Each school has some unique
characteristics that influence its education quality and sometimes it is nearly
impossible to collect data which reflect such differences.[1] However,
if we neglect them and do the regression, we would get biased results because
these unobserved variables affect quality and expenditure decisions of an
institution. To cope with omitted variable bias, we can use fixed effect model
provided we have a panel data and assume that these characteristics are
constant over years. Namely, we collect data from the same institutions in different
time periods, factor out unobserved yet constant variables over time for each
school, and get unbiased results.
I assumed that the size and reputation of
an institution are constant over time. These two characteristics can affect
expenditure of a school: a school with larger size might consider hiring more
administrative staff, spending more on facility maintenance and so on. A school
not that well-known might want to spend more to send a signal to market that it
is of high quality[2].
That reputation remains constant over years is a little bit strong assumption:
Emory University and New York Universities are counter-examples that become
academically famous over the past two decades. Some people might also argue
that according to college ranking released annually by US News & World
Report, many schools have had ups and downs (Wake Forest, University of
Rochester, etc.), and such fluctuation should imply that academic reputation
changes quickly over time. However, in terms of college ranking, academic
reputation is not the sole consideration. Factors like retention rate,
matriculation rate, rejection rate and endowment income all take some weights
over time, and institutions to some extent are capable of manipulating them. [3] So
college ranking at best partly reflects academic reputation. In addition,
considering that the data set has over 3000 institutions, a few noted exception
will not jeopardize the assumption.
[1]
For
example, University of Chicago is known for its intellectual atmosphere, which
has positive effect on educating promising students, but such impact is hard to
be quantified in data.
[2] In
a perfectly competitive market, higher price of a product means higher cost of
inputs, which indicates inputs are of good quality. However, Gerstner (1985)
found that empirically for many products the relation between price and quality
is weak.
[3]
Factor weights vary over years, so an institution can experience rank
fluctuation with no intrinsic change. (Cornell) Besides, some critics contend
that the ranking undervalues public universities because their endowment
revenue is small, and one piece of evidence is that top 20 schools in the
ranking are all private. As for factor manipulation, an example is that
institutions can attract more applications by posting lower application
standards on website to increase rejection rate. For a more detailed and
entertaining discussion, refer to Tuition
Rising by Ronald Ehrenberg. ( 978-0674009882)
No comments:
Post a Comment