Introduction-Classification

of multivariate techniques
Multivariate Analysis
• Many statistical techniques focus on just one or two variables
• Multivariate analysis (MVA) techniques allow more than two variables
to be analysed at once
• Multivariate analysis is a term which is used for algorithms that have
the ability to analyze multiple variables.
EXAMPLE
• Consider a researcher who is trying to understand the factors which
influence the use of self-service banking.
• After conducting an exhaustive review of literature, the researchers
narrowed down technology acceptance model in order to study the
factors which influence self-service banking. Therefore, using this
model he defines that he wants to study effect of technology
discomfort, perceived risk, perceived ease of use and perceived
usefulness on the adoption of self-service banking by a consumer.
• The number of independent and dependent variables being studied
by the researcher is more than two.
Classification of Multivariate Techniques

• Selection of the appropriate multivariate technique depends upon-

• a) Are the variables divided into independent and dependent classification?
• b) If Yes, how many variables are treated as dependents in a single analysis?
• c) How are the variables, both dependent and independent measured?
• Multivariate analysis technique can be classified into two broad categories viz.,
This classification depends upon the question: are the involved variables
dependent on each other or not?
• If the answer is yes: We have Dependence methods.
If the answer is no: We have Interdependence methods.
INTERDEPENDENCE AND
DEPENDENCE
• Interdependence, refers to a fundamental where we can say that the variables
influence the amount of variance in each other to a varying extent.
• For example in certain cases perceived ease of use influences perceived
usefulness and vice versa. Therefore there is a mutual interaction between
these two variables and this is called as interdependence.
• Dependence, refers to a fundamental way we can say that the variables can be
categorised into dependent and independent variables and the study tries to
find the relationship or the influence of independent variables on dependent
variable.
• For example a simple regression analysis to find the effect of perceived
usefulness, perceived ease of use, perceived risk and technology discomfort on
the adoption of self-service banking is an dependence analysis.
• The techniques which try and find out interdependence are called as
interdependence techniques. These kind of techniques are used in order to
provide some sort of structure to the dataset.
• For example, the factor analysis and cluster analysis are the most common
interdependence techniques which are applied on metric data.
• The techniques which try and find out the effect of independent variables on
dependent variable are referred to as dependence techniques.
• Dependence techniques further can be classified on basis of the number of
dependent variables. If there is only one dependent variable and metric data
then multiple regression analysis and algorithms based on regression analysis
can be used.
• If several dependent variables are to be analyzed and researcher can move
towards Canonical correlational analysis or multivariate analysis of variance
(MANOVA). If a researcher wants to study multiple relationships of dependent
and independent variables then techniques like structural equation modelling
can be used
Multiple Regression Analysis
• Multiple regression is a statistical technique that can be used to
analyze the relationship between a single dependent variable and
several independent variables.
• The objective of multiple regression analysis is to use the independent
variables whose values are known to predict the value of the single
dependent value. Each predictor value is weighed, the weights
denoting their relative contribution to the overall prediction.

