Tuesday, June 24, 2014

Can You Use Math To Figure Out Whether Someone Wants You?

Yes. (I'm linking out to an external page because this post has interactive visualizations which Blogger won't let me include.)

Tuesday, June 10, 2014

The Statistics of Crisis Counseling

Trigger warning: this post discusses mental health and suicide. Out of respect for the privacy of the callers to the counseling center, it is at times deliberately vague. Huge thanks to the counselors for their patience and thoughtfulness in working with me to make this data available.

My senior year at Stanford I lived at Stanford's Bridge Peer Counseling Center, where from 12 - 9 AM I took calls from anyone who needed to talk. I liked counseling because, like statistics, it let me