We think these matches are ... mostly attributable to this initiative
As someone whose donation was partially matched ($3k of $5k), I can attest that this is correct, I would not have participated without at least some of these efforts from this group of people.
We’ve received information from nonprofits on donation and match amounts for ~81% of the estimated amount donated, and I've updated the Follow-up with nonprofits section with some information on this. In general, we’ve found that the amounts received by the nonprofits have been similar to or greater than the amounts we had estimated.
Thanks Avi.
I’m posting very late but wanted to mention that this was a blast and was great community building.
Thanks for writing this up! Collecting knowledge about what did or didn't work is really important for making progress.
Thank you, this is a really useful write up of what sounds like a great project.
by Avi Norowitz
For #GivingTuesday this year, Facebook had announced that Facebook and the Gates Foundation would match donations made on Facebook up to $2 million. A number of us saw this as a rare opportunity to get donations to EA nonprofits counterfactually matched, similar to employer matching programs that match employee donations to any registered nonprofit. We created a Facebook group, a Facebook event, and a spreadsheet to help coordinate our efforts to capture as much of the match as possible. In the end, we had ~$379k in donations. Unfortunately, because the matching funds were exhausted at 1 minute and 26 seconds into the match, only ~$48k (~13%) of these donations were matched. As a result, it’s plausible that the most important effects of our efforts may have been indirect, and these effects may have been a mix of positive and negative. We consider some of the lessons we learned, whether we should try this again in 2018 (polls suggest yes), and some questions regarding the implementation in 2018. We also briefly discuss our follow-up work with nonprofits, which show that, in general, nonprofits received amounts similar to or greater than the amounts we had estimated.
The Facebook #GivingTuesday matching program
Our strategy to capture the matching funds
Estimated amounts donated and matched, by cause area and nonprofit
~$48k in counterfactually valid donation matches
Some EAs may have been encouraged to donate more
Some non-EAs may have been encouraged to donate to EA nonprofits
Increased openness about donating
Facebook newsfeeds were flooded with fundraisers
Some donations may have been made suboptimally
Some productivity may have been lost
EAs can work impressively well as a community
We should have focused more on donating fast
We should have prepared more for payment problems
We should have began working on this earlier
We should have suggested that people not donate privately
EAs outside of the US ran into problems
It might be worthwhile for the EA community to mobilize on other similar opportunities
Should we try again in 2018? Polls suggest yes
Questions for #GivingTuesday 2018
How should we keep track of fundraisers?
What other unexpected problems might we face?
The following other members of the EA #GivingTuesday organizer team (listed in alphabetical order by last name) were enormously helpful in making this all happen, including providing valuable assistance with this post.
I’m also grateful to the many other members of the EA community for their valuable contributions, and the 183 participants who created fundraisers, pledged donations to fundraisers, and did an admirable job maintaining the spreadsheet.
We also greatly appreciate the multiple nonprofits who provided us valuable information for our follow-up work.
On October 27, 2017, Facebook had announced that Facebook and the Gates Foundation would match donations made on Facebook on #GivingTuesday (November 28, 2017) up to $2 million:
On #GivingTuesday, Facebook and the Bill & Melinda Gates Foundation will be matching up to $2 million of funds raised on Facebook for US nonprofits. Facebook is also waiving its fees for donations made to nonprofits on Facebook this #GivingTuesday.
Donations to nonprofits made through Facebook’s charitable giving tools on November 28th will be matched up to $50,000 per nonprofit or $1,000 per fundraiser or donate button, until the $2 million in matching funds run out. The match will begin at 8AM EST (5AM PST).
Facebook ran a similar matching program for the previous year, and reported that the initial match of $500,000 was reached “within hours,” and that $6.79 million had been raised by Facebook fundraisers overall. Facebook also reported that after the initial $500,000 match had run out, the Gates Foundation had increased the matching funds to $900,000.
