Season 19 of Grey’s Anatomy premieres later this week, and I am one of the many dedicated fans who has been with the show since Day One. I remember in college, in the pre-streaming era, I bought a cheap VCR for the sole reason of recording Grey’s episodes during my evening chem lab. I can tell you exactly where I was when I watched the bazooka bomb episode guest starring Christina Ricci and Kyle Chandler. This show has been a constant for me over the past 18 seasons (wow!) spanning 17 years (WOW!).

And in these glorious 18 seasons, the Grey’s universe has accumulated an enormous amount of mess among its characters. There are love stories, heartbreaks, one night stands, family drama, spinoff shows, workplace feuds, physical fights, natural disasters, tragedies – I could (and might) go on.

Earlier this summer, I was brainstorming infographic ideas for this blog and was drawing inspiration from the abundance of data out there about The Bachelor franchise. My most popular blog post is a Bachelor data interactive; the Instagram account @BachelorData has a huge following; there is even an entire book about Bachelor game statistics called How to Win The Bachelor. It occurred to me that, with 18 seasons of Grey’s under our belts, there is so much data to extract about Grey’s storylines – plus a huge fan base out there who might find it interesting to nerd out on. Just like there is this huge community in Bachelor Nation, is there perhaps an underserved market of Grey’s dataheads?

So now, three months after I began this research project, and with only a few days left before more romantic entanglements ensue in the S19 premiere, I am extremely proud to present my very first Grey’s data infographic. This is essentially a network diagram which visualizes the various romantic connections (“links”) among the characters (“nodes”). Note that the Grey’s universe is huge, and as such this graphic is best seen in full-screen, zoom-in, high-res mode here.

There are so many different flavors of romantic connections I considered – past vs present, unrequited vs consummated, on-screen vs off-screen. I developed these rules below to determine which relationships were important enough to qualify for this infographic:

What kinds of “nodes” and “links” are eligible to appear here?

• Characters who are eligible to appear as “nodes” must be at least one of the following:
  • (a) series regulars with at least 1 romantic link
  • (b) 1st degree romantic links to series regulars
  • or (c) 2nd degree links to a series regular as long as this person’s 1st degree link is with any Grey-Sloan Memorial (GSM) doctor.
• Characters and any of their eligible connections also must also meet all of the following criteria:
  • (a) credited in at least 1 Grey’s Anatomy main show episode
  • (b) physically appears in in a “present day” storyline
  • and (c) has at least a first name.
• Characters are given a romantic connection if they, at minimum, have either had sex or have a dating history that goes beyond a one-time date.
• The occurrence of two characters’ romantic connection can happen at any point (on-screen or off-screen, historical or future to their appearance, or even in separate spinoff show episodes) — as long as each character at one point appears in a “present day” Grey’s Anatomy storyline.
• Familial links (parent/child or siblings) are shown between characters who have other romantic entanglements, to visualize the family drama also afoot.

What kinds of “nodes” and “links” were excluded?

• Two characters are not given a romantic connection in these scenarios:
  • (a) they have only been on one sex-less date (e.g., Derek Shepherd’s awkward date with Sydney Heron)
  • (b) they only kiss briefly then one or both parties puts a stop to it (e.g., Jordan Wright kisses Miranda Bailey but she’s having none of it; Alex Karev and April Kepner kiss once, but when he tries to go further she stops him and he storms off)
  • (c) one or both characters have romantic feelings but never act on it (e.g., Lexie Grey’s crush on George O’Malley, Taryn Helms’s crush on Meredith Grey)
  • (d) one character never appears on a present day Grey’s Anatomy storyline (e.g., Teddy Altman’s late ex-girlfriend Allison who dies in a 9/11 flashback)
  • or (e) they are only romantically connected to someone who is not a Grey-Sloan doctor (e.g., Thatcher Grey’s second and third wives).
• If a romantic partner is only discussed but the character never appears present day, this person/link is not shown.
  • This “disqualifies” exes who never appear or only appear as flashbacks (e.g., due to estrangement or death) — which unfortunately disqualifies important backstories, like Tom Koracick’s late wife Dana.
  • This was a tough decision, but I had to draw the line somewhere. This prevents the chart from getting clogged up any time a character offhandedly mentions an off-screen/historical ex-lover.
• While familial links are shown where they exist, family members are not listed here unless they have a primary or secondary romantic connection to a series regular themselves

