Professional Documents
Culture Documents
The subject on which research is to be done may be conceptualized as a system in its simplest form. A system is generally an individual entity (usually non
isolated) or an assemblage of interacting or interdependent entities (real or abstract) forming an integrated whole and that may have one or more inputs and one
or more recognizable outputs. The input (independent variable) and outputs (dependent variable) associated to the system may be related with known or unknown
(yet to be determined) rules.
In addition to the input and output variables, a system may be subjected to variables of external nature (exogenous) generally called undesired disturbances. The
system itself may be described using system variables (endogenous) or as usually called system parameters.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
A case in point is Newtonian physics, which is highly useful except for the very small, the very fast, and the very massive phenomena of the universe. However,
a fit to empirical data alone is not sufficient for a model to be accepted as valid. Other factors important in evaluating a model include:
One may attempt to quantify the evaluation of a model using a utility function.
Simulation and visualization
For the scientist, a model is also a way in which the human thought processes can be amplified. For instance, models that are rendered in software allow scientists
to leverage computational power to simulate, visualize, manipulate and gain intuition about the entity, phenomenon, or process being represented. Such computer
models are called in silico (an expression used to mean "performed on computer or via computer simulation.").
Other types of scientific model may be in vivo (living models, such as laboratory rats), in vitro (in glassware, such as tissue culture) and a lot more types in other
disciplines such as pilot plants.
Simulation
A simulation is the implementation of a model. A steady state simulation provides information about the system at a specific instant in time usually at
equilibrium. A dynamic simulation provides information over time. A simulation brings a model to life and shows how a particular object or phenomenon will
behave. Such a simulation can be useful for testing, analysis, or training in those cases where real-world systems or concepts can be represented by models.
Simulation approach involves the construction of an artificial environment within which relevant information and data can be generated. This permits an
observation of the dynamic behaviour of a system (or its sub-system) under controlled conditions.
The term simulation in the context of business and social sciences applications refers to the operation of a numerical model that represents the structure of a
dynamic process. Given the values of initial conditions, parameters and exogenous variables, a simulation is run to represent the behaviour of the process over
time.
Simulation approach can also be useful in building models for understanding future conditions.
Visualization is any technique for creating images, diagrams, or animations to communicate a message. Visualization through visual imagery has been an
effective way to communicate both abstract and concrete ideas since the dawn of man. Examples from history include cave paintings, Egyptian hieroglyphs,
Greek geometry, and Leonardo da Vinci'
s revolutionary methods of technical drawing for engineering and scientific purposes.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Mathematical model
A mathematical model is a description of a system using mathematical language. That is, a mathematical model relates the input variables and output variables of
a system representing a real world phenomenon in order to describe it in the form of one or a set of equations. Variables are abstractions of quantities of interest
in the described systems.
The process of developing a mathematical model is termed mathematical modelling.
Mathematical models are used not only in the natural sciences (such as physics, biology, earth science, meteorology) and engineering disciplines, but also in the
social sciences (such as economics, psychology, sociology and political science).
Physicists, engineers, statisticians, operations research analysts and economists use mathematical models most extensively.
Mathematical models can take many forms, including but not limited to dynamical systems, statistical models, differential equations, or game theoretic models.
These and other types of models can overlap, with a given model involving a variety of abstract structures.
The purpose of the mathematical model of the system may be at least either of the following:
i) Description: i.e. to sufficiently (accurately but simply) describe the characteristics of the real world physical phenomenon described as the model
when subjected to input as well as exogenous variables
ii) (Inferential) to explore how the system parameters can be varied/manipulated in order to obtain an optimal performance (response) for a given set
of input variables and a known set of external factors. Studying predictive (interpolating and extrapolating) accuracy is also another purpose of this
category.
In an optimization problem, the function to be optimized is called an objective function. The conditions that should be satisfied during optimization are called
constraints of the system. Objective functions will depend on the perspective of the model'
s user. Depending on the context, an objective function is also known
as an index of performance, as it is some measure of interest to the user. Although there is no limit to the number of objective functions and constraints a model
can have, using or optimizing the model becomes more involved (computationally) as the number increases.
