Scribd
Upload
35 views4 pages

Model Specification

The document discusses model specification errors in regression analysis, highlighting the importance of correctly specifying a model to avoid bias and inefficiencies. It outlines various types of specification errors, such as omitting relevant variables, including irrelevant variables, wrong functional forms, and measurement errors, along with their consequences. Additionally, it raises questions about model selection criteria, detection of specification errors, and potential remedies.

Uploaded by

goitseoneseribe
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
35 views4 pages

Model Specification

The document discusses model specification errors in regression analysis, highlighting the importance of correctly specifying a model to avoid bias and inefficiencies. It outlines various types of specification errors, such as omitting relevant variables, including irrelevant variables, wrong functional forms, and measurement errors, along with their consequences. Additionally, it raises questions about model selection criteria, detection of specification errors, and potential remedies.

Uploaded by

goitseoneseribe
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Model specification

Violation of the classical linear regression model (CLRM) that the regression model is “correctly”
specified

If not correctly specified leads to problem of model specification error/model specification bias

Relevant questions to ask:

1. How does one go about finding the “correct” model? i.e criteria in choosing a model for
empirical analysis?
2. Types of model specification errors likely to be encountered in practice?
3. Consequences of specification errors?
4. Detection of specification error? i.e. diagnostic tools used?
5. Remedies?
6. Evaluation of performance of competing models?

Model selection criteria

How do we go about choosing of a “good” model? [Group assignment]

2.Types of specification errors

2.1 Omitting a relevant Variable

Let correct model be:

𝑌 =𝛽 + 𝛽 𝑋 + 𝛽 𝑋 +𝛽 𝑋 +𝑢 (2.1)

Where Y=total cost of production

X=Output

Suppose following model is used instead

𝑌 =𝛼 + 𝛼 𝑋 + 𝛼 𝑋 +𝑢 (2.2)

Equation (2.2) is a specification error omitting a relevant variable (𝑋 .). Therefore, error term
in (2.2) is
𝑢 =𝑢 + 𝛽 𝑋 (2.3)

2.2.1 Consequences?

(a) If the omitted variable X3 is correlated with the included variable X2 (i.e. r23=0) and 𝛼 and
𝛼 are biased as well as inconsistent and the bias doesn’t disappears as the sample size gets
larger.

(b) What if X2 and X3 are uncorrelated?

𝛼 is biased but 𝛼 is now unbiased

© The disturbance variance 𝜎 is incorrectly estimated

(d) The variance of 𝛼 is a biased estimator of the true 𝛽

(e) Usual confidence interval and hypothesis testing procedures are likely to give misleading
conclusions about statistical significance of estimated parameters

(f) Forecasts based on the incorrect model and the forecast (confidence) intervals will be
unreliable

2.3 Including an Irrelevant Variable

Assume another researcher uses

𝑌 =𝛽 + 𝛽 𝑋 + 𝛽 𝑋 +𝛽 𝑋 +𝛽 𝑋 +𝑢 (2.4)

Also a specification error (i.e. including an unnecessary/irrelevant variable)

New error term

𝑢 =𝑢 − 𝛽 𝑋 =𝑢 (2.5)

2.3.1 Consequences?

(a) OLS estimators of the parameters of the “incorrect” model are unbiased and consistent

(b) Error variance 𝜎 is correctly estimated

(c) Confidence interval and hypothesis testing procedures remain valid

(d) Estimated 𝛽s will be generally inefficient, i.e. variances larger than estimated 𝛽s in true
model
2.4 Wrong Functional Form

Assume another researcher uses the following model:

𝑙𝑛𝑌 = 𝛽 + 𝛽 𝑋 + 𝛽 𝑋 +𝛽 𝑋 +𝑢 (2.6)

Log-linear vs.linear

2.5 Measurement Error

Assume a researcher uses the following model:

𝑌 ∗ = 𝛽 ∗ + 𝛽 ∗ 𝑋 ∗ + 𝛽 ∗ 𝑋 ∗ + 𝛽 ∗ 𝑋 ∗ + 𝑢∗ (2.7)

Where:

𝑌∗ = 𝑌 + 𝜀

𝑋∗ = 𝑋 + 𝑊

𝜀 and 𝑊 are errors of measurement

N.b. Equation (2.7)=measurement error where the true Yi and Xi use proxies 𝑌 ∗ and 𝑋 ∗ (which
may contain errors of measurement)

2.6 Incorrect specification of the stochastic error

Given 𝑌 = 𝛽𝑋 𝑢 (2.8)

i.e. the stochastic error enters in multiplicatively. However, taking logs satisfies the CLRM
against the following model:

𝑌 = 𝛽𝑋 +𝑢 (2.9)

Where error term enters additively. If equation (2.8) is “correct”or holds ‘true” would the 𝛽 in
2.9 be an unbiased estimate of one in (2.8)? If not, then we have a specification error.

Group work: How do you go about detecting for misspecification error?

You might also like