Multivariate Analysis

Definition of Multivariate Analysis

Multivariate analysis or multivariate techniques may be defined as the collection of methods for analysing data in which a dependent variable is represented in terms of several independent number of observations which are available to define such relationship. In brief, techniques that take account of the various relationships among variables are termed multivariate analysis or multivariate techniques.Mathematically, multivariate analysis was defined by Takeuchi, Yanai and Mukherjee as, forming a linear composite vector in vector space, which can be represented in terms of projection of a vector on to certain specified subspaces.

Objectives of Multivariate Analysis

The basic objectives of multivariate analysis are: (i) To represent the collection of large set of data in a simplified way, by transforming large number of observations into smaller composite scores. (ii) To predict the variability of the dependent variable based on its covariance with all the independent variables. (iii) To classify individuals or objects into one of the two or more mutually exclusive and exhaustive groups on the basis of a set of independent variables.

Advantages of Multivariate Analysis

The main advantage of multivariate analysis is that since it takes into account more than one factor/element of independent variables which affect the variability of dependent variable, the conclusions drawn are more accurate. The conclusions are more realistic and nearer to the real life situation.

Disadvantages of Multivariate Analysis

(i) It requires rather complex computations to arrive at a satisfactory conclusion. (ii) Due to the above fact that a large number of observations for large number of variables need to be collected and tabulated, it is a rather time consuming process. (iii) Obviously due to (i) and (ii) above, at times multivariate analysis proves to be an expensive proposition, in terms of cost. (iv) Specialized trained staff is required to process and analyze the complex data utilizing multivariate techniques.

Applications of Multivariate Analysis

These techniques are successfully employed in the following areas: (a) Econometrics or decision-making in economics: Such as impact of inflation, money circulation, lowering of tariffs, etc., on price rise. (b) Sociological decision-making: Such as divorce rates,and their cause-effect relationships, with marriage, social demographics and income levels. (c) Agrarian predictions: Such as impacts of rain,fertilizers and mechanization on agricultural yields per acre. (d) Drug testing: Such as impact of new drugs on the main disease and other side effects. (e) State fiscal decision-making: Such as impact of tax structure, duties, penalty rates, etc., on the government revenues. (f) Industrial decision-making: Such as plant location which depends on infrastructure, availability of raw materials, distribution channel, etc.

Factor Analysis
It is one of the more popular "analysis of interdependence"techniques. In studies of interdependence, all the variables are on an equal footing, and the analysis is concerned with the whole set of relationships among the variables that characterize the objects.Unlike regression or discriminant analysis, factor analysis would focus on the whole set of interrelationships displayed by the variables. At the conceptual level, a factor is a qualitative dimension of the data that attempts to depict the "way in which entities differ, much as the length of an object or the flavour of a product defines a qualitative dimension on which objects may or may not differ." Factor analysis is a procedure that takes a large number of variables or objects and searches out factors in common which account for their inter-correlation. For example, we can attribute the high association between grades in computer courses to the factor of intelligence or the association between certain attributes of coffee and factor of acidity.

Application of Factor Analysis in Marketing Research

There are a number of applications in marketing research.These includes data reduction, structure identification and scaling.

Data Reduction: Factor analysis is used for reducing a mass of data to a manageable level. For example, the marketing researcher who has collected data on 40 attributes of a brand. This analysis and understanding of this data may be aided by reducing 30 attributes and having only 10 attributes.Factor analysis has been employed to purify original sets of scale items by isolating those items that do not reflect a common core and also to name the dimensions captured by measure. It is also used in the study of lifestyles and psychographic research problems in which it is used to develop consumer profiles that reflect peoples attitudes,activities, interests, opinions, perceptions and preferences,so as to better predict their consumption and purchase behaviour. One of the major impetuses of lifestyles and psychographic research has been the generally inadequate nature of market segments defined using traditional demographic measures. Structure Identification: Factor analysis may be used to understand the basic structure underlying a set of measures. For example, the above 30 attributes may be reduced to ten factors identified by the researcher. Scaling: A marketing researcher may develop a scale on a research topic. One problem that may occur in developing any scale is in weighing the variables being combined to form the scale.

Limitations of Factor Analysis

(1) The factor analysis is not strongly supported by statistical methods. (2) The different results can be obtained by using loading factors in different manners. (3) A large number of attributes are required to study factor analysis. (4) In some cases, complex factors are used which respondents find difficult to understand.

Cluster Analysis
Cluster analysis is used by the researcher to place variables or objects into subgroups or clusters. These clusters are not defined by the researchers but are formed by the cluster analysis procedure itself. The following factors are considered for cluster analysis: 1. They form subgroupings and assign variables or objects to these groups. 2. They take as input a matrix of association between variables or objects. 3. They assume that natural clusters exist within the data.

Example of Cluster Analysis

A two-dimensional perceptual Figure 8.5 has been drawn on the basis of data relating to 15 individuals A to O, on the basis of (i) Number of vacation days (ii) Expenditure on vacations during a given year.