Description and performance optimization are, thus, two important purposes of a mathematical model of a real world physical or abstract system.
Thus, often when engineers are interested to study a system to be controlled or optimized, they use a mathematical model. In analysis, engineers can build a
descriptive model of the system as a hypothesis of how the system could work, or try to estimate how an unforeseeable event could affect the system.
Similarly, in control of a system, engineers can try out different control approaches in simulations.
The mathematical model of a system may be classified in accordance to the type of system or operator. Thus we may have linear or non-linear models;
deterministic or non-deterministic (stochastic/probabilistic) model; static or dynamic model etc.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Nonlinearity, even in fairly simple systems, is often associated with phenomena such as chaos and irreversibility. Although there are exceptions, nonlinear
systems and models tend to be more difficult to study than linear ones. A common approach to nonlinear problems is linearization, but this can be problematic if
one is trying to study aspects such as irreversibility, which are strongly tied to nonlinearity.
Mathematical models can be divided into either analytical models or numerical models.
An analytical model is a mathematical model that has a closed form expression, i.e. the output and input variables can be related in such a way that the two sets
of variables are disassociated from themselves or expressed as a mathematical analytic function.
On the other hand, a numerical model is a mathematical model that uses some sort of numerical time-stepping procedure to obtain the models behavior over time.
Eg recursive methods
A priori information
Mathematical modelling problems are often classified into black box or white box models, according to how much a priori information is available of the system.
A black-box model is a system of which there is no a priori information available. A white-box model (also called glass box or clear box) is a system where all
necessary information is available. Practically all systems are somewhere between the black-box and white-box models, so this concept is useful only as an
intuitive guide for deciding which approach to take.
Usually it is preferable to use as much a priori information as possible to make the model more accurate. Therefore the white-box models are usually considered
easier, because if you have used the information correctly, then the model will behave correctly. Often the a priori information comes in forms of knowing the
type of functions relating different variables.
For example, suppose we want to make a model of how a medicine works in a human system knowing the fact that (i.e. a priori information) usually the amount
of medicine in the blood is an exponentially decaying function. But we are still left with several unknown parameters; how rapidly does the medicine amount
decay, and what is the initial amount of medicine in blood? This example is therefore not a completely white-box model. These parameters have to be estimated
through some means (say experiment) before one can use the model.
In black-box models one tries to estimate both the functional form of relations between variables (input and output) and the numerical parameters (of the system)
in those functions. Using a priori information we could end up, for example, with a set of functions that probably could describe the system adequately. If there is
no a priori information we would try to use functions as general as possible to cover all different models.
An often used approach for black-box models are neural networks which usually do not make assumptions about incoming data. The problem with using a large
set of functions to describe a system is that estimating the parameters becomes increasingly difficult when the amount of parameters (and different types of
functions) increases.
Any model which is not pure white-box contains some parameters that can be used to fit the model to the system it is intended to describe. If the modelling is
done by a neural network, the optimization of parameters is called training. In more conventional modelling through explicitly given mathematical functions,
parameters are determined by curve fitting.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Population Growth. A simple (though approximate) model of population growth is the Malthusian growth model, sometimes called the simple
exponential growth model, is essentially exponential growth based on a constant rate of compound interest. The model is named after the Reverend
Thomas Malthus, who authored An Essay on the Principle of Population, one of the earliest and most influential books on population and exponential
growth.
where
P0 = initial population, r = growth rate, sometimes also called Malthusian parameter, t = time.
Model of a particle in a potential-field. In this model we consider a particle as being a point of mass which describes a trajectory in space which is
modeled by a function giving its coordinates in space as a function of time. The potential field is given by a function V : R3 R and the trajectory is
a solution of the differential equation
Note this model assumes the particle is a point mass, which is certainly known to be false in many cases in which we use this model; for example, as a
model of planetary motion.