A number of us saw this as a rare opportunity to get donations to EA nonprofits counterfactually matched, similar to employer matching programs that match employee donations to any registered nonprofit. In particular, this opportunity appeared different than most matching opportunities in the following respects:
We saw this donation matching opportunity as one that avoided most of the pitfalls described by this old GiveWell blog post on donation matching.
The matching program had a limit of $1000 per Facebook fundraiser or donate button and a limit of $50,000 per nonprofit. Our plan was to encourage people in the EA community to create fundraisers for EA nonprofits, and encourage EAs to donate to up to $1000 to each of those fundraisers as soon as possible after 8am EST (5am PST) on #GivingTuesday.
To help coordinate these efforts, we created an EA #GivingTuesday Fundraiser Matching[2] group on Facebook. We realized that using a Facebook group alone to handle coordination between people creating fundraisers and donating to them would be difficult to manage and scale, so we created a Google Sheet where some people could post fundraisers and other people could pledge to donate to them. We also created a Facebook event for people to easily add to their calendars and invite their EA friends.
To create awareness about the opportunity, we posted to the EA Forum and a few EA groups on Facebook. We also invited our EA friends to our Facebook event, and encouraged people to invite their other EA friends as well.
On Monday morning, we had a total of $67k in pledges on our Google Sheet. By Tuesday morning, that amount had increased to $365k.
At Tuesday starting at 8am EST, more than 100 EAs -- including a large number of EAs awake at 5am on the west coast -- rushed to donate hundreds of thousands of dollars to EA nonprofits to capture as much of the matching funds as possible. EA Facebook was flooded with one $1000 donation after another. The mood was well captured in this post by Tyler John:
By 8:49am EST, Facebook had begun reporting to users that the $2 million in matching funds had run out.[3]
By 11:47am EST, we had finished manually collecting donation data from the fundraisers on the spreadsheet. We documented $338k in donations, an impressive 93% of the amount pledged.[4]
We had expected Facebook to begin reporting match amounts to fundraisers and nonprofits the following day. Unfortunately, Facebook kept us in suspense and didn’t report match amounts until Friday afternoon.
Although the amounts pledged and donated were very impressive, the amount matched was fairly disappointing. Through some impressive detective work, some EAs had found that the matching funds ran out at 1 minute and 26 seconds into the match. We attempted to estimate the match amounts from a combination of information reported by people who had created fundraisers and our observations of donation timestamps. The following are our current estimates:
Most of the donations went to animal (42%) and global poverty (33%) nonprofits. This differs from donations and cause area preferences reported in the EA Survey 2017 results. The differences appear to be partially the result of a small number of large donors and nonprofits who were actively involved in attempting to capture the matching funds.
Donations to long-term future nonprofits were unusually successful at generating matching funds, with 22% of donations matched, compared to the average of 13% (weighted by donations).
Machine Intelligence Research Institute (MIRI) received more donations than any other nonprofit ($49k),[8] followed by the Against Malaria Foundation ($43k), GiveWell ($39k),[9] and The Humane League ($38k).
The Animal Charities Evaluators’ Recommended Charities Fund had a larger percent of donations matched than any other nonprofit, with 50% of donations matched, compared to the average of 13% (weighted by donations).
To ensure that nonprofits receive the expected amounts, and to help us estimate the effectiveness our our work, we’ve been following up with all nonprofits that received donations according to our fundraiser worksheet. We’ve received information from nonprofits on donation and match amounts for ~81% of the estimated amount donated. In general, we’ve found that the amounts received by the nonprofits have been similar to or greater than the amounts we had estimated. (One notable exception is that Animal Ethics, which should have received a total of $150 in donations through one fundraiser, reported to us that they had received only $104. We haven’t been able to determine the cause of this discrepancy.)
