Monday, August 28, 2017

Duration of Under-Graduate Career and Student Debt

Duration of Under-Graduate Career and Student Debt

Originally, this post was motivated by a general interest in student debt levels.  I wanted to present some numbers related to the impact of on-time graduation rates on student debt levels and how this relationship changed in recent years.  I then read a Wall Street Journal article about how some elite colleges are deciding to be stingy on AP credits


I am concerned that a stingy AP exam credit policy will delay graduation and increase debt levels for future students.

Question:   How do student debt levels of students who graduate in four years or less differ from the student debt levels of students who take more than four years to finish their undergraduate career?

Data and Methodology:  The database used for this question is the NPSAS 2012 undergraduate file accessed through the NCES DATALAB.   I am looking at all under-graduates who received a four-year bachelor degree in 2012 and all undergraduates who received a Bachelors degree in 2004.    

Statistics on the percent of Bachelor degree recipients with positive loan debt and the average debt for degree recipients with debt are presented for students who took fiver or more years to graduate and four or fewer years to graduated.  

Statistics are also presented on the share of students taking five or more years to graduate in 2004 and in 2012.

Results  The first chart below has information on incidence of debt and average debt for students receiving a BA degree in 2012.

Loan Debt and Years in School for Bachelors Degree
 Recipients in 2012
% Graduates With Positive Loan Debt
Average Loan Debt for Student Borrowers at Graduation
(%>0)
(Avg>0)
Five or More Years in College
73.33
31,639.32
Four or Fewer Years in College
62.60
25,527.86
% Diff.
17.1%
23.9%
Total
68.97
29,384.20

Sample consists of people receiving a Bachelors degree from a 4-year institution in 2004

The chart below here provides information on incidence of debt and average debt for students receiving a BA degree in 2004.

Loan Debt and Years in School For Bachelors Degree Recipients in 2004
%  Graduates With Positive Loan Debt
Average Loan Debt for Student Borrowers at Graduation
(%>0)
(Avg>0)
Five or More Years in College
67.58
19,978.00
Four or Fewer Years in College
59.70
16,513.74
% Diff.
13.2%
21.0%
Total
64.23
18,607.88
Sample consists of people receiving a Bachelor’s Degree from a 4-year institution in 2004.


The chart below provides information on share of people graduating in four years or less and share of people graduating in five or more years in both 2004 and 2012.

Years to Graduation for BA Recipients in 2004 and 2012
Year
% of BA Recipients who Graduate in 5 years or More
% of BA Recipients who Graduate in 4 Years or Less
2004
57.45
42.55
2012
59.35
40.65
Diff in Percentage Points
1.89
-1.89
% Diff.
3.3%
-4.4%


Discussion of Results:  The results above highlight the impact of finishing college on time on student debt after college.

In 2012, around 73 percent of BA recipients who took five or more years to graduate had debt.   The comparable percentage for BA recipients who finished in four of fewer years was around 63 percent.

Average debt for BA recipients who finish in five or more years was $6,111 higher or 23.9% higher than debt for students who took four years or less.

In 2004, around 68 percent of BA degree recipients who took five or more years to graduate had debt compared to around 60 percent for people who graduated in 4 years or fewer. 

The difference in debt levels between the two groups was around $3,500.

The percent of people taking five or more years to receive their BA degree went from 57.45 percent in 2004 to 59.35 percent in 2012.

Concluding Remarks:  The results presented here indicate that people who take more than four years to complete an four-year undergraduate program are more likely to take on debt and tend to have more student debt than people who finish their undergraduate program in four years or less.  Moreover, the percent of people who take more than four years to complete their undergraduate degree program appears to be rising.

These results suggest that part of the solution to the problem of increased student debt levels should involve better preparing people for college so they can graduate on time.   In addition, the results presented here indicate that the decision by some elite colleges to reduce amount of credit given to people that pass AP exams could delay graduation and increase debt levels of future cohorts of students.   Subsequent posts will address this issue in greater detail.








Using Survey Data to Consider Out-of-State Student Debt

Using Survey Data to Consider Out-of-State Student Debt

Cumulative Debt for In-State and Out-of-state Students at Public Universities in 2012

Question:   Use NCES POWERSTATS software to analyze differences in cumulative debt for in-state students versus out-of-state students at public universities during the 2011/2012 academic year.

What further analysis should be considered based on the results presented here?

Comment on the limitations of the software used to conduct this analysis?


Database and Methodology:  The data used for this problem was obtained from the NPSAS 2012 survey.   

