1. Revenue Generation in Hospital Foundations: Neural Network versus Regression Model Recommendations Mary E. Malliaris Loyola University Chicago Maria Pappas Thorek Memorial Foundation, Chicago, IL

2. Problem How should a nonprofit foundation supporting a hospital allocate its resources in order to optimize income? Will a regression model and a neural network model give the same recommendations?

3. Foundation Structure Personnel – board of directors, staff, professional fundraisers Strategy followed to collect funds – fundraising, investing, or mixed Types of events – memorial, athletic, party, giving Spending Categories – the associated hospital, the community, charity care, and research/education

4. Data Form 990 from Guidestar.org for each foundation Variables: Net Assets Amount for Expenses Number of Hospital Beds (an indicator of size) Total Contributions Area of Spending: Hospital, Charity, Community, Research/Education Compensation: Board, Staff, Professional Fundraisers Type of Events: Giving, Party, Athletic, Memorial Target Variable: Revenue

5. Foundations Number per Region and by Size RegionCountHospSizeCount Midwest50Large22 Northeast50MedLarge20 South40Medium72 West42MedSmall43 Small25

6. Models Regression Neural Network – 15 inputs – 1 hidden layer with 15 nodes – 1 output Software: SPSS Predictive Analytics Software

7. Models Accuracy Results RegressionNeural Network Minimum Error-2.21-2.42 Maximum Error18.104.16 Mean Absolute Error1.710.45 Standard Deviation3.370.72 Linear Correlation0.960.99

8. Regression

9. Neural Network

10. Variable Importance Variable importance values in PASW are relative. The sum of the values for all input variables in each model is 1.0. Variable importance is not model accuracy. It compares the importance of each variable in making the prediction, not whether or not the prediction is correct. It is calculated by looking at how much the dependent variable changes when the lowest and highest values of the variable are fed through the model and all other variable values are held constant.

11. Comparative Variable Importance VariableNeural NetworkRegressionAbs Difference Net Assets0.4370.0000.437 Expenses0.2860.6990.413 Contributions0.021 0.001 Beds0.0240.0200.004 Board Comp0.0210.0030.019 Staff Comp0.0430.0340.010 Fundraiser Comp0.0190.0000.019 Giving0.0120.0440.032 Party0.0110.0220.011 Athletic0.0190.0220.003 Memorial0.0250.0230.003 Hospital0.0130.0670.054 Charity0.0250.0000.025 Community0.0200.0170.003 Research0.0230.0300.007

12. Conclusions Regression: – Focus on fundraising events with activities that directly raise money such as auctions and radiothons – Spend money first on the associated hospital Neural Network: – Raise money through personal memorials or naming opportunities – Spend money on charity care