• Here Y is the dependent variable, and X1,…,Xn are the n independent

variables. In calculating the weights, a, b1,…,bn, regression analysis
ensures maximal prediction of the dependent variable from the set of
independent variables. This is usually done by least squares
estimation.
Uses of Multiple Regression
Analysis
• It is the most widely used multivariate technique and is
often used in forecasting.
• Multiple regression has two or more independent
variables instead of one. For example, instead of
showing only the relationship between sales and
advertising, it can show other variables, such as price,
the day of the week or changes to the GDP. Multiple
regression can show you which of these variables, or a
combination of variables, is most closely tied to
increases in sales.
Multiple Discriminant Analysis
And Logistic Regression
• In order to study the effect of multiple independent variables (which are metric
in nature) on one dependent variable (which is categorical in nature) then the
appropriate technique would be multiple discriminant analysis. In this scenario
multiple regression would not work as it assumes all data to be in metric scale.
• Therefore, when the total sample can be divided into groups or classes and the
primary objective is to understand the group differences based on multiple
independent variables than the technique used is discriminant analysis.
• For example, if researcher wants to study the difference in the perception of
perceived ease of use, perceived usefulness, technology discomfort and
perceived risk across users and non-users of self-service banking, then
discriminant analysis would be an appropriate technique to be used.
• Discriminant function analysis is used to classify observations or
people into different groups. A good example of this is classifying
potential customers based on how much they are likely to spend.
• Suppose you want to identify who is more likely to buy from you
within the next year after subscribing to your weekly newsletter.
Using discriminant function analysis, you can apply several variables
to your current customers, such as their gender, age and income
level, all compared to the amount they spent in the past year.
• This analysis should give you different combinations of variables that
make one person more likely to become a major customer than
another. Then, when you get new subscribers who are in the "big
spender" category, you can offer them different incentives than
someone who is in the "little spender" category.
Logistic Regression
• Logistic regression-based algorithms and models are used to predict
relationships amongst multiple independent variables and dependent
variable which might be nonmetric. It is a nonparametric option to
multiple regression.
• It is only used when you are looking for a binary
outcome, like "yes or no" or "Brand A or Brand B."
Canonical Correlation Analysis
• A researcher might be faced with a situation where he desires to find effect of
multiple independent variables on multiple dependent variables, where both are
measured on a metric scale.
• In such circumstances a multivariate analysis technique referred to as Canonical
Correlation analysis can be used by the researcher. The principle behind this
particular algorithm is to develop linear combination between the dependent and
independent variables so as to maximise the correlation.
• For example, you might want to know how customer satisfaction
and loyalty are influenced by product quality and service. CCA can
help you answer questions like: Which aspects of quality and
service are most related to satisfaction and loyalty? How strong are
these relationships? How can you use this information to improve
your business decisions?
MANOVA
• Multivariate analysis of variance, MANOVA, is a commonly used
multivariate technique. MANOVA assesses the relationship between two
or more dependent variables and classificatory variables or factors. It is
similar to ANOVA but with the added ability to handle several dependent
variables simultaneously. It uses special matrices to test for differences
among groups.
• The uniqueness of the algorithm is that it is used to state the relationship
between those independent variables which might be categorical in
nature and multiple dependent variables which are on metric.
• The F ratio, generalized to a ratio of the within-group variance and total-
group variance matrices, tests for equality among treatment groups.
• If, for example, you want to change how you deliver services to
customers, you can compare the new service to the old service in
how long the work took to complete, the levels of customer
satisfaction and the effects on employee morale.
• It helps identify consumer behaviour patterns by examining variables
such as demographics, purchase history and preferences. It aids in
segmenting markets, developing targeted marketing strategies and
predicting consumer trends.
Conjoint Analysis:
• It is one of the emerging dependence multivariate analysis techniques. This is a technique which is
most commonly used in the discipline of marketing as it has its applications in evaluation of objects
like new products, new services or new marketing mix is developed by the organization. It is form
of statistical analysis that firms use in market research to understand how
customers value different components or features of their products or services.
• It is typically conducted via a specialized survey that asks consumers to rank the
importance of the specific features in question. This technique allows the researcher to
find the relative importance of various attributes being studied. It is a technique which makes
subsets of the various levels of independent variable being studied by the researcher and gives an
evaluation in terms of which one of those combinations is best accepted by the customers. This is a
technique which has highest applications and development of proposed marketing mix.
• Conjoint Analysis, also known as trade-off analysis, is useful for identifying how
people like or dislike different attributes of a product or service. You can use this
analysis to find the ideal combination of attributes, such as features, benefits
and colors.
Principal Component And