The codebook for the survey is here:

This problem studied cumulative debt incurred by undergraduate students seeking a Bachelor’s degree at public universities.   The analysis compares cumulative debt levels for in-state students and out-of-state students. 

The sample includes all students enrolled at some point during the 2011/2012 year.  Debt levels at completion of the program will be higher for both groups.

I don’t have access to the micro file for this data.   There is no public-use file for these databases.  I don’t currently qualify for access to the restricted-use file because I don’t have access to a secure data room.

However, I can examine statistics from the database using POWERSTATS software available at 



Tabulations:  The statistics obtained for this problem include the average cumulative debt for undergraduates with debt, the percent of undergraduates incurring debt, and the percent of undergraduates with more than $25,000 in debt for the two groups.

Tabulations are presented below:


NPSAS institution sector (4 with multiple) = Public 4-year
Cumulative amount borrowed for undergrad
Cumulative amount borrowed for undergrad
Cumulative amount borrowed for undergrad
(Avg>0)
(%>0)
(%>25000)
Estimates
Legal Resident
18,651.69
65.0493
17.2814
Not Legal Resident
20,351.63
62.1141
16.5029
Standard Error (BRR)
Legal Resident
172.5303
0.42951
0.33366
Not Legal Resident
669.04395
1.6342
1.2266




Discussion of Results:

Percent of in-states student with debt is larger than percent of out-of-state students with debt, by nearly 3 percentage points.  This probably occurs because out-of-state students are more affluent than in-state students.

Percent of in-state students with more than $25,000 in debt is larger than percent of out-of-state students with $25,000 in debt.  The differential is 0.7 percentage points.

The average debt for students with debt is higher for out-of-state students than in-state students.   The differential is around $1,800.

The difference in average debt levels is statistically significant using a large-sample test of difference in means under the assumption of unequal variances.  

The difference in average debt levels appears to be the result of  a relatively small number of out-of-state students with a lot of debt.

Implications of results:

From other publications we know that debt levels for undergraduate students have risen substantially between 2004 and 2012. 


The average debt level out-of-state students is higher than for in-state students but this result was driven by a relatively few students with a lot of debt.

I suspect that access to out-of-state schools for lower-income and middle-income students may have fallen between 2004 and 2012.    This is worth looking into. 

Some limitations with POWERSTATS:

The POWERSTATS software does not allow me to conduct non-parametric procedures including Wilcoxon rank sum tests and the Kolmogorov-Smirnoff procedure.   The software gave me means and standard errors, which I inputted into a t-test calculator.   

The POWERSTATS software does contain regression and logistical regression models but does not appear to allow the user to create models and tests from multiple years of NPSAS or other databases.   I might want to a logistical model that considers whether students from lower-income household are less likely to go out-of-state in 2012 compared to 2004.  

 This and similar questions could be more easily addressed with public use databases similar to the ones that available for health care issues at AHRQ.   

For now, I will continue to use POWERSTATS to address some of these issues.








Measuring Performance on AP Exams

Issue:   A recent Washington Post columnist advocated expanding the number of students taking AP exams even in high schools where the fail rate is high.  


Do you agree with the contention that the number of people in a high school that take an AP exam is an appropriate way evaluate and compare different high schools? 

Develop and discuss different ways to measure the effectiveness of high school AP policy that incorporates information on both participation and performance.

 Short Answer:  An evaluation of the number of high schools or academic programs on the basis of the number of exams taken is similar to an evaluation of a quarterback based on the number of passes attempted regardless of the outcome of the passes.   It reminds me of the ribbons for participation in “About Schmidt.”

The alternative approach proposed and discussed in this post incorporates information on the ratio of the exams passed to total number of students and the average AP score.

Analysis:   Hypothetical data in the table below on high school AP participation rates and average AP score is used to illustrate the need for including both participation and performance in a high school AP rating performance index.

High School Number
Number of AP Exams Taken
Average for all Three AP Exams
Rank of AP Exams Taken
Rank of AP Grade Average
Sum of ranks for exams taken and average grade
1
750
3.6
5
6
11
2
750
3.4
5
4
9
3
750
2.6
5
1.5
6.5
4
500
3.5
2
5
7
5
500
2.8
2
3
5
6
500
2.6
2
1.5
3.5


Some thoughts on how one might rank high school AP performance from the above table.  For simplicity all high schools in this example have 400 students.

High School One has the highest performance level among the six high schools because it has the highest exam to student ratio and the highest average exam grade among the schools.   We can say that high school one is Pareto Superior to the other high schools.

High School six with the least number of exams taken and the lowest average grade will have the lowest ranking among the schools.  It is Pareto Inferior.

