Peer-to-peer fundraising: count the gifts as well as the dollars

Researchers Ike Silver and Deborah Small worked with DonorsChoose, a platform supporting classroom projects in the United States. From August 13 to September 9, 2020, people who donated online saw a request to share the cause. Participants received either a standard sharing request or a request emphasizing the potential benefit to the cause. Repeat donors kept the same message assignment.^[1] The original paper tested the effects of those messages; we ask how the resulting donations and dollars were distributed among participants with a credited referral.

The released dataset contains one row per donor participant. It records the number of downstream donations credited through that participant’s referral link and the dollars those donations brought in. Those are different from the participant’s own gifts. A donation count also does not establish how many different people gave.^[2]

We start with all 77,485 participants to show how many had a credited referral. We then keep the participants with at least one referral donation for both concentration calculations. Someone with no credited referral belongs in the participation context but does not enter the ranked comparison. This keeps the question focused on how results were distributed among people who brought in a gift.

Peer-to-peer fundraising often refers to supporter-created fundraising pages. This was a more specific donor-sharing experiment, involving requests made after a donation. It offers a concrete case for examining referral results, rather than a representative survey of every kind of fundraising campaign.^[3]

The two percentages in this table answer a participation question. They are not the shares of gifts or dollars raised by those participants. In the rest of the article, the relevant groups are the 692 and 783 donors with a credited referral. We keep the message groups separate throughout.

In the standard-request group, the top fifth accounted for 40.70% of referral donations when ranked by donation count. Ranked separately by dollars, the top fifth accounted for 54.47% of referral dollars. In the social-impact-request group, the corresponding shares were 40.63% and 56.51%.^[4]

The practical contrast appears in both groups: the dollar total was more concentrated at this cutoff than the number of gifts. A team reading only its dollar report would see a different distribution from a team counting donations. Neither report replaces the other.

A fifth of a group rarely produces a whole number of people. We rounded the headcount up, so the standard group uses 139 of 692 participants and the social-impact group uses 157 of 783. The table shows those actual shares rather than labeling them as exactly 20%.

Each measure has its own ranking. The people contributing the most dollars need not be the people credited with the most donations. We did not calculate whether the same individuals appeared in both shortlists. We also do not interpret a difference between the two selected groups as an effect of the sharing message; the people who achieved a referral can differ between groups.

No: among participants with a credited referral, the top fifth supplied roughly two fifths of the donations and a little over half of the dollars. Neither measure was close to an 80% share at that cutoff. That is a finding about this historical case and this denominator, not a replacement rule for other fundraisers.

The denominator matters to that answer. A comparison among all participants, including those with no credited referral, would ask a different question. Do not compare these positive-referral-group percentages with a rule quoted for all donors, all campaign participants or all supporters of an organization.

Looking beyond a single cutoff makes the pattern easier to assess. We fixed the following five points before examining the ranked outcomes. The tables show every point, including the amounts behind the percentages. The last row contains the full total for each measure.

For example, the 70 participants at the standard group’s top-10% cutoff accounted for 282 donations when ranked by count. A separately ranked set of 70 accounted for USD 19,699.98. The first comparison uses a total of 1,064 donations; the second uses USD 51,623.89. The measure determines both the ranking and the total against which it is compared.

The groups had different numbers of successful referrers and different totals. Reading each series within its own group avoids turning a larger dollar total into a claim about better fundraising performance. These results also cover a fixed calendar period; participants did not necessarily have equal time to generate referrals.

Another way to read the same distributions is to ask how many ranked participants were needed to reach a share of the observed total. In the standard group, 117 participants accounted for at least half the dollars, while 189 were needed to reach half the donation count. In the social-impact group, those headcounts were 120 and 211.

Reaching 80% of dollars required 47.40% of successful referrers in the standard group and 45.59% in the social-impact group. Reaching 80% of donation counts required larger shares: 69.36% and 70.63%, respectively.

These are the smallest whole-person headcounts that reach each threshold in the recorded results. A threshold can be exceeded by the final contribution. They are not recruitment targets: the table cannot identify who will contribute next time, or what would happen if a particular supporter were absent.

For a business supporting a charity or a creator bringing an audience to a cause, the useful next step is to decide what a successful report should explain. Money matters to the project being funded. Donation counts describe another part of the response. When unique-donor information is available and relevant, that is a separate measure again; it should not be inferred from the number of gifts.

A report can show the campaign’s credited referral dollars alongside the number of referral donations and the number of participants who brought in at least one gift. The concentration calculation then adds a specific question: how much of each result came from the participants at the top of that measure’s ranking?

This distinction can make a planning meeting more useful. If the team’s goal is funding a classroom project, it will need the money total. If it is trying to understand how many donation events its supporters helped generate, it needs the count. Before comparing campaigns, the team should also check that the reporting period and definition of referral credit are comparable.

