rev 12 Six Sigma Project Charter Name of project: An Examination of Cohort Default Rates: A Causal Analysis for Prevention Green belt: Jennifer Howells Submitted by: Jennifer Howells e-mail: jhowells@purdue.edu Date submitted: Draft 5/21/12 Final 10/4/12 I. Project Selection Process Item Yes No Comments Key business issue x Default rates are a very hot topic in the world of Financial Aid (at Purdue and nationally). Linked to a defined process x Loan default rates are sent to Purdue annually. Customers identified x Purdue University (rankings), students currently accepting loans, and DFA employees. Defects clearly defined x We do not have a process to keep our loan default numbers in check. I have described how and why the project was selected below and referenced the tools used. II. Project Description Project Title Date Charted Target Completion Date Actual Completion Date 5/16/12 10/4/12 N/A Project Leader Team Facilitator Team Champion Jennifer Howells Jennifer Howells Joel Wenger Estimated Cost Savings Actual Cost Savings Costs of implementing project Although not a clear monetary savings for N/A The costs incurred with the office, this project will help reduce the this project will be in work risk of going over the 10% default rate, which hours and minimal office keeps us in good graces with the Department of Education. If our rate would happen to supplies (paper, toner, skyrocket, we could lose eligibility for Title IV etc.) aid completely, costing the University more than 200 million dollars/year. Also, by having a low rate, we can offer early disbursement options to all students, including new freshman who would otherwise have to wait 1
rev 12 30 days before getting their loan funds. The default rate is a comparison point between the Big Ten Universities and Peers, so any decrease would benefit Purdue s reputation. Team members Jennifer Howells (analyst) Joel Wenger (supervisor/will provide the initial default cohort list) Ted Malone (final approval at each stage) Stephanie Fiddler (area expert, compiles cohort default rates for Peers & Big Ten) Problem Statement Our draft cohort default rate increased more than expected this year. We need a process to prevent students from appearing on the loan default list (lowering our cohort default rate). To this point, we have not identified common characteristics of these defaulters; we simply receive the names and SSNs for the borrowers who have defaulted. I have the ability to run the names through our Student and Financial aid databases (using Cognos) to add as many characteristics as needed. I will use the characteristics that appear most often in the defaulters to identify an at-risk population. If we can focus education/intervention on the identified at-risk populations while they are still in school, we could reduce the number of borrowers who will default (lowering the cohort default rate). I will also attempt to identify loan servicer issues that may have an impact on the default rate. Project Goal and Metrics The goal of the project is to prevent the cohort default rate from increasing. More specifically, identify common characteristics of loan defaulters using the NSLDS default list. We can use those characteristics to identify current loan borrowers who may benefit from additional financial literacy or additional counseling/intervention. Information gathered might also be used to alleviate loan servicer disparities. Describe the challenges and support required Time available for the project outside of Six Sigma class could be an issue. I will need my supervisor s support and will need to ask for assistance with the Six Sigma project and/or work duties as needed to ensure complete of this task. Project Schedule D1. Select the output characteristic. Date: 5/21/12 Y = Reduce the Cohort Default Rate To make sure that there was a need for a Six Sigma Project, I used Pareto charts to plot the number of loan defaulters over the past 6 years and the number of 2
rev 12 people in loan repayment over the same time period. I lined the bars for defaulters by year with the tallest bar on the left. Then to compare, I lined the bars for the number in repayment by year in the same order. They do not match, so there are other factors causing the increased number of loan defaults. It is not as simple as the number of defaulters increasing as the number in repayment increases (Attachment 1). D2. Define the output performance standard. Date: 5/22/12 Our department would like to reduce next year s cohort default rate as reported by NSLDS by one standard deviation. This is a one-sided spec for reduction only. The project passed the RUMBA questions 5 for 5 (Attachment 2). D3. Describe the process. Date: 7/1/12 Required tools: SIPOC, Detailed process map I used a SIPOC worksheet to define my process, inputs, outputs, suppliers, and customers (Attachment 3). Process Students accept loans for educational expenses, then drop out of school or graduate. The loan recipient enters repayment of their loans and after two years if they stop