Discriminant Analysis
Discriminant analysis is a technique that is appropriate with a nominal dependent variable and interval independent variables. Nominal dependent variables are very common in marketing. For example, a satisfied customer, an unsatisfied customer, service seeker customer, non-service seeker customer. This is the reason why discriminant analysis has received extensive application in marketing research.The basic idea is to find out the linear combination of the independent variables that make the mean scores across categories of the dependent variable on this linear combination maximally different. This linear combination is called the discriminant function. It can be represented in the following way: DF = V1 X1 + V2 X2 + ... Vm Xm The criterion used to decide when group means are maximally different is the familiar ANOVA (F-Test) for the differences among means thus, the Vs are derived such that FSS between SS within is maximised Confusion matrix for customer satisfaction.
Actual category Satisfied Unsatisfied Satisfied 400 30 Unsatisfied 20 50

The output of discriminant analysis includes the values of the Vs plus what is called a confusion matrix. This matrix compares the category of the dependent variable that the discriminant functions predict a subject will be in with the category involved in it. The example of pain relievers explains the complete analysis of discriminants. Discriminant Analysis in Marketing: This analysis has been used quite extensively in marketing. The following aspects can be studied by this type of analysis: 1. Identification of new buyer group. 2. Consumer behaviour toward new products or brands. 3. Brand loyalty study. 4. Relationship between variables. 5. Checklist of properties of new products. Discriminant analysis technique can be used by the marketing manager who wants to have an idea about the

market in terms of certain profile of pain relievers from a data about particular types of pain relievers. The method derives the relative weights to the different types of pain relievers needs. Thus, the discriminant functions show the common patterns of pain relievers for various types of pain relievers. In this example, the market is initially segmented on the basis of product types and then we attempt to discriminate each of the products on the basis of the specific combination of pain relievers needs that they may be serving.

Conjoint Analysis
Conjoint analysis is concerned with the measurement a combination of two or more attributes that are important in marketing decisions from customers point of view. In this analysis, marketing research suggests a new combination on the basis of existing attributes. For example: A bank would like to know which is the most desirable combination of attributes to frequent borrowers rate of interest, instalment amount, date of payment of instalment. Conjoint analysis is useful for products which can be offered in different combinations for which the consumer has to make a trade-off. As such it is more useful for products which require a longer buying process for consumers. The attributes for which the different levels are to be tested might not be the ones that the consumer considers important while making a choice. It is not always right to add the utility scores across respondents unless the researcher is sure that they belong to a homogenous group.

Numerical Example
Consider a situation in which a manufacturer different colours of paints is interested in measuring consumers trade offs among the following attributes: *Brand *Price Alpha 50 per litre Gamma 60 per lire Beta 70 per litre *Life *Colours 3 Years Pink 4 Years Green 5 Years Potential replacement paint purchasers are recruited and presented with the 18 alternatives. Each respondent is asked to rate each of the alternatives on a 0-10 scale, where 0 indicates absolutely no interest in purchasing and 10 indicates extremely high interest in purchasing. Illustratively, Figure 8.13 shows the 18 evaluations made by one respondent. Although a total of 3x3x3x2 = 54 combinations of attribute levels could be made up, the respondent needs to evaluate only 18 of these. However, this specific set of 18 should be selected in a particular way, as will be noted later. Figure 8.14 shows graphs of the implied values for each of the attribute levels; these can be obtained from an ordinary multiple regression program using dummy variable coding. All one needs to do to estimate the respondents original evaluations is to add each separate value (The regressions intercept term may be added in later if there is interest in estimating the absolute level of purchase interest.) For example, to obtain the respondents estimated evaluation of card 1, one reads off the part worths: Value of alpha = 2.4 Value of 3 years = 0 Value for Rs 50 per litre (a Rs 20 cost saving) = 1.6 Value for green colour = 1.2 = 55.2 In this instance we obtain an almost perfect prediction of the persons overall response to card 1. Similarly, we can find the estimated total evaluations for the other 17 options and compare them with the respondent's original evaluations. The regression technique guarantees that the (squared) prediction error between estimated and actual response will be minimized.The information also permits the researcher to find estimated evaluations for all combinations including the 5418=36 options never shown to the respondent. Moreover,all respondents separate part-worth functions as illustrated for person can be compared in order to see if various types of respondents (e.g. high versus low income respondents)differ in their separate attribute evaluations.In short, while the respondent evaluates complete bundles of attributes, the technique solves for a set of part-worthsone for each attribute level that are computed from the overall trade offs. These part-worths can then be combined in various ways to estimate the evaluation that a respondent would give to any combination of interest. It is this high leverage between the options that are actually evaluated and those that can be evaluated (after the analysis) that makes conjoint analysis a useful tool.

Reliability and Validity Checks

In carrying out a conjoint analysis it is useful to include the following ancillary analysis: (1) Test-retest reliability,

(2) A comparison of actual utilities with those of random respondents and (3) An internal validity check on modelbased utilities. The test-retest reliability can be conducted by including a few replicate judgements at a later stage in the interview. The purpose here is to see if the judgements are highly correlated, on a test-retest basis, to justify the analysis of the respondents d