On Correlation Coefficient. The correlation coefficient indicates the degree of linear dependence of two random variables.



Similar documents
Ilona V. Tregub, ScD., Professor

Mechanics 1: Work, Power and Kinetic Energy

The Binomial Distribution

Skills Needed for Success in Calculus 1

An Introduction to Omega

Mechanics 1: Motion in a Central Force Field

Semipartial (Part) and Partial Correlation

Questions for Review. By buying bonds This period you save s, next period you get s(1+r)

Standardized Coefficients

Coordinate Systems L. M. Kalnins, March 2009

Graphs of Equations. A coordinate system is a way to graphically show the relationship between 2 quantities.

2. TRIGONOMETRIC FUNCTIONS OF GENERAL ANGLES

Questions & Answers Chapter 10 Software Reliability Prediction, Allocation and Demonstration Testing

Spirotechnics! September 7, Amanda Zeringue, Michael Spannuth and Amanda Zeringue Dierential Geometry Project

YARN PROPERTIES MEASUREMENT: AN OPTICAL APPROACH

Financing Terms in the EOQ Model

Saturated and weakly saturated hypergraphs

Model Question Paper Mathematics Class XII

Experiment 6: Centripetal Force

Vector Calculus: Are you ready? Vectors in 2D and 3D Space: Review

UNIT CIRCLE TRIGONOMETRY

AN IMPLEMENTATION OF BINARY AND FLOATING POINT CHROMOSOME REPRESENTATION IN GENETIC ALGORITHM

Supplementary Material for EpiDiff

Figure 2. So it is very likely that the Babylonians attributed 60 units to each side of the hexagon. Its resulting perimeter would then be 360!

Problem Set # 9 Solutions

Symmetric polynomials and partitions Eugene Mukhin

CHAPTER 10 Aggregate Demand I

Physics 235 Chapter 5. Chapter 5 Gravitation

The Detection of Obstacles Using Features by the Horizon View Camera

2 r2 θ = r2 t. (3.59) The equal area law is the statement that the term in parentheses,

The Role of Gravity in Orbital Motion

VISCOSITY OF BIO-DIESEL FUELS

Continuous Compounding and Annualization

ON THE (Q, R) POLICY IN PRODUCTION-INVENTORY SYSTEMS

Comparing Availability of Various Rack Power Redundancy Configurations

Deflection of Electrons by Electric and Magnetic Fields

Nontrivial lower bounds for the least common multiple of some finite sequences of integers

Chris J. Skinner The probability of identification: applying ideas from forensic statistics to disclosure risk assessment

The force between electric charges. Comparing gravity and the interaction between charges. Coulomb s Law. Forces between two charges

Forces & Magnetic Dipoles. r r τ = μ B r

Episode 401: Newton s law of universal gravitation

Do Bonds Span the Fixed Income Markets? Theory and Evidence for Unspanned Stochastic Volatility

Lecture 16: Color and Intensity. and he made him a coat of many colours. Genesis 37:3

Chapter 3 Savings, Present Value and Ricardian Equivalence

STUDENT RESPONSE TO ANNUITY FORMULA DERIVATION

Chapter 4: Matrix Norms

Gauss Law. Physics 231 Lecture 2-1

Trading Volume and Serial Correlation in Stock Returns in Pakistan. Abstract

Comparing Availability of Various Rack Power Redundancy Configurations

Valuation of Floating Rate Bonds 1

Risk Sensitive Portfolio Management With Cox-Ingersoll-Ross Interest Rates: the HJB Equation

est using the formula I = Prt, where I is the interest earned, P is the principal, r is the interest rate, and t is the time in years.

Optimal Capital Structure with Endogenous Bankruptcy:

Contingent capital with repeated interconversion between debt and equity

A Capacitated Commodity Trading Model with Market Power

Open Economies. Chapter 32. A Macroeconomic Theory of the Open Economy. Basic Assumptions of a Macroeconomic Model of an Open Economy

TARGET ESTIMATION IN COLOCATED MIMO RADAR VIA MATRIX COMPLETION. Shunqiao Sun, Athina P. Petropulu and Waheed U. Bajwa

Do Vibrations Make Sound?

