1
Phase 4-Individual Portion of Group Project
MGMT600-1104A-05
By
DemetriosGavrilos
Colorado Technical University Online
Ladies and Gentlemen of upper management I have been assigned by our group project research team to explain the chosen multivariate technique for our organization. Our company is committed to continuously improving its products. Naturally, this goal requires the use of mathematical and statistical tools. One of these statistical tools is multivariate. Multivariate uses more than one variable to examine data. (the free dictionary, 2011) I think multivariate regression may help one comprehend the idea behind what multivariate is. If, for example, a real estate agent records for various listings of houses their size in square feet, the number of bedrooms, the average income of each respective neighborhood and a rating of the appeal of each house. (statsoft.com, 2011) Now one can learn if the number of bedrooms is a better predictor of price for a house than how pretty it is or use all of the variables to forecast price. (statsoft.com, 2011)
Our group has focused on three multivariate techniques. This is because the group was trying to determine which of the three techniques would be most feasible for the company. I must note that the group was split. One member offered no input regarding which technique he preferred. As far as the other four were concerned two of them preferred cluster analysis and the other two preferred multidimensional scaling. One of the two members expressed concern that multidimensional scaling could not be applied at WidgeCorp. Personally, I disagree with this view. I think that all three techniques are applicable. Our personal experiences, likes and interest tend to affect our decisions. I am a Geography buff. This is part of the reason I was initially attracted to the multidimensional scaling technique. The distance between cities chart or, distance matrix, intrigued me. When the ALSCAL software was applied and the derived stimulus configuration was obtained I was impressed. (Babinec, 2011) The position of the cities in the chart resembled their position on the map. I started thinking that the ease at which one could discern how far apart cities were without having knowledge of Geography was considerably powerful. Perhaps this technique could be used to simplify situations WidgeCorp faced in marketing its products or developing new ones. This simplification would take a negative, such as ignorance with Geography, and minimize it so our organizations decision makers could develop more sound decisions. Since our group was split on the preferred technique and I liked multi-dimensional scaling better than cluster analysis. I will treat multi-dimensional scaling as the chosen multivariate technique.
For example, the stimulus coordinates in the stimulus configuration show Miami’s stimulus number on the lower right portion of the configuration. The stimulus configuration looks like a ‘y’ and ‘x’ axis intersecting with different point plotted around them. Seattle’s stimulus number is on the upper left portion of the configuration. This this corresponds to the way the two cities are arranged on a map of the United States. Miami is on the southeastern edge of the country and Seattle is on the northwestern end of it. One doesn’t need to know the information a map shows if all that is needed is for one to know relative information. That is to say, where the cities are in relation to each other and how far they are from each other relatively speaking and not in terms of measured distance. One doesn’t even need to know what a map of the U.S.A. looks like to make certain sound logistic decisions. One could tell at a glance it would a little more than twice as long to travel from Miami to Seattle than it would to go from Miami to Chicago regardless.