For the nonprofits that reported amounts for both donations and matches to us, the table below shows the nonprofit name our estimated amounts, and the amounts reported by nonprofits.
|
Total donated (estimated via fundraisers) |
Total donated (reported by nonprofit) |
Total matched (estimated via fundraisers) |
Total matched (reported by nonprofit) |
|
|
MIRI |
$48,526 |
$48,422 |
$11,272 |
$11,372 |
|
Against Malaria |
$43,126 |
$67,716 |
$5,979 |
$7,986 |
|
GiveWell |
$39,225 |
$40,484 |
$0 |
$1,254 |
|
Humane League |
$37,915 |
$48,228 |
$4,380 |
$5,260 |
|
Good Food Inst. |
$25,681 |
$33,444 |
$1,100 |
$3,100 |
|
ACE Rec. Fund |
$24,292 |
$33,460 |
$12,220 |
$13,220 |
|
CIWF USA |
$23,076 |
$23,081 |
$1,500 |
$1,500 |
|
GiveDirectly |
$22,719 |
$44,786 |
$100 |
$4,600 |
|
Sentience Inst. |
$15,358 |
$14,177 |
$1,249 |
$2,249 |
|
New Harvest |
$7,475 |
$7,491 |
$0 |
$0 |
|
Innov Poverty |
$6,000 |
$6,000 |
$0 |
$0 |
|
BERI |
$4,400 |
$5,400 |
$1,000 |
$1,000 |
|
One Step |
$3,779 |
$4,179 |
$2,000 |
$2,000 |
|
Faunalytics |
$2,599 |
$2,599 |
$500 |
$500 |
|
Cool Earth |
$1,674 |
$3,900 |
$0 |
$0 |
|
Reducetarian |
$150 |
$150 |
$0 |
$0 |
|
Animal Ethics |
$150 |
$104 |
$0 |
$0 |
|
Total |
$306,145 |
$383,621 |
$41,300 |
$54,041 |
The following graph shows the donated amount we had estimated via fundraisers compared to the donated amount reported to us by nonprofits. (We’ve omitted Reducetarian Foundation and Animal Ethics from the graph because the donated amounts are too low to be visible.)
And the following graph shows the match amount we had estimated via fundraisers compared to the match amount reported to us by nonprofits.
Although in general the total donation and match amounts were higher than our estimates, we don’t know how much of difference was caused by our initiative.
The EA #GivingTuesday Fundraiser Matching project may have had mixed effects, some positive and some negative. In this section, I’ll provide an overview of the various effects the project may have had. The views in this section (some of which is very speculative) should be seen as my own, and may not necessarily reflect the views of other people involved.
Though our estimated match percent of 13% is lower than I had expected, I consider the absolute estimated match amount of $48k to be fairly impressive. We think these matches are counterfactually valid and mostly attributable to this initiative, for the following reasons:
The matching incentive and the excitement around the matching may have encouraged some EAs to make donations they would have not made otherwise, or donate more than they would have donated otherwise.
The hundreds of fundraisers for EA nonprofits flooding Facebook newsfeeds and timelines, combined with Facebook’s advertising about the matching, may have encouraged some non-EAs to donate to some EA fundraisers.
We attempted to estimate the amount of non-EA donations by looking at fundraisers with “surplus” donations, where the total donation amount exceeded total pledge amount.[10] As we began looking at the fundraisers with surplus donations more closely, however, we found that a lot of these surplus donations appeared to be from EA’s. We went through each fundraiser with surplus donations, and based on our subjective judgment, we excluded fundraisers where most of the surplus donations appeared to originate from EA’s. We also reviewed all fundraisers with non-zero donation amounts that undershot their pledges to look for non-EA donations, and made a subjective judgment about the non-EA donation amounts for each of those fundraisers. We estimate that around $11k (or 3% of donations) originated from non-EA’s, but we feel very uncertain about this estimate.
The methodology used to generate the above estimate can be found here.
The donors participating in our initiative were unusually open about their donations, by pledging donations on the publicly shared spreadsheet, and by making (mostly) public donations on Facebook fundraisers.
This openness may help inspire people to give, so I’m considering this a positive. On the other hand, it’s possible that donating large amounts publicly may have been perceived negatively by some non-EAs.
