What makes an active timebank?Identifying predictors of an active timebank using linear regressionDawn GrahamBlockedUnblockFollowFollowingFeb 12In this article, I explain my process for collecting data, feature engineering, and using linear regression to identify predictors of active timebanks.
But first…What’s a timebank?Without fail, when I say I am working on a project about timebanks, the first question someone asks is, “What’s a timebank?”As explained on the TimeBanks USA website:“Timebanking is a time-based currency.
Give one hour of service to another, and receive one time credit.
”You may have heard of timebanks or similar systems before, associated with terms like mutual aid networks, skill shares, local exchange trading systems (LETS), or sharing economies.
The Cincinnati Time Store was an experimental retail store that operated from 1827 to 1830.
It used “labor notes” as currency and is considered a precursor to other LETS, including timebanks, and modern cryptocurrency.
(Source: “Cincinnati Time Store”, Wikipedia)In the 1990s, Edgar Cahn trademarked the term “TimeBank” and founded TimeBanks USA, pushing the movement in the U.
He explained his motivation in “The Banks That Run on Time Instead of Money” episode of the People Fixing the World podcast.
After experiencing a heart attack that limited his ability to work, he reflected on capitalism as: “A monetary system which valued what was scarce and devalued what was more abundant and treated as worthless anything that was truly abundant.
And I suddenly realized that meant it devalues being a human being because we are not scarce.
Maybe we needed a kind of money that valued what it meant to be a human being.
”Cahn identified five core values of timebanking:Asset: Every one of us has something of value to share with someone else.
Redefining Work: There are some forms of work that money will not easily pay for, like building strong families, revitalizing neighborhoods, making democracy work, advancing social justice.
Time credits were designed to reward, recognize and honor that work.
Reciprocity: The question: “How can I help you?” needs to change so we ask: “Will you help someone too?” Paying it forward ensures that, together, we help each other build the world we all will live in.
Community/Social Networks: Helping each other, we reweave communities of support, strength & trust.
Community is built by sinking roots, building trust, creating networks.
Respect: The heart and soul of democracy lies in respect for others.
We strive to respect where people are in the moment, not where we hope they will be at some future point.
Why should we care?Currently in the U.
, many people are concerned about seemingly increasing social and political division.
Meanwhile, people are still dealing with the effects of the longest government shutdown in U.
history, with another potential shutdown ahead.
Timebanks offer the possibility of strengthening ties within communities, (re)building trust and relationships, and creating support systems to help meet daily needs and recover from crises (political, natural, or otherwise) — without relying on money, markets, or the state.
Following the 1995 heat wave in Chicago that claimed hundreds of lives, Eric Klinenberg researched the disparate death rates between neighborhoods.
Little Village had one of the lowest rates in the city with 4 heat-related deaths per 100,000 residents.
Meanwhile, the directly adjacent neighborhood of North Lawndale had one of the highest rates with 40 deaths per 100,000 residents.
Klinenberg attributed the disparity to differences in social and spatial context:Most of the African American neighborhoods with high heat wave death rates had been abandoned — by employers, stores, and residents — in recent decades.
The social ecology of abandonment, dispersion, and decay makes systems of social support exceedingly difficult to sustain.
Of course forces of nature played a major role.
But these deaths were not an act of God.
Hundreds of Chicago residents died alone, behind locked doors and sealed windows, out of contact with friends, family, and neighbors, unassisted by public agencies or community groups.
There’s nothing natural about that.
At its core, timebanking is about building strong systems of support.
Before the Canterbury earthquakes in 2010 and 2011, the Lyttelton Time Bank had already organized 10% of the town’s residents and 18 organizations.
Thanks to the network that had already been developed, the timebank was able to mobilize community members quickly when the earthquakes hit.
They already knew who to check in on and how to best utilize people’s skills.
Significantly, the timebank also enabled ongoing support and recovery long after emergency workers left.
Community members organize using resources provided by the Lyttelton Time Bank after the Canterbury earthquakes in New Zealand (Source: Developing Local Partners in Emergency Planning and Management)While we don’t know that the Chicago heat wave would have had different outcomes with a timebank or similar system in place, the Lyttelton example suggests that timebanks can be a powerful contributor to community safety, resilience, and recovery.
Cahn has shared other examples of successes with timebanks:In Chicago, 127 schools nearly eliminated special ed by having fifth graders help the third graders learn the alphabet.
Teens serving on a youth court jury in Washington, D.
, helped reduce recidivism.
The program reduced rearrests from 34 to 6%.
