Know your Constituents Through Effective Data Mining

1
Know your Constituents Through Effective Data
Mining

• Karen Matheson
• MR Strategic Services
• CASE District VIII Conference, February 23, 2006

2
Overview

• Definition
• Benefits
• Data Organization
• Data Collection Methods
• Data Profiling vs Data Modeling
• RFM Analysis
• Data Modeling Case Study
• Questions

3
Definition of Data Mining
Looking for meaning and patterns within the
information in your organizations database so
you can unlock potential value - Jennifer
Shimp-Bowerman, Bucknell University
4
Whats So Hot About Data Mining?

• It provides a way to organize your fundraising
  time and resources to cultivate the most
  promising major gift candidates
• With data mining, youre able to identify traits
  and characteristics of your top donors
• It allows you to identify the events and
  activities that most contribute to developing
  positive relationships with your University or
  College
• Its a more reliable way of assessing
  constituents than subjective impressions

5
Data Mining Example

• University of Oregon - major gift predictive
  model and test outcome

6
How Can You Make a Numerical Judgment About a
Person?

• Its all about the data collection and storage
• Successful data mining involves having data to
  from which to mine and being able to
  quantitatively analyze it

7
Data Storage Show Me the Data!

• Best Database Practices
• Centralize all advancement constituent data into
  one database
• Use a database with which you can create custom
  fields
• Coordinate yearly data dumps with organizations
  across campus
• Graduation Records
• Scholarships
• Student Life/Activities
• Individual Schools and Colleges
• Use the central database to track events and
  constituent contact information
• Build web forms to make the database more user
  friendly

8
My Shoes Are in the Dresser and My Clothes Are on
the Floor!

• Its extremely important for the data mining
  process that data is stored in the correct place
• Data in the wrong places could lead to the
  following problems
• Data could be difficult to find
• Data could be in the wrong format
• Data could be easily entered incorrectly, leaving
  hours of data clean-up
• Data could be completely worthless

9
Common Data Storage Errors

• Inconsistent data entry methods
• Using place holders with data for empty fields
• Storing numeric variables in text format
• Storing variables in the wrong fields

10
Common Data Storage Error Examples

• Storing financial information in a text field
  allows all types of formatting on a variable that
  should only allow a currency value
• Data entry errors such as leaving off the dollar
  sign or substituting a period for a comma can
  lead to hours of data clean-up
• How do we quantify 1,000,000 or lt500,000?

11
Common Data Storage Error Examples

• The Alumni Lunch does not belong in the student
  activities variable because it was not an
  activity that a student could have participated
  in
• Because the student activities field contains
  both student and non-student activities, we are
  not able to tally the number of student
  activities as a way of looking at a constituents
  record
• A data miner, database manager, or database user
  might not know or remember which activities were
  student activities

The Alumni 2003 Lunch was an event for Alumni
Association members (not current students)
12
What Types of Data Are Useful For Data Mining?

• Constituents history with the institution
• Graduation year
• Student activities
• Current involvement
• Demographics
• Age
• Gender
• Income
• Households with children
• Professional and Community Involvement
• Board memberships
• Club/Activity memberships
• Donor history

13
Where Do I Find All This Data?

• Alumni Database
• Donor Database
• Individual Schools and Colleges
• Online/Offline Surveys
• Census Data
• Voter File Data
• Data Vendors

14
Overview

• Definition
• Benefits
• Data Organization
• Data Collection Methods
• Data Profiling vs Data Modeling
• RFM Analysis
• Data Modeling Case Study
• Questions

15
Data Profiling vs. Data Modeling

• Data Profiling
• Can be used to describe segments from your
  organizations database
• All medical school alumni
• Donors who have made gifts of 25,000 or more
• Uses descriptive statistics such as means
  (averages), medians (mid-points), and modes (most
  common data points)
• Data Modeling
• Uses inferential statistics to draw inferences
  (conclusions)
• Uses a random sample from your organizations
  database