Model of rational behavior of a consumer. In this model we assume a consumer faces a choice of n commodities labeled 1,2,...,n each with a market
price p1, p2,..., pn. The consumer is assumed to have a cardinal utility function U (cardinal in the sense that it assigns numerical values to utilities),
depending on the amounts of commodities x1, x2,..., xn consumed. The model further assumes that the consumer has a budget M which is used to
purchase a vector x1, x2,..., xn in such a way as to maximize U(x1, x2,..., xn). The problem of rational behavior in this model then becomes an
optimization problem, that is:
subject to:
This model has been used in general equilibrium theory, particularly to show existence and Pareto efficiency of economic equilibria. However, the
fact that this particular formulation assigns numerical values to levels of satisfaction is the source of criticism (and even ridicule). However, it is not
an essential ingredient of the theory and again this is an idealization.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Descriptive statistics: summarizes the population data by describing what was observed in the sample numerically or graphically. Numerical
descriptors include mean, variance and standard deviation for continuous data types (like heights or weights), while frequency and percentage are
more useful in terms of describing categorical data (like race). This method is useful in research, when communicating the results of experiments.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Inferential statistics: uses patterns in the sample data to draw inferences (predictions) about the population represented, accounting for randomness.
These inferences may take the form of:
modeling relationships within the data (regression), extrapolation, interpolation, or other modeling techniques like ANOVA (Analysis of
variance), time series, and data mining. This means patterns in the data may be modeled in a way that accounts for randomness and
uncertainty in the observations, and are then used to draw inferences about the process (system) or population being studied;
Inference is a vital element of scientific advance, since it provides a prediction (based on data) for where a theory logically leads. To further prove the guiding
theory, these predictions are tested (i.e hypothesis testing) as well, as part of the scientific method. If the inference holds true, then the descriptive statistics of the
new data increase the soundness of that hypothesis.
The concept of correlation is particularly noteworthy for the potential confusion it can cause. Statistical analysis of a data set often reveals that two variables
(properties) of the population under consideration tend to vary together, as if they were connected. For example, a study of annual income that also looks at age
of death might find that poor people tend to have shorter lives than affluent people. The two variables are said to be correlated; however, they may or may not be
the cause of one another. The correlation phenomena could be caused by a third, previously unconsidered phenomenon, called a lurking variable or confounding
variable. For this reason, there is no way to immediately infer the existence of a causal relationship between the two variables. Correlation does not imply
causation.
Experimental and observational studies
A common goal for a statistical research project is to investigate causality, and in particular to draw a conclusion on the effect of changes in the values of
predictors or independent variables on dependent variables or response.
There are two major types of causal statistical studies: experimental studies and observational studies. In both types of studies, the effect of differences of an
independent variable (or variables) on the behavior of the dependent variable are observed. The difference between the two types lies in how the study is actually
conducted. Each can be very effective.
An experimental study involves taking measurements of the system under study, manipulating the system, and then taking additional measurements using the
same procedure to determine if the manipulation has modified the values of the measurements.
In contrast, an observational study does not involve experimental manipulation. Instead, data are gathered and correlations between predictors and response are
investigated.
Experiments
The basic steps of a statistical experiment are:
1.
Planning the research, including finding the number of replicates of the study, using the following information:
Preliminary estimates regarding the size of treatment effects, alternative hypotheses, and the estimated experimental variability.
Consideration of the selection of experimental subjects and the ethics of research is necessary.
Statisticians recommend that experiments compare (at least) one new treatment with a standard treatment or control, to allow an unbiased estimate of
the difference in treatment effects.
2.
Design of experiments, using blocking to reduce the influence of confounding variables, and randomized assignment of treatments to subjects to
allow unbiased estimates of treatment effects and experimental error.
At this stage, the experimenters and statisticians write the experimental protocol that shall guide the performance of the experiment and that specifies
the primary analysis of the experimental data.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Performing (execute) the experiment and analyzing the data following the experimental protocol.
