping systems Demands and solutions for DEX methodology European Journal of Operational Research 189 594 608 2008 Bohanec M Mess an A Scatasta S Angevin F Griffiths B Krogh P H nidar i M Dzeroski S A qualitative multi attribute model for economic and ecological assessment of genetically modified crops Ecological Modelling 215 247 261 2008 Rozman Poto nik M Pa ek K Borec A Majkovi D Bohanec M A multi criteria assessment of tourist farm service guality Tourism Management 30 629 637 2009 nidar i M Bohanec M Lavra N Cestnik B Project self evaluation methodology The Healthreats project case study Proc Information Society IS 2009 Ljubljana 85 88 2009 Bohanec M nidar i M Izku nje z ve parametrskimi odlo itvenimi modeli pri podpori odlo anja o gensko spremenjenih organizmih DAES 2010 Sodobni izzivi mened menta v v
64. mbers usually normalized to the sum or maximum of 100 which define the contribution of the corresponding attribute to the final evaluation In Decision Analysis utility functions are commonly defined using some form of the weighted sum for example AX X X Ww XX w5XX w XX Here w denote weights and X denote attributes In qualitative multi attribute models there is natively no room for weights attributes are symbolic and utility functions are defined by decision rules However to bridge the gap between qualitative and quantitative models it is possible to introduce weights in a very approximate and imprecise way also into the qualitative models 2 10 3 1 Principle k unace low good medium acc ae TECH CHAR 9 PRICE The figure above illustrates the basic approach It shows the CAR utility function represented by points blue dots in a three dimensional space Each point represents one defined decision rule To find out the weights DEXi places a hyper plane shown in red into this space so that it matches the points as closely as possible using the least squares measure Once done weights can be approximated directly from the slopes of the hyperplane the higher the slope in the direction of an attribute the higher the corresponding relative weight In the above figure the weights of PRICE and TECH CHAR are almost identical 53 and 47 respectively These are local normalized we
65. mediate size up to 50 or maybe 100 a useful approach is to combine your data entry with DEXi s Use scale orders strategy In this case you jump across the table and enter values only for some important rules such as rules having extreme values of attributes or extreme function values Indeed it takes some practice to learn which rules are important At the same time you let DEXi calculate values of non entered rules In this way you may quickly get a highly determined function You may follow this stage by several iterations of finding weak unsatisfactorily defined points of your definition and entering your values there until the function has been completely defined 36 For even larger tables you should additionally employ DEXi s Use weights strategy Start again with entering important values Then open Weight Editor and define your required attribute weights If successful this will fill all the non entered cells with single values corresponding as closely as possible to your weights and already entered rules Afterward do not forget to check what DEXi has done automatically for you Review the assigned values and verify if they make sense to you Also check the actual weights achieved in this way for example run Weight Editor again and see the Current weights Although DEXi can work with less than completely defined utility functions it is highly recommended to define the functions completely Incompletely define
66. mn Undo Undo the last option editing operation Redo Repeat the last undone options editing operation 3 12 3 Remarks Duplicating columns is very useful for what if analysis duplicate an option then leave the original intact and modify only its copy In this way you can easily compare the effects of changes on evaluation results 3 13 Evaluation Page The Evaluation Page of DEXi model window displays evaluation results Each option created previously in the Options Page is evaluated by the model created in the Model Page You can change option descriptions values of basic attributes and see the effects of changes and copy and paste all option data In addition the Analysis Menu and corresponding toolbar buttons provide commands for option analyses Plus minus 1 analysis Selective explanation and Compare options 38 i Model S Options Z Evaluation IE Chans Ha W ti 3 A w ca Option Carl Car CAR exc good PRICE J law medium BUTY PRICE medum meda mn MAINT PRICE bo medn TECH CHAR exc good _ COMFORT high high PERS mone more 4D00A5 4 3 _ LUGGAGE tio big ze SAFETY hagh medu lattributas 10 6 basic aggregate Scales 10 Functions 4 Options 2 3 13 1 Workspace The workspace is similar to Options Page s one in that it provides a table showing option data The difference is that all option data is shown here including the values of