16
Donor Profiling Steps

• Figure out the set of constituents we want to
  study
• Example All donors who made past gifts totaling
  25,000 to the music school
• What demographic information do we want to look
  at?
• Example Age, Gender, Average Income
• What constituent information do we want to
  include?
• Example Type of degree, event attendee
• Analyze demographic values according to giving
  factors such as
• Gift total for group, average lifetime giving,
  median gift total, last campaign giving

17
Demographic Profile Age

• Examining the age of the 1,000 25,000 music
  school donors

18
Recency, Frequency, and Monetary Value (RFM)
Analysis
RFM is a useful strategy that can be used to
predict the future value of donors and their
likelihood to respond to direct mail marketing
efforts.

• A scoring or ratings system based on
• How recently has a donor contributed?
• How often does a donor give?
• How much does a donor give?

19
Limitations of RFM

• Works best on larger databases
• It only rates donors
• Not designed to rate non-donors
• It is a fluid rating system that needs to be
  continually updated

20
RFM Scoring System

• RFM analysis produces "scores" that rank donors
  relative to each other for the likelihood that
  they will repeat whatever action is being
  profiled.
• Sample Scoring System
• Each category Recency, Frequency, Monetary
  Value receives a score of 1 to 5 based on their
  quintile ranking
• Bottom 20 of each category would receive a 1,
  top 20 would receive a 5

21
Predictive Modeling

• Predictive modeling uses inferential statistics
  to predict future behavior
• It is a useful tool for segmenting and ranking
  large amounts of data in order to
• Identify annual/major/planned giving prospects
• Ensure strategic planning

22
Before You Even Start

• You should ask the following questions
• What do we want to predict?
• Whats the budget?
• Do I have the statistical ability?
• What other offices/departments need to be
  involved?
• How much time will this take?
• How will this accomplish our goals?
• What does our donor database look like?
• How would we implement the results of a model?

23
University of Oregon Model Example

• What do we want to predict?
• Who is most likely to give a major gift to the
  University
• Whats our budget?
• Didnt have one
• Do I have the statistical ability?
• Sure, I took some statistics courses in college
  and have some dusty text books
• What other offices or departments need to be
  involved?
• Information Services
• Records and Receipting

24
University of Oregon Model Example

• How much time do we have?
• One year
• How will this accomplish our goals?
• Identification of new prospects
• Assist fundraisers with prospect prioritization
• What does our database look like?
• Not pretty, but workable
• How would we implement the results of a model?
• Fill in regional fundraiser appointment schedules
  with top rated prospects
• Conduct research on top rated individuals not
  currently managed

25
Data Modeling Steps

• Obtain random sample
• Split the sample into two and set second half
  aside
• Using the first half of the sample, identify
  variables that affect major giving
• Develop a scale of measurement (model)
• Test model on second half of the sample

26
Obtain Random Sample

• Decide what data youd like to pull out of the
  database so you can test it against the outcome
  you want to predict
• Figure out how many people youll need in your
  sample Id recommend using the sample size
  calculator on this website
• http//www.surveysystem.com/sscalc.htm
• When you get the random sample back, put half of
  the sample aside for testing the model

27
Random Sample
Out of the University of Oregons 300,000 person
database, a random sample of 10,000 individuals
with their subsequent data was drawn
28
Random Sample
29
Variables Galore!

