18.07.2014 Views

Panel Data Analysis, Josef Brüderl - Sowi

Panel Data Analysis, Josef Brüderl - Sowi

Panel Data Analysis, Josef Brüderl - Sowi

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>Panel</strong> <strong>Data</strong> <strong>Analysis</strong>, <strong>Josef</strong> Brüderl 8<br />

delta(wage)<br />

-200 -100 0 100 200 300 400 500 600<br />

0 .2 .4 .6 .8 1<br />

delta(marr)<br />

Within-person change<br />

The intuition behind the FD-estimator is that it no longer uses the between-person<br />

comparison. It uses only within-person changes: If X changes, how much does Y<br />

change (within one person)? Therefore, in our example, unobserved ability<br />

differences between persons no longer bias the estimator.<br />

Fixed-Effects Estimation<br />

An alternative to differencing is the within transformation. We start from the<br />

error-components model:<br />

y it 1 x it i it .<br />

Average this equation over time for each i (between transformation):<br />

y i<br />

1 x i i i .<br />

Subtract the second equation from the first for each t (within transformation):<br />

y it − y i<br />

1 x it − x i it − i .<br />

This model can be estimated by pooled-OLS (fixed-effects (FE) estimator). The<br />

important thing is that again the i have disappeared. We no longer need the<br />

assumption that i is uncorrelated with x it . Time-constant unobserved heterogeneity<br />

is no longer a problem.<br />

What we do here is to "time-demean" the data. Again, only the within variation is<br />

left, because we subtract the between variation. But here all information is used, the<br />

within transformation is more efficient than differencing. Therefore, this estimator is<br />

also called the within estimator.<br />

Example<br />

We time-demean our data and run OLS:<br />

egen<br />

mwage mean(wage), by(id)

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!