This document discusses heteroscedasticity and various methods to detect it. Heteroscedasticity occurs when the variance of the error terms is not constant. The author outlines graphical methods, Park's test, Glejser test, Spearman's rank correlation test, Goldfeld-Quandt test, and Breusch-Pagan-Godfrey test to detect heteroscedasticity. These tests analyze residuals, model error variance, rank correlations, and sum of squares to determine if heteroscedasticity is present in the data.
This document discusses heteroscedasticity and various methods to detect it. Heteroscedasticity occurs when the variance of the error terms is not constant. The author outlines graphical methods, Park's test, Glejser test, Spearman's rank correlation test, Goldfeld-Quandt test, and Breusch-Pagan-Godfrey test to detect heteroscedasticity. These tests analyze residuals, model error variance, rank correlations, and sum of squares to determine if heteroscedasticity is present in the data.
This document discusses heteroscedasticity and various methods to detect it. Heteroscedasticity occurs when the variance of the error terms is not constant. The author outlines graphical methods, Park's test, Glejser test, Spearman's rank correlation test, Goldfeld-Quandt test, and Breusch-Pagan-Godfrey test to detect heteroscedasticity. These tests analyze residuals, model error variance, rank correlations, and sum of squares to determine if heteroscedasticity is present in the data.
This document discusses heteroscedasticity and various methods to detect it. Heteroscedasticity occurs when the variance of the error terms is not constant. The author outlines graphical methods, Park's test, Glejser test, Spearman's rank correlation test, Goldfeld-Quandt test, and Breusch-Pagan-Godfrey test to detect heteroscedasticity. These tests analyze residuals, model error variance, rank correlations, and sum of squares to determine if heteroscedasticity is present in the data.
By: Dr. Akash Asthana Assistant Professor, Dept. of Statistics, University of Lucknow, Lucknow HETEOSCEDASTICITY
In the linear regression model, it is assumed that
the variance of error terms should be constant and also independent of each other. i.e. If this assumption is not fulfilled then heteroscedasticity is said to be present. In presence of heteroscedasticity the estimates of regression coefficients not remain BLUE. DETECTION OF HETEROSCEDASTICITY
There are several methods used for the detection of
heteroscedasticity among which most commonly used methods are: 1. Graphical Method 2. Park’s test 3. Glejser Test 4. Speaeman’s rank correlation test 5. Goldfeld – Quandt Test 6. Breusch – Pagan – Godfrey Test RESIDUALS Residuals (ri) are nothing but the difference between the observed and predicted value of the dependent variable. It can be considered as estimate of the error term in the regression model. ri (or Ûi) = Yi – Ŷi Residuals are of following types: 1. Simple Residual 2. Standardized Residual 3. Studentized Residual SIMPLE RESIDUAL
The simple residual (ri) is obtained by the difference between the
observed and predicted value of the dependent variable. STANDARDIZED RESIDUAL
The simple residual has a drawback that its distribution is not
Normal therefore standardized residuals are used these are given by:
STUDENTIZED RESIDUAL
The standardized residual has a drawback that its values are
unbounded. To overcome this difficulty the standardized residuals are used these are given by:
Where hii is the ith diagonal element of hat matrix (H) = X(X’X)-1X’ It follows the student’s t-distribution. GRAPHICAL METHOD In this method Residual square (ri2 or ri2’)is plotted against the predicted value of the dependent variable. If the plot doesn’t show any pattern then heteroscedasticity is said to be absent otherwise it is said to be absent. In the figures on next slide first figure (a) shows case of homoscedastic data whereas other figures (b, c, d & e) are the examples of heteroscedastic data [1]. PARK’S TEST
Park had modeled the error variance as a function
of explanatory variables defined as:
or, Where νi is homoscedastic error term. However, since σi2 is unknown Park had been suggested the use of ri2 in its place. If β comes out to be significant then heteroscedasticity is said to be present in the data. GLEJSER TEST
The Glejser test is similar to Park’s test. Instead of
one Glejser had used different functional forms to model error variance (or its estimate) over explanatory variables. If any of the model comes out to be significant then heteroscedasticity is said to be present. The functional forms used by Glejser are on next slide. GLEJSER TEST
However this test has a drawback that in last two models
parameters are nonlinear and therefore can’t be estimated with the help of OLS method. Also these two method considered the term νi as white noise, which is not necessary according to Goldfled and Quandt i.e. it may be heteroscedastic itself. SPEARMAN’S RANK CORRELATION TEST
It has following steps:
1. Fit the regression of Y on X and otain the residuals. 2. Compute the Spearman’s rank correlation between absolute value of residuals and Xi (or Ŷi) 3. Test the null hypothesis that population correlation coefficient is zero using t-test. If the hypothesis is rejected then heteroscedasticity is said to be present. The t-statistics is given by:
rs = Spearman’s rankcorrelation coefficient.
GOLDFLED – QUANDT TEST This method is applicable only if the heteroscedastic variance (σi2) is positively related with one of the explanatory variable. In this method it is assumed that σi2 is proportional to the square of the explanatory variable. Goldfled and Quandt had suggested a number of steps to detect the heteroscedasticity which are on next slide. GOLDFLED – QUANDT TEST 1. Order (or arrange) the observations in the ascending order of values of X. 2. Omit c central values (c is a specified a priori) and divide the remaining c central values into two equal halves having (n-c)/2 observations. 3. Fit separate OLS regression for both the halves and compute residual sum of squares for each (say RSS1 and RSS2) having (n-c-2k)/2 degrees of freedom. (k is no. of parameters estimated.) 4. Compute the ratio λ = RSS1/RSS2. 5. Heteroscedasticity is said to present if : λ > F(n-c-2k)/2, (n-c-2k)/2, α BREUSCH–PAGAN–GODFREY TEST
The goldfled–Quandt test depend upon the
correct selection of the value of c and the correct explanatory variable according to which observations are to be arranged. To overcome this difficulty Breusch–Pagan– Godfrey defined another test. This test has a number of steps which are on next slide. BREUSCH–PAGAN–GODFREY TEST 1. Fit the regression model Y = Xβ + є using OLS method and obtain the residuals r1, r2, …, rn. 2. Obtain the estimate of σ2 using:
3. Construct the variable pi using:
4. Regress the pi over the Zj’s. Some or all Xj’s may serve as Zj’s.
Where νi is the homoscedastic error term.
5. Obtain the Explained (Regression) Sum of Square (ESS) and define: Θ = ESS/2 6. If Θ exceeds the at given level of significance then heteroscedasticity is said to be present. REFRENCES
1. Gujrati DN, Basic Econometrics, 4th edition
(2004) , The McGraw-Hill Companies. 2. Draper NR & Smith H, Applied Regression Analysis, 3rd edition (1998), John Wiley & Sons Inc. 3. Johnston J & Dinardo J, Econometric Methods, 4th edition (1997), McGraw-Hill Companies.