4.
Further examining the data set in secondary analyses, to suggest new hypotheses for future study.
5.
Experiments on human behavior have special concerns. The famous Hawthorne study examined changes to the working environment at the Hawthorne plant of
the Western Electric Company.
The researchers were interested in determining whether increased illumination would increase the productivity of the assembly line workers. The researchers first
measured the productivity in the plant, then modified the illumination in an area of the plant and checked if the changes in illumination affected productivity.
It turned out that productivity indeed improved (under the experimental conditions). However, the study is heavily criticized today for errors in experimental
procedures, specifically for the lack of a control group and blindness. The Hawthorne effect refers to finding that an outcome (in this case, worker productivity)
changed due to observation itself. Those in the Hawthorne study became more productive not because the lighting was changed but because they were being
observed.
Observational study
An example of an observational study is one that explores the correlation between smoking and lung cancer. This type of study typically uses a survey to collect
observations about the area of interest and then performs statistical analysis. In this case, the researchers would collect observations of both smokers and nonsmokers, perhaps through a case-control study, and then look for the number of cases of lung cancer in each group.
Levels of measurement
There are four main levels of measurement used in statistics:
ratio
interval,
ordinal and
nominal,
Type I errors:
where the null hypothesis is falsely rejected giving a "false positive".
That is, it is when we accept the alternative hypothesis when the null hypothesis is in fact correct.
Type II errors:
where the null hypothesis fails to be rejected and an actual difference between populations is missed.
That is, when we reject the alternative hypothesis even if the null hypothesis is wrong.
Error also refers to the extent to which individual observations in a sample differ from a central value, such as the sample or population mean. Many statistical
methods seek to minimize the mean-squared error, and these are called "methods of least squares."
Measurement processes that generate statistical data are also subject to error. Many of these errors are classified as random (noise) or systematic (bias), but other
important types of errors (e.g., blunder, such as when an analyst reports incorrect units) can also be important.
Confidence intervals
Most studies will only sample part of a population and then the result is used to interpret the null hypothesis in the context of the whole population. Any
estimates obtained from the sample only approximate the population value.
Confidence intervals allow statisticians to express how closely the sample estimate matches the true value in the whole population. Often they are expressed as
95% confidence intervals. Formally, a 95% confidence interval of a procedure is a range where, if the sampling an analysis were repeated under the same
conditions, the interval would include the true (population) value 95% of the time. This does not imply that the probability that the true value is in the confidence
interval is 95%. (From the frequentist perspective, such a claim does not even make sense, as the true value is not a random variable. Either the true value is or is
not within the given interval.)
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Statistical computing
The rapid and sustained increases in computing power starting from the second half of the 20th century have had a substantial impact on the practice of statistical
science.
Early statistical models were almost always from the class of linear models. However, nowadays, powerful computers, coupled with suitable numerical
algorithms, caused an increased interest in nonlinear models (such as neural networks) as well as the creation of new types, such as generalized linear models
and multilevel models.
Increased computing power has also led to the growing popularity of computationally intensive methods based on resampling, such as permutation tests and the
bootstrap, while techniques such as Gibbs sampling have made use of Bayesian models more feasible.
The computer revolution has implications for the future of statistics with new emphasis on "experimental" and "empirical" statistics. A large number of both
general and special purpose statistical software are now available such as gretl, which is an example of an open source statistical package.
Misuse of statistics
Misuse of statistics can produce subtle, but serious errors in description and interpretation subtle in the sense that even experienced professionals make such
errors, and serious in the sense that they can lead to devastating decision errors. For instance, social policy, medical practice, and the reliability of structures like
bridges all rely on the proper use of statistics. Even when statistics are correctly applied, the results can be difficult to interpret for those lacking expertise.
The statistical significance of a trend in the data which measures the extent to which a trend could be caused by random variation in the sample may or
may not agree with an intuitive sense of its significance. The set of basic statistical skills (and skepticism) that people need to deal with information in their
everyday lives properly is referred to as statistical literacy.