67. n commercial applications For information on commercial applications and for any other questions please contact Marko Bohanec marko bohanec iijs si Any feedback on your experience with DEXi will be greatly appreciated 1 2 Functionality DEXi supports two basic tasks 1 development of qualitative multi attribute models 2 application of these models for the evaluation and analysis of decision options The models are developed by defining e attributes qualitative variables that represent decision sub problems e scales ordered or unordered sets of symbolic values that can be assigned to attributes e tree of attributes a hierarchical structure representing the decomposition of the decision problem e utility functions rules that define the aggregation of attributes from bottom to the top of the tree of attributes In the evaluation and analysis stage DEXi facilitates e description of options defining the values of basic attributes terminal nodes of the tree e evaluation of options a bottom up aggregation of option values based on utility functions e analysis what if analysis plus minus 1 analysis selective explanation and comparison of options e reporting graphical and textual presentation of models options and evaluation results DEXi differs from most conventional multi attribute decision modeling tools in that it uses qualitative symbolic attributes instead of quantitative numeric ones Also aggregati
68. n printouts table headings etc e Description usually a longer string providing further documentation about the attribute e Scale Aggregate attributes also have a utility function 2 8 Tree of Attributes In a multi attribute model attributes are organized hierarchically into a tree of attributes A model can have one or more root attributes Each attribute can be decomposed into one or more descendant attributes that appear one level below that attribute in the tree Decomposed attributes are called aggregate attributes Attributes that do not have descendants and appear as leaves of the tree are called basic attributes 2 8 1 Interpretation A tree of attributes can be interpreted in three ways 1 Decomposition It represents a decomposition of a decision problem into sub problems To solve a problem which is represented by a a higher level attribute one has to solve sub problems represented by its lower level descendants 2 Dependency A higher level attribute depends on its immediate descendants in the tree This dependency is modeled by a utility function that corresponds to the higher level attribute 3 Aggregation Tree structure defines the bottom up aggregation of option values The value of a higher level attribute is calculated as an aggregation of the values of its immediate descendants in the tree Again this aggregation is defined by the corresponding utility function The interpretation of attribute types
69. nM SAFETY TAM g njam 3 The same as above but with text values and transposed orientation BUY PRICE MAINT PRICE PERS DOORS LUGGAGE SAFETY Cari medium low more 4 big high Car2 medium medium more 4 big medium 3 4 2 Function Data File Function data files are tab delimited text files for storing utility functions which are exported and imported through Export function and Import function File Menu commands respectively One file contains one utility function whose format can be specified with Settings Import Export 3 4 2 1 Examples 1 TECH CHAR function from the Car Evaluation model exported in tab delimited format using base 1 values and exporting all rules The character represents an entered rule The TAB character is denoted ICOMF ORT SAFETY TECH CHAR oe ts 7 22275 414 iS fa an 31 471717175 502 317 12 914 pas TO TP Toup XD 87121373 933343434 2 The same function exported using text values and exporting only entered rules in this case there is no need to display the or entered status ACCOMFORT SAFETY TECH CHAR l bad bad bad 27acc bad bad 32good bad bad 25 4 bad bad bad STacctacc7acc 671good good good 7 bad bad7bad 8 accigoodigood 97good exc exc 3 5 Edit Menu The DEXi Edit Menu provides commands for editing the DEXi model that is shown in the currently active model window The available ed
