In the second half of part 1 of this series, we looked at the type of projects that donors prefer by studying the projects that are more likely to become fully funded. An equally important factor to consider is return donorship. About one in three donors make subsequent donations within one year of their first. Having returning donors mean that donors were happy about the impacts they made, and that future projects are more likely to be funded. In part 3, we look at factors affecting whether a first-time donor would return and continue to contribute to DonorsChoose.
For the purpose of this analysis, new donors are considered to have “returned” if they made new contributions on DonorsChoose within the next 365 days.
Percentage return donorship have declined a little in the past few years, dipping down from 35% in 2008 to a little less than 30% in 2010. (There is some abnormality in the year 2006; it is unclear why this is the case.)
The chart below shows various factors and how they affects the odds of return donorship. The factor that affects return donorship the most is whether or not the donors themselves are teachers: teachers have a 200% increased odds of returning to DonorsChoose. This is hardly surprising, since teachers know best the impact made by supporting classroom projects.
Other than being return donors themselves, teachers have another way of increasing the odds of return donorship, and that is by mailing “thank you” packets to those who supported their projects. When donors receive “thank you” packets for the first project they donate to, their odds of making subsequent donations increase by about 95%. Donors who made their first donation via the “Giving” page is also more likely to return.
The bottom item in the above chart is slightly surprising: if the first project that a donor supported eventually becomes fully funded, then the donor is less likely to return. This can be interpreted a little differently: perhaps when the project a donor supported isn’t fully funded, the donor responds by trying again and donating to a different project.
Those are not the only factors affecting return donorship. The amount of money donors gave and their payment methods also affect whether or not they would return: the more money the donor gave, the less likely they are to return, and payment via paypal is most correlated with return donorship. Donors who first contributed with a gift card (or via some other method where the actualy payment was already made to DonorsChoose) are less likely to return.
School metro and poverty level affect return donorship almost the same way as they affected project completion: rural schools fare the worst in terms of getting first-time donors back, whereas urban and suburban schools do equally well. Schools with higher poverty levels do better, and schools with unknown poverty levels are surprisingly good at getting donors back.
And once again, music & the arts continues to be donors’ favourite subjects. Donors who first donated to a project in music & the arts are slightly more likely to return, as are donors who first chose a project in math & sciences. Donors whose first projects were related to health & sports are the least likely to return.
The target grade level of donors’ first projects affect return donorship slightly less than the other factors we’ve mentioned: donors who chose to fund projects geared towards younger students are slightly more likely to return than those who funded projects geared toward high school students.
Lastly, teacher prefix affects return donorship slightly as well. Although we’ve seen that donors are less likely to fully fund projects posted by a “Mrs”, donors who first chose to give money to such a project are slightly more likely to return. Finally a win for the Mrs!
Most of the analysis was done on projects posted between 2004 and 2010. I used R and ggplot2 to generate graphs, and python for data manipulation. To find factors affecting return donorship I used logistic regression on donations data between 2005 and 2010. For most parts of the analysis, causation should not be implied — even though I wasn’t too careful with the language when causation seemed extremely likely. All the code used for the above analysis is in github. (So if I did something wrong, please let me know.)