There is a general perception that statistical knowledge is all-too-frequently intentionally misused by finding ways to interpret only the data that are favorable to
the presenter.
The famous saying, "There are three kinds of lies: lies, damned lies, and statistics". which was popularized in the USA by Samuel Clemens, has come to
represent the general mistrust [and misunderstanding] of statistical science.
Harvard President Lawrence Lowell wrote in 1909 that statistics, "...like veal pies, are good if you know the person that made them, and are sure of the
ingredients."
If various studies appear to contradict one another, then the public may come to distrust such studies. For example, one study may suggest that a given diet or
activity raises blood pressure, while another may suggest that it lowers blood pressure. The discrepancy can arise from subtle variations in experimental design,
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Statistical modelling
A statistical model is a formalization of relationships between variables in the form of mathematical equations. A statistical model describes how one or more
random variables are stochastically related to one or more random variables.
Statistical models include issues such as statistical characterization of numerical data, estimating the probabilistic future behavior of a system based on past
behavior, extrapolation or interpolation of data based on some best-fit, error estimates of observations, or spectral analysis of data or model generated output.
In mathematical terms, a statistical model is frequently thought of as a pair (Y, P) where Y is the set of possible observations and P is the set of possible
probability distributions on Y. It is assumed that there is a distinct element of P which generates the observed data. Statistical inference enables us to make
statements about which element(s) of this set are likely to be the true one.
Most statistical tests can be described in the form of a statistical model. For example, the Student'
s t-test for comparing the means of two groups can be
formulated as seeing if an estimated parameter in the model is different from 0. Another similarity between tests and models is that there are assumptions
involved. Error is assumed to be normally distributed in most models.[1]
More formally a statistical model, P, is a collection of probability distribution or density functions. A parametric model is a collection of distributions, each of
which is indexed by a unique finite-dimensional parameter:
of parameters, which is a subset of d-dimensional Euclidean space.
, where
is a parameter and
A statistical model may be used to describe the set of distributions from which one assumes that a particular data set is sampled. For example, if one assumes that
data arise from a univariate Gaussian distribution, then one has assumed a Gaussian model:
.
Performance of a student class, for instance, can best be modelled using normal distribution.
Arrival of a customer in a bank queue or call making to a telephone network is modelled using Poison (exponentially distributed) distribution.
Other models include Rayleigh distribution, etc.
A non-parametric model is a set of probability distributions with infinite dimensional parameters, and might be written as
A semi-parametric model also has infinite dimensional parameters, but is not dense in the space of distributions. For example, a mixture of Gaussians with one
Gaussian at each data point is dense is the space of distributions. Formally, if d is the dimension of the parameter, and n is the number of samples, if
as
and
as
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Quadratic: y = ax2 + bx + c: Appropriate for fitting data with one minimum or one maximum. To find out if equally spaced data is quadratic, check if the
differences of the successive differences of consecutive y-values are constant. If a > 0, this function is concave up; if a < 0, it is
concave down.
iii.
Cubic: y = ax3 + bx2 + cx + d: Appropriate for fitting data with one minimum and one maximum.
iv.
Quartic: y = ax4 + bx3 + cx2 + dx + e: Convenient for fitting data with two minima and one maximum or two maxima and one minimum. When working
with polynomial models a thing to keep in mind: balance between complexity and precision. For examples, see section Testing
the effectiveness of a model. Also, monitor the long term behavior and check how realistic it is. For example, see sections
Testing the validity and Choosing the right model.
v.
Exponential: y = abx or y = aekx: While a linear function has constant average rate of change (a constant difference of two consecutive y-values), an
exponential function has constant percent (relative) rate of change (a constant quotient of two consecutive y-values). Thus, an
easy test to check if data is linear: if independent variable values are equally spaced, simply check if quotient of consecutive yvalues is the same. If k > 0, then the function is increasing and concave up. If k < 0, then the function is decreasing and
concave up. This model is appropriate if the increase is slow at first but then it speeds up (or, if k < 0 if the decrease is fast at
first but then slows down).
vi.
