Data Mining and Alumni Association Membership

hideousbotanistData Management

Nov 20, 2013 (3 years and 10 months ago)

78 views

Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 1


Data Mining and Alumni Association Membership
By Peter B. Wylie and John Sammis
It was a stifling hot day in Chicago in July of 2007. I was at the CASE Summit looking out at an audience
of some 100 folks clearly older than the 30-somethings John Sammis and I regularly work with. After
being introduced, I paced back and forth in front of them, not uttering a word for a good 20 seconds. Then
I stopped abruptly, peered out at them, and said:
“I work with the young professionals who report to you guys. Heads of the annual fund. Directors of
prospect research and alumni relations. And those overworked and underpaid IT folks in advancement
services.
“With them, I feel I have to be pretty nice. After all, they’re in the trenches working away every day
trying to keep the fundraising machinery of your schools running as smoothly as they can. But …”
I paused for a little dramatic effect as frowns and grimaces crawled onto their faces.
“But with you guys I can be more blunt and candid. That’s part of what you get paid for.”
The frowns and grimaces hardened.
“Here’s the deal. If your counterparts in the private sector—and I’m talking mostly people in direct
marketing—ignored the meager information they store on their customers the way most of you ignore the
vast information you store on your alums, ya know what would happen? They’d go out of business in six
months.”
Now I was getting some blank stares and more than a few expressions that said, “Who is this arrogant
geezer who has the gall to talk to us like this?” But I was also getting more than a few nods of approval
and agreement.
If John Sammis and I had both been giving that talk, I would have taken a softer approach. John is a
gentler, more diplomatic fellow than I who has successfully filed some of the crustiness off my
curmudgeon side. However, I know he agrees wholeheartedly with me that most schools do

ignore too
much of their alumni data when they go about making appeals, whether for the annual fund or major
Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 2

giving or planned giving or what have you. In fact, the two of us spend gobs of time empirically
demonstrating this fact with the data mining and predictive modeling work we do.
Fortunately, we think the tide is beginning to turn. More and more schools are starting to pay attention to
their alumni data as a way to save money and generate more revenue on appeals. (See “
Does Data Mining
Really Work for Higher Education Fundraising
?”).But is that the case with schools with dues-based
alumni associations? Are they using data mining and predictive modeling to thin down the hundreds of
thousands of calls they make each year trying to sign up new members? We don’t think so. We haven’t
done a scientific survey to support our opinion, but we’ve talked to enough schools with dues-based
associations to know that such data driven decision-making is barely on their radar screens.
Let’s say we’re right. Then a reasonable question becomes: Would data mining and predictive modeling
be useful in identifying alums who are more likely than their peers to respond positively to an invitation
to join? We think the evidence offered in this paper supports at least a tentative “yes” in answer to this
question.
Here’s what we’ll cover:
• The data we used to answer the question
• What the data showed us
• Some conclusions we’ve drawn from the data
The Data We Used
Before asking any schools to participate in this study, we needed to make a decision about what data to
request. Our hunch was that the basic variables we always look at as possible predictors (when we build
experimental sample giving models) made good sense:
• Whether or not a home phone was listed for the alum in the database
• Whether or not a business phone was listed
• Whether or not any kind of an e-mail address was listed
• Whatever code for the marital status field was listed for the alum in the database
• The alum’s preferred year of graduation
But building a model requires an outcome variable, sometimes called a “dependent” variable. More
simply put, it’s the variable you’re trying to predict. To build giving models, we almost always ask for the
lifetime hard credit dollars given by each alum. With this study, of course, we weren’t trying to predict
giving; predicting alumni membership was our goal. While alumni membership sounds concrete (an alum

Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 3

is a dues-paying member or not), we found things to be a bit more complicated when we started asking
questions. For example, alums can be classified as:
• Never having been a dues-paying member
• A lifetime member where one large fee takes care of all the dues
• A lapsed member (not an active member but someone who may have been an active member
several times over the course of a decade or more)
• An active member whose dues are up to date but who may or may not be a member next year
To arrive at a common definition of alumni membership across schools, we decided on this: An alum was
or was not an active member of the school’s alumni association. Not a perfect definition, but good
enough.
Four public higher education universities spread across the country agreed to provide us with the data for
the study. The graduate and undergraduate enrollment of each school ranged from approximately 4,500 to
27,000. Each school sent us a file that included all their solicitable alums (living and having good
addresses) with fields that allowed us construct the variables described above.
What the Data Showed Us
In this project we did a lot of number crunching and generated a bunch of charts, almost all of which we
found interesting. However, if we show you all those charts, we risk losing you, of having you exclaim,
“Guys, just show us the forest, we don’t need to see every maple, oak, and birch!” So we thinned the
herd. Let’s start with Table 1, whose title speaks for itself.
Table 1. Percentage of Current Active Alumni Association Members for Each of Four Schools
SCHOOL A
(TLDO)
SCHOOL B
(ECU)
SCHOOL C
(CSUMB)
SCHOOL D
(CSUSB)
% current active
alumni association
members
9.9% 3.7% 21.0% 3.1%
% nonmembers 90.1% 96.3% 79.0% 96.9%

Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 4

It’s pretty clear that none of the schools, even School C, has a very high percentage of its alums as active
association members.
Now we’ll take you through the charts for School A, which show the relationship between each of our
candidate predictor variables and alumni membership. (These charts, in the figures that follow, don’t look
a whole lot different from the comparable charts for the other three schools.)
Figure 1. Percentage of Active Members by Home Phone Present for School A








The chart in figure 1 shows that there is a clear relationship between whether or not an alum has a home
phone number listed in the database and alumni membership. Notice that almost 11 percent of those
alums with a home phone listed are active members; only slightly more than 5 percent of alums without a
home phone listed are active members. To put this another way: An alum is more than twice as likely to
be an active member if he or she has a home phone listed than if no home phone number is listed.

Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 5

Figure 2. Percentage of Active Members by Business Phone Present for School A

Figure 2 shows there is also a clear relationship between whether or not an alum has a business phone
number listed in the database and alumni membership. Of those alums with a business phone listed, 11
percent are active members; less than 5 percent of alums without a business phone listed are active
members. Again, we can say it this way: An alum is more than twice as likely to be an active member if
he or she has a business phone listed than if no business phone number is listed.
Figure 3. Percentage of Active Members by Home Email Present for School A

Figure 3 is kind of interesting. It certainly shows there is a clear relationship between whether or not an
alum has a home e-mail listed in the database and alumni membership. But the relationship appears to be
somewhat stronger than it does for home phone and business phone listed. Almost 23 percent of those

Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 6

alums with a business phone listed are active members; less than 9 percent of alums without a business
listed are active members. So an alum is two and a half times more likely to be an active member if he or
she has a home e-mail present listed than if no home e-mail is listed.
Figure 4. Percentage of Active Members by Marital Status Is “Married” for School A

By now we’re pretty sure you’ve gotten the hang of these charts, so we’ll be brief here. An alum whose
marital status code is “Married” is two and a half times more likely to be an active member than an alum
who is not listed as married.
Figure 5. Percentage of Active Members by Class Year Quartile for School A

We found the chart in figure 5 intriguing. We expected to see much the same trend as we see with alum
giving. That is, the longer alums have been out of school, the more total money they’ve given. Therefore,

Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 7

it seemed reasonable to assume that the longer alums had been out of school, the more likely it would be
that they’d be a member of the alumni association. As you can see here, that’s not necessarily the case.
Yes, the oldest 25 percent of alums (those who graduated in 1980 or earlier) have the highest membership
participation rate. But the youngest 25 percent (those who graduated in 1999 or later) have a higher
participation rate than the middle two quartiles (alums who graduated between 1981 and 1998). We found
similarity irregularities in the corresponding charts from the other three schools.
Our next step was to build a very simple score for each school where we expected to see a reasonably
strong relationship between score level and alumni association membership. Here’s the formula we used:
SCORE = Home Phone Present (0/1) + Business Phone Present (0/1) + Home E-mail Present
(0/1) + Marital Status Is “Married” (0/1) + Oldest Grad Class Quartile (0/1) + 1
We used the constant of 1 in the formula to avoid negative zero scores that can be confusing. The formula
may look a bit daunting, but it’s really pretty straightforward. We’ll work through a couple of examples.
Let’s say an alum:
• has a home phone present,
• does not have a business phone present,
• has a home e-mail present,
• has a marital status code other than “married,” and
• is not in the oldest class year quartile.
What’s the alum’s score? We say it’s 3. How did we arrive at that? The alum gets a 1 for having a home
phone present plus a 1 for having a home e-mail present plus a 1 for the constant. That adds up to 3.
Now let’s say an alum:
• has a home phone present,
• has a business phone present,
• has a home e-mail present,
• has a marital status code of “married,” and
• is not in the oldest class year quartile.
What’s the alum’s score? We say it’s 5. The alum gets a 1 for having a home phone present plus a 1 for
having a business phone present plus a 1 for having a home e-mail present plus a 1 for having a marital
status code plus a 1 for the constant. That adds up to 5.
Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 8

We computed this score for every alumni record that each of the four schools sent us. Figures 6–9 below
show the relationship between score level and active alumni membership for the four schools. After figure
6 we’ll offer a detailed interpretation of what we see. For the remaining figures we’ll be briefer.
Figure 6. Percentage of Active Members by Mini Score for School A

The big picture in this chart, of course, is pretty clear: the higher the score, the greater the percentage of
alums who are active members of the association. If we go back to Table 1, we see that the overall
percentage of alums who are active members of the association for School A is 9.9 percent. Up to score
level 3 in figure 6, the participation rates are well below or just at this overall level. Then what do we see
at score levels 5 and 6? A pronounced jump up in participation rates: almost 25 percent for score level 5
and over 46 percent for score level 6. (We’ll talk about the implications of these trends in the next
section.)

Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 9

Figure 7. Percentage of Active Members by Mini Score for School B

Compared to figure 6, this chart may not look all that impressive to you. However, remember that the
overall percentage of alums who are active members of the association for School B is only 3.7 percent.
Therefore, we think the participation rates for score levels 4, 5, and 6 look pretty good.
Figure 8. Percentage of Active Members by Mini Score for School C

What can we say? We’re impressed with this chart.

Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 10

Figure 9. Percentage of Active Members by Mini Score for School D

Remember what we said about the chart in figure 7? If you don’t want to go back and look, we’ll say it
again here:
Compared to figure 6, this chart may not look all that impressive to you. However, remember that
the overall percentage of alums who are active members of the association for School B is only
3.7 percent. Therefore we think the participation rates for score levels 4, 5, and 6 look pretty
good.
Except for substituting “3.1 percent” for “3.7 percent,” we’d say exactly the same thing about chart 9.
Some Conclusions We’ve Drawn from the Data
We’ve drawn three major conclusions from this study:
• There is compelling evidence that some information in alumni databases is strongly related to
active alumni association membership.
• When these pieces of information are combined together into even a very crude score, this
relationship between information and active membership appears even more compelling.
• We think schools with dues paying alumni associations should be using this information to save
money on membership appeals and to substantially up their membership.

Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 11

There is compelling evidence that some information in alumni databases is strongly related to
active alumni association membership.
For this project we looked at only a very few fields (we call them variables). Again, so you don’t have to
go back and hunt for them, they were:
• Whether or not a home phone was listed for the alum in the database
• Whether or not a business phone was listed
• Whether or not any kind of an e-mail address was listed
• Whatever code for the marital status field was listed for the alum in the database
• The alum’s preferred year of graduation
With the exception of preferred year of graduation, each of these variables showed a strong relationship
with active alumni membership. But what about all the other information that is stored in alumni
databases—things like number of reunions attended, state of residence, Greek affiliation as an
undergraduate, number of degrees received from the school, and on and on? We have to believe that some
of these variables will show an even stronger relationship with active alumni membership than the ones
we’ve displayed here.
The only way to find out if we’re right is to start foraging through these databases to hunt down these
predictive gems.
When these pieces of information are combined into even a very crude score, this relationship
between information and active membership appears even more compelling.
If you doubt our contention, take another gander at Charts 6–9. By the way, if we weren’t so concerned
about throwing too much information at you, we’d show you something else. We’d show you some
scores we developed using multiple regression (on these same data) to produce even more impressive
scores than the ones we’ve laid out here. (If anybody would like to see those scores, give us a shout and
we’ll e-mail them to you.)
We think schools with dues-based alumni associations should be using this information to save
money on membership appeals and to substantially increase their membership.

Copyright © 2009 Peter B. Wylie and Data Description Inc. All rights reserved. Used by permission.

Brought to you by CASE Books • www.case.org • 1-800-554-8536 12

Here’s another look at figure 6:
Figure 6. Percentage of Active Members by Mini Score for School A

For each score level, there are more non-active members than there are active members. Let’s say you’re
in charge of the call center where students are making calls into the tens of thousands (maybe more) to
these nonmembers. Where do you think you’re going to get the most bang for your budget buck: calling
non-active members with scores of 1 or 2, or those with scores of 5 and 6?
Yeah, it really is that simple. Of course, when you get into all the logistical details, it doesn’t seem so
simple—and we appreciate that even though we never managed a call center. But conceptually, it is that
simple. You want to be calling non-active members who your internal data say “They look like active
members even though we haven’t yet converted them.”
As always, we appreciate feedback, even if it’s negative and critical and puts us back on our defensive
heels. So let us know what you think of what we’ve said here.
About the Authors
Peter Wylie is a national recognized advancement consultant who has worked with major colleges and
universities and taught many advancement professions how to mine donor data to find simple predictors
of giving. You can reach Peter at pbradwylie@aol.com.
John Sammis is a predictive modeling expert who has helped all types of organizations around the world
analyze and understand their data, including the development organizations of many academic
institutions. You can reach John at jsammis@datadesk.com.