College Enrollment, Dropouts and Option Value of Education

Moment and couple. In 3-D, because the determination of the distance can be tedious, a vector approach becomes advantageous. r r

Displacement, Velocity And Acceleration

Experiment MF Magnetic Force

Things to Remember. r Complete all of the sections on the Retirement Benefit Options form that apply to your request.

LATIN SQUARE DESIGN (LS) -With the Latin Square design you are able to control variation in two directions.

Week 3-4: Permutations and Combinations

How To Find The Optimal Stategy For Buying Life Insuance

FI3300 Corporate Finance

Excitation energies for molecules by Time-Dependent. based on Effective Exact Exchange Kohn-Sham potential

Fluids Lecture 15 Notes

1240 ev nm 2.5 ev. (4) r 2 or mv 2 = ke2

Efficient Redundancy Techniques for Latency Reduction in Cloud Systems

Model-based clustering of longitudinal data. McNicholas, Paul D.; Murphy, Thomas Brendan. Canadian Journal of Statistics, 38 (1):

Lesson 8 Ampère s Law and Differential Operators

Cloud Service Reliability: Modeling and Analysis

Concept and Experiences on using a Wiki-based System for Software-related Seminar Papers

Uncertain Version Control in Open Collaborative Editing of Tree-Structured Documents

Investigation of advanced data processing technique in magnetic anomaly detection systems

Chapter 30: Magnetic Fields Due to Currents

Data Center Demand Response: Avoiding the Coincident Peak via Workload Shifting and Local Generation

CLOSE RANGE PHOTOGRAMMETRY WITH CCD CAMERAS AND MATCHING METHODS - APPLIED TO THE FRACTURE SURFACE OF AN IRON BOLT

Experimentation under Uninsurable Idiosyncratic Risk: An Application to Entrepreneurial Survival

Module Availability at Regent s School of Drama, Film and Media Autumn 2016 and Spring 2017 *subject to change*

There is considerable variation in health care utilization and spending. Geographic Variation in Health Care: The Role of Private Markets

NURBS Drawing Week 5, Lecture 10

Supply chain information sharing in a macro prediction market

Summary: Vectors. This theorem is used to find any points (or position vectors) on a given line (direction vector). Two ways RT can be applied:

UNIVERSIDAD DE CANTABRIA TESIS DOCTORAL

Chapter 22. Outside a uniformly charged sphere, the field looks like that of a point charge at the center of the sphere.

Tracking/Fusion and Deghosting with Doppler Frequency from Two Passive Acoustic Sensors

Patent renewals and R&D incentives

4a 4ab b (count number of places from first non-zero digit to

Determining solar characteristics using planetary data

MULTIPLE SOLUTIONS OF THE PRESCRIBED MEAN CURVATURE EQUATION

AMB111F Financial Maths Notes

Approximation Algorithms for Data Management in Networks

The Supply of Loanable Funds: A Comment on the Misconception and Its Implications

Lab #7: Energy Conservation

Reduced Pattern Training Based on Task Decomposition Using Pattern Distributor

Software Engineering and Development

Definitions and terminology

Transcription:

C.Candan EE3/53-METU On Coelation Coefficient The coelation coefficient indicates the degee of linea dependence of two andom vaiables. It is defined as ( )( )} σ σ Popeties: 1. 1. (See appendi fo the poof of this popet.). If then and ae called uncoelated andom vaiables. (Note that two independent vaiables ae guaanteed to be uncoelated; but the evese is not tue in geneal. So thee can be two andom vaiables which ae uncoelated, but dependent.) 3. 1 a + b Hee a and b ae non-andom paametes, i.e. scalas. This elation shows that when 1, then the andom vaiable is a lineal elated to is sufficient to detemine the othe and vice-vesa. If 1, knowing o one though a + b. So knowing one of two andom vaiables is as good as knowing the both of them. (See appendi fo the poof of this popet.) 4. In man applications, we can estimate the coelation coefficient between two andom vaiables b conducting epeiments. In pactice we use the coelation coefficient to pedict the value of not luck to obseve (something of inteest) when we can onl obseve diectl in man applications. If then we ma epect that we can eliabl pedict Lets sa we ae inteested in coefficient between and fom and.. We ae ae closel elated, ; but have onl and we know the coelation. You will lean in some othe couses that we can σ pedict as follows ˆ ( ) +. This is the best linea pediction of in σ the mean squae sense. (You will also hea about mean squae sense at these couses.)