70. ng of e adata entry field used to edit the value of a single cell in the table and e command buttons To enter an option name select a corresponding name cell in the table and edit it contents in the data entry field 3 To enter a cell value other than name you can e Select an item from the data entry field e right click and select an item from the pop up menu or e press one of the keys 1 2 on the keyboard which represent consecutive scale values or press the key which denotes an undefined value 3 12 2 Commands To invoke a command you may either e press a button shown in the toolbar e select an item from the Edit Menu All commands refer to the option column whose cell is currently selected in the table Add option Inserts a new column to the right of the currently selected option The new option is named New and all its value cells are set to undefined Delete option Deletes the currently selected option Move left Moves the currently selected column one place to the left Move right Moves the currently selected column one place to the right de Cut Eguivalent to Copy followed by Delete option Copy Copies the currently selected option column of cells into the clipboard for further use E Paste Inserts previously Cut or Copied column into the table positioning it to the right of the currently selected column Ei Duplicate Duplicates the currently selected colu
71. o ei 58 962 BUG ee O n C MUR as 58 E 59 5 7 1 New program eatur Sieci aaa anni alenka obna lene eo o bana aaa vini 59 WA MEE 5 Ta ie O EE 59 90 VESON OF eion ea eo oa ee oa tensa A me 59 5 8 1 New program Te ales p abb e el ea Aid ees 59 9 0 2 BUG TRES eiiis diia pt ida 59 lp O eae 60 DEXI Version 3 04 Program for multiattribute decision making Copyright 1999 2012 Developed in collaboration Jo ef Stefan Institute Ljubljana Faculty of Organisational Sciences Kranj Ministry of Education scence and Sport of the Republic of slovenia 1 Introduction DEXi is a computer program for multi attribute decision making It is aimed at interactive development of qualitative multi attribute decision models and the evaluation of options This is useful for supporting complex decision making tasks where there is a need to select a particular option from a set of possible ones so as to satisfy the goals of the decision maker A multi attribute model is a hierarchical structure that represents the decomposition of the decision problem into sub problems which are smaller less complex and possibly easier to solve than the complete problem 1 1 Availability DEXi is implemented in Borland Delphi and runs on Microsoft Windows platforms The latest version can be downloaded from http kt ijs si MarkoBohanec dexi html DEXi can be used free of charge for no
72. o handle In this case try to restructure the tree below that attribute Attribute linking works only when enabled in Settings In that case it is done automatically while you edit your model on the Model Page As this may appear confusing you may temporarily disable linking When enabled again your links will be restored automatically 2 9 Scale Scale represents a set of values that can be assigned to an attribute In DEXi scales are qualitative and discrete They consist of a set of words such as excellent acceptable inappropriate etc 14 Scales can be ordered or unordered and ordered scales can be either increasing or decreasing An unordered scale is just a collection of values whose relation with each other is unknown or undefined In contrast the values of an ordered scale are ordered preferentially that is according to their contribution to the quality of options The values of increasing scales are ordered from bad to good values and the value of a decreasing scales are ordered from good to bad values In both cases bad represents a value that is disadvantageous for the option and is least preferred by the decision maker Analogously good represents an advantageous and most preferred value The ordering of scales plays an important role in the definition of utility functions where it simplifies the definition of decision rules and facilitates checking of their consistency With ordered scale
73. o product portfolio evaluation 3 Companies o business partner selection o performance evaluation of companies 4 Personnel Management o personnel evaluation o Selection and composition of expert groups o evaluation of personal applications for jobs 5 Medicine and Health Care o risk assessment o diagnosis and prognosis 6 Other Areas o granting personal corporate loans o assessment of technologies o assessments in ecology and environment o assessments in agronomy 1 4 Development and history DEXi has been developed in collaboration of the e Department of Knowledge Technologies Jo ef Stefan Institute Ljubljana Slovenia and e University of Maribor Faculty of Organizational Sciences Kranj Slovenia The initial development was financially supported by the Ministry of Education Science and Sport of the Republic of Slovenia within the Ro Computer Literacy Programme 1999 2000 The development of DEXi started in 1999 The motivation was twofold First it was conceived as a successor to DEX a successful program for multi attribute modeling Until 1999 DEX had been used in several tens of real life decision situations see publications on the DEX web page However as a MS DOS program DEX was becoming outdated and less convenient for its users Second in that time in Slovenia the DEX methodology was taught in several university courses and there were ideas to introduce it into secondary schools as well Thus there was a strong n
