What is

Multivariate Analysis?
An introduction to the principles and common
models used in multivariate data analysis
 What is Multivariate Analysis?
Imagine out of the five senses you only had Essentially, multivariate analysis is a tool to
sight. From your perspective you could see find patterns and relationships between
Introduction the world but you would not be able to hear several variables simultaneously. It lets us
the sounds around you, smell, taste or feel predict the effect a change in one variable will
things. Your understanding of the world would have on other variables.
The world is multivariate. Your analysis be more limited.
should be as well. In the event of danger, provided you were
Again using the example of our senses, while
taste and smell are two separate measures,
Most of the problems in the world are multivariate in nature - meaning that there are pointed in the direction of the danger, you they are not independent of each other. For
many variables that contribute to them. We cannot simply use the season of the would have some chance of avoiding it. But if example, if it smells bad, it often tastes bad.
year to predict the weather, as many other variables are part of the relationship to it was behind you, you would never hear the This gives multivariate analysis a decisive
weather. The same holds true with other scientific, economic, and consumer danger, or know that it was coming. With the advantage over other forms of analysis.
preference relationships, such as product quality, which is influenced by numerous combination of sight and sound it is easier to
avoid danger. Even with two of the five Multivariate analysis is also highly graphical in
process variables, or consumer preference for a product which can be based on
senses, your view of the world is still limited. its approach. This allows an analyst to
traits such as color, taste and price.
examine the inner or hidden structure of large
This short guide gives you an introduction to the principles of multivariate analysis, Most of us use all of our senses to understand data sets, and to visually identify the factors
some broad applications for this technology, how it differs from classical (univariate) the world around us i.e. not just one which influence the results. The expression
statistics and an overview of common multivariate models. “measurement” but the combination of several ‘a picture is worth a thousand words’ is
senses working together. This is like particularly relevant when trying to interpret
multivariate analysis. In multivariate analysis large, complex data sets!
we use the information from many sources
simultaneously to get a better picture of our
surroundings.
 How are multivariate
methods used?
Multivariate analysis is used widely in many relationships. From there a model could be
industries, from raw material analysis and developed to help select the right car for the
The use of multivariate analysis drug discovery in the pharmaceutical industry, right conditions. Wherever many
early event detection and gasoline blending in measurements on many variables are
refineries, right through to predicting future performed, this data is ideal for multivariate
Multivariate methods are typically used for: market trends in business intelligence analysis.
applications. It can be used for measuring
Pattern recognition by providing an understanding of the important patterns For most real-world problems we need to use
data sets with many input variables or for
and underlying relationships in data multivariate analysis to model the relationship
investigating the trends in time series data,
all of which provide a better understanding of to the response. We need tools to mine our
Getting deeper insights into data allowing you to model and visualise complex
a given issue and often result in resource and data, and understand the relationships in
data for more insightful analysis
time savings. them, whether for pattern recognition, or to
Find relationships between variables in data to predict likely outcomes develop models that can be used to predict
Using one more example, if we consider the values for new samples.
rate of fuel used by a car (based on the time it
takes for the indicator to go from full to empty), Multivariate analysis (MVA) has wide
this is not just a simple function of driving the application to data including instrumental data,
car. It is a function of the size of the engine, medical diagnostics, census data, economic
whether the car was driven on flat or data, marketing data, or even a sports team’s
mountainous roads, the load in the car, performance. MVA gives us a means to find
whether the car is in a good state of repair, the relationships in the data, and provides
the grade of gasoline used etc. To fully tools to visualize the relationships between
understand this problem, a number of cars samples and variables. It can be used for both
would have to be tested under varying qualitative and quantitative analysis.
conditions and analysed using multivariate
methods to reveal the patterns and
 Multivariate versus
univariate analysis
Most people have heard of the mean, median, Multivariate analysis provides a more accurate
standard deviation, normal distribution etc. depiction of the behavior of data that are
The Multivariate difference These are univariate - or classical - statistics. highly correlated, and can indicate when there
Univariate statistics can be useful, but are are potential problems in a system or process.
limited by only looking at one variable at a While this might not seem to be a big problem,
Get the full story from your data time. In reality, there’s often more than one in many situations important decisions and
Follow the red-dots: Looking at the variable involved, so univariate analysis can products we use every day are often based on
variables individually there are no lead to the wrong conclusions. It is often univariate analysis that does not necessarily
apparent issue but combining them necessary to sample, observe, study or tell the full story or might not even be
in a multivariate view immediately measure more than one variable accurate.
reveals an issue
simultaneously to understand a process or
any set of samples with numerous This example also illustrates how important
measurements. visualization of data is. Making various plots of
Univariate view our data helps us to see trends and
An illustration of why this is important is relationships that are not evident when looking
Only with multivariate analysis can the fault be detected shown in the figure to the left, where the at lists or tables of numbers.
The univariate limits are too wide to detect a univariate plots of two variables, pH and
multivariate fault temperature are shown along with a
The two variables under consideration are not multivariate plot of these two variables.
independent Because there is a correlation between these
The “sweet spot” is defined by the ellipse univariate measurements, the univariate plots
You can not see the outlier without looking at both do not capture the fact that there is a sample
variables at the same time for which there is a problem in the relationship
Univariate view between the pH and temperature values,
shown by the red dot.
 Common multivariate methods
There are numerous tools used in multivariate One thing to keep in mind is that modeling is
analysis, from descriptive statistics to an iterative process. As we work though our
Multivariate methods exploratory data analysis, and onwards to data, we may need to remove outliers, focus
quantitative regression models. We often use on just some of the measurements, and make
descriptive statistics, principal component other adjustments to optimize our analysis and
Multivariate method overview analysis, and clustering in our initial give us a picture that is descriptive of our
explorations of our data. system.

