Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
72 views

Assumptions For Regression Analysis: MGMT 230: Introductory Statistics

This document discusses the key assumptions of ordinary least squares (OLS) regression analysis: linearity, independence of the error term, normality of the error term, and stationary variance of the error term. It notes that independence of the error term and normality are not strictly necessary. Additional important assumptions are independence of the explanatory variables from the error term and stationarity of the explanatory variables. The document provides examples and ways to evaluate whether the assumptions hold for a given dataset and regression model.

Uploaded by

Poonam Naidu
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
72 views

Assumptions For Regression Analysis: MGMT 230: Introductory Statistics

This document discusses the key assumptions of ordinary least squares (OLS) regression analysis: linearity, independence of the error term, normality of the error term, and stationary variance of the error term. It notes that independence of the error term and normality are not strictly necessary. Additional important assumptions are independence of the explanatory variables from the error term and stationarity of the explanatory variables. The document provides examples and ways to evaluate whether the assumptions hold for a given dataset and regression model.

Uploaded by

Poonam Naidu
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Assumptions for Regression Analysis

Mgmt 230: Introductory Statistics

1
Goals of this section

• Learn about the assumptions behind OLS estimation.


• Learn how to evaluate the validity of these assumptions.
• Introduce how to handle cases where the assumptions may be violated.

Assumptions behind OLS


You’re textbook says the four assumptions behind valid OLS estimation are:

1. Linearity.
2. Independence of error term. ← not completely necessary.
3. Normality of the error terms i . ← not true.
4. Stationary variance of i . ← not completely necessary.

You’re textbook fails to mention some very important assumptions:


• The explanatory variables must be independent of the error term.
• The explanatory variables must be stationary. Often not true in financial
and economics.

Linearity

• Linear model must be an accurate description of the true relationship


between the variables.

yi = β0 + β1 x1,i + β2 x2,i + ... + βk−1,i xk−1 + i

• Evaluating Linearity:
– Scatter plot with a trend line.
– Scatter plot of the residuals.

1
Linearity

• Quadratic relationship. yi increases as xi increases, but then decreases (or


vice versa).
– To account for this possibility, also put x2i into the model.
– Example: worker productivity and age.
• Exponential relationship. yi increases as xi increases, but at an increasing
rate.
– To account for this possibility, put exi into the model instead of xi .
– Example: Total costs and total output.
• Logarithmic relationship. yi increases as xi increases, but at a decreasing
rate.
– To account for this possibility, put ln(xi ) into the model instead of
xi .
– A quadratic relationship may appropriately capture this relationship
as well.
– Example: earnings and experience.

Independence of error term

• This assumption states that an error from one observation (i ) is indepen-
dent of the error from another observation (j ).
• This often happens in financial and economic time series data.
• Satisfying this assumption is not necessary for OLS results to be consis-
tent. But, better methods than OLS are possible.

• Consistency: An estimate is consistent if as the sample size gets very


large, the sample estimates for the coefficients approach the true popula-
tion coefficients.
• If the residuals are not independent, this most likely indicates you mis-
specified the model (i.e. linearity assumption is violated).

Normality of error term

• Why was the central limit theorem so cool?


• Correct assumption: the sample size is sufficiently large or the population
error term is normally distributed.
• If this assumption holds:

2
– The sampling distribution of the estimates for the coefficients (b’s)
will be normal.
– The residuals will be normal.
• Forget about rules of thumb like n > 30 for regression.
• To evaluate if this assumption holds, can do a histogram of the residuals.

Stationary variance

• The population error term should have a constant variance.


• The variance should not increase as xi increases.
• This often happens with data related to income or wealth.
– Suppose you are predicting how much people spend on luxury goods.
– Larger errors are going to be made for people with larger incomes.
• Satisfying this assumption is not necessary for consistency, although bet-
ter methods than OLS exist for estimating models with this problem.
• To evaluate if this assumption holds, do a scatter plot of the residuals on
the y-axis, and the x variable on the x-axis.

Independence of explanatory variables

• For consistent and unbiased results, the X variables must be independent


of the population error term (i .
• That is, the errors made in the regression cannot be related to your vari-
ables.
• Omitted variable bias: when there are possible explanatory variables (that
may not even be measurable) not included in the regression that are cor-
related with the included explanatory variables.
• The error term accounts for anything not included in the regression.

Stationary Explanatory Variables

• The explanatory variables must be stationary.


• Economics and time series data are often not stationary, rather they grow
as time goes on.
• Examples: GDP, income, price level, wages.
• It can be very tough to handle this problem.

You might also like