74. of the corresponding attribute iBix Quality of a car unacc unacceptable car acc acceptable but won t like it good satisfies my needs exc excellent car 3 11 2 Toolbars The horizontal toolbar which is shown above the table consists of Rule values field used to select a value and assign it to the current cell Delete value Deletes the contents of the currently selected cell Complex rules Makes and previews complex rules of the current function Weights Open Weight Editor to define attribute weights Use scale orders and Use weights are two check boxes that define how to handle non entered values The horizontal toolbar below the table contains Undo Undo the last function editing operation 32 Redo Repeat the last undone function editing operation Copy Copies the current utility function into the clipboard for further use The copied data format is the same as used in a function file E Paste Imports previously Copied function into the current function After the operation it reports the number of imported rules rules whose values matched between the clipboard and the current functions and the number of rules whose values actually changed The buttons Undo and Redo appear only if activated in Settings It is strongly recommended not to change Import Export Settings between copying and pasting a function 3 11 3 Pop up Menu This menu appears after right clickin
75. on utility functions in DEXi are defined by if then decision rules rather numerically by weights or some other kind of formula However DEXi does support weights indirectly In comparison with its predecessor DEX DEXi has a more modern and more convenient user interface Also it has better graphical and reporting capabilities and facilitates the use of weights to represent and assess qualitative utility functions On the other hand DEXi is somewhat less powerful than DEX in dealing with incomplete option descriptions DEX employs probabilistic and fuzzy distribution of values while DEXi facilitates only the use of crisp or unknown option values 1 3 Applications DEXi is particularly suitable for solving complex decision problems which typically involve many say 15 or more attributes many options 10 or more judgment which prevalently requires qualitative reasoning rather than numerical evaluation inaccurate and or missing data group decision making which requires communication and explanation For successful application DEXi requires sufficient resources in particular expertise and time for developing a DEXi model Some typical application areas and decision problems in which DEX and DEXi have been used so far are the following 1 Information technology o evaluation of computers o evaluation of software o evaluation of Web portals 2 Projects o evaluation of projects o evaluation of proposals and investments
76. ry options displaying all values of the primary option and only those values of the secondary options that differ from the primary s ones Attribute Cart Car CAR exc good PRI E low medium BUY PRICE medium MAINT PRICE low medium TECH CHAR good exc COMPORT high PERS more DOORS 4 L LUGGAGE big SAFETY high medium 91 This example compares Car primary option with Car2 secondary option Car2 differs from Car in the values of basic attributes MAINT PRICE and SAFETY which cause different evaluations of TECH CHAR PRICE and CAR 4 6 Some Car Evaluation Charts The following examples show some charts that can be obtained in DEXi from the evaluation of cars using the Car Evaluation Model The charts differ in the number of evaluation dimensions actually attributes selected for presentation in DEXi s Chart Page 4 6 1 Bar Chart This chart displays evaluation results according to one evaluation dimension In this case this is the root attribute CAR so the chart shows the overall evaluation of two cars unacc dcc good exc CAR 4 6 2 Scatter Chart A scatter chart displays evaluation results according to two selected evaluation dimensions In this case the selected dimensions are PRICE and TECH CHAR that is the two attributes that occur just below the root attribute CAR see the tree structure high medium low PRICE 92 4 6 3 Radar Chart Radar chart displays evaluation results according to
77. s the lowest value is considered bad and the highest good by default This can be changed in Scale Editor where you can individually declare the status of each value In this way not only a single value but a whole subsequence of values can be declared as bad or good For emphasis and better visualization extreme values of ordered scales are printed in different fonts and colors By default bad values appear in bold red and good values appear in talic green These can be changed in Settings Report 2 9 1 Example scales no yes low medium high e g for Quality high medium ow e g for Price unacceptable acceptable good excellent 2 9 2 Recommendations On scale size number of values e For basic attributes Use the least number of values that is still sufficient to distinguish between importantly different characteristics of options Usually this means two to four values e For aggregate attributes The number of values should gradually increase from basic attributes towards the root of tree of attributes For example three four valued attributes might be aggregated into a five valued attribute Five valued root attributes usually work quite well On scale ordering e Use increasing scales whenever possible they really help in the definition of utility functions e Avoid decreasing scales They are much less comprehensible than increasing scales Increasing and decreasing scales do not work when mix