• There are a couple of types of variables were
  going to come across in our sample
• Quantitative Variables
• Variables that are measured as a number for which
  meaningful arithmetic operations make sense
• Examples Height, age, GPA, salary, temperature
• Categorical Variables
• Variables with values that are one of several
  possible categories. Categorical variables have
  no numerical meaning
• Examples gender, political affiliation, field of
  study
• Ordinal Variables
• A special type of categorical variable for which
  the levels can be naturally ordered
• Examples Taco Bell Hot Sauce (medium, hot, fire)

30
Statistics Software

• Slap all of the data from the obtained random
  sample into statistics software
• Excel Data Analysis (Microsoft Office)
• DataDesk (www.datadesk.com)
• SPSS (www.spss.com)
• SAS (www.sas.com)
• Minitab (www.minitab.com)

31
Start Testing All of the Variables

• How do I do that? Heres where those dusty
  statistics text books come in handy
• Correlation
• Use this test for quantitative variables
• Examples age, number of gifts
• Independent Samples t-Test
• Use this test for categorical variables with only
  two categories
• Examples gender, membership in the Greek system
• One-Way ANOVA
• Use this test for categorical variables with more
  than two categories
• Examples different states, taco bell hot sauce
• Linear Regression

32
So We Tested and Found

• Through correlation testing it was found that UO
  major giving was moderately correlated with age,
  number of UO events attended and number of gifts
  and pledges.
• Through t-tests it was found that there were
  significant differences in the major gift
  averages of the following groups Greek
  membership, Alumni Association membership, former
  membership in one or more student activities, and
  current or past membership in one or more post
  college UO volunteer organizations/boards.

33
Identify Scale of Measurement
A simple way of developing a rating system is by
using a nominal scale

• In this method, a one or a zero is assigned for
  inclusion/exclusion of each category included in
  the model
• Add each individuals categories of ones and
  zeros together for a total score

34
Assigning Values for Categorical Data

• In the UO example, an individual would receive a
  score of one for each of the following
• Greek membership
• Alumni Association membership
• Former membership in at least one student
  activity
• Current or past membership in at least one post
  college university volunteer/board membership

35
Assigning Values for Quantitative Data

• Split the Quantitative Variable into groups
• a. Example Age groups - 45-54, 55-64
• Determine of the value against the total amount
  of individuals in the random sample (A)
• 3. Determine of how many in the value are
  20,000 donors (B)
• 4. Assign a score of 1 to those values in which
  the B is higher than the A
• 5. Assign a score of 0 to those values in which
  the B is lower than the A

36
Quantitative Data Example
37
Testing Predictive Model
After assigning values to all predictor
variables, add up the scores for each individual
and report on your findings. UO Example
38
Testing Predictive Model
39
Implement Predictive Model

• Make predictive model rating easy to understand
• Example A great major donor prospects, B
  good major donor prospects, C unlikely major
  donor prospects, D extremely unlikely major
  donor prospects
• Present model to staff
• Keep presentation simple, avoid presenting
  confusing jargon leadership and/or fundraisers
  typically just want to know predictors how they
  impact giving
• Work with Information Services department to put
  rating where staff can easily access it
• Test model in a year or two to evaluate how well
  it worked

40
Questions?

• Contact Information
• Karen Matheson
• MR Strategic Services
• 615 2nd Avenue Suite 550
• Seattle, WA 98104
• [phone]
• kmatheson_at_mrss.com

41
Sources/Bibliography
Lisa Howley and Karen Matheson. Data Mining
Using Queries and Statistics to Discover New
Donors (2005). APRA International
Conference. StatSoft, Inc. (2004). Electronic
Statistics Textbook. Tulsa, OK StatSoft.
lthttp//www.statsoft.com/textbook/stathome.htmlgt A
rchambault, Susan (2000). Psychology Department,
Wellesley College. Wellesley, MA.
lthttp//www.wellesley.edu/Psychology/Psych205/inde
pttest.htmlgt Information Technology Services
(2002). The University of Texas at Austin.
lthttp//www.utexas.edu/its/rc/tutorials/stat/spss/
spss2/gt Preston, Scott (2005). Oswego State
University of New York. lthttp//www.utexas.edu/it
s/rc/tutorials/stat/spss/spss2/gt Howell, David
(2002). Statistical Methods for Psychology.
Wadsworth Group Pacific Grove, CA.