Logarithmic: y = a + b ln x: If b > 0, then the function is increasing and concave down. If b < 0, then the function is decreasing and concave up. If the data
indicates an increase, this model is appropriate if the increase is fast at first but then it slows down.
vii.
Logistic: y = c/(1+ae-bx) : Increasing for b > 0: In this case, the increase is slow at first, then it speeds up and then it slows down again and approaches the yvalue c for x
.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Power y = axb: If a > 0, y is increasing for b > 0 and decreasing for b < 0. It is called a power model since an increase of x by factor of t causes an increase of
y by the power tb of t (for b > 0). Increasing power function will not increase as rapidly as an increasing exponential function.
Connection with linear model: If y = axb, then ln y = ln a + b ln x. So, if y is a power
function, ln y is a linear function of ln x. Note: If y = abx, then ln y = ln a + x ln b. So, if y is
an exponential function, ln y is a linear function of x.
In conclusion:
ix. Sine y = a sin(bx+c)+d: Appropriate for periodic data. In the formula, a is the amplitude, b/360 the period, c/b the horizontal shift, and d the vertical shift.
Modeling with piecewise defined functions
A piecewise defined function is a function defined by different formulas for different values of the independent variable. Sometimes, it is better to model a given
data using one piecewise defined function than with a single function.
Analytic Methods of Model fitting
In this section, we look into the mathematics behind obtaining regression models and curve fitting.
Least-Squares Criterion
This criterion is the most frequently used curve- fitting criterion. It is based on an idea that in order to fit a set of data points (xi , yi), i = 1, , m onto a curve y =
f(x), we again want the differences between yi and f(xi) to be as small as possible.
Recall that the square of distance of the vector between the observed value yi and the predicted value f(xi) is given by
Thus, a way to make the differences yi - f(xi) small is to minimize the sum
and
We can view the equations above as two linear equations in the unknowns a and b. Solving them for a and b, we obtain
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
An important fact to note is that if the partial derivative equations are used to find the unknown coefficients a and B in y = Beax, y = a ln x + b or y = Bxa without
transforming the original data, we would end up with different models than the models that we get when we transform the data and find a linear regression for the
new data. It is important to keep in mind how different software applications handle the data.
Transforming the original data often simplifies the formulas even for models with less parameters than exponential, power or logarithmic. The next example
illustrates this.
Example 1: Find the formula computing the coefficient a for the quadratic fit y = ax2 and use it to find a transformed quadratic fit for the data below:
Take ln of y = ax2. Obtain then ln y = ln a + 2 ln x. Let us denote ln a by A. Thus, we need to consider ln y = A + 2 ln x and to minimize
The partial derivative is
ri
r for i = 1, , m i.e.
Example 2. Suppose that a point B is somewhere inside a line segment AC. Assume that measuring the distances AB, BC and AC, we obtain that AB = 13, BC =
7 and AC = 19. As 13 + 7 = 20 19, we need to resolve the discrepancy.
We can do that using the Chebyshev approximation criterion. Let x1, x2 and x3 stand for exact lengths of AB, BC and AC respectively. Then r1 = x1 13, r2 = x2
7, and r3 = x1 + x2 19.
Let r stands for the maximum of the absolute values of r1, r2 and r3 (note that we don'
t know which one is r). We would like to minimize r subject to
r - x1 + 13
0, r + x1 - 13
0,
r - x2 + 7
0, r + x2 - 7
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
0, r + x1 + x2 - 19
Using the methods of linear programming, we obtain that r = 1/3, x1 = 12 &2/3 and x2 = 6 & 2/3.
Although in this example linear programming is used, note that in some cases the constraints will not be linear equations and other optimization methods should
be used.