---

And of course I did more than just draw out the connections; I crunched those numbers, too. Here you can see, of the 21 “main characters,” who has the most 1st degree romantic partners (i.e., direct connections) as well as who have the most 1st degree + 2nd degree partners (i.e., your partners plus your partners’ partners).

For Grey’s fans, it’s not surprising to see Alex Karev and Mark Sloan at the top of the “most number of partners” ranking. They both were known for being promiscuous around the hospital, and they both were given longer-term romantic storylines. I was pleasantly surprised to see Jo Wilson so high – this means the writers are giving her a complicated personal life (great for her screen time!) and as one of the only main characters who is presently single, I could see her increasing her partner count in coming episodes. Her romantic connectedness also means that anyone who newly becomes linked to Jo will immediately have high drama potential.

One of the most interesting parts of this analysis for me was the “1st + 2nd degree hookups.” This tells us not just how many partners the character had, but how entangled their chosen partners are with other characters on the show. For example: While Meredith is #3 in direct partners, her partners aren’t likely to have relationships with others on the show; as Meredith is the lead, I think this protects her from getting into petty interpersonal drama.

Speaking of protection, Miranda Bailey is completely isolated from the drama. While she is given 3 romantic partners over the past 18 seasons, none of these 3 partners ever have an on-screen romance with anyone else. Bailey is often the moral compass and the authority figure; again, this feels like a choice to protect her storylines and her reputation.

---

One last thing I made for this analysis: If you follow me (@huskerdana on Instagram), you’ll see I posted a main grid carousel for the first time with an “explainer” of this infographic! Here is the slide show I’m putting on social:

Why have I never broken up my analyses into a carousel like this?! It’s so cute! I’ve been honestly inspired by @BachelorData, who creates fun, engaging, easy-to-share videos and multi-slide posts to really make sure the world sees all the hard work she does to capture data about The Bachelor. Plus I’ve been learning Adobe Illustrator and Adobe XD, so I feel like my toolkit is growing in a really creative way. This entire effort took me several months, and I’m honestly so proud of it. I hope dipping my toes into some new design software & new social media techniques will help this infographic reach as many of my fellow Grey’s enthusiasts as possible.

There is so much untapped potential in this Grey’s Anatomy data, and so much to unpack with my fellow Grey’s dataheads. Should I dive deeper into the romantic connection data? Collect new storyline data points? Quantify the ways in which the show has evolved over time? Comment below!

-Dd

I was reeled into the Bachelor universe in the winter of 2010. As a graduate student paying non-Nebraska(!) rent for the first time, cable was definitely out of my budget. Instead, I had a crappy second-hand antenna that could only get the local ABC channel. 

When it was time to unwind, I would turn my brain off, adjust my precarious antenna, and let whatever was on ABC wash over me. My escapes from the stress of school weremy daily Jeopardy, my Grey’s Anatomy Thursdays, and my Bachelor Mondays.

I had never watched a Bachelor show before, even though it’d been in the cultural zeitgeist since 2002. When I finally tuned in eight years later that fateful January, I had the pleasure of watching 25 ladies (of varying mental stability) compete for the love of airline pilot Jake Pavelka. In a series of questionable decision-making, Jake proposed to Vienna in the taped final rose ceremony, only to have a bitter, uncomfortable, on-air post-breakup blowout in the live finale. (BREATHTAKING.) And then, because the franchise upcycles previous contestants to become the next lead, I was instantly invested in the next Bachelorette’s second-chance storyline. It’s at the same time a vicious cycle and a riveting pop culture machine.

