“Bod”-havioral Economics Part One: Experiment Design

In exactly 6 weeks and one day (plus the duration of one red-eye flight with a tragically long layover in Orlando), I will be on the island of Jamaica. In a villa. On the ocean. As a bridesmaid. This is a vacation where photos will abound.

I already know I’ll be joy-crying nonstop, disheveling my makeup in every shot. But for the photos where I can KEEP-IT-TOGETHER-DANA? I want to feel great and look good. Enter: Operation Beach Bod.

Fortunately, I’m not the only wedding-goer hoping to be at their best for the occasion. My friend Kelsey texted me the following at the end of February (I’m paraphrasing and removing profanities below):

I have an idea with fitness motivation. What if we start a Facebook group and we have a challenge for March. Everyone puts in $10, and if you’re successful, then you get your name put into a drawing at the end of the month. Winner gets the pot.

As a junkie of the behavioral economics non-fiction genre (think Freakonomics, Nudge, Switch, Blink, and other catchy one-word titles with lengthy, wonky subtitles), I was totally on board. Not only is this a financial incentive, but there’s a social disincentive at stake. Add in the fact that I’m intrinsically competitive? Let’s do this.

After some back and forth, Kelsey and I teased out the parameters – and even recruited three other Jamaica-bound souls to join this experiment with us. Here’s what we came up with:

Kelsey and I made some key decisions in the experiment design that are worth calling out.

  • Each person should set their own individualized goal (something challenging and meaningful), as opposed to all of us sharing a group goal. Goals should be about doing something positive, like exercise, rather than something restrictive, like dieting.
  • Instead of designing a one-month goal with a one-month prize, each person sets a goal for each week, with one raffle entry per successful week. That way, if you get off track in week 2, you still have motivation to get back on track for weeks 3 and 4. Plus, the shorter cycle for positive feedback and milestone achievement seemed smart.
  • We needed to recruit at least 4 people to the group to make the final prize seem, well, prize-worthy.

It’s a variation on stickK.com, a behavioral economics-inspired web platform where you make a “commitment contract,” recruit a “referee,” and set your “financial stakes.” But, to me, something was always weird about the unidirectionality of this model. If I’m the person who wants to achieve a goal, I have to get others to agree to support me and hold me accountable. What do the supporters and referees get out of this?

With the DataDana variation, the challenge, support, and accountability flows in all directions. It’s multiple people asking each other. Plus, my variation has the added conceit of the gamble – a chance at multiplying your financial stakes. Yes, most participants will lose their money. But $10 is low enough that it won’t be missed, practically speaking; and $50 is high enough that it feels like a prize. If I win, I’m treating myself and one lucky dining companion to drinks and cheese at my local supermarket’s on-site bar. (One night of rosé and manchego won’t unravel all that hard work. Right?)

We are currently in Week 4 (of 4) in the Bod-havioral Economics experiment. After Sunday? I’ll update you dataheads with the results and some bonus analytics.


Say Rent to the Tent?

I’m at the age where all my travel plans this year involve wedding-related events. I personally love it. I’m a sap for romance, I love a good open bar, and weddings are endless fonts of quantitative inspiration. (For example, FiveThirtyEight has a fun post about most-played songs at wedding receptions.)

So when my friends Eling and Dan shared their wedding tent rental dilemma, I jumped at the chance to wrap my brain around some data. Here’s the situation, paraphrased from an email from Eling:

Most tent companies require you to order your tent at least five days before the wedding. Tents cost upwards of $1,500. The forecast is looking good as of now, but forecasts can be so unreliable in Miami. What should we be willing to pay to ensure our guests don’t get rained on? Is there a magical formula that can calculate when it’s worth it?

I love this question. There are actually two parts to it. One, how reliable is a 5-day forecast? And, two, how can the couple use the information available to decide on this big ticket item?

For #1, I gathered data about Saturdays (i.e., most popular wedding days) from 2011-2017 in Miami in March. I got data on whether it actually rained from the Weather Underground’s weather history site. The trickier part was figuring out what had been predicted for that day five days earlier. (I’m guessing for-profit weather sites don’t want people criticizing them for poor predictions, so don’t make these easy to mine.) But after some deep Googling just shy of the Dark Web, I found a workable resource from the U.S. National Oceanic and Atmospheric Administration (NOAA). Using their Weather Prediction Center Medium Range Archive, I could look up the date of the Monday that preceded the wedding and see what the predicted precipitation had been for the Miami area.