Remembe that we have noted in item 3 the following: If 1, the knowing o is as good as knowing both of them. Theefoe we epect to have zeo pediction eo in this case. Fo othe immediatel clea. values, the value of the pediction eo is not The gaph given below shows the mean squae eo (appoimation eo) fo a geneal value of. As epected, the mean squae eo is zeo, when 1 and as the magnitude of coelation coefficient deceases, the eo inceases. The eo eaches its maimum when two andom vaiables ae uncoelated. Mean Squae Eo of the pediction σ (1- ) σ - -1.5-1 -.5.5 1 1.5 linspace(-,,1).*ect(linspace(-,,1),-1,1); plot(linspace(-,,1),*(1-.^).*ect(linspace(-,,1),-1,1)); gid on; label('_{}'); title('mean Squae Eo of the pediction \sigma_^(1-^_{})'); ais([-.5]) [Fo moe info Haes, Statistical Digital Signal Pocessing and Modeling, p. 7] Eamples with Scatte Plots: Lets sa that we want to lean be given as ; but we can onl obseve. Let the obsevation model

Hee + n n is the effect of noise. (You can assume zeo mean noise without an ham o loss of genealit.) The coelation coefficient between σ. σ + σ and n can be calculated as Lets stat with the case of little noise When noise is little, i.e. vaiance of noise is small; is close to 1. 1 Scatte plot fo (,) pais + n; n is the andom noise on,.9999 8 6.999 4-1 3 4 5 6 7 8 9 1 The plot given above is called scatte plot and it is dawn b andoml geneating and n and calculating though + n. If thee wee no noise, ; but unfotunatel thee is noise in an obsevation. The scatte plot is dawn b putting coss maks () whee the andoml geneated calculated ae on the (,) plane. Thee ae 1 cosses in the given figue. So we conclude fom this figue, when thee is little noise, knowing knowing, which is wondeful. and can be as good as

Below we have some othe scatte plots. The noise level is highe in these plots, theefoe thee is a bigge spead aound the line. 1 Scatte plot fo (,) pais + n; n is the andom noise on,.99 1 8.99 6 4-1 3 4 5 6 7 8 9 1 14 Scatte plot fo (,) pais + n; n is the andom noise on,.9 1 1 8.9 6 4 - -4 1 3 4 5 6 7 8 9 1

Scatte plot fo (,) pais + n; n is the andom noise on,.75 15.75 1 5-5 -1 1 3 4 5 6 7 8 9 1 5 Scatte plot fo (,) pais + n; n is the andom noise on,.5 15.5 1 5-5 -1 1 3 4 5 6 7 8 9 1

4 Scatte plot fo (,) pais + n; n is the andom noise on,.5 3.5 1-1 - -3-4 1 3 4 5 6 7 8 9 1 8 Scatte plot fo (,) pais + n; n is the andom noise on,.1 6 4.1 - -4-6 -8-1 1 3 4 5 6 7 8 9 1

So as a conclusion, the coelation coefficients show how much two andom vaiables ae elated to each othe in a linea wa. Matlab code fo scatte plots: 1*and(1,1);.5; sigma1^/1; % sqt(sigma/(sigma+sigman) sigman (1/^-1)*sigma, + sqt(sigman)*andn(size()); plot(,,''); title(['scatte plot fo (,) pais' cha(1) ' + n; n is the andom noise on,' cha(1) ' _{}' numst()]); label(''),label(''); Non-Lineal Related Random-Vaiables and Coelation Coefficient In the pevious section, we have tied to intepet the coelation coefficient fo a linea obsevation model. Linea obsevation model means that the signal of inteest is mapped to the output though a linea function. In the eample pesented in the pevious section, the model is etemel simple (but useful) one, + n. In this section, elaboate futhe on the same topic; but we switch to the non-linea obsevation models such as + n. As in the pevious eample, lets assume that is unifoml distibuted in [, ]. Then 3 3 E { }, E { }, E { } and so on. We will constuct a non-linea function 3 4 of in the fom f ( ) a b such that the coelation coefficient of and is zeo! (Note that, we ae not adding an noise to the obsevations. The coelation is zeo in the absence of noise!) The coelation coefficient is epessed as follows: ( )( )} σ σ } σ σ It is clea that fo the coelation coefficient to be zeo, E { }.