How I’ve made it this far as a data enthusiast without doing a Bachelor analysis is a mystery. I think I was intimidated by the amazing analyses already out there from the likes of FiveThirtyEight and The Ringer. What could I possibly add?

As I started roping in Bachelor newbies to drink wine, eat cheese, and enjoy Bachelor Mondays with me, I found myself trying to explain the origin story of each character — whose season they were on, how long they lasted, their off-camera instagram feuds, what crazy occupation they purported to have. And I realized, while there are a number of great meta-analyses and deep dives, there aren’t a lot of person-level interactives that let you explore the entire franchise alumni network, from the first contestant out of the limo to the last final rose.

So, for the first time that I’m aware of, here is an infographic of every single Bachelor cast member of all time, interactive and hover-able for detailed deep-diving. This is what I’ve dubbed, The Bachelor Family Tree. Disclaimer: with a franchise that’s 17 years old and counting, it’ll take a few scrolls.

Creating this, I found myself smiling at the “careers” attributed to the contestants. The word former is often prefixed, presumably because the person quit their job to pursue reality TV or was conveniently already unemployed. With my new Bachelor database, I could explore my curiosity. I decided to break down what career types we see most often, in the interactive below. Go ahead! Type in your job, and see what previous contestants had it. (Only 1 contestant had the word “data” in their occupation. Hello, John Wolfner from Emily Maynard’s season, ya fellow datahead!)

Here, you can definitely see the gender-normative jobs for women (child care, teaching) as well as men (construction, engineering). But there is some unexpected parity. For instance, there is a fairly equal number of male/female lawyers and personal trainers. It’s also interesting to see, overall, what job types are common or uncommon. A lot of people in Sales, which makes sense if you think of a stereotypically outgoing (read: camera ready) salesperson. Not a lot of people in politics, probably because of the exposure downside; hope your tequila company works out, Luke Stone. Younger women are “Students” (because they’re on the show before starting a career), older men are Doctors (because it takes a post-graduate education). It’s a fun albeit skewed sample of the millenial job pool. Funeral director. Jumbotron operator. Italian prince. Not so different from your LinkedIn network, right?

The last thing I thought folks would want to explore is age. Our most recent Bachelorette, Hannah Brown, is the youngest lead ever at age 24. She picked the youngest ever male suitor (Jed Wyatt, age 25). Our oldest ever lead, Byron Velvick (age 40) in a surprising good-for-you move had the oldest cast (average of 31 years) and the oldest female winner (Mary Delgado, age 35). So I thought I’d not only visualize the age distribution of Bachelor Nation, but the ages of contestants relative to their lead for some matchmaking context. 

As I expected, every single Bachelor had a cast on average younger than him. All but three Bachelorettes had a cast older than her. Except for Byron’s season, the average age is predictably between 24 and 29 for both genders of contestants. While this is interesting from a data perspective, it also makes a certain amount of common sense. The number of beautiful, compelling, eligible people (who are also willing to compete on a reality dating show) likely drops off with age.

I could make a dozen more interactives with this data. And I might, later. Much later. This was probably my most intensive project thus far for the blog. Creating a database of all these contestants, organizing a timeline, reading decade-old articles; I felt like Michelle McNamara researching the Golden State Killer but waaaaay more pointless. But I know I learned a lot (one of the O’Connell brothers was the Bachelor, guys! nope, not the one you know!), and I hope my beloved readers enjoy the fruits of my nonsensical labors.

Happy Bach-ing, dataheads.
-Dd

It’s almost Halloween, dataheads. And I came across quite the fright as I scrolled through Twitter today.

I got a promoted tweet for Texas Senate candidate Beto O’Rourke, soliciting campaign donations. And this data viz? Insert scary movie score here. Take a look.

What grinds my gears about this is that it violates a data visualization tenet: never be visually misleading. But here, it both misrepresents Beto’s position in the race and confounds his call-to-action for donations.