The stimulus configuration may be obtained even if the actual distances between cities are not known. The distance between Seattle and Miami may ranked as being a 9 on scale from 1 to 10 and then the ALSCAL software may be used to obtain a multi-dimensional scaling stimulus configuration like the one that was derived from the distance chart. (Babinec, 1989) Similar rankings may be used to obtain a chart which pertains to dissimilarities between snack foods produced by various companies. These rankings may be obtained from consumers that have been providing researchers with their input. If our organization’s snacks are located near several other products on the stimulus configuration then our organization will need to find ways to differentiate the products from competitors if it is determined that this differentiation will likely increase customer satisfaction. Multiple-dimensional scaling works effectively in matters where subjectivity is involved. (Babinec, 1989) These are issues that involve taste, smell, and perceptual judgments. (Babinec, 1989) This is one of the strengths of multi-dimensional scaling. MDS assumes that the subject evaluates the object in all relevant characteristics without the researcher having to list them. (Babinec, 1989)
The subject may taste the snack and rank its similarity based on saltiness, sweetness, or cherry-ness. (Babinec, 1989) This is what makes MDS different from a lot of other multivariate techniques where all relevant variables need to be enumerated and an omissionof them can bias results. (Babinec, 1989) Another advantage of multidimensional scaling is it doesn’t need the input to be quantitative data. (Babinec, 1989) As I pointed out in the snack food example dissimilarities were used as with one of the city distance examples. These were ordinal assumptions and not precise numerical inputs. (Babinec, 1989) Sometimes researchers can only measure at the ordinal level. Multiple dimensional scaling enables researchers to overcome this constraint and rate items in terms of similarities and dissimilarities and then assess how products relate to other products. SPSS Inc., a company that writes and markets software for marketing research and analysis has performed multidimensional scaling in the soft drink industry so as to show similarities and dissimilarities between different brands of soft drinks. (Babinec, 1989)
An example of another real life company that has used multidimensional scaling would be Sam’s club. Sam’s club has used multidimensional scaling to determine how research participants categorize logos. (sandsresearch, 2008) In a world where consumers are constantly seeing logos of various types, from sports team to automobiles, organizations need to know how consumers will respond to the logos used on their products. Sands research performed research for Sam’s club pertaining to which logo the chain of stores could use to promote its “simple steps for saving green” campaign. (sands research, 2008) The brain responses of the various logos which the organization was interested in were analyzed. Basically, statistically significant data was used to determine the level of engagement in response to the logos. (sands research, 2008) This data was obtained by using neurobiological preferences of various logos. (sands research, 2008) Brain activity was recorded and displayed using special equipment so that the yellow colored areas on the brain images on monitors showed high emotional engagement by using the color yellow. (sands research, 2008) A perceptual grid or stimulus configuration was developed by using software in a similar manner to the examples already mentioned. Much like in the snack food examples similarities and dissimilarities were taken into account and clarity and unbiased preferences were plotted on a grid using software. (sands research, 2008) The candidates evaluated all of the relevant characteristics. By looking at the outputted grid or configuration researchers could tell which logo excited or was most engaging to the group. The result was the light green leaf touching water and making the ripple on it.
(sands research, 2008)
The other two group members preferred cluster analysis. Cluster analysis is different from multi-dimensional scaling in that in divides data into groups or clusters and doesn’t involve using similarities or dissimilarities so as to develop a grid or configuration where they may be observed. (Kumar, 2004) These groups may be meaningful, useful or both. (Kumar, 2004) It is very common for people to divide things into groups or clusters and assign objects to these groups, or classify them. (Kumar, 2004) For example children easily cluster things that they see in pictures such as cars, buildings or people. (Kumar, 2004) Businesses tend to use clustering analysis. They collect large amounts of informationon current and potential customers. (Kumar, 2004) Clustering can be used for segmenting customers into smaller groups for further analysis and marketing activities. (Kumar, 2004) Also, cluster analysis didn’t develop within a particular field of study as MDS did. However cluster analysis is similar to multidimensional scaling in that it makes no distinction between dependent and independent variables. (Sheppard, 1996)
Blue Cross Blue Shield of Iowa performed a cluster analysis to develop groups of customers with different preferences of insurance coverage and this enabled BCBSI to develop products that are most appropriate for customers. (Thomas, 1990) This shows that bundling service and products and service features may be used to develop optimal products. (Thomas, 1990) Five clusters were created. Some of them were made up of people who were willing to pay fully for certain doctors with all services covered others for all doctors with no out of pocket and others for with automatic claims filed at $40 per month. (Thomas, 1990) One of our group members used a zoo clustering example to show if customers wanted this particular zoo to be educational for their children or merely a zoo where animals were kept. I think WidgeCorp can apply cluster analysis as well. It may be used group people into those who preferred morning snacks, health conscience subjects, those who prefer sweet over salty snacks etc. This may enable WidgeCorp to produce products more adapted to the needs of consumers.
