GiveCrypto Monthly Update — August 2020
An update on marketplace development and final RCT data
Update on Marketplace Development
We are deep in development mode of the crypto donation marketplace and on track to launch the week of September 21st. This will be a web app that allows anybody to apply for and anybody to provide help. It will be similar to DonorsChoose or GoFundMe, but all help will be in crypto. As is the case with any product development process, we’ve had to make some trade-offs.
UX vs. Fraud Prevention/Detection
We’ve spent quite a bit of time considering the balance between maintaining a simple on-boarding process and dealing with fraud. This trade-off boils down to how much information we collect from users. Fraud prevention/detection is generally more robust when more user information is required, but each piece of data creates more friction for the user.
The first way we are dealing with this challenge is to use Google for authentication. This provides a number of benefits:
- Outsources most of the data collection/retention challenges
- Makes fraudulent account creation more difficult
- Gives users an easier authentication path
The primary drawback to this approach is that all users will need a Google account. This isn’t ideal, but we think this is an acceptable trade-off.
On the fraud detection front, we are focusing efforts on a number of passive data collection tactics that were described in last month’s post.
We will closely monitor fraud rates and adjust the prevention/detection efforts as we evolve the service.
Structuring Payments
The initial design of the service allowed the recipient to specify how much money they would receive. After going through a few rounds of prototyping, we realized that the payment structure needed to be simplified. There were some UX challenges with this level of control and we were worried about potential fraud resulting from user defined amounts.
We decided to constrain v1 so that all recipients would get a predefined amount of money over six weeks. The amount of money will depend on the recipient’s country of residence and the World Bank’s PPP index. The PPP index measures the amount of goods and services that can be purchased using a unit of country A’s currency in country B. It is a factor that corresponds to the cost of living in each country. We set the initial amount at $1.90 per person per day (the World Bank’s international poverty line) and then adjusted using the PPP index. The table below shows some example amounts in different countries.
Incorporating Community Feedback
Since the app will allow users to help each other, we feel that community should be baked into its DNA. For v1, this will involve users voting on the content created by other users. There will be two types of content created in v1.
- Requests for funding from recipients
- Updates on what was done with the funds
Like Reddit, users can vote on these content items (up and down). And, the item’s vote score will determine the ranking in the feed.
Incorporating voting data into an algorithm that determines content importance is a non-trivial problem. A poorly constructed algorithm can have unintended consequences and there is a lot written about how voting mechanisms have evolved.
Without going into details, we are leveraging lessons from sites like Reddit and are using an algorithm that involves both vote scores and a time decay element. Our hope is that higher quality and fresher content will rise to the top.
Final RCT Data
The RCT was completed in mid July. We have collected all survey responses and are currently writing the report that we plan to publish before the end of September. As part of this analysis, we have had a chance to perform regressions on the data and wanted to share the preliminary results.
As a reminder, the RCT involved three groups; one received $10 per week in cash, one received $10 per week in crypto and one control group. All of the groups were asked impact questions before, during and after the study. The table below shows the number of respondents in each arm of the RCT at different stages of follow-up.
In the table below, we present the results of regression estimates of causal impact. For each outcome we pool the impact across the three waves of follow-up surveys, so our estimates reflect the average effect sizes across them. In each regression, we control for the baseline value of the outcomes (except for social distancing which was not measured at baseline) as well as demographic variables (gender, age, education level and household size). We also include wave dummy variables to estimate the wave-specific level effects and cluster the standard errors at the Ambassador level (the level of treatment). Stars indicate the statistical significance level of the estimate (* for 10%, ** for 5%, and *** for 1%). A lower significance level (i.e. more stars) indicates a lower likelihood that the estimated value is due to chance.
We see that both the cash and the crypto transfers have a sizable impact on food security (0.35 and 0.40 times its baseline standard deviation) and both estimates are statistically significant at the 1% level (which is well above the conventional 5% level set by most researchers). Crypto transfers seem to have a higher impact, though there is no statistically significant difference between the crypto and cash (possibly due to insufficient sample size to detect the difference). Both crypto and cash also have a positive impact on psychological well being and are statistically significant at the 5% level. Finally, there seems to be very little impact on social distancing as measured by how many times people leave home.
Both crypto and cash transfers have a sizable positive impact on food security
The control group reported an average of 7.23 days out of the past month where someone in the respondents’ household skipped a meal or cut meal size. Crypto transfers reduced this number by 2.35 days to a total of 4.88 days (~48% reduction). Cash transfers reduced this number by 1.77 days to a total of 5.46 days (~32% reduction). These are meaningful changes to the food security of the participants in the treatment arms.
Both crypto and cash transfers have a sizable positive impact on psychological well being
58.6% of the control group reported being often or extremely optimistic about their future. The crypto transfer increased this by 7.4 percentage points and the cash transfer increased it by 8.8 percentage points. These are meaningful changes to the psychological well being of the participants in the treatment arms.
Staying Connected
As always, we love to hear from supporters outside of our updates and events. Please follow along on Twitter and our blog for updates on our programs.
