chapter3

pdf

School

University of South Africa *

*We aren’t endorsed by this school

Course

3704

Subject

Mathematics

Date

Nov 24, 2024

Type

pdf

Pages

14

Uploaded by ChancellorFang8501

Report
Chapter 3 The AHP and SMART 3.1 Introduction In the previous chapter we discussed the parts from which methods are constructed in order to select one particular item from among many. Two of these methods, with variations, are the subject of this chapter. The best-known method is probably Thomas L Saaty’s [7] analytical hierarchical process (AHP). This method is based on Saaty’s eigenvector method for calculating weights from an A matrix, but any other method for calculating weights may also be used. However, the AHP is susceptible to several serious errors, inter alia two types of rank reversal. (These problems are discussed in Chapter 4.) F A Lootsma [6] proposed an alternative, namely the geometric (or multiplicative) AHP, which is increasingly accepted as an alternative to the AHP because it is not subject to rank reversal. The most attractive method for large practical problems is SMART, the abbreviation of a somewhat forced name, the Simple Multi-Attribute Review Technique. SMART also uses a method to calculate weights, and although the earliest versions proposed quite primitive methods, methods based on a A matrix have become the norm. SMART may also be seen as a methodology that is independent of the method used to determine weights. When one item has to be selected from many, we usually find one item that is the strongest in a particular area whereas another item is better in another area. This immediately tells us that there is more than one criterion or objective. All these methods allocate weights to all the different criteria, and values are assigned to the different items under each criterion. The criteria weights are then utilised to weigh the values assigned to the items under the different criteria before they are added together. Put differently: The criteria weights are used to calculate a weighted average of the values allocated to an item under the different criteria. In Example 2.2 of the previous chapter, values are allocated to six criteria. Assume that these values are encountered in an actual problem and that the figures presented in the last four columns present the performance of four items under the six criteria. Higher values are better. In this instance the values may be seen as percentages (or scores out of one hundred). The question is: Which item should be selected? 51
DSC3704 CHAPTER 3. THE AHP AND SMART Scores for Scores for Scores for Scores for Criterion Item 1 Item 2 Item 3 Item 4 1 85 70 65 70 2 55 90 70 85 3 35 60 85 60 4 68 65 80 55 5 65 60 30 60 6 70 65 45 65 The only correct answer is probably: It all depends! However, something that can be deduced from a direct comparison of the items is that Item 4 is outperformed by Item 2 in every instance. The scores given to Item 2 for each criterion are either the same or higher than those allocated to Item 4. Item 2 dominates Item 4 and the latter can be removed from the list of candidates because it offers nothing that Item 2 cannot deliver to the same extent or better. Definition 3.1. Item A is dominated by item B if B’s performance equals or exceeds A’s in every area (or for every criterion). In the table above item performance was presented as columns (or vectors) and the decision maker had no clear way of ranking the items. The bottom row of the table below presents a single value that summarises the column above it and this facilitates item comparison. Inability to compare the items in the above table has been eliminated by the weighted average. Criterion Weight Scores for Scores for Scores for i w i Item 1 Item 2 Item 3 1 0,15 85 70 65 2 0,18 55 90 70 3 0,25 35 60 85 4 0,21 68 65 80 5 0,10 65 60 30 6 0,11 70 65 45 Weighted average 59,88 68,50 68,35 The table can be produced by any one of the methods if the allocated scores were discounted. The principle of using a weighted average on the basis of criteria weights remains the same. The value functions of SMART and AHP are weighted arithmetic means of their scoring functions, while the value function of the multiplicative AHP is a weighted geometric mean of its scoring functions. SMART is discussed first because it is the simplest method and because its value function is so visible – which is not the case with AHP. It is therefore easier to understand AHP once you have understood SMART. 52