None of our group members chose factor analysis as their preferred multivariate technique. Although Shawna, our small group member, did some research on factor analysis, she seemed to support cluster analysis more than factor analysis.Factor analysis is basically a term for techniques that analyze interrelationships between variables. (Hanson, 1988) Factor analysis’ main purpose is to reduce a large set of variables into a smaller set of “factors” or unifying concepts. (Hanson,1988) It is seen as an alternative to multidimensional scaling. (statssoft.com, 2011) Multidimensional scaling tends to offer faster more and more interpretable than MDS because factor analysis requires more dimensions be extracted as I will try to show later.(statssoft.com, 2011)
MDS can be applied to any kind of distances or similarities but factor analysis requires the use of a correlation matrix. (statssoft.com, 2011) In the previous Sam’s club I showed how multidimensional scaling (MDS) used brain stimuli to determine which logos subjects engaged in most. In factor analysis the stimuli would have to be rated on some list of attributes. (statssoft.com, 2011) Also factor analysis requires that underlying data is distributed as multivariate normal and that the relationships are linear. (statssoft.com, 2011)
Statistical models are used to explain the correlation between the variables, and software such as SPSS and SAS are used to conduct this factor analysis. (Hanson, 1988) Maritz Marketing Research Inc. has applied factor analysis to ratings respondents give pertaining to products. (Hanson, 1988) Some of these attributes may measure similar facets of the products. (Hanson, 1988) For example, let us assume that respondents are asked to rate a product or service on 20 attributes. (Hanson, 1988) Things such as cost, quality or how useful the product is are made up of different levels. These dimensions or levels are not easy to measure but factor analysis may help determine the factors that underlie the 20 original variables. (Hanson, 1988) It can also make it easier to understand complicated topics such as purchase intentions and consumer evaluations. (Hanson, 1988) There are basically four steps in factor analysis: 1. Compute the Correlation matrix, 2. Extract the factors, 3. Rotate the factors and 4. Calculate the factor scores. (Hanson, 1988) In the chart below one will see 3 variables per customer for analysis rather than the original 10. (Hanson, 1998) some of the dimensions making up the factors, such as good value for the price and reliable, may be applied to WidgCorp products and the retailers who sell them.
The eigenvalues at the bottom are the portion of total variation explained by each factor and since they are greater than “1” they are retained. (Hanson, 1988) It is important to note that in step 1 if the correlations in the correlations matrix are small values then the analysis will not be carried out since there is no point in doing so. The methods involved in steps 2 and 3 may be beyond the scope of upper management’s level of knowledge.I believe all three techniques are valuable. As the summary has indicated multi-dimensional scaling, is more interesting, versatile and powerful than factor analysis and cluster analysis. At times there is subjectivity involved in terms of which factor to retain, the extraction method, and the rotation method. (Hanson, 1988) This means that researchers may come up with different conclusions even though they are researching the same data. Factor analysis seems to be useful only when the situation is right for it. MDS is more flexible.
References:
Babinec, T. (1989) Multidimensional Scaling for Market Research. Retrieved November 3,
2011 from
Hanson, R. (1988) A Useful Tool but Not a Panacea. Retrieved November 5, 2011 From
Kumar, (2004) Cluster Analysis: Basic Concepts and Algorithms. Retrieved November 5, 2011
From
Sands Research (2008) Sam’s Club: A Case Study on Logo Evaluation. Retrieved November 4,
2011 From
Sheppard, A. (1996) The Sequence of Factor Analysis and Cluster Analysis: Differences
In Segmentation and Dimensionality Through the Use of Raw Factor Scores. Tourism Analysis, 1 (Inaugural Volume), 49-57 Retrieved November 7. 2011 From
Stat Soft, Multi-Dimensional Scaling. Retrieved November 6, 2011 From
Stat Soft, Cluster Analysis. Retrieved November 5, 2011 From
Stat Soft, Principal Components and Factor Analysis. Retrieved November 4, 2011
From
Thomas, T. (1990) New Findings with Old Data Using Cluster Analysis Retrieved November, 7
2011 From