paying, they appear on the default list. NSLDS provides the report of loan defaulters and our official default rate to our data area. Purdue would have severe ramifications if the cohort default rate were to increase drastically. I will request the default list from the data area, convert it to a flat file, Query for student ID s, pull all financial aid & student information using Cognos, combine that data with the default list. The remainder of the project will be to analyze and identify similar characteristics using the Cognos and Banner Systems. Inputs The default list, Cognos for data retrieval, Banner to confirm Cognos reports and FAFSA information, TIME to complete the project. Later, I will be examining characteristics such as gender, age, race, college, SAP status, full time vs. part time, degree vs. dropout, EFC, GPA, etc. (BRAINSTORMED with team Attachment 4). Outputs The information from NSLDS and the at risk populations identified by common defaulter characteristics. Suppliers NSLDS, data area, champion, team Customers Purdue University (rankings among the Big Ten & Peers), The Division of Financial Aid, and Purdue students (loan borrowers). M1. Validate the measuring system. Date:8/23/12 Required tools: Gage R&R/Attribute Agreement Analysis I will use existing documents (ten years of the default reports). It is in a form that can be used easily (I have converted the flat file to an Excel spreadsheet, and downloaded the Department of Education s Cohort Default Rate Guide). It is the most current data available. It has a consistent collection time each year. The data is representative. The data comes from a federal agency so repeatability and reproducibility will be difficult to test. 3
rev 12 I used the Attribute Agreement Analysis to check the last date of attendance for the students on the reports to confirm cohort reproducibility. The Cohort was correctly chosen with all students selected not having enrollment past the expected date. I completed two reports to check for enrollment and used Minitab to confirm the results (Attachment 5). M2. Establish current process capability for the output. Date:8/23/12 Required tools: Process capability, Control chart To show where we were before the project, I created a Binomial Process Capability Analysis of Defaulters using Minitab showing the number of defaulters and the number entering repayment over the years (Attachment 6). The process is out of control as seen by the points falling outside of the control limits. M3. Determine project objectives. Date:8/23/12 I determined the project objectives based on the results of M2. We know that there is more to the story, there is not a direct correlation between the number of defaulters increasing as the number in repayment increases. The 50-90 Rule does not work in this case, it would not be realistic to reduce the number of defaulters by 90%. We have agreed that a slight reduction (approximately 0.5%) is more reasonable. A1. Identify and list all potential causes (inputs). Date:9/28/12 Required tools: Process map, Brainstorming, Fishbone diagram, Cause and effect matrix, Potential X matrix In the root cause analysis, I will identify potential student characteristics that may have an impact on loan defaults. My inputs X s will be determined by the most popular characteristics of the 2010 Two Year Cohort Defaulters. I have used a Fishbone diagram (Attachment 7). My brainstorming attachment (#4) is referenced in Section D3. A2. Screen potential causes. Date:10/1/12 Required tools: See A1 I will try to use a Cause & Effect Matrix to prioritize the buckets of Xs (specific characteristics) to a potential reduction in the Cohort Default Rate. The Cause & Effect Matrix is not an ideal tool for this project due to having only one output; a default on a loan, but I have attached one to show understanding of the tool (Attachment 8). I have used practice data to this point as the actual default results for the 2010 Two-Year Cohort Defaulters were released at the end of September. These processes will be redone using actual data and will take several months. 4
rev 12 A3. Determine the f(x) key input variable(s) Date:10/1/12 Required tools: One factor at a time experiment I will try to use a Potential X Matrix to analyze the remaining Xs (specific characteristics) to further drill down to a potential reduction in the Cohort Default Rate. (Attachment 9). I have used practice data to this point as the actual default results were released at the end of September. These processes will be redone using actual data and will take several months to complete. I-1. Establish operating tolerances for key inputs and output. Date:10/1/12 Describe how the solution was derived and how it will be implemented. Describe the operating tolerances and how they were selected. The team will brainstorm and build a Solution Prioritization Matrix in December to determine our action. Ideas such as 1)New method of borrower education? 2)Contact with advisors? 