CHAPTER 3. THE AHP AND SMART DSC3704 3.2 SMART SMART (Von Winterfeldt and Edwards [9]) makes use of a scoring function f i to allocate values between 0 and 100 to items on behalf of criterion i . The value function z is then z = w i f i , a typical weighted average of the values provided by the scoring functions. The task is therefore to prepare a scoring function for each criterion, and it should preferably convert an objectively measurable quantity (e.g. distance in kilometres, or income in rands per year) into a score out of one hundred. Not all criteria can be measured objectively (e.g. appearance or status) and generally depend on human judgement. SMART requires scoring functions for these criteria as well, converting the initial assessment to a score between 0 and 100. The example illustrates both possibilities. Example 3.1. A man of thirty-four who has two children of seven and nine years old is looking for a house in a price category that would reflect the fact that both he and his wife are rapidly advancing in their chosen careers. They have already inspected a dozen houses and then decide to compile a SMART model. The first task is to determine the criteria, and they decide to concentrate on five characteristics: distance from the house to the local primary school and to the wife’s place of work, age, appearance, and the area of the house. They determine the weights of the criteria as 0 , 15 (school), 0 , 12 (work), 0 , 21 (age), 0 , 22 (appear- ance), and 0 , 30 (area). Scoring functions are prepared, and the two distances (to school and to the workplace) are represented by the functions below where x is the distance in kilometers. f School ( x ) = f Work ( x ) = 80 + 20 x with 0 x 1 100 with 1 < x 3 115 5 x with 3 < x 23 0 with x > 23 The distances are scored in the same way. However, this is not necessary and the graphs may differ. The graph is also piece-wise linear, which makes expression in algebraic form much easier. Once again this is not necessary, and the values may just as well be read from the graph itself. The score for the age of the house is presented in the table below. Age Age in years Score 0 + to 1 70 1 + to 2 100 2 + to 5 80 5 + to 8 60 8 + to 50 0 50 + 100 The score for the appearance of the house is presented in the following table: 53
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help
DSC3704 CHAPTER 3. THE AHP AND SMART Appearance Description Score Will impress yuppies 100 Will impress boss 80 Will impress parents 40 An ordinary house 0 The scoring function for the surface area, x , is presented by the following: f Area ( x ) = 0 with 0 x 200 0 , 2 x with 200 < x 500 100 with x > 500 The five functions above are not all in an accustomed form and the one that has to provide a score under appearance is a function of a subjective input. If the couple in the example were to select the description that is consistent with their value system, the table would nevertheless allocate the desired score under this criterion. Allow f School , f Work , f Age , f Appearance , f Area to represent the five scoring functions. The value function would then be: z = 0 , 15 f School + 0 , 12 f Work + 0 , 21 f Age + 0 , 22 f Appearance + 0 , 30 f Area . The details of the next three houses they view are presented in the table, and their values have to be calculated in accordance with the value function z . Characteristic House A House B House C School (km) 4 1 12 Work (km) 9 10 1 Age (years) 0 6 69 Appearance (category) yuppies yuppies boss Area (m 2 ) 230 420 390 The next table presents the values of the scoring functions in the columns below the houses, with the weighted averages at the bottom – the values assigned to the houses by the value function above. Characteristic Weight House A House B House C School (score) 0 , 15 95 100 55 Work (score) 0 , 12 70 65 100 Age (score) 0 , 21 70 60 100 Appearance (score) 0 , 22 100 100 80 Area (score) 0 , 30 46 84 78 Value 73 , 15 82 , 60 82 , 25 54
