Linearization

•In mathematics linearization refers to finding the

linear approximation to a function at a given point.
•In the study of dynamical systems, linearization is
a method for assessing the local stability of an
equilibrium point of a system of nonlinear
differential equations or discrete dynamical systems
.
•This method is used in fields such as engineering,
physics, economics, and ecology.
 Linearization of a function are lines —usually
lines that can be used for purposes of calculation.
Linearization is an effective method for
approximating the output of a function at any
based on the value and slope of the function at ,
given that is differentiable on (or ) and that is
close to .
In short, linearization approximates the output of
a function near .
For example, . However, what would be a
good approximation of ?
• To find , we can use the fact that .
The linearization of at is
, because the function
defines the slope of the function at
. Substituting , the linearization at 4 is
. In this case , so is
approximately .
The true value is close to 2.00024998, so the
linearization approximation has a relative error
of less than 1 millionth of a percent.
For any given function , can be
approximated if it is near a known
differentiable point. The most basic requisite is
that, where is the linearization of at
. The point-slope form of an equation forms
an equation of a line, given a point and
slope . The general form of this equation is:
Using the point , becomes .
Because differentiable functions are
locally linear, the best slope to substitute in
would be the slope of the line tangent to at
.
While the concept of local linearity applies the
most to points arbitrarily close to , those
relatively close work relatively well for linear
approximations. The slope should be, most
accurately, the slope of the tangent line
at .
The final equation for the linearization of a
function at is:

For , .The derivative of is ,

and the slope of at is .
 Uses of linearization
Linearization makes it possible to use tools for
studying nonlinear systems to analyze the
behavior of a nonlinear function near a given
point. The linearization of a function is the first
order term of its Taylor expansion around the
point of interest.
For a system defined by the equation:
the linearized system can be written as

where is the point of interest and is the Jacobian

of evaluated at .
 Bivariate Linearization
• Many statistical indices summarize information
about particular phenomena under study.

• For example, the Pearson (r) summarizes the

magnitude of a linear relationship between pairs
of variables.

• However, one major scientific research objective

is to “explain”, “predict”, or “control” phenomena.
• To explain, predict, and control phenomena, we
must not view variables in isolation.

• How variables do or do not relate to other variables

provide us with valuable clues which allow us to:
Explain Predict, and Control

• The examination of these relationships leads to the

formation of networks of variables that provide
the basis for the development of theories about a
phenomenon.
• Linearization are statistical procedures which allow
us to move from description to explanation,
prediction, and possibly control.

• Bivariate linearization analysis is the simplest

linearization procedure.

• The procedure is called simple linearization

because the model:
 explores the predictive or explanatory relationship
for only 2 variables, and
 Examines only linear relationships.
• Simple linearization model focuses on
explaining/ predicting one of the variables on
the basis of information on the other variable.

• The model thus examines changes in one

variable as a function of changes or
differences in values of the other variable.
• The model labels variables according to their
role:
 Dependent Variable (Criterion Variable): The
variable whose variation we want to explain or
predict.
 Independent Variable (Predictor Variable):
Variable used to predict systematic changes in
the dependent/criterion variable.
• To summarize:
 The Bivariate linearization aims to determine
how, and to what extent, the criterion variable
varies as a function of changes in the predictor
variable.
 The criterion variable in a study is easily
identifiable. It is the variable of primary
interest, the one we want to explain or predict.
Several points should be remembered in conceptualizing
simple linear regression:
 Data must be collected on two variables under
investigation.

 The dependent and independent variables should be

quantitative (categorical variables need to recoded to
binary variables).

 The criterion variable is designated as Y and the predictor

variable as X.
 The data analyzed are the same as in correlational analysis.
 The test still examines covariability and variability but with
different assumptions and intentions.
The mathematical equation for the general linear model using population
parameters is
Y= β0+ β1X + ε
Where :
 Y and X represent the scores for individualion the criterion and predictor
variable respectively.
 The parameters β0and β1are constants describing the functional
relationship in the population.
 The value of β1 identifies the change along the Y scale expected for
every unit changed in fixed values of X (represents the slope or degree
of steepness).
 The values of β0identifies an adjustment constant due to scale
differences in measuring X and Y (the intercept or the place on the Y axis
through which the straight line passes. It is the value of Y when X = 0).
 Σ(Epsilon) represents an error component for each individual. The
portion of Y score that cannot be accounted for by its systematic
relationship with values of X.
• For a bivariate linearization data are collected on a
predictor variable (X) and a criterion variable (Y) for
each individual.

• Indices are computed to assess how accurately the Y

scores are predicted by the linear equation.

• The significance test evaluates whether X is useful in

predicting Y.

• The test evaluates the null hypothesis that:

 the population slope = 0, or
 the population corrélation coefficient = 0
There are two sets of assumptions to be
considered for the:
 Fixed-effects model (appropriate for
experimental studies).

 Random-effects model (more appropriate for

non-experimental studies).
Plots are very important in simple linearization:

• They can validate the assumptions of normality,

equality of variance and linearity.

• They also help in detecting unusual observations,

outliers, and other types of relationships.

• Typically scatter plots are used especially plotting

of Y versus X and plots of the Residuals against
Predicted Values of the model.
 Multivariate Linearization
• Multivariate linearization extends bivariate linearization by incorporating
multiple independent variables (predictors).

 Y= β0+ β1X + ε(The simple linear model with 1 predictor)

• When adding a second predictor, the model is expressed as:

 Y= β0+ β1X1+ β2X2+ ε

• When adding more than 2 predictors, the model is expressed as:

 Y= β0+ β1X1+ …+ βpXp+ ε

• ε:In model building, a residual is what is left after the model is fit. It is the
difference between the observed values and the values predicted by the
model.
 Screening
• In order to identify problems, the data needs to
be screened first.

• Look for missing data.

• Look for outliers.

• Examine if the relationships are other than linear.

 Missing Data
• If some variables have a lot of missing values, you may not want
to include those variables in the analysis if possible.

• If only a few cases have missing values, then delete those cases.

• If there are missing values for several cases on different

variables, then retain those cases to avoid data loss.

• If there are not too much missing data, and you are satisfied
that the missing data are random (there is no pattern in terms
of what is missing), then there should be no worry.
 Outliers
• Outliers are a typical points suspiciously different from others
which have a substantial effect on the model’s goodness of fit.

• An outlier is frequently defined as a value that is at least 3

standard deviations above or below the mean.

• Examine outliers carefully to see if they result from errors in

gathering, coding, or entering data. Correct if it is the case.

• Consider interaction with other variables in case of no

apparent reasons for the outlier.
 Goodness of Fit
• An important part of multivariate linearization is to establish how
well the model actually fits the data.

• This includes the detection of possible violations of the required

test assumptions.

• A commonly used measure of goodness of fit of a linear model is

or the coefficient of determination.

• If all observations fall on the regression line = 1.

• In case of absence of linear relationship = 0.