The other schools cannot easily be compared because in many cases one school has an advantage in the number of exams taken while the other school has the higher average AP grade.   For example, AP performance at school two is not unambiguously better than AP performance at school four.  (Students at school two take more AP exams while students at school four have a slightly higher average grade.)

A composite AP performance index incorporating information on both participation and grades can be constructed.  There is no unique ranking when participation is higher in one school and grades are higher in another.  In this instance, the ranking of AP high school performance is determined by the design of the index.

One way to create a composite AP performance index is to use the rank function.  The fourth column in the table provides the rank of schools based on AP exams taken.   The fifth column of the table provides the rank of schools based on average AP grade.   The sixth column  -- the sum of the two ranks  -- is my composite AP performance rating.

 It ranks school 1 first and school 6 last.  

It ranks school 4 behind school 2 but ahead of school three.  

A different rating scheme and/or more and different data could have changed the order of these ratings.


Concluding thoughts:

It is clearly inappropriate to base the performance of a high school AP program based on the ratio of tests given to the number of students.  Such a formula provides a strong incentive to put more students in multiple AP exams, regardless of likely outcome.  It sends the message that failure is okay.

A simple and more permissible measure of AP performance might be the ratio of tests passed to total number of students.  This ratio depends upon a large number of students both taking and passing tests.  An increase in tests given could actually decrease the number of tests passed if it causes the pass rate to fall.   First, an increase in the number of AP test takers can dilute teaching resources.   Second, (and probably more importantly) an increase in the number of students taking multiple exams will cause some students to short circuit and fail all tests.



This post is not the last word on the measurement of AP test performance.  A more sophisticated composite index might penalize schools where many students take and perform poorly on a large number of exams.  Also, it may be reasonable to expect most student to take at least one AP exam.   Hence; it might be appropriate to penalize schools where a large percent of students take no exams.

This post is not just an academic investigation.   Many AP students enroll in and fail AP exams.  Was this an appropriate use of their time and scarce resources?  Some high schools place students in many AP subjects.   Would many of these students have been better off if they had concentrated on a single exam rather than many exams?    Should there be criteria limiting the eligibility to take an AP course.   For example, it might be reasonable to base entry into an AP course on performance in a specialized SAT subject exam.


A tangent conveniently placed near the end of the post:

An alternative approach to the one presented here could be patterned after the NFL QB rating system.  The NFL QB rating system does not consider attempts or even just completions or yards.   It covers four factors – pass completion percentage, yards per attempt, touchdowns per attempt and interceptions per attempt.

The index also places a cap on each component.  A decrease in the interception rate to 0% is not optimal because interceptions are part of the game, especially in the fourth quarter when a risky pass is needed to win the game.  

A 0% interception rate would indicate that the team’s QB or game plan is too risk averse.   Similarly, a 4.0 AP grade average would indicate that a school is not encouraging students to challenge themselves more.

Here are links to two of my posts on NFL QB rating that motivated this new AP performance rating system.

Manning Versus Wilson

A post written BEFORE the Super Bowl indicating that Seattle’s offense was as potent as Denver’s offense.


Evaluating the QBs from the class of 2004.

Omission of information on sack from the QB rating formula:


A post of interest on education reform:

People interested in this post on AP exams might also be interested in my analysis of the economics of education reform.   My post on competition in the market for education is by far the most popular post I have ever created.  


A question on additional research:


I would like to create a larger paper on methods that might be used to rank high school AP performance.   In the past most of my work has been geared towards academic conferences and journals.  However, for this project I am willing to work as a consultant for an organization that needs to study this issue.  People who are aware of data needed for this research and an organization interested in sponsoring this work should contact me through inmail at linkedin or at Bernstein.book1958@gmail.com




PLUS Loans for Parents and Parent Income



PLUS Loans for Parents and Parent Income

Question:  How has the use of PLUS loans for parents changed over time for parents of student attending undergraduate institutions and for students attending graduate schools?   What is the share of PLUS loans taken out by parents with income in the bottom quartile?

Does it appear that parents taking out PLUS loans for students have adequate income to repay their obligations?

Why this issue is important:  Parents who have problems repaying PLUS loans are not allowed to default on the loan.   Increasingly, many parents with PLUS loan obligations have had problems repaying and in some cases the government has garnished Social Security benefits from these borrowers.   It is possible that many of the financial problems caused by use of PLUS loans could have been prevented if lenders had considered the adequacy of parent income prior to making the loan.

Data and Methodology:

I addressed this issue with TRENDSTATS from the NCES DATALAB.