The project provided a number of useful learning experiences that will be useful if we try again in 2018, and may also be useful for similar EA community projects.
As an unintended side effect of our project, the newsfeeds of people with EA friends who had created fundraisers were flooded with these fundraisers. Some people (EAs or outsiders) may have found this spammy or annoying, and it’s possible this could have reflected negatively on the image of EA as a community or individual EAs. Other people may have found it inspiring to see so many people trying to raise money for EA nonprofits, and to see one fundraiser get fully funded after the other.
It’s also possible that the flood of fundraisers in newsfeeds resulted in some productivity loss, though it’s similarly possible that it encouraged people to get off of Facebook and do something more productive instead.
The US tax law changes beginning in 2018 will reduce the tax benefits of donating for many people by increasing the standard deduction and reducing tax rates. If some people were encouraged to front-load their 2018 donations into 2017 to try to capture the match, they may receive some tax benefits, which could potentially later translate into increased donations in the future.
On the negative side, one or more people in the UK donated and lost the opportunity to claim Gift Aid for their donations. In one such case though, Facebook has been willing to refund these donations.
There are a number of reasons to believe that some donations may have been made suboptimally:
That being said, to try to mitigate this risk, we did add the following line to our instructions: “Be careful about donating to an organization you think is less effective just for the match. Some EA organizations may be 4x or more effective as others so it might be better to forgo a match instead of donating to a less effective organization.”
If we consider the hours of work on this project by the core people involved, as well as the 183 EAs involved in creating fundraisers and donating to fundraisers, this may have taken up 500 hours or more of people’s time.
There may have also been some additional productivity loss caused by the sleep deprivation of people on the west coast waking up before 5am to donate on time.
As noted above, the EA #GivingTuesday Fundraiser Matching project provided a number of lessons learned that may be useful if we try again in 2018, and for other potential similar projects that involve the EA community.
I estimate that we had 183 EAs creating fundraisers and pledging to donate to them. Despite the large numbers of people involved, EAs worked together impressively well. I was worried that our spreadsheet would be unmanageable with so many people working on it, but this risk did not materialize, and the spreadsheet was well maintained by the participants.
I did not anticipate that the matching funds would have run out in less than 2 minutes. If we had anticipated this, we could have emphasized the need to donate fast, and provided more instructions on donating fast, such as:
A number of people experienced payment problems when trying to donate. We should have provided more instructions on preparation to mitigate these problems, such as:
I only became aware of the Facebook donating matching when Arushi Gupta informed me about it on Thanksgiving morning, less than a week before #GivingTuesday. If we had begun working on this earlier, we could have seen a number of benefits:
Facebook had announced they were matching donations for nonprofits in the UK as well. So early on, we created a separate Google Sheet and Facebook event for people in the UK. It turns out, however, that for nonprofits in the UK to be eligible, it was necessary that they were registered with Facebook Payments. Unfortunately, it turned out that no EA nonprofits in the UK were registered with Facebook Payments, and our early efforts may have been confusing to people.
We also found that while some EAs outside the US were able to create fundraisers for US nonprofits, some of them did not see banners from Facebook with information about the amount matching. We interpret this to mean that donations to these fundraisers were not eligible for the match.
We should take these problems into account if we try this again in 2018.
The opportunities to capture counterfactually valid donation matches to EA nonprofits without distorting donor behavior (too much) seem fairly limited, but perhaps there are others that we’re missing? More broadly, perhaps we should be looking for other kinds of opportunities where the EA community may be able to raise counterfactual donations to EA nonprofits. A number of EAs already do effective work each year raising counterfactually valid donations through Project For Awesome, but here are some other ideas that come to mind:
If Facebook runs a similar donation matching program for #GivingTuesday 2018, should we try again? To get the views of the EA community, I conducted a number of Facebook polls.
|
% Yes |
67% |
60% |
72% |
|
% Maybe |
33% |
24% |
24% |
|
% No |
0% |
16% |
3% |
|
Total responses |
43 |
25 |
29 |
In general, the polls appear to show that EAs both involved and not involved in the #GivingTuesday donation matching are generally supportive of trying again in 2018.