3)Communication with Dept of Ed? 4)Limiting loan awards to upper student levels? (Attachment 10). I-2. Re-evaluate the measuring system. Required tools: Gage R&R/Attribute Agreement Analysis Date: FUTURE In October of 2013, I will use an Attribute Agreement Analysis to confirm the last date of attendance with the new Cohort Default List, as I did in 2012. Once I am confident that the population is accurate, I will analyze the student characteristics of the defaulters. Using a Cause & Effect Matrix and a Potential X Matrix, I will ensure that we are still focusing our efforts on the right students to keep them from defaulting in the future. Most of our efforts for current students will not be evident for many years. If we target enhanced loan entrance awareness, financial literacy education, and changes to academic counseling for freshman we will not see the results for seven years (four years until they graduate and three additional years until they would appear on the default list). Changes to loan servicing would have a faster impact on our default rate, but may be out of our control. I-3. Establish final capability for key input(s) and the output. Date: FUTURE Required tools: Process capability, Control chart I will not have the results at the end of the Six Sigma time frame. We are hoping to see an impact before the 2013 and 2014 release of default rates, but will not likely see large changes for seven years. C1. Implement process controls for the key inputs. Date: FUTURE Required tool: Four levels of control, error proofing List controls including error proofing. Utilize highest level of control possible. Categorize controls 5
rev 12 0, 1, 2, or 3. Our process control will be Level 1 Our department needs to continue to monitor the default rate and adjust the solution as needed. If we see increases in the default rate we will need to screen for new X s and adjust accordingly. We need to remain dedicated to reallocating resources where necessary and using Six Sigma tools will validate the cost and effort. Follow-up to ensure effectiveness. Date: FUTURE We will continue to analyze the characteristics of future loan defaulters every year. 6
Comparative Pareto Charts Comparing the growing number of students in default to the number of students in loan repayment Attachment 1 The Division of Financial Aid receives an annual cohort default rate from NSLDS. There would be repurcussions if our cohort default rate were to skyrocket, but to date we have not considered any causes for defaulting on loans. Before beginning a project, I wanted to rule out that the number of people in default on their loans was simply a reflection of the number of people in default. Purdue West Lafayette 2009 2008 2007 2006 2005 2004 Number in Default 87 51 81 67 68 39 Number in Repayment 5,081 4,138 4,914 7,471 8,220 4,495 Rate 1.7 1.2 1.6 0.8 0.8 0.8 Sorted for Number in Default 2009 2007 2005 2006 2008 2004 Number in Default 87 81 68 67 51 39 Number in Repayment 5,081 4,914 8,220 7,471 4,138 4,495 Rate 1.7 1.6 0.8 0.8 1.2 0.8 Sorted Number in Repayment to Match Number in Default 2009 2007 2005 2006 2008 2004 Number in Repayment 5,081 4,914 8,220 7,471 4,138 4,495 Number in Default 87 81 68 67 51 39 Rate 1.7 1.6 0.8 0.8 1.2 0.8 Number in Default 2004 2009 Number in Repayment 2004 2009 100 9,000 80 8,000 60 40 20 7,000 6,000 5,000 4,000 0 2009 2007 2005 2006 2008 2004 3,000 2009 2007 2005 2006 2008 2004 Number in Default Number in Repayment I used the Pareto charts to compare two sets of data over 6 years (the number of loan defaulters and the number in repayment). This disproves that the number of loan defaulters is simply in line with the number of people in loan repayment.
RUMBA Reduce next year's Two Year Cohort Default Rate by one standard deviation Attachment 2 Reasonable? Understandable? Measurable? Believable? I believe this project is realistic and necessary. The project and the goals are defined in the problem statement. The default rate is measurable with a clear formula that is provided by the Department of Education. The goal is believable and attainable if we invest the time needed Attainable? The goal is believable and attainable if we invest the time needed Our draft cohort default rate increased more than expected this year. We need a process to prevent students from appearing on the loan default list (lowering our cohort default rate). To this point, we have not identified common characteristics of these defaulters; we simply receive the names and SSNs for the borrowers who have defaulted. I have the ability to run the names through our Student and Financial aid databases (using Cognos) to add as many characteristics as needed. I will use the characteristics that appear most often in the defaulters to identify an at-risk population. If we can focus education/intervention on the identified at-risk populations while they are still in school, we could reduce the number of borrowers who will default (lowering the cohort default rate). I will also attempt to identify loan servicer issues that may have an impact on the default rate.