CHAPTER 3. THE AHP AND SMART DSC3704 The ranking for SMART is B > C > A . The value for B is just a little higher than the value for C. Before B is chosen, we have to carefully consider the reasons for B’s high value. An error in the scoring function or an overenthusiastic weight can sometimes be the underlying reason. A sensitivity analysis of each case would be a good idea. A valid question in this case is what the smallest change, θ , would be in a weight to increase the value of C above that of B. C would benefit most from an increase in the weight for age because the score for C exceeds that of B most in the case of this characteristic. Indeed, if this weight were to increase by θ , the value for C would increase by 40 θ more than the value for B would increase. The gap is therefore reduced by 40 θ . In the same manner a reduction in the score for school would decrease the value of B quite quickly relative to C’s value, namely 45 θ . The extent to which the gap would decrease when the weight for age were to increase by θ and that for school were to decrease by the same amount, is therefore 85 θ and the gap to be closed is 0 , 35 . The equation 85 θ = 0 , 35 gives θ = 0 , 004118 , a very small figure. The choice between B and C is therefore quite sensitive to changes in the weights. Moreover, the weights are obtained in a process that is subject to quite extensive errors. For practical purposes, the two are equal. Activity 3.1. Briefly describe the scoring functions for: 1. Distance from the school. 2. Area of the house. Activity 3.2. Calculate the values of the houses described in the table Characteristic House X House Y House Z School (km) 6 5 12 Work (km) 3 10 15 Age (years) 1 10 65 Appearance (category) Ordinary Yuppies Parents Area (m 2 ) 530 520 300 under the value function z = 0 , 20 f School + 0 , 17 f Work + 0 , 16 f Age + 0 , 22 f Appearance + 0 , 25 f Area . Activity 3.3. The outcome of the previous activity places Y before X, which is in turn before Z. What would be the smallest weights change to give Z and Y the same values? Use the solution of Activity 3.2. 55
DSC3704 CHAPTER 3. THE AHP AND SMART 3.3 AHP The AHP differs from SMART only in the scoring functions. The scoring functions of AHP are not prepared beforehand. The scoring takes place under each criterion by comparing the various competitors pair-wise and by using the methods discussed in Chapter 2 to solve the resultant A matrix. (The AHP, as it was originally described, uses the eigenvector method to determine the preference vectors.) A preference vector is determined under each criterion, and when the value of the first item is calculated, the first component of each preference vector is used as the score for that criterion. The AHP cannot evaluate a competitor on its own; all of them are compared, a preference vector is derived from the comparison, and the components of the preference vector represent the scores of the competitors. Because the components of a preference vector add up to one (normalisation), it means that the more competitors there are, the lower would be the score of each. The values allocated to each competitor by the value function consequently also depend on the number of competitors. In principle this is not a problem because the competitors are only compared with each other, but problems can occur as can be seen from the example. Example 3.2. Look again at the above example where different houses have to be compared with each other. Assume the same weights were determined for the criteria, then the value function would again be z = 0 , 15 f School + 0 , 12 f Work + 0 , 21 f Age + 0 , 22 f Appearance + 0 , 30 f Area . Also assume that there is an external value process that assigns values to the three houses (as in the table). This same table has been used in the example and is selected to compare the outcome of the AHP with that of SMART. Characteristic House A House B House C School (score) 95 100 55 Work (score) 70 65 100 Age (score) 70 60 100 Appearance (score) 100 100 80 Area (score) 46 84 78 The A matrices are compiled by using the values in the table in place of the u i . For the first characteristic, A is A School = 1 , 00 95 100 95 55 1 , 00 100 55 1 , 00 and its preference vector is w School = 0 , 3800 0 , 4000 0 , 2200 . 56
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help
