Flashcards
Heteroskedasticity part 1
Cards in this deck (7)
- 1What does heteroskedasticity mean? [Homoskedasticity's literal meaning = "equal spread"]The variances of the error terms are not constant
- 2Why is it a problem? (4 problems)1. It affects the reliability of the regression *location* as the variance changes
2. Variances (standard errors) of the estimated regression parameters (coefficients) will be biased and inconsistent, so you may not correctly test for significance of estimated parameters
3. With heteroscedasticity, OLS becomes inefficient, i.e. it does not provide the minimum variance for estimated coefficients
4. The t- and F-tests will be incorrect in general - 3What are the typical causes of heteroskedasticity?1. Omitted important variables, e.g. food consumption does not depend on consumer income only
2. Mis-specification of model
3. Measurement errors in the variables of the model
4. Values of observations change a lot within a sample (e.g. spending pattern of poor v rich people)
5. Skewness? - 4How can we test for heteroskedasticity?1. Inspect scatter plots of the residuals, try to detect an irregular pattern
2. Systematically plot the least squares residuals against each IV and each DV to see if they vary in a clear pattern [e.g. the errors may exhibit greater variations for larger values]
3. Apply formal statistical tests - 5What are the 3 formal tests for heteroskedasticity?1. Goldfeld-Quandt test
2. Breusch-Pagan test
3. White test - 6What are the steps for the Goldfeld-Quandt test?1. Sort all regression variables in ascending or descending order (of variable Xi e.g. income)
2. Split the sample of observations into half and run two separate regressions: one for large values of Xi and another for smaller values
3. Run the regression for each sub-sample, get the residual sum of squares SSR2 from the subsample with large variance, N2 observations and SSR1 from the subsample with smaller variance, N1 observations
4. Conduct Goldfeld-Quandt test - 7How do we interpret the Goldfeld-Quandt test?
Information
- Description
- Flashcards on Heteroskedasticity part 1, created by Felix Culas on 12/04/2017.
- Created by
- Felix Culas
- Views
- 1
- Rating
More on Regression Assumptions and Diagnostics