SIPOC Loan Default Process Attachment 3 SUPPLIERS INPUTS PROCESS OUTPUTS CUSTOMERS List Suppliers, internal and external. List Inputs to Process: Data, information, materials, manpower, environment, equipment, resources. Map Process Below. Do not get led by the form! List as many steps as necessary to descriibe the List Outputs to Process: Data, information, materials, manpower, environment, equipment, resources. MACRO process. The purpose of his exercise is to Lendors Student Data from Cognos Lendor Info from NSLDS Students examine scope, to list primary Financial Aid Data from inputs and outputs, and to list NSLDS at risk populations Cognos high-level customer expectations. Students Loan App Info Purdue Financial Aid Office Banner DFA Cognos Big Ten DFA Time FAFSA info List customers, internal and external. Department of Education (they publish rates) Students accept loans for educational expenses Those students drop out of school or graduate from Purdue The loan recipient enters repayment of their loans If after two years in repayment the recipient stops paying their loan, they are in default NSLDS provides a report of loan defaulters (and default rate) to Purdue annually Purdue would have severe ramifications if the cohort default rate were to increase drastically # of defaulters # of borrowers = default rate
Brainstorming for potential inputs on SIPOC Attachment 4 Residency Indiana Resident Non residents Demographics Fin Aid Characteristics Loan Characteristics College/Advising Other Gender Age Race Marital Status Number of Dependents First Generation Student (keep in mind for future years) Legacy? SAP Status EFC SSACI Levels Parent Income Student Income Dependency Number of Budget Adjustments Pell Eligible Indicator Education Tax Credits Vet Benefits Child Support Paid Amount unmet need Family Size Number in Household Number in College SWT? Merit Scholar? Work Study Other Purdue Employment Funds Lender Servicer Type of Loan Amount of Loan (set levels) College Major GPA Advisor Name Dean's List CODO (undecided in the beginning) Total Credits Student Level Part Time vs. Full Time Degree vs. Drop Out SAT/ACT High School GPA Course Credits Earned in HS Extra Curricular Coop/Frat/Sorr
Attachment 5 8/23/2012 2:40:45 PM Attribute Agreement Analysis for Last Date of Attendance Between Appraisers (two Cognos reports by Jennifer Howells) Assessment Agreement # Inspected # Matched Percent 95% CI 5567 5567 100.00 (99.95, 100.00) # Matched: All appraisers' assessments agree with each other. Fleiss' Kappa Statistics Identical assessments. Cannot compute kappa. * NOTE * Single trial within each appraiser. No percentage of assessment agreement within appraiser is plotted.
Attachment 6
The Fishbone Diagram Attachment 7 College/Advising Demographics Residency College Gender Indiana Resident Major Age Non residents GPA Race Advisor Name Marital Status Dean's List First Generation Student CODO (undecided in the beginning) Legacy? Total Credits Number of Dependents Student Level Part Time vs. Full Time Degree vs. Drop Out Cohort Default Rate SAP Status EFC SSACI Levels Parent Income SAT/ACT Lender Student Income High School GPA Servicer Dependency Course Credits Earned in HS Type of Loan # of Budget Adjusts Extra Curricular Amount of Loan (set levels) Pell Eligible Indicator Coop/Frat/Sorr Loan Characteristics Education Tax Credits Other Vet Benefits Child Support Paid Amount unmet need Family Size Number in Household Number in College SWT? Merit Scholar? Work Study Other Purdue Employment Funds Fin Aid Characteristics
The Cause & Effect Matrix Attachment 8 This will be updated as further examintation of defaulters continues. Rating of Importance 1 The stength of the relationship of each Input will be rated 0, 1, 3 or 9 based on the number of defaulters with those characteristics. Default List Overall Rating Indiana Resident 3 3 Non residents 3 3 Male 3 3 Female 3 3 URM 1 1 Married 1 1 Single 1 1 First Generation Student (keep in mind for future years) 1 1 Legacy? 0 0 Negative SAP 9 9 zero EFC 3 3 SSACI Level 1 1 Dependent 1 1 Budget Adjustments >1 9 9 Pell Eligible Indicator 1 1 Child Support Paid 1 1 Unmet need 9 9 Number in Household >5 0 0 Number in College >2 0 0 SWT 0 0 Merit Scholar 1 1 Work Study 1 1 Other Purdue Employment 1 1 DL Servicer 9 9 FFEL Servicer 9 9 Type of Loan 9 9 Amount of Loan>10,000 9 9 Amount of Loan>20,000 9 9 Amount of Loan>30,000 9 9 College 3 3 Major 3 3 High GPA 3 3 Low GPA 3 3 Dean's List 1 1 CODO (undecided in the beginning) 1 1 Total Credits 1 1 Student Level 1 1 Part Time 1 1 Degree vs. Drop Out 9 9 SAT/ACT 0 0 Course Credits Earned in HS 0 0