Now let us come back to Measuring the Validity of a Model
In the previously seen examples, two different quadratic power models y = 3.1368x2 and y = 3.1869x2 are obtained for the same set of data. If Chebyshev
criterion is to be used for the same set of data, it would yield yet another model y = 3.17073x2 for the same set of data.
A natural question is: how can we choose the best model?
Before we attempt to answer, let us introduce some notation. If the data (xi, yi), i = 1, . . ., m is to be fit on the curve y = f1(x) obtained using the Chebyshev
criterion, let ci denote the absolute deviations | yi - f(xi)|. If the same data is to be fit on the curve y = f2(x) obtained using the least-squares criterion, let di denote
the absolute deviations |yi - f2(xi)|. Also, let cmax denote the largest of ci and dmax denote the largest of di. Since the least-squares criterion is such that the squares
of the deviations are minimal, we have that
Thus,
Also, since the Chebyshev criterion is such that the maximum of the absolute deviations is minimal, we have that cmax dmax. Thus,
D cmax dmax
This last equation can help us determine which model to use: the least-squares criterion is convenient to use but there is always to concern that the difference
between D and dmax might be too large. Balancing these two conditions might help us decide which model to use.
Let us go back to the example with three different quadratic power models. Computing the deviations for all three models we conclude
1. For the model y = 3.1869x2 obtained using the least-squares fit. The largest absolute deviation is 0.347. The smallest is 0.0976.
2. For the model y = 3.1368x2 obtained using transformed least-squares fit. The largest absolute deviation is 0.495. The smallest is 0.0842. This model was the
easiest to compute but has the largest absolute deviation.
3. For the model y = 3.17073x2 obtained using Chebyshev criterion. The largest absolute deviation is 0.282. The smallest is 0.0659. Computationally, this model
was the hardest to get but has the smallest absolute deviation.
However, considering the sum of squares of deviations instead of the maximum of the absolute deviations, the least-squares fit model is better than the
Chebyshev (.2095 for the least squares versus .2256 for Chebyshev).
The conclusion that we can draw from this example is that there is not a single right answer when trying to decide which model is the best. Thus, in each of the
following cases, our choice of models would be different.
1. If it is more important to minimize the sum of squares of deviations than the maximal absolute deviation, the least-squares criterion should be used.
2. If it is more important that the model is computationally simple than to have small maximum or sum of squares of absolute deviations, the
transformed least-squares criterion should be used.
3. If it is more important to minimize the maximal absolute deviation than the sum of squares of deviation, the Chebyshev criterion should be used.
So, we need to decide which model is the best on case-by-case basis, taking all the specifics into account (what is the purpose of the model, how precise should it
be, how accurate is the data, etc).
Linear Regression
Note: regression is a measure of the relation between the mean value of one variable and corresponding values of other variables.)
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
where
is the average of the y-values of the data points. TSS reflects the variation in the y values about the line y =
3. The regression sum of squares is RSS = TSS ESS.
Thus defined R2 is independent on the units of x and y and is also independent on which of the two variables is labeled as independent and which as dependent
(in other words, data (yi, xi), i = 1, , m, would produce the same value of R2).
If R2 = 0.9 for example, then 90% of the total variation of y-values from the line y =
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
Pathways to Validation
Cases
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012
External Validity
Internal Validity
Convergent and Discriminant Validity
Test Validity
Criterion Validity
Concurrent Validity
Predictive Validity
A definition of reliability may be "Yielding the same or compatible results in different clinical experiments or statistical trials" (the free dictionary). Research
methodology lacking reliability cannot be trusted. Replication studies are a way to test reliability. Consider the center point in the set of concentric circles is the
real world fact
Types of Reliability:
Test-Retest Reliability
Internal Consistency Reliability
Instrument Reliability
Statistical Reliability
Reproducibility
Both validity and reliability are important aspects of the research methodology to get better explanations of the world.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Extractions: by Tezazu Bireda, AAU, Addis Ababa Institute of Technology, ECE Dept May 2012