CHAPTER 3. THE AHP AND SMART DSC3704 The other A matrices are A Work = 1 , 00 70 65 70 100 1 , 00 65 100 1 , 00 A Age = 1 , 00 70 60 70 100 1 , 00 60 100 1 , 00 A Appearance = 1 , 00 100 100 100 80 1 , 00 100 80 1 , 00 A Area = 1 , 00 46 84 46 78 1 , 00 84 78 1 , 00 and their preference vectors are w Work = 0 , 2979 0 , 2766 0 , 4255 , w Age = 0 , 3043 0 , 2609 0 , 4348 , w Appearance = 0 , 3571 0 , 3571 0 , 2857 and w Area = 0 , 2212 0 , 4038 0 , 3750 . The i -th component of a preference vector is the score the scoring function assigns to the i -th item. In this example it means that f School provides the score 0,3800 to house A and that f Age = 0 , 2609 for house B. The value of house A is therefore z A = 0 , 15 f School + 0 , 12 f Work + 0 , 21 f Age + 0 , 22 f Appearance + 0 , 30 f Area = 0 , 15(0 , 3800) + 0 , 12(0 , 2979) + 0 , 21(0 , 3043) + 0 , 22(0 , 3571) + 0 , 30(0 , 2212) = 0 , 3016 . The value of house B is z B = 0 , 15(0 , 4000) + 0 , 12(0 , 2766) + 0 , 21(0 , 2609) + 0 , 22(0 , 3571) + 0 , 30(0 , 4038) = 0 , 3477 and that of house C is z C = 0 , 15(0 , 2200) + 0 , 12(0 , 4255) + 0 , 21(0 , 4348) + 0 , 22(0 , 2857) + 0 , 30(0 , 3750) = 0 , 3507 . The AHP ranking is C > B > A . 57
DSC3704 CHAPTER 3. THE AHP AND SMART Activity 3.4. The values of competitors M 1 , M 2 , M 3 , M 4 have to be calculated with the AHP. There are three criteria: A, B and C, and a 3 × 3 matrix of pair-wise ratios is obtained from a group of decision-makers. After applying one of the solution methods from the previous chapter, the preference vector is determined as w criteria = 0 , 2080 0 , 4305 0 , 3615 . Three 4 × 4 matrices are also prepared by the group and their preference vectors are found to be w A = 0 , 1719 0 , 1990 0 , 3166 0 , 3125 , w B = 0 , 3004 0 , 2909 0 , 2109 0 , 1978 , and w C = 0 , 2710 0 , 2571 0 , 2571 0 , 2148 . Calculate the AHP score of each competitor. 3.4 Multiplicative AHP Another way of approaching the value function is by asking: Why the weighted sum of the scores? Why not the weighted product? Why z from z = w i f i and not from z = ( f i ) w i ? Lootsma (1999) is strongly in favour of this (multiplicative or geometric) approach as an alternative to the (normal additive) AHP because it naturally arises from the nature and structure of some problems. The w i are the criteria weights and the f i values are calculated as for the normal AHP scores. We again use the example of the person looking for a house. Example 3.3. You have already been introduced to the table below. It contains the weights of the five criteria in the second column and the performance of the three houses in the columns below the houses. Characteristic Weight House A House B House C School (score) 0 , 15 95 100 55 Work (score) 0 , 12 70 65 100 Age (score) 0 , 21 70 60 100 Appearance (score) 0 , 22 100 100 80 Area (score) 0 , 30 46 84 78 In order to determine the value of A in terms of the multiplicative AHP, the AHP scores under each of the criteria have to be calculated first. Under School it gets 95 out of the total allocation of 95 + 100 + 55 = 250 , therefore f School = 95 250 = 0 , 38 . The scores under the other criteria are: f Work = 70 235 = 0 , 2979 ; f Age = 70 230 = 0 , 3043 ; f Appearance = 100 280 = 0 , 3571 ; and f Area = 46 208 = 0 , 2212 . 58
CHAPTER 3. THE AHP AND SMART DSC3704 The value function of the multiplicative AHP allocates to A the value z A with z A = 0 , 3800 0 , 15 × 0 , 2979 0 , 12 × 0 , 3043 0 , 21 × 0 , 3571 0 , 22 × 0 , 2212 0 , 30 = 0 , 2954 . Strictly speaking, this is not yet the final z A because the AHP always demands normalisation. The values z B and z C also have to be calculated and the three of them have to be normalised. Activity 3.5. Complete the process by calculating z B and z C and normalising the three weights. Activity 3.6. Which values does the value function of the multiplicative AHP assign to the four items of Activity 3 . 4 ? 