Note from ProspectResearch.com: Kevin MacDonell from Dalhousie University kindly agreed to write a guest post showcasing his predictive modeling and prospect research knowledge. You may visit his blog at http://cooldata.wordpress.com.
Guest Post from Kevin MacDonell:
Lacking databases full of alumni information, nonprofits can’t do predictive modeling at the same level as their peers in higher education. I’ve heard this frequently from nonprofits, and for a while I even believed it.
About a year ago, I gave a presentation on predictive modeling to fundraisers who work for nonprofits in my city. Probably no one in the audience was prepared to carry out anything I showed them. Some worked for organizations too small to have any need, and some might have had a need but hadn’t collected the data. Others had data, but it was a mess. Several had databases, but they contained donation data only: Entirely transactional in nature, with no indicators of affinity to one’s organization or cause.
I personally work in the Annual Giving office of a large university, sitting on a rich database of our interactions with more than 100,000 living alumni from their student days to the present. In other words, I may as well have been visiting from another planet.
But another conference, and another presentation, proved me wrong. Last fall, I co-presented a workshop with a data mining consultant in Toronto for members of the Canada Chapter of the Association of Professional Researchers for Advancement (APRA-Canada). We demonstrated the building of two predictive models, one for a university alumni database, and one for a performing arts nonprofit.
Our two “data partners” each provided an anonymized sample of constituent data for analysis. The consultant worked primarily with the University of Manitoba, and I worked with the Toronto Symphony Orchestra. They gave us the data, and we gave them score sets for their database constituents.
The Symphony’s data was an eye-opener. There were plenty of non-donors in the file — people who had either purchased single tickets for performances or had subscribed to one or more concert series. The data didn’t come from a neat, central database; it had to be brought together in my stats software. This was easily done, though, because every individual had a unique ID.
I found this modeling project novel and fascinating. The top three predictors for giving to the Symphony were related to concert series subscriptions: The number of years a person had subscribed, the total dollars they spent on subscriptions and, most interesting, the type of subscription. This third variable took some work, but yielded big results. It turns out that if a person has subscribed to at least one concert series that is traditional and “serious,” i.e. that is part of a Masters series or devoted to a specific composer, they are more likely to be a donor. People who have subscribed exclusively to lighter fare (pops concerts and more recent works) were still more likely to be donors than non-subscribers, but less likely than the “serious” concert-goers.
My experience with the Symphony proved to me that any organization of a certain size can profit from mining their data like any institution of higher education. The TSO model differed from what I’m used to, but had as good a fit as any model created for university advancement.
What nonprofits lack in historical, engagement-related data, they can make up by gathering current data. Think of survey data (which obviously has to be non-anonymous), volunteer involvement, purchases of tickets and merchandise, and event attendance. Organizations that are membership-based have an opportunity to gather demographic data. Having home addresses opens the door to buying external data by ZIP code (postal code in Canada).
Nonprofits may have a long way to go generally, but I would encourage them to do two things that the Toronto Symphony Orchestra has done.
- Enlarge your data set to capture the full spectrum of your constituency (members, volunteers and trustees, attendees, ticket buyers, and so on).
- Ensure that you can connect an individual’s involvement behaviour (say, ticket buying) with the same individual’s giving behaviour, via the use of unique IDs.