Potential X Matrix Attachment 9 This will be updated as further examintation of defaulters continues replacing values with recent data. (X) Rating from C&E Measurement, Technique and Units Currently Collected? Statistical Test 1 Indiana Resident 3 Attribute Yes/No Yes Chi-square 2 Non residents 3 Attribute Yes/No Yes Chi-square 3 Male 3 Attribute Yes/No Yes Chi-square 4 Female 3 Attribute Yes/No Yes Chi-square 5 URM 1 Attribute Yes/No Yes Chi-square 6 Married 1 Attribute Yes/No Yes Chi-square 7 Single 1 Attribute Yes/No Yes Chi-square 8 First Generation Student (keep in mind for future years) 1 Attribute Yes/No Yes Chi-square 9 Legacy? 0 Attribute Yes/No Yes Chi-square 10 Negative SAP 9 Attribute Yes/No Yes Chi-square 11 zero EFC 3 Attribute Yes/No Yes Chi-square 12 SSACI Level 1 Attribute Yes/No Yes Chi-square 13 Dependent 1 Attribute Yes/No Yes Chi-square 14 Budget Adjustments >1 9 Attribute Yes/No Yes Chi-square 15 Pell Eligible Indicator 1 Attribute Yes/No Yes Chi-square 16 Child Support Paid 1 Attribute Yes/No Yes Chi-square 17 Unmet need 9 Attribute Yes/No Yes Chi-square 18 Number in Household >5 0 Attribute Yes/No Yes Chi-square 19 Number in College >2 0 Attribute Yes/No Yes Chi-square 20 SWT 0 Attribute Yes/No Yes Chi-square 21 Merit Scholar 1 Attribute Yes/No Yes Chi-square 22 Work Study 1 Attribute Yes/No Yes Chi-square 23 Other Purdue Employment 1 Attribute Yes/No Yes Chi-square 24 DL Servicer 9 Attribute Yes/No Yes Chi-square 25 FFEL Servicer 9 Attribute Yes/No Yes Chi-square 26 Type of Loan 9 Attribute Yes/No Yes Chi-square 27 Amount of Loan>10,000 9 Attribute Yes/No Yes Chi-square 28 Amount of Loan>20,000 9 Attribute Yes/No Yes Chi-square 29 Amount of Loan>30,000 9 Attribute Yes/No Yes Chi-square 30 College 3 Attribute Yes/No Yes Chi-square 31 Major 3 Attribute Yes/No Yes Chi-square 32 High GPA 3 Attribute Yes/No Yes Chi-square 33 Low GPA 3 Attribute Yes/No Yes Chi-square 34 Dean's List 1 Attribute Yes/No Yes Chi-square 35 CODO (undecided in the beginning) 1 Attribute Yes/No Yes Chi-square 36 Total Credits 1 Attribute Yes/No Yes Chi-square 37 Student Level 1 Attribute Yes/No Yes Chi-square 38 Part Time 1 Attribute Yes/No Yes Chi-square 39 Degree vs. Drop Out 9 Attribute Yes/No Yes Chi-square 40 SAT/ACT 0 Attribute Yes/No Yes Chi-square 41 Course Credits Earned in HS 0 Attribute Yes/No Yes Chi-square
Solution Prioritization Matrix Attachment 10 Our team will meet in December after looking at the results of the analysis to populate this matrix. Solution Selection Matrix Rank each solution from 1 10 based on the criteria in the left hand column 1= very low, 10=very high Solution Number Criteria for solution selection #1 #2 #3 #4 #5 #6 SUM Reduction in Default Rate 0 Reduction in Defaulters with Fin Aid Characteristics 0 Reduction in Defaulters with Advising Characteristics 0 Reduction in Defaulters with Loan Characteristics 0 Other 0 SUM 0 0 0 0 0 0 0 Solution Identification DRAFT IDEAS #1 New Method of Borrower Education loan counseling #2 Finacial Literacy Courses for Borrowers #3 Training for Advisors #4 Communication with the Department of Ed #5 Changes to Loan Awarding Procedures #6 Other Solution Description More information, education and hoops to jump through before the loan is disbursed. Students that fall into at risk groups could be referred to online courses. Encourage students to finish on time, not take unnecessary courses, etc. Point out concerns about Direct Loan Servicing. Look at not awarding loans to Cost of Education until a student is on target to graduate (no SAP issues). Ideas to be generated in December.