Why? Well, Beto’s “rainbow” takes up more real estate on the graphic than Cruz’s. It has more volume, more length, and has a longer arc. My brain looks at the picture, and before it has a chance to process the labels at the ends of the arcs (46% vs 50%), I might conclude that Beto has overtaken Cruz. So not only is it an inaccurate rendering, but it might actually make me think Beto doesn’t need my ten bucks.

Yes, I see that Cruz’s arc is more “complete,” I guess, than Beto’s (and I do like, visually, stacking Beto on top of Cruz to give a sense of victory). But this is marketing disguised as data.

The point of data visualization is that it capitalizes on our brain’s “pre-attentive attributes.” If designed well, when we look at an image, our eyes go immediately to the pattern we’re supposed to recognize — or the discrepancy we’re supposed to notice.

 

So I played around with a couple different options for getting Beto’s message across while still representing the data truthfully and compellingly. I tried a few options with donut charts (which is what the original designer used, essentially). But differences in angle just aren’t that powerful to the eye, in my opinion. That’s why I’m loathe to use pie charts.

 

But then I realized, you can do some subtle things with font sizes and colors to make Beto stick out, without distorting the chart itself. In fact, simplifying the chart could actually make it more powerful!

Getting rid of the unnecessary curvature effect actually makes this better. And it gave me more options for playing around with the label size and colors. My eye can go to Beto (because he’s the important one in this ad), but my eye can also see that gap in the polling numbers.

Any other thoughts on ways to improve Beto’s message? Anyone else want to toss $10 Beto’s way? Comment away, dataheads.

-Dd

I’ve gone down the Lifetime Movie rabbit hole. In a manner not unlike so many Lifetime female protagonists obsessed with solving the case, uncovering a conman (or conwoman!), and/or surviving being trapped under a pool cover for an entire holiday weekend.

Honestly, I’ve always enjoyed the plot audacity and general digestibility of a Lifetime movie. They’re not even a guilty pleasure, because I don’t feel guilty for watching. Then, a few weeks ago, my friend told me he worked on a recently-premiered Lifetime movie (A Dangerous Date) — AND, he and two other friends are pitching ideas to the network for new original movies. This got my data juices flowing. The rabbit hole was dug.

I wanted to use data to help my friends become successful Lifetime movie creators. (And I fully expect to be credited in the end-scroll.) Primarily, I wondered:

1. What are the general “kinds” of Lifetime movies? Is the network interested in expanding existing genres, or putting a new spin into an under-explored area?
2. What Lifetime movies are considered “successes”? Is it a property that’s generally well-reviewed? Or a property that’s notorious, regardless of what critics say?
3. Can math help me decide what Lifetime movie to watch this weekend?

Enter, my Complete Taxonomy of Lifetime Movies, based on every single property in the March 2018 Lifetime Movie Club streaming catalog:

This is a network diagram that’s inspired by a poster I’ve seen by Wine Folly, that diagrams all the major types of wine. I like that it’s basically a qualitative arrangement of the topics, the simple purpose being visual entertainment. That is, as opposed to “true” network diagrams that use statistics to determine the size of each circle (“node”), the relationship/distance between circles (“tie”), and all that good nerdy stuff. (Here’s an awesome overview from a Violence Reduction Network presentation about using network analysis to quantify gang dynamics, which crime-fighters can apply to intervene on high-risk and high-influence nodes [based on their “centrality” to the network]. I AM GEEKING OUT HARD. Note that my Lifetime work is not even close to that level of statistical rigor or subject mater importance.)

I like this Taxonomy as an exploratory tool. An introduction to the Lifetime universe. One of my objectives of this analysis was to help my friends identify opportunities. So let’s see. It seems like stalking movies are a big area of interest, but that there aren’t many female-being-stalked-by-an-unknown-person stories (usually the stalker is known to the protagonist). Or, it seems like stories about Traumatic Violence, like wrongly-accused-child-abuse and victim-escaping-domestic-violence, are actually pretty well-received; it’s a heavier subject area than, say, a woman being stalked by the man who saved her from a shark attack. But a good writing team could handle it.

