Measuring Risk of too
Much Debt in College
Situation: The official College Scorecard provides
information on the median debt of student borrowers who finished college. The median provides information on a typical
person in the middle of the distribution. It would be interesting to know what percent
of student borrowers at a school borrowed more than a certain amount, perhaps
$40,000 or $50,000. The official web
site of the college scorecard does not contain such information.
The college scorecard web site has a section with data files
that contains additional information, which is not used in the online
statistics page available to most users.
The link to the additional data is presented below.
A file called Most Recent Data has a variable called
CUML_DEBT_P90. This variable is the 90^{th}
percentile of debt for people. (I believe the debt percentile pertains to
people who finish school but the documentation is not clear on this point.)
Question: Consider 4year public universities with more
than 5,000 students. (I found 324 such
universities.) Compare the median debt levels and the 90^{th}
percentile of student borrowers at these universities. Why is the 90^{th} percentile a
better measure of the ability of universities to control debt of their students
than the median?
Analysis:
Statistics on the median debt levels and the 90^{th} percentile debt
levels for student borrowers at 324 large 4year public universities are
presented below.
Statistics on the median reflect the debt accumulated by a
typical student at a particular school.
Statistics on the 90^{th} percentile reflect the
experience of a student borrower who borrows more than 90% of all student
borrowers at a school.
The typical or middle of the distribution experience at a
school that has typical debt over 75 percent of schools is $18,298.
The above 90% of debt experience at a school that has a 90^{th}
percentile greater than 75% of schools is $37,309.
Median Debt and 90th
Percentile of Debt
at 324 Public
Universities


Median Debt

90th Percentile of Debt


n

324

324

Mean

15896

34478

Standard Error

177

274

LB 95% CI

15549

33939

UB 95% CI

16244

35017

Min

5500

13750

25th

14177

31049

75th

18298

37309

Max

24250

49400

Discussion: Why should a prospective student be
concerned about the 90^{th} percentile of debt at a university, perhaps
more so than the median debt level at the university?
The median debt does not consider risk. The 90^{th} percentile of debt
incurred by borrowers at a school provides more information about what can go
wrong because of risk.
Some students come from a household where a parent might be
able to contribute more even if the aid package requires substantial debt. Perhaps students from households with modest
means should shun universities where a large number of people exit school with
a lot of debt. The 90^{th}
percentile is a better measure of a lot of debt than the median.
There are other sources of risk. Some schools do not
maintain a full financial aid package over all four years.
In addition, many students take more than four years to
complete their degree. It would be
useful to know whether some schools are better than other schools at getting
students done on time.
In any event, I believe the 90^{th} percentile of
debt figures provide additional information on risk and that these 90^{th}
percentile debt figures should be provided in the main College Scorecard web
site and not be relegated to an obscure file in the data section of the web
page.
Notes:
Note One: The statistics presented in the chart above
are not weighted by the size of the institutions. All institutions in this particular sample
are public universities offering fouryear degrees with at least 5,000
students.
Note Two: The 90^{th} percentile statistics by
institution cannot be used to measure the number of students in the country
with high debt levels. Even weighting
by number of students would not resolve this issue. The 90^{th} percentile of the total
population of students can be disproportionately impacted by a relatively small
number of bad institutions. The table
below illustrates that the mean of schools may be inconsistent with the 90^{th}
percentile for the entire populations
Example Showing Weighted
Average of 90th Percentile
is not Equal to 90th
Percentile of Total


School One

School Two

Total


Borrowers

5,000

10,000

15,000

Number of people with
$30,000 in Debt

2,000

500

2,500

Number of people with
$10,000 in Debt

3000

9500

12,500

90th Percentile

$30,000

$10,000

$30,000

Note Three: The data files on the Department of Education
web page are not easily analyzed. Many
of the variables in the file have alphabetic entries for some observations and
these string variables are not easily read.
It would be highly useful if the Department of Education created a SAS
Export file with data that was readable in SAS and other packages including
STATA.
Readers of this post may want to learn more about President Obama's effort to rank colleges. See the post below.
http://policymemos.blogspot.com/2016/04/rankingcollegesbasedonvalueand.html
Also, readers might be interested in my plan to front load aid to firstyear students.
http://policymemos.blogspot.com/2016/03/eliminationofcollegedebtforfirst.html
Also, readers might be interested in my plan to front load aid to firstyear students.
http://policymemos.blogspot.com/2016/03/eliminationofcollegedebtforfirst.html
I agree. The focus on the 50% quantile is convention. One can regress on any quantile, as long as there is sufficient data to support it.
ReplyDeleteThis principle can be generalized to other contexts. For instance, it is probably much more appropriate to plan based upon the 90% quantile of sea level rise than it is upon the 50% quantile (or some approximation, like the mean) because those risks are what want to be minimized. There is probably a methodical way of picking the quantile point based upon relative losses.