78. some option characteristics change e How sensitive is the evaluation to small changes of the model such as addition or deletion of an attribute modification of some decision rules In other words analysis is a creative and possibly repetitive application of decision models aimed at better understanding of the decision problem better understanding of options their characteristics and consequences and better justification of the decision In general this involves techniques such as what if analysis sensitivity analysis stability analysis etc In DEXi analyses are mostly carried out on the Evaluation Page where you can e Review intermediate and overall results of option evaluation In order to clarify and justify the results you may focus on particularly bad or good evaluations e Change individual option values and immediately see the effects on evaluation results e Employ commands on the Analysis Menu or corresponding toolbar buttons to perform the analyses Plus minus 1 analysis Selective explanation and Comparison of options Some analyses can also take place elsewhere e Onthe Options Page you can duplicate an option description and prepare it for what if analysis e On the Model Page you can change any model component and retry the evaluation by opening the Evaluation Page e Various charts and reports may provide further insight into the evaluation process and achieved results 21 3 DEXi components and comman
79. stage the objective is to understand the decision problem and its components Some typical questions asked in this stage are What is the decision problem about Is it difficult and important Why Who is the stakeholder decision owner Who is responsible and who will be affected by the decision Who are other possible participants in the decision process What in general are the options alternatives in this case Can we define some specific ones Which goals objectives should be achieved by the decision Which are the criteria to be met by the decision What are the uncertainties involved What are the goals of the decision process Should we select a single option or evaluate or rank more of them What are the expected consequences of this decision process Do we need to justify the decision To whom and how 11 To be suitable for multi attribute modeling a decision problem must have some specific properties Primarily it should deal with options which need to be evaluated analyzed and compared with each other It is important that the decision problem can be decomposed into smaller less complex sub problems and that the options can be described by their basic features that correspond to the problem decomposition Thus we should also ask questions such as e Can we think of decomposing the problem into sub problems Can we define the relationship between factors that affect the decision e Can we think about representin
80. t v odvisnosti od konteksta Model za vrednotenje stre nih kritin Proceedings of the 15th International Conference Information Society IS 2012 8 12 10 2012 Ljubljana 76 79 2012 Bohanec M Bertheau Y Brera C Gruden K Holst Jensen A Kok E J L croart B Mess an A Miraglia M Onori R Prins T W Soler L G nidar i M The Co Extra decision support system A model based integration of project results Genetically modified and non genetically modified food supply chains Co existence and traceability ed Bertheau Y 461 489 Wiley Blackwell 2013 56 DEXi Versions This is a detailed summary of additions and improvements in different versions of DEXi 5 3 Version 2 0 5 3 1 5 3 3 5 3 4 9 4 5 4 1 New components Installation package English help Reconstructed DEXi Web Page New program features Program settings which control various aspects of DEXi particularly reports and data import export Importing and exporting option data through File Menu commands and Copy Paste operations on the Evaluation Page Exporting utility functions through a File Menu command Moving option data on the Options Page Attribute linking Slovene and English user interface Changed program features Modified DEXi file format o full conformance with XML UTF 8 standard o Saving program settings in DEXi files o Saving DEXi files in version 1 and version 2 formats Scale Editor
