[go: up one dir, main page]
Scribd
Upload
73 views1 page

Steps of Panel Data Analysis

The document outlines the 4 main steps of panel data analysis: 1. Choosing between a pooled OLS, fixed effects, or random effects model using the Hausman test. 2. Estimating the model using pooled OLS, fixed effects, or random effects. 3. Evaluating the model for heteroscedasticity and autocorrelation. 4. Interpreting the estimated coefficients if the model is correctly specified.

Uploaded by

Gerad Teo
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
73 views1 page

Steps of Panel Data Analysis

The document outlines the 4 main steps of panel data analysis: 1. Choosing between a pooled OLS, fixed effects, or random effects model using the Hausman test. 2. Estimating the model using pooled OLS, fixed effects, or random effects. 3. Evaluating the model for heteroscedasticity and autocorrelation. 4. Interpreting the estimated coefficients if the model is correctly specified.

Uploaded by

Gerad Teo
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/ 1

Steps of Panel Data Analysis:

Step 1: Choice of Model


unobserved characteristics or
Types of Models Hausman Test
unobserved heterogeneity (𝒄𝒊 )
Ideal only if there are not omitted 𝑦$,&'( = 𝛼 + 𝑥$,&'( 𝛽 + 𝛿$ 𝑑$ + 𝑢$,&'(
Pooled (OLS)
variable biased
Ideal if 𝑐$ is related to the 𝐻2 : 𝑑( = 𝑑4 = ⋯ = 𝑑6 = 0 → no difference, which would imply that the random effects
Fixed effects model
explanatory variables specification was adequate

𝐻( : 𝑎𝑡 𝑙𝑒𝑎𝑠𝑡 𝑜𝑛𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑑( ≠ 0, 𝑑4 ≠ 0 ⋯ 𝑑6 ≠ 0 → the fixed effects specification should


be used It is possible to estimate several models and test
for differences between them, which will be
If 𝐻2 cannot be rejected then this implies that the individual dummy variables are indicative of the appropriate specification.
Ideal if 𝑐$ is unrelated to the
Random effects model unnecessary, i.e., there are no substantial idiosyncratic terms and the fixed effects
explanatory variables
specification is not required. We could then estimate a random effects model.

If 𝐻2 can be rejected then we could estimate both the fixed effects model and the
random effects model and test whether the parameter estimates were significantly
different.
Step 2: Estimation
(pooled) OLS Fixed effects model Random effects model
𝐸F𝑢$,& G𝑋, 𝑐I = 𝐸F𝑢$,& I ∀ 𝑖, 𝑡 𝐸[𝑐$ |𝑋] = 𝐸[𝑐$ ] ∀ 𝑖 &
𝐸F𝑢$,& G𝑋, 𝑐$ I = 𝐸F𝑢$,& I ∀ 𝑖, 𝑡
Unbiased & consistent 𝐸[𝑐$ |𝑋] = 𝐸[𝑐$ ] ∀ 𝑖, 𝑡 𝐸F𝑢$,& G𝑋, 𝑐$ I = 𝐸F𝑢$,& I ∀ 𝑖, 𝑡.
No omitted variables
The efficiency of the estimators depends upon the properties of the errors.
Efficiency
If autocorrelation and/or heteroskedasticity is present then we can account for this by using robust standard errors.
Step 3: Evaluation
(pooled) OLS Fixed effects model Random effects model
Heteroscedasticity Breusch-Pagan test (regressing squared estimation residuals on the explanatory variables).
In practice, one simple approach is to estimate the model with and without robust standard errors. If the calculated values of the standard errors change substantially between the two
Autocorrelation
then that can be taken as an indication that there is autocorrelation present (this could apply equally to the case of possible heteroskedasticity).
Step 4: Interpretation
(pooled) OLS Fixed effects model Random effects model
Suppose that we have a correctly specified model (no omitted variable bias, appropriate functional form and appropriate panel specification) then the interpretation of our estimates is
the same as previously: each estimated coefficient gives the marginal expected effect of that variable, other things equal, and we can conduct hypothesis tests and construct confidence
intervals in the usual manner, provided that we also have appropriately calculated standard errors.

You might also like