3.5 Ranking issues Consider again the rankings that were obtained for the three houses in Example 3.1 using SMART and in Example 3.2 using AHP. The ranking that was obtained for SMART is B (value 82 , 60), C (82 , 25), A (73 , 15). This is contrary to the AHP ranking of C (value 0 , 3507), B (0 , 3477), A (0 , 3016) and an explanation has to be found. The difference cannot be explained by the criteria weights because exactly the same value function is used, and the only difference is in the scoring functions. The difference therefore has to lie in the scoring functions. The scoring functions of the AHP do not provide values out of one hundred as SMART’s scoring functions do. In itself this should not make any difference, because the relative quantities of the values that give rise to their ranking are at issue here and not their absolute quantities. The scores of the scoring functions of the AHP are weights – which means that they add up to one – and this is where the problem lies. An example would clarify this point, but first note that it is sometimes unnecessary to compile A matrices. In this study guide and in the literature it is sometimes necessary to construct examples to explain a concept. Values are then postulated for the competitors and from these the weights are directly calculated by way of normalisation. For example, the above example uses the scores from the SMART exercise. The A matrices were then compiled and solved to determine the preference vectors. However, this is unnecessary. Consider for example the first characteristic (the distance to school) of the three houses. The scores are 95, 100, 55 with a total of 250. Normalisation provides 95 250 ; 100 250 ; 55 250 and we again have w School = 0 , 3800 0 , 4000 0 , 2200 . Returning to the argument, assume there are five items and that their actual performance cannot be improved. SMART recognises that they are all good by allocating 100 to all of them. The AHP (which provides scores to all and where the scores add up to one), allocates 0 , 20 to each. Suppose the same five items are all equally poor with respect to a subsequent criterion. The AHP will again allocate 0 , 20 to all of them, merely because they recorded the same performance, whereas SMART 59
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help
DSC3704 CHAPTER 3. THE AHP AND SMART might allocate 35. This difference between SMART and the AHP can easily give rise to different rankings. Once again consider five items and two criteria. Suppose they all perform quite well on the first criterion, with an excellent score for the fifth item. SMART’s scoring function for the first criterion allocates values of say 76; 76; 76; 76; 96. Normalisation provides the following weights: 0 , 190; 0 , 190; 0 , 190; 0 , 190; 0 , 240, and these are the AHP’s scores for the five items under the first criterion. Further assume that SMART allocates values of 48; 38; 38; 38; 38 under the second crite- rion. The weights to be used as scores for the AHP are then 0 , 240; 0 , 190; 0 , 190; 0 , 190; 0 , 190. Also suppose that the two criteria carry the same weight. The value function of the AHP then allocates the values 0 , 215; 0 , 190; 0 , 190; 0 , 190; 0 , 215 and the SMART values are 62; 57; 57; 57; 67. The example could refer to five students who each writes two papers. The students who are com- peting for the best scores are One and Five. In the first paper, Five beats the other four by 20 points, but in the second paper One is the winner with 10 points more than the rest. The SMART approach uses the average of the two papers and the 67 average of Five is the highest. This is not unexpected, because he beat the other four by 20 points in the first paper whereas One earned 10 points more in the second paper. The AHP – where the points allocated by the scoring function have to add up to one – shows 0 , 24 for both top students in the paper against 0 , 19 for the rest. The fact that one student achieved 20 points more in one paper compared to the other’s 10 points more in another paper is lost in the process. The problem lies in normalisation. SMART provides 400 points (for the five competitors together) for the first, and 200 for the second criterion. The AHP always allocates a total of one. In the comprehensive criticism of AHP (which is discussed later in the guide), “normalisation at every level” is repeatedly mentioned as the reason for and the cause of specific problems. 60
CHAPTER 3. THE AHP AND SMART DSC3704 3.6 Answers to activities Activity 3.1. 1. The graph of the scoring function appears below. D i s t a n c e i n k m S c h o o l ( a s w e l l a s w o r k ) S c o r e 2. The graph of the scoring function appears below. A r e a A r e a i n m 2 S c o r e 61