Our experiences this year raises some questions that are relevant if we try this again next year.
Facebook did not have rules in place to prevent people from creating multiple fundraisers for the same nonprofit, and we’ve seen no evidence that it prevented any donations from getting matched. I had concerns that encouraging people create lots of fundraisers for the same nonprofit could be perceived as an unethical attempt to circumvent the $1000 limit per fundraiser, and may go too far in the direction of violating the spirit of the matching program. Therefore, our instructions did not encourage people to create multiple fundraisers. On the other hand, perhaps it’s fine as long as we’re following the rules.
The matching funds ran out at 1 minute and 26 seconds. Is it reasonable to expect we’ll have 1-2 minutes in 2018? Or should we expect a much shorter period of time, i.e. because other people might optimize to make their donations faster as well?
Our use of a Google Sheet to keep track of fundraisers could have led to problems, including accidental or malicious edits that could have made the spreadsheet unusable. Fortunately, none of these problems materialized, but we may have just been lucky. Should we manage the security of the spreadsheet differently in 2018? Or should we use a different tool to keep track of fundraisers?
Most of us failed to predict the matching funds would run out so fast. What other unexpected problems might we face in 2018?
[1] A few nonprofits reported to us that, based on the information they received from Network for Good, it appears the match was a 50/50 split between Facebook and the Gates Foundation.
[2] After #GivingTuesday had ended, we found that the group was attracting people outside the EA community who were confused about the purpose of the group. To mitigate this, we removed references to #GivingTuesday in the name, description, and group image. We’ll restore these references if/when we resurrect the group for #GivingTuesday 2018.
[3] We later learned that the matching funds had run out much earlier, at 1 minute and 26 seconds into the match.
[4] The $338k donated included some donations that did not have pledges in the spreadsheet.
[5] Both our pledges and donation amounts in our spreadsheet had increased as people added and updated fundraisers in the spreadsheet.
[6] The $379k donated included some donations that did not have pledges in the spreadsheet. We also think this is an underestimate, because a number of nonprofits gave us donation and match amount numbers substantially higher than what we had estimated, and we’d guess that a lot of these additional donations are still a result of our initiative.
[7] This estimate is very uncertain, and is based on limited and messy data. We think it’s in the right ballpark though.
[8] The high donation amount to MIRI was largely the result of an effort by MIRI to try to capture matching funds from Facebook.
[9] GiveWell did report a match amount of $1,254 to us. To remain consistent though, we’re reporting the amount we estimated from our fundraisers spreadsheet, which is $0.
[10] Special thanks to Angelina Li who spearheaded this work.
[11] It’s possible though that some of this was caused by people waiting for GiveWell or Animal Charity Evaluators to release their recommendations, which they did within those 24 hours.
Great to see such thorough analysis!
Do you think the time costs will go down next year? That strikes me as the most significant drawback to this project. "500 hours or more of people’s time" for ~$48k matched is ~$96/hr, which is good, but appears to be on the lower end of fundraising ROI that I've tracked.
Still, this project seems to at least be competitive with Project for Awesome. I also suspect the value of bringing EAs together to work on a clear, common goal is an underrated form of impact through building community bonds and cohesion.
Yes, if we try again in 2018, I think we can avoid some of the learning curve and improve efficiency. I'd also hope we can use what we learned to get more than 13% matched.
Hopefully...since it's a zero-sum game though, I'm not necessarily convinced that we can improve efficiency and learn from our mistakes more than other groups. In fact, I'd expect the %matched to go down next year, as the % of the matching funds directed by the EA community was far larger than the % of total annual donations made by EAs (and so we're likely to revert to the mean)
I think we are much more organised than most, and hence more able to learn from our mistakes.
But would this view have predicted we'd only get 13% matched, well below the EA consensus prediction?
I agree that this is a risk.