I kind of love the behavioral economics lessons in this graph. Let’s say you’re a couple who faced a forecast of 0-10% and, therefore, you “gambled” not to rent. On 9 out of 10 of these Saturdays, their gamble worked out. Spend that money on your honeymoon, lovers! But does that mean the  3/29/14 couples who ended up with 0.36 inches of rain make a bad choice if they, like the other 0-10%ers, didn’t rent? Not necessarily! They had the exact same information as the other 0-10% couples. The chips just didn’t fall their way.

I mean, think about it the other way. If the forecast was for 80-90%, most people would be safe and rent a tent. But there are probably couples who “gambled” not to rent, and it turned out not to rain. Does that mean they’re meteorological savants? Unlikely. I would argue that they made a bad choice even though they had a good outcome. There’s a principle in behavioral economics that “one should be judged not by the outcome of a decision, but by the process that led to that decision.”

That brings us back to #2: what is a good process (a “magical formula”) for making this decision? I propose using the tool below, a Tableau interactive which finally puts all those years of econometrics to good use:

Fortunately, March is one of the least rainy months in Miami; so chances are, Eling and Dan will have a clear forecast and will only have to make decisions on the left-hand, sunny-forecast side of this graph. But this “fortune” is also, unfortunately, the grey area. Choosing not to get a tent will always be a gamble.

How willing you are to make this gamble depends on whether you’re willing to accept a little risk. And this risk-averseness is harder to model without doing legit tent consumer research. While I added some guidance into the interactive, I couldn’t figure out a way to make this more cut-and-dried. Sorry, dudes.

Would love to hear your thoughts, dataheads! Is there a way to make the “grey area” more black-and-white? Do you have any good weather data resources?


Splitwise’s Math Problem

Math puzzle nerds, like me, love the challenge of splitting expenses. Whether it’s roommate bills or a restaurant check, figuring out who owes who what can be a weirdly gratifying duty. (My dream job, as depicted on Portlandia.)

So when I heard about this new app called Splitwise, I thought my skill would be made totally obsolete. Basically, you enter group expenses into the app, Splitwise does all the calculations, and uses your PayPal/Venmo/etc account to “settle up.”

While all you people with real lives and real hobbies probably love this, I was initially reluctant to forsake my beloved spreadsheet matrices. But I caved. I gave Splitwise a whirl at a (delightful!) nine-lady bachelorette party last weekend.

And I have to confess: Splitwise is pretty great. The user experience is intuitive, it’s easy to make different subsets responsible for different expenses, and users do zero calculations to figure out the transaction requirements/amounts.

But, dude. The way Splitwise does this is, to put it in technical terms, inefficient AF:

The way I’ve always done this math is a little different. I basically try to minimize the number of transactions needed. This is probably a relic of pre-Venmo days, when we all actually had to write physical paper checks to one another. Instead of wasting checks for every one-on-one balance, I’d basically say: OK. If I put $541.78 into the “group balance sheet,” but I only owed $225.87 to the group, all that matters is that I get $315.91 back from the group somehow. I give zero craps about who it actually comes from. Similarly, if Joslyn owes $218.87 to the group, it doesn’t matter to her who she pays it to.

So, to balance the cash flow while minimizing transactions, I just MATCH people who owe the group (owe-ers) to people who are owed by the group (owe-ees).

Why does this matter? Some users may have transaction fees associated with their bank account or with the payment app they associate with Splitwise. Plus, all these transactions clog up my Venmo feed, which is normally a very fascinating window into my social network’s financial habits. And also, efficiency for efficiency’s sake!

But I could see why Splitwise does this. For one, it promotes their app multiple times over across multiple Venmo feeds. Free publicity, baby. But importantly, if there’s a stinker in the group who doesn’t pony up? The effect is spread among everyone. For example, if Joslyn snaps and goes off the grid and never settles her tab, five of us share the brunt rather than just Claire.

What do y’all think? Any other bill splitting methodologies out there? Anyone think they can get fewer than 8 transactions out of this scenario? Download the data and explore my Tableau Public workbook.