While the Taxonomy is a good way to organize all these individual movies into one big picture, I wanted to play around with some other ways to represent this information. Mostly because I needed some guidance on which movie to stream this weekend. But also, it could help my friend pitch to the business folks. Let me show you what I mean:

Here’s professional me at the Lifetime pitch. “I realize stalking movies are your biggest genre. Among these movies, regardless of how well it was reviewed, only 300 people on average are generating reviews. Whereas Fight-For-Survival movies by far get the most viewers, even though it’s a smaller genre and generally less well-reviewed. I want to combine the strength of stalking movies with the draw of survival movies with [I’m making this up but I love it] our new movie where a woman and her female stalker neighbor must fight together for survival.”

Here’s lazy me on Sunday afternoon. “Oooh, the most-viewed Teen Drama movie stars Jenna Dewan Tatum? I’m in for Fab Five: The Texas Cheerleading Scandal.” [Cue me on the couch with a bloody mary for the next 90 minutes.]

Right now, this analysis only includes the 105 movies in the Lifetime Movie Club streaming catalog. Lifetime has A TON more properties than what’s listed here. So let’s consider the streaming catalog my “training set” for the Taxonomy, and hopefully down the road I can get access to additional Lifetime movie data; the idea is that the categories that emerged from the “training set” should persist even when new datapoints are added. For example, the Lifetime movie we watched in high school health ed class would fall into Mental Health-Compassionate Drama-Eating Disorder-Mother Helps Child Overcome; my friend’s movie, A Dangerous Date, would fall into Crime-Deceit-Protagonist Victim. I hope the taxonomy holds up, but I also would fully-welcome a genre-defying datapoint.

Enjoy the shows, Dataheads.

-Dd

I usually love College Football Bowl season. It’s a magical time at the end of the regular season where strange intra-conference match-ups happen, where talented athletes actually get to rest for a couple weeks so they really shine, and where sponsors insist on longer- and longer-named bowl titles (one of this year’s longest and most ridiculous: the Cheribundi Tart Cherry Boca Raton Bowl … it’s apparently a juice?). There is even a bowl that proclaims it has become sentient, with an amazing twitter account and Reddit presence; I applaud you and miss you, dear sweet Belk Bowl.

This year, however, was bittersweet. Nebraska’s losing record made us bowl-ineligible for only the third time in 48 years. (Whoa, I just realized that stat corresponds to our season record of 4-8; damn you, former head coach Mike Riley, have you been trolling us?). Most of us Husker fans entertained ourselves by bandwagoning onto our newly-hired head coach’s current team. But you know what else gave me holiday cheer?

Data visualizing, baby!  This past week, I created my own dataset about each of the 40 bowl games that have occurred since December 16th; the only one left, by the way, is the Championship Game between Georgia and Alabama in two days. (GO DWAGS.)

I wanted to represent every game in an easy-to-digest, one-page snapshot. As part of this, I thought it would help to sort the bowls so the “most entertaining” were first. Of course one cannot simply eyeball the games to make this determination. There’s no math (and, therefore, no fun) in that. So, I made a simple algorithm to calculate what I call the Entertainment Value Ranking. More on that later.

You dataheads out there are probably wondering what goes into the Entertainment Value Ranking (henceforth: EVR) that I calculated. I chose four metrics that I could calculate based on the scoring data I collected. The factors I included were:

• Total Scoring = Team 1 Final Score + Team 2 Final Score. All things being equal, I enjoy higher-scoring games more.
• Final Score Difference: Team 1 Final Score – Team 2 Final Score. The closer the final score, the more entertaining it is. This also helps to downweight games like Cheribundi Tart Cherry Boca Raton Bowl, where FAU laid down a solid 50 in a blowout against Akron (who only scored 3).
• Lead Change Events: Number of Times the Leader Status Changes. If one team is ahead, and the other team ties it up, that’s one event. If one team takes the lead from the other, that’s one event. I just counted these and summed ’em up.
• Lead Lost: If Losing Team Was Ever Ahead, Amount of That Lead. It’s fun to see a come-from-behind situation. So this helps upweight those games.