81. to weights required by the user and that its surface lies as close as possible in the least squares sense to the already entered values ZE Decision rules TECH CHAR q dl x le Use scale orders d ni E goo v Use weights small 2 small medium bad 3 small high had 4 medium small bad By medium medium acc B medium high good high small bad 8 high medium good J high high exc Above the requested weights were 50 for COMFORT and 50 for SAFETY Using these weights and the already defined rules 3 5 7 and 9 DEXi constructed a hyperplane and used it to determine the values of non entered rules 1 2 4 6 and 8 At least a few rules have to be entered by the user before this method could construct a hyperplane The exact number of needed rules depends on function dimensions and geometric positions of entered rules but until this condition has not been met DEXi deactivates this strategy and disables the Use weights checkbox 3 11 8 Function Editing Editing utility functions requires quite some skills and experience particularly when the corresponding tables are large You may want to try some of the following function editing approaches For small tables which consist of about up to 10 or 20 rules it is probably most effective just proceeding sequentially through the table and entering all the values in one turn For fast data entry use the keyboard keys 1 2 etc and Delete For tables of inter
82. um From this it immediately follows that for any rules having COMFORT or SAFETY better than medium the TECH CHAR value should be acc or better this is exactly what is shown with rules 6 and 8 above Similarly the entered bad value of rule 7 implies that the value of non entered rules 4 and 1 is bad too This strategy fails whenever the user enters values that violate monotonicity When the user attempts to enter a violating inconsistent value into a monotone rule set DEXi issues a warning and requests confirmation In the case that the user confirms such entry DEXi deactivates the scale order strategy and disables the Use scale orders checkbox The checkbox remains disabled as long as the entered rules violate monotonicity If monotonicity is reestablished later the checkbox is enabled 35 again but it remains unchecked the user must explicitly check it to reestablish the scale order strategy Hint Observe the status of Use scale orders checkbox When it is disabled grayed out your function is non monotone When it is enabled the function is monotone However the scale order strategy is in effect only when the checkbox is enabled and actually checked 3 11 7 2 Weights This strategy calculates the values of non entered rules using a hyperplane linear function which is constructed using weights as defined in the Weight Editor and other already entered rules The hyperplane is constructed so that its slopes correspond
83. umber of decimal places and to select individual weight types Evaluation These controls specify the display of evaluation results e Number of columns refers to the number of options shown together on the right hand side of the Evaluation results report element e Trim characters specifies the maximum number of characters that can be used to display option values in this report element the shorter the strings the more options can be displayed together Reports in html When checked all reports are prepared in html and shown in a browser without activating the standard preview Use default system browser When checked html reports are shown using the default system browser that is externally from DEXi In this case DEXi writes html files to the temporary system folder Otherwise a browser based on Internet Explorer is invoked internally without writing files to disk 44 3 19 2 Import Export Page This Settings page specifies the format of external data files which are imported and exported through File Menu commands Option data format These controls specify the format and contents of imported and exported option data files e Option values specifies the format of values which can be either text strings or ordinal numbers starting at 0 base 0 or 1 base 7 e Attributes import export the values of all or only basic attributes e Orientation specifies whether data file rows correspond to attributes normal
84. vent portals Multi attribute model and case study Knowledge management in electronic government ed Wimmer M A 4th IFIP International Working Conference KMGov 2003 Rhodes Greece 25 36 2003 Mladeni D Lavra N Bohanec M Moyle S eds Data mining and decision support Integration and collaboration Kluwer Academic Publishers 2003 Chapters e Bohanec M Decision support 23 35 e Bohanec M Rajkovi V Cestnik B Five decision support applications 177 189 e Moyle S Bohanec M Ostrowski E Large and tall buildings A case study in the application of decision support and data mining 191 202 Moyle S Ostrowski E Bohanec M Knowledge development using data mining A specific application in the construction industry Leveraging corporate knowledge ed Truch E Gower pp 181 197 2004 Vintar M Grad J ur E uprava Izbrane razvojne perspektive Univerza v Ljubljani Fakulteta za upravo 2004 e Leben A Bohanec M Vrednotenje portalov ivljenjskih situacij 123 140 e Bohanec M Odlo anje in ve parametrsko modeliranje 205 219 o4 Bohanec M Mess an A Angevin F nidar i M SMAC Advisor A decision support tool on coexistence of genetically modified and conventional maize Proc Information Society IS 2006 Ljubljana 9 12 2006 Verdev M Bohanec M DZeroski S Decision support for a waste electrical and electronic equipment treatment syste