 TRENDSTATS allowed me to get data on use of parent plus loans by income quartile for five different survey years  -- 1996, 2000, 2004, 2008 and 2012.   

I created separate analysis for parents of undergraduate students and parents of graduate students.

The table on PLUS loans for undergraduates only involves parents of dependent students.   

The table on PLUS loans for graduate students uses the combined income of the student and the parent.


Results:  Two tables on PLUS loan use and income quartiles over time are presented below.


Percent of Dependent Parents with PLUS Loans by Income Quartile
Year
Lowest 25th  Percent
Lower Middle 25th  Percent
Lower Upper 25th  Percent
Upper 25th  Percent
Total
1996
2.96
5.56
6.38
5.65
5.06
2000
3.56
5.48
8.61
6.76
6.07
2004
3.92
6.53
9.34
8.34
6.98
2008
4.33
6.73
9.37
8.86
7.25
2012
6.22
9.17
11.33
10.87
9.27
Percentage Growth 1997 to 2012
109.91%
64.96%
77.57%
92.49%
83.27%
Sample is all parents of dependent undergraduate students


Parent Plus Loans for Graduate Student by Quartile of
Sum of Parent and Student Income
Year
Lowest 25th  Percent
Lower Middle 25th  Percent
Upper Middle 25th  Percent
Upper 25th  Percent
Total
1996
6.83
3.94
2.36
0.80
3.48
2000
7.37
5.75
4.14
2.90
5.07
2004
7.98
6.18
3.44
3.87
5.51
2008
9.82
8.14
5.48
3.88
6.76
2012
11.47
7.87
5.23
3.27
7.13
% Change
67.85%
99.63%
121.73%
308.91%
104.82%


Analysis of Percent of Plus Loans Across Income Quartiles:

Undergraduate Students:

The lower upper 25th percentile had the highest share of students dependent on PLUS loans for parents in all years.

Growth rate in use of PLUS loans for parents is highest in the lowest 25th percentile.

Graduate Students:  

The lowest 25th percentile consistently had the highest percent of people dependent on PLUS loans for parents.

The upper 25th percentile had the highest growth rate in the use of PLUS loans for parents; although, the PLUS loan share for this quartile remained lower than all other quartiles in 2012.

Share of PLUS Loans Taken Out by Parents in First and Second Income Quartile:

Above I discussed the percent of students in each quartile that used a PLUS loan.

Here I look at the percent of students using PLUS loans that are in particular quartiles in each income quartile.  


PLUS Loans for Parents Usage
Number out of 1,000 per income quartile
Q1
Q2
Q3
Q4
Total
Undergraduates
62.2
91.7
113.3
108.7
375.9
Graduates
114.7
78.7
52.3
32.7
278.4
Share in Each Quartile
Q1
Q2
Q3
Q4
Total
Undergraduates
16.5%
24.4%
30.1%
28.9%
100.0%
Graduates
41.2%
28.3%
18.8%
11.7%
100.0%

Calculations above are for 2012

Observations on use of Parent PLUS Loans Across Income Quartiles:

Lower-income people take out a lot of PLUS loans.

16.5 percent of PLUS loans taken out by parents of undergraduates are in the lowest income quartile.

41.2 percent of PLUS loans taken out by parents of graduate students are in the lowest income quartile.


Methodological Note:

I wanted the software to provide numbers of students in each income quartile based on population weights.   I would have obtained contingency tables based on population weights in SAS or STATA if I had access to the raw data files.  TRENDSTATS does not appear to have this capability.   Alas, I don’t have access to the raw data so this could not happen.

I attempted to switch the row and column variables in TRENDSTATS but the TRENDSTATS software does not allow for automatic creation of income quartiles when parent income of dependent variable is the column variable.

How then did I get the share of loans for all income quartiles?

By definition, each quartile has the same number of observations so I assumed each group had 1000 students.   I multiplied 1000 by share of students using PLUS loans for each quartile to get PLUS loan use per 1,000 students.

The sum of these numbers is total PLUS loan use across all students.   I divided PLUS loan use by income quartile by total PLUS loan use in the population to get quartile shares.

I am very interested in understanding the advantages and limitations of the POWERSTATS and TRENDSTATS education department software and will continue to make comments that might lead to improvements in the on-line databases.

Concluding Thought:

Barring really exceptional circumstances, student debt including PLUS loans obtained by parents is not forgiven or discharged even in bankruptcy.   Lenders happily give PLUS loans to lower-income parents because the loans are guaranteed even if the lender cannot make repayments.

The combination of government guarantees for loan payments and a prohibition on discharge of loans in bankruptcy has led to a thriving debt market geared towards people with little chance of repayment.