Descriptive Regression & Classification To make quantitative models, regression Depending on the objective of data analysis,
methods Predictive methods methods methods such as partial least-squares multivariate data can be used to understand
regression (PLSR) are used. Regression is and model numerous outcomes. An overview
Principal Component Multiple Linear SIMCA (PCA, PLSR) used to develop a model using known of the different model types is in the table on
Analysis (PCA) Regression (MLR) samples and responses that can provide a the left.
Support Vector Machine predictive model. To identify or classify
Basic Statistics Principal Component (SVM) samples into groups with similar To find out how multivariate analysis can be
Regression (PCR) characteristics, there are various classification used in your industry, please visit our website
Clustering Linear Discriminant models such as Soft Independent Modeling by or community for more information.
Partial Least Squares Analysis (LDA) Class Analogy (SIMCA) or discriminant
Regression (PLSR) analysis.
Partial Least Squares -
Discriminant Analysis
(PLS-DA)
 Examples of multivariate
analysis in action

Free 30-day trial

Try the power of multivariate analysis

Using multivariate analysis Nidar is HarvestLab from ZEISS and John
Unscrambler is the most easy to use all-in-one tool with powerful multivariate
able to resolve product quality Deere uses multivariate analysis for
analysis, interactive graphics and unique capabilities for spectroscopy and
chemometrics. It is the preferred tool for 25000 data analysts, researchers
problems and optimise their on-the-go measuring of moisture, dry
and engineers who need to analyze data quickly, easily and accurately. manufacturing processes from matter, protein, starch, fiber, neutral
better process understanding. detergent fiber, acid detergent fiber,
and sugar.
DOWNLOAD FREE TRIAL READ MORE

Multivariate analysis helps SAB Miller

create leading brands and get greater
value from their manufacturing and
market research data.
More information

To find out how multivariate analysis can be used in your industry, please visit our
website or community for more information. We also offer training in multivariate
analysis suited to different learning preferences, skills and user roles.

VISIT WEBSITE TRAINING OVERVIEW

Camo is the leader in industrial analytics and the preferred partner for industry leaders digitising their value chain. With a world class industrial analytics platform, we help
companies optimise process and product quality through innovative analytical solutions. Founded in 1984 by Norwegian scientists, Camo has applied analytical science to
process and product quality problems for decades. The Camo Unscrambler suite and industrial analytics platform is built on our outstanding analytical pedigree, proprietary
software and commercial success. © 2019 Camo Analytics. All rights reserved.