Let s calculate E {} : } ( a 3 a } b 3 a b 4 3 b)} } Similal we can calculate as follows: ae ead to evaluate } : 3 a b 3 a b 6 4. Now we 3 3 } a 4 6 b 3 Fom the last elation, we can conclude that when and is zeo. 4 1 ( a b) a 1, the coelation coefficient of b The following figue pesents, the scatte plot fo the linea obsevation model and the non-linea obsevation model. We ae assuming that both obsevation models have the same noise, i.e. noise is zeo-mean Gaussian and have identical vaiance. Fo the plot given in top pat of the figue, the coelation coefficient fo linea model is set to.99. Fo the non-linea obsevation model, it is equal to since we have set a 1 and b 1. Fom these figues, we can see the coelation coefficient of two andom vaiables having a non-linea elation between them should be teated with cae. Fom these figues, it is clea that when the effect of noise is little, it is possible to sa something about given obsevation fo both models. At least, it is possible to educe the set of possibilities fo the unknown given. Unfotunatel, the coelation coefficient of the non-linea obsevation model is equal to zeo iespective of the noise level coupting the obsevations. Hence coelation coefficient and elated ideas ae especiall useful fo linea obsevation models.

Scatte plot fo (,) pais + n (blue) and 1/ - + n (ed) 1.99 1 1 8 1.99 6 4 - -4 1 3 4 5 6 7 8 9 1 15 Scatte plot fo (,) pais + n (blue) and 1/ - + n (ed) 1.9 1 5 1.9-5 -1 1 3 4 5 6 7 8 9 1

The following is the Matlab code geneating the figue pesented above. Delta1;MCnum1e3; Delta*and(1,MCnum);.9; sigmadelta^/1; % sqt(sigma/(sigma+sigman) sigman (1/^-1)*sigma, 1 + sqt(sigman)*andn(size()); 1/Delta*.^ - + sqt(sigman)*andn(size()); plot(,1,''); hold all; plot(,,''); hold off; title(['scatte plot fo (,) pais' cha(1) ' + n (blue)'... ' and 1/\Delta ^ - + n (ed) ' cha(1)... ' _{_1}' numst()... ' _{_} ' ]); label(''),label(''); co_coef1 1/MCnum*sum((-mean()).*(1-mean(1)))/sqt(va()*va(1)), co_coef 1/MCnum*sum((-mean()).*(-mean()))/sqt(va()*va()), The last two lines of the Matlab code geneates an estimate fo the coelation coefficient. The estimate is poduced b estimating the mean, vaiance and coss-coelation of andom vaiables fom the epeimental data. When MCnum is set to 1, and the scipt is un, we get the following esult: co_coef1.91 co_coef -.85 This esult shows that the coelation coefficient fo the linea model is almost equal to.9 (as epected) and thee is indeed ve little coelation between and fo the nonlinea model.

Appendi: Poof of 1 a + b It can be noted that that fom the definition of that does not depend on the mean values of and. Without loss of an genealit, we assume that and ae zeo mean. Then the esult to be poved and the definition of educes to 1 a, } espectivel, fo zeo-mean andom vaiables and. } } i. Poof of 1 a If a, then } } } E aσ, { } aσ a σ σ 1. E a σ, and { } E σ, then { } ii. Poof of 1 a Let P (ψ ) be a quadatic polnomial in ψ. P (ψ ) is defined as follows: ( ψ ) } } ψ } ψ + } aψ + b c. P ( ψ ) ψ + It is clea that P ( ψ ) fo all ψ values. Then the disciminant of P (ψ ), i.e. b 4ac, should be eithe o negative valued. The disciminant can be calculated as } 4 }. Since, ( } ) } } and then ( ) } If 1 } } 1, which is the fist popet. }, then ( E { } ) 4 } } value called ψ fo which ( ). last elation shows that.. Theefoe thee is a specific ψ P ψ This leads to P( ψ ) ( ψ ) } ψ. The