Because these are each on difference scales, I normalized each of the raw values (i.e., put them all into a 0 – 1 scale), added them up, then ranked those sums. And, based on the handful of games that I watched, I think it actually does a good job representing which games were competitive and fun to watch. The Rose Bowl, which was a double-overtime shootout, should of course be #1. I swear it’s just a happy coincidence that my sweetheart Belk Bowl came in at a solid #2. I was a little surprised to see the Peach Bowl come in at a low-seeming #10; adding in a Scott Frost “bump” was tempting, but ultimately seemed self-pandering.

In the spirit of transparency, here’s a table showing how each bowl’s EVR was derived:

Last, I need to give a shout-out to Tableau blogger Chris Demartini, whose viz on golf tournaments taught me how to create step lines, and inspired the “small multiples” chart layout. Awesome work, dude.

What do you think, dataheads? Any other factors you would incorporate into an EVR? Any thoughts on bowl game sentience? Comment away.

-Dd

For being such a self-proclaimed data geek, I am surprisingly inept at fantasy football. Since 2012, I’ve been in the same ESPN Fantasy League, named “Game of Throwin’s” (applaud our commissioner Megan on this amazing pun). My cumulative record is an abysmal 24-43-3.

As a believer in statistics, I’ve always used the projections to optimize my roster – i.e, for drafting and setting my weekly lineup. I start the season off with a decent team. My starters generally outscore my bench. But I realized that, basically, once my starters get injured or underperform, I’m scraping the bottom of the waiver wire barrel to fill my roster. Ergo, my team always gets worse as the season progresses.

This week, I’m trying something new. I’m going to attempt a trade!!!! I’ve never actually initiated one before. Never? It’s true! I blame this on the ESPN app’s subpar information design. Fantasy apps in general do not a dang thing to suggest trades that might be beneficial. You just have these lists of names and projection numbers. There’s no way to look at all the information, all at once, in a meaningful way. It’s impossible for me to (1) see which of my positions are strong vs weak, and (2) simultaneously which of my league-mates have complementary strengths/weaknesses.

So I took our Game of Throwin’s rosters and joined it with publicly available fantasy football statistics, using data as of Week 4. My hope was that “seeing” it all in one place would help me more easily evaluate potential win-wins (and, ultimately, propose a trade that my league-mate would actually accept).

But this viz started to make me curious. How is it possible that Breitbart Starr has the highest performing roster in the league, but hasn’t won a matchup yet? Is it poor starting lineup management? How come Zeke Is Free looks average, but is undefeated? (And is en route to beating me this week…) Perhaps it’s more important that you have high-scoring stars who drive your starting lineup, versus a stockpile of better-than-average players, one-third of whom just sit on the bench anyway?

I played around with the data a bit more. I took all of the players who appear on our rosters and, by points-per-game, assigned each to a “scoring quartile.” So, at one end, the top 25% of players with the highest points-per-game will be in the Upper Quartile; at the other end, the lowest 25% will be in the Lower Quartile. That’s what you see here:

Then, I looked to see how each team stacked up in terms of how many high-performing players they had. Maybe this is a better predictor of our league standings?

Welp. Inconclusive, huh?

OK, I do think I can make sense out of these analyses. I see the trend. Fantasy football performance is a freakin’ crapshoot. Maybe record and effort are unrelated. (Hey, Dana. Rationalize much?)

Readers, I promise to embark into the wild west of trade proposals. ESPN’s app didn’t make this easy, so I hope my data viz efforts pay off.

Any other ideas how I can use data and visualization to improve my performance? ANYTHING? PLEASE??? Comment below, y’all.

-Dd