Common Factor Analysis
• Principal component analysis, or PCA, is a dimensionality reduction method
that is often used to reduce the dimensionality of large data sets, by
transforming a large set of variables into a smaller one that still contains
most of the information in the large set.
• The technique basically helps to extract a common underlying factor on basis
of interdependence or commonality of variance among the variables with
minimal loss of information. It is important to note that whenever any sort of
data condensation technique is applied there is a loss of sensitivity of the
data.
• It is up to the researcher to determine what is more important for this study
i.e. sensitivity of the data or an in-depth analysis (which might be
compromised due to large number of variables).
EXAMPLE
• For example, a researcher wants to study what are the various components of a
print advertisement. Therefore, he collects data regarding various components
present in a print advertisement i.e. brand-name, trademark, copyright, model,
model details, backdrop, product, adjectives used etc. he in total has 58 such
components for which he has collected data for more than 1000 advertisements.
Therefore. researcher lands up with 58,000 data points. Analysis of data across 58
that components in detail is very difficult.
• Therefore, for the ease of data analysis the researcher can reduce the 58
components on basis of factor analysis. Factor analysis on basis of Covarinace will
cluster the components into Factors. For the present for example two factors were
generated for 58 components i.e. information cues and attractiveness use. This
made an in-depth analysis as well as conversion of data into information by the
researcher easier.
Multidimensional Scaling
• Multidimensional scaling is a visual representation of distances or
dissimilarities between sets of objects
• For example, given a matrix of perceived similarities between various
brands of air fresheners, MDS plots the brands on a map such that
those brands that are perceived to be very similar to each other are
placed near each other on the map, and those brands that are
perceived to be very different from each other are placed far away
from each other on the map.
Structural Equation Modelling And
Confirmatory Factor Analysis
• Confirmatory factor analysis, is a variation of factor analysis. In circumstances where the
structure of the covariance, among the variables being studied, is not known to the
researcher the researcher prefers to use common factor analysis it is also referred to as
exploratory factor analysis. In this technique the researcher tries to explore the plausible
structures, in the variables, which can be developed and then accepts the best one. However,
there are certain circumstances where the researcher, based on review of literature, already
knows the structure of covariance, among the variables being studied.
• In this particular case applying an exploratory factor analysis might result in results which are
counter-productive to a predetermined structure. In these situations, the researcher is
advised to use confirmatory factor analysis, where this starting point is the structure of
covariance, as defined by the researcher. Confirmatory factor analysis, is a model-based
assessment of the proposed options. Structural equation modelling as a technique uses
confirmatory factor analysis as an data preparation and data editing step. Only when a model
has converged and passed the confirmatory factor analysis, is it ready to apply the technique
of structural equation modelling.
CFA
• Confirmatory factor analysis (CFA) is a multivariate statistical procedure
that is used to test how well the measured variables represent the
number of constructs. Confirmatory factor analysis (CFA)
and exploratory factor analysis (EFA) are similar techniques, but in
exploratory factor analysis (EFA), data is simply explored and provides
information about the numbers of factors required to represent the data.
• In exploratory factor analysis, all measured variables are related to every
latent variable. But in confirmatory factor analysis (CFA), researchers can
specify the number of factors required in the data and which measured
variable is related to which latent variable. Confirmatory factor analysis
(CFA) is a tool that is used to confirm or reject the measurement theory.
SEM
• Structural equation modelling as a technique allows development of
paths/relationships for each set of dependent variables. It is one of
the best techniques which allows a simultaneous assessment of
multiple regression equations at the same time.
• It is important for the readers to know that a model in which the
paths re defined in terms of covariance is referred to as confirmatory
factor analysis. While the model in which the paths are defined in
terms of regression is referred to as measurement model and the
technique is structural equation modelling
Structural Equation Modeling (SEM)

Model
Model Specification
Specification

Estimation
Estimation

Evaluation
Evaluation of
of Fit
Fit

Respecification
Respecification of
of the
the Model
Model

Interpretation
Interpretation and
and Communication
Communication
19-24
Structural Equation Modeling (SEM)

19-25
Process of Conducting
Multivariate Analysis
Objectives of MVA
• 1) Data reduction or structural simplification: This helps data to get simplified as possible
without sacrificing valuable information. This will make interpretation easier.
• (2) Sorting and grouping: When we have multiple variables, Groups of “similar” objects or
variables are created, based upon measured characteristics.
• (3) Investigation of dependence among variables: The nature of the relationships among
variables is of interest. Are all the variables mutually independent or are one or more
variables dependent on the others?
• (4) Prediction Relationships between variables: must be determined for the purpose of
predicting the values of one or more variables based on observations on the other
variables.
• (5) Hypothesis construction and testing. Specific statistical hypotheses, formulated in
terms of the parameters of multivariate populations, are tested. This may be done to
validate assumptions or to reinforce prior convictions.