TimeBanks USA has also recently formalized a partnership with Seva Exchange, “a platform using artificial intelligence and blockchain technologies to galvanize global volunteerism by reinventing timebanking services for the modern digital economy.
” Seva Exchange is looking into how time-based cryptocurrency could provide a form of Universal Basic Income, operating separately from state welfare systems.
What makes an active timebank?Now that we’ve gone over some background information, let’s dive into my project.
Data CollectionThis project is focused on data from TimeBanks USA.
While there are other platforms with more timebanks listed (such as hOurworld and Community Exchange System), they do not have as much centralized and publicly accessible information about individual timebanks.
I first scraped all pages of the TimeBanks.
org directory to get an initial listing of all timebanks on the platform.
I then created a daily scraper to get updates on the following at the beginning of each day:Numbers for exchanges, hours, members, offers, requests, last exchangeOffers, requests, and talents by categoryAll offer and request listingsI manually gathered data from Facebook and Twitter accounts that were included on timebank pages.
This decision was originally prompted by Facebook’s restrictions on automated data collection.
However, this became an opportunity to learn more about the different timebanks and was feasible given the relatively small number of timebanks with social media accounts.
Geolocation and census data for locations of U.
timebanks was compiled using Geocodio.
I gathered population estimates for New Zealand from Stats NZ.
These will be used for future developments with the project.
Processing & Feature EngineeringSee the Cleaning & Feature Engineering notebook for full details.
In this notebook, I generate new features and combine collected data into a single file to use for modeling.
I got compound sentiment scores (using VADER’s SentimentIntensityAnalyzer) and word counts for each timebank's mission statement and notes.
Dummy variables indicate if each timebank has a phone number, sponsor, secondary website, Facebook account, or Twitter account listed.
I also calculated the averages for the number of daily exchanges, number of hours exchanged each day, hours per exchange, number of offers, number of requests, ratio of offers to requests, and number of new members each day.
Additional features included the number of parent categories for each timebank, total categories (parent and child), the percent of categories with offers and requests, and total talents per parent category per member.
Exploratory Data AnalysisWorking on this project has been a highly iterative process.
I used exploratory data analysis throughout data collection, processing, feature engineering, etc.
The EDA notebook contains analysis and visualization of the combined data that will be used for modeling.
This includes a look at data for ALL timebanks compared to INACTIVE and ACTIVE timebanks.
From this, I anticipated that creating separate models for all timebanks and active timebanks would provide helpful insights.
ModelingI created separate models for all timebanks and active timebanks only.
Using GridSearchCV to find the best parameters, I tried models using LinearRegression, RidgeCV, LassoCV, ElasticNetCV, and DecisionTreeRegressor.
Ultimately, I selected linear regression with backward feature selection as the best model for both.
The best model for ALL timebanks explained 54.
3% of the variability in the data, relative to a model with no predictors.
The model predicted that the average number of daily exchanges would increase with:more requestsmore hours per exchangea phone number being listedhigher offer to request ratiohigher percentage of categories with requestsThe best model for ACTIVE timebanks explained 39.
8% of the variability in the data, relative to a model with no predictors.
The model predicted that the average number of daily exchanges would increase with:more Facebook followersmore requestsa sponsor being listedSummaryConsidering both the relatively small number of timebanks in this data set (156 total, 48 active) and how much they differ from each other, I was not expecting to get very robust models.
However, they do still provide useful insights into differences between active and inactive timebanks, as well as potential predictors of more exchanges among active timebanks.
I am hopeful that adding data about the communities the timebanks are based in will strengthen the models and provide more insights into what makes an active timebank.
See the Process Log for additional documentation of my process, challenges, questions, ideas, and references.
Future DevelopmentsThis project is very much a work in progress.
Future developments include:Collecting updated directory info regularly to gather new timebank info.
Doing timeseries analysis of listings once more data over a longer period of time is collected.
Using natural language processing to take a deeper dive into what requests and offers are exchanged, what goes unmet, and other patterns.
Deeper look at language use, framing, descriptiveness, and active/passive voice in mission statements and other timebank info.
Investigating if areas with more timebanks in closer proximity also have more activity.
Looking at demographic info in relation to timebank locations, including population size and density, unemployment rates, home ownership, etc.
Setting up the daily scraper to run automatically via AWS.
An InvitationI am actively working on this project and am hoping to provide updates here as it progresses.
Constructive critiques, suggestions, and thought partners are welcomed with open arms.
Please feel free to get in touch.
Get up-to-date contact info at dawngraham.
You can also check out the GitHub repo for this project.
Thanks for reading!.Thoughts, questions, and feedback are always appreciated.