DSC3704 CHAPTER 3. THE AHP AND SMART Activity 3.2. Characteristic House X House Y House Z School (km) 6 5 12 Work (km) 3 10 15 Age (years) 1 10 65 Appearance (category) Ordinary Yuppies Parents Area (m 2 ) 530 520 300 Characteristic Weight House X House Y House Z School (score) 0 , 20 85 90 55 Work (score) 0 , 17 100 65 40 Age (score) 0 , 16 70 0 100 Appearance (score) 0 , 22 0 100 40 Area (score) 0 , 25 100 100 60 Value 70 , 20 76 , 05 57 , 60 Activity 3.3. The gap between the values for Y and Z is 18 , 45 and it can be closed by increasing the weight of a characteristic so that Z exceeds Y and simultaneously decreasing to the same extent the weight of a characteristic for which the opposite is true in order to keep the sum of the weights equal to one. The two characteristics that show the highest score differences are selected to ensure the smallest weight change. The difference under age between Z and Y’s scores is the biggest, and for an increase of θ the gap narrows to 100 θ . Under appearance , the difference between the scores for Z and Y is the smallest (most negative), and the gap narrows by a further 60 θ . The overall narrowing of the gap is therefore 160 θ and the θ that makes 160 θ = 18 , 45 has to be found. The answer is θ = 0 , 1153 and this is the smallest change in weights (one weight up and one down) that is necessary to equalise the values for Z and Y. Activity 3.4. The value function is z = 0 , 2080 f A + 0 , 4305 f B + 0 , 3615 f C . The value of M 1 is therefore z 1 = 0 , 2080 f A + 0 , 4305 f B + 0 , 3615 f C = 0 , 2080(0 , 1719) + 0 , 4305(0 , 3004) + 0 , 3615(0 , 2710) = 0 , 2630 . The calculations can be summarised in a table: 62
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help
CHAPTER 3. THE AHP AND SMART DSC3704 Criterion Weight M1 M2 M3 M4 A 0 , 2080 0 , 1719 0 , 1990 0 , 3166 0 , 3125 B 0 , 4305 0 , 3004 0 , 2909 0 , 2109 0 , 1978 C 0 , 3615 0 , 2710 0 , 2571 0 , 2571 0 , 2148 Value 0 , 2630 0 , 2596 0 , 2496 0 , 2278 Activity 3.5. The table below has been given. The first task is to calculate the weights for house B and house C and the last column has been added for this reason. Characteristic House A House B House C Total School (score) 95 100 55 250 Work (score) 70 65 100 235 Age (score) 70 60 100 230 Appearance (score) 100 100 80 280 Area (score) 46 84 78 208 Normalisation requires each of the values allocated under a particular criterion to be divided by the total of those values (the value on the right). For the sake of completeness, the figures for house A are also shown. Characteristic Criterion House A House B House C Weight (Weight) (Weight) (Weight) School 0 , 15 0 , 3800 0 , 4000 0 , 2200 Work 0 , 12 0 , 2979 0 , 2766 0 , 4255 Age 0 , 21 0 , 3043 0 , 2609 0 , 4348 Appearance 0 , 22 0 , 3571 0 , 3571 0 , 2857 Area 0 , 30 0 , 2212 0 , 4038 0 , 3750 The value function of the multiplicative AHP allocated the value z A to A with z A = 0 , 3800 0 , 15 × 0 , 2979 0 , 12 × 0 , 3043 0 , 21 × 0 , 3571 0 , 22 × 0 , 2212 0 , 30 = 0 , 2954 . The value of B is z B = 0 , 4000 0 , 15 × 0 , 2766 0 , 12 × 0 , 2609 0 , 21 × 0 , 3571 0 , 22 × 0 , 4038 0 , 30 = 0 , 3422 and that of C is z C = 0 , 2200 0 , 15 × 0 , 4255 0 , 12 × 0 , 4348 0 , 21 × 0 , 2857 0 , 22 × 0 , 3750 0 , 30 = 0 , 3415 . These three values add up to 0 , 9791 and they are normalised by dividing each by this figure. According to the multiplicative AHP, the values of houses A, B and C are then 0 , 3017 ; 0 , 3495 ; 0 , 3488 . 63
DSC3704 CHAPTER 3. THE AHP AND SMART Activity 3.6. The weights vector of the three criteria A, B and C w Criteria = 0 , 2080 0 , 4305 0 , 3615 has already been calculated, and the weights (acting as scores) of the four competitors (M 1 , M 2 , M 3 and M 4 ) under the various criteria are presented by the preference vectors w A = 0 , 1719 0 , 1990 0 , 3166 0 , 3125 , w B = 0 , 3004 0 , 2909 0 , 2109 0 , 1978 , and w C = 0 , 2710 0 , 2571 0 , 2571 0 , 2148 which have also already been determined. The value function of the multiplicative AHP allocates the following values to M 1 , M 2 , M 3 and M 4 . z 1 = 0 , 1719 0 , 2080 × 0 , 3004 0 , 4305 × 0 , 2710 0 , 3615 = 0 , 2577 , z 2 = 0 , 1990 0 , 2080 × 0 , 2909 0 , 4305 × 0 , 2571 0 , 3615 = 0 , 2571 , z 3 = 0 , 3166 0 , 2080 × 0 , 2109 0 , 4305 × 0 , 2571 0 , 3615 = 0 , 2465 , z 4 = 0 , 3125 0 , 2080 × 0 , 1978 0 , 4305 × 0 , 2148 0 , 3615 = 0 , 2241 . After normalisation, the values of M 1 , M 2 , M 3 and M 4 are 0 , 2615 ; 0 , 2609 ; 0 , 2502 and 0 , 2274 . 64