85. y defined options You can check individual options to be displayed in the current chart This selection also determines which options are displayed in reports or exported to files The chart type is determined according to the number of checked attributes see chart examples No attributes No chart is displayed One attribute Bar chart Two attribute Scatter chart Three to six attributes Radar chart Radar charts can display only one option at the same time To choose which option is displayed select it highlight its name on the Options sub page 3 15 2 Commands To invoke a command you may either e select an item from the Chart Menu e right click and select an item from the pop up menu Show all options Select all options for display Hide all options Hide all options from display Usually this is followed by selecting individual options from the Options sub page Show values When this item is checked full scale value names are shown in charts Otherwise only their ordinal numbers are displayed This is typically used to save space or to avoid overlapping text Show option names This item controls whether option names are displayed in charts or not This may also save space or avoid overlapping text 41 Four options This item is available only for radar charts When it is checked four radar charts and thus four options are shown simultaneously Otherwise there is only one large chart t
86. yed in a report Report Create and preview a report Settings View and edit DEXi program settings Exit Close all windows and terminate the execution of DEXi 3 4 1 Option Data File Option data files contain option data which is imported and exported through File Menu commands In DEXi the contents and format of option files are controlled at three points e On Charts Page Options Sub page where you can select options to be actually exported all options are selected by default e n Import options and Export options dialogues that are used to specify file names and basic file format which is either Tab delimited or Comma separated CSV e In Settings Import Export where you can specify further details of Option data format 24 3 4 1 1 Examples 1 Tab delimited option data file containing both options from the Car Evaluation model and exported using the default Settings using base 1 values displaying all attributes using indentation normal orientation The TAB character is denoted aCarl cCarZ CAR7473 w PRICE 3332 BUY PRICE 7272 MAINT PRICE 7372 TECH CHAR W473 COMFORT 3 73 PERS7373 DOORS7373 LUGGAGE7373 SAFETYa3a2 2 The same data as above but using comma separated format base 0 values and including only non indented basic attributes in normal orientation m Gari UGar2 BUY PRICE ib n MAINT PRICEY 729 HIT PERS TL n DOORS on mM LUGGAGE ie iain
87. ze The second measure determination is somewhat more complex but better It takes into the account that in general cell values are intervals rather than single values A cell is 100 determined if it is assigned a single value and is 0 determined if it is completely undefined that is it contains the complete interval of values denoted For smaller intervals intermediate values are calculated proportionally A function is fully determined when all its cells both entered and non entered are 10096 determined This is in general achievable with less than 100 entered rules due to DEXi s handling of non entered values Therefore when editing a function the primary aim is to make it 100 determined regardless on the ratio of entered rules 33 The display of utility function status also includes a frequency distribution of function values shown in square brackets For instance high 5 medium 1 low 3 denotes that five decision rules map to the value high one rule to medium and three to low For functions that contain intervals the sum of frequencies is greater than the total number of rules 3 11 6 Weight Editor In Weight Editor you can view and define weights of the utility function that is currently edited in Function Editor Primarily these weights are intended for weight based calculation of non entered function values x Attribute 0 50 100 Required Current PRICE Lo B 50 EL TECH CHAH

DEXi: A Program for Multi-Attribute Decision Making

Contents

Download Pdf Manuals

Related Search

Related Contents