Concentration is a description of a completed campaign, not a judgment about the value of a supporter. Someone with a small credited result may have contributed time, attention or other help that these fields do not measure. The findings do not justify withdrawing assistance from smaller fundraisers or expecting last campaign’s leading participants to carry the next one.

For the creator and nonprofit preparing their report, one concrete choice is to put two separately ranked columns beside the campaign total: share of donation events and share of dollars. Label the group and reporting period once, then explain why the columns differ. That gives the team a clearer account of what happened without turning a past result into a forecast.

Suppose ten people receive a sharing request in an invented fundraiser. Five have no credited referral donation. The other five produce the figures below. Every number in this example is fictional; it is here to show the calculation in a small group.

Participation is five out of ten, or 50%. For the concentration calculation, we use the five people with a positive donation count. Together, their referrals account for 13 donations and USD 100.

The top fifth of five people is one person. Ranked by donation count, A is first: six of the 13 donations, or 46.15%. Ranked by dollars, C is first: USD 60 of USD 100, or 60%. A and C lead different measures, even though both calculations use the same five-person group.

That example shows why “our top fundraiser” needs a definition. It could mean the person credited with the most donation events or the person whose referrals brought in the most money. These results do not tell us who worked hardest, persuaded people most effectively or reached the largest audience.

The five people with no credited referral are still part of the participation report. Moving them into the concentration ranking would change the question and its denominator. Neither choice should happen silently. The real study uses the same positive-donation-count rule in both groups and for both outcomes.

Copy the following fields into the campaign document your team already uses. Start with existing, consistently defined records. If a field is unavailable, leave it unfilled and describe the gap; a missing record is not evidence of a zero result.

Campaign and period: [name] · [start date] · [end date]. State whether everyone had the same opportunity to generate a credited referral, or whether the report covers a common calendar window.

Referral definition: [what earns credit in this report]. Record which gifts are included and how the report attributes them. Keep a participant’s own donation separate from donations credited through that participant’s referrals.

Participation: [all eligible participants] · [participants with one or more credited referral donations] · [the second count divided by the first]. Name the eligible group explicitly; for example, people receiving a sharing request or people who created a fundraising page.

Recorded outcomes: [total credited referral donations] · [total credited referral dollars] · [currency]. Keep donation events separate from unique donors. Do not combine amounts in different currencies without an explicit, appropriate conversion method.

Concentration: [fixed cutoff] · [actual headcount] · [actual share of the positive-referral group] · [share of donations] · [share of dollars]. Rank the two outcomes separately. Choose the cutoffs before looking for a headline. Multiply the cutoff proportion by the group size, round the headcount up to a whole person and report its actual share of the group.

Review note: [what the team learned] · [what remains outside the available records] · [one reporting or planning decision]. Keep observations about the completed campaign separate from predictions about the next one.

A short review can then focus on concrete questions. Does the chosen measure match the campaign’s objective? Are the comparisons using the same definitions? Which forms of supporter contribution are absent from this report? The worksheet helps make those choices explicit without requiring a new tracking setup.

We used the versioned public ResearchBox 105 archive mirrored on Zenodo. The released file contains 77,485 unique participant identifiers, with 38,621 standard-request participants and 38,864 social-impact-request participants. These counts match the primary paper’s methods and Table 4. A different total printed in another table heading does not match the release.^[1][2]

Before calculating concentration, we fixed the method and checked the file’s schema, participant identifiers, message codes and outcomes. Donation counts had to be complete, nonnegative integers; dollars had to be complete, finite and nonnegative. We retained the released cohort without additional exclusions or outlier removal. The concentration groups contain participants credited with at least one donation.

We required at least 200 participants with a credited referral in each message group before proceeding. That ensured the smallest 5% cutoff included at least ten people. This was a display check, not a claim about statistical precision. Both groups passed; no particular concentration result was required.

Within each group, we sorted donation counts from largest to smallest and did the same separately for dollars. At each preselected percentage, we rounded the participant headcount up, added the contributions of that many participants and divided by the group’s corresponding total. Equal values at a cutoff do not change that sum, regardless of which tied participant is listed first.

We also found the smallest headcount whose cumulative contribution reached at least 50% or 80% of each total. Calculations used exact integers and decimal-derived fractions; displayed percentages are rounded to two decimal places. Dollar amounts remain historical USD. The reproduction files retain the exact fractions behind the rounded display values.

These are descriptive calculations on the released records. The study did not give all participants identical follow-up time, directly observe their shared messages or supply a representative sample of fundraising campaigns. The original experiment’s 30-minute measurement window applies to sharing-option clicks, not to the referral totals analyzed here. We make no causal comparison between the selected successful-referral groups.

The downloads contain the aggregate tables, prospective method and reproduction code, with source attribution. The localized results file also includes the fictional example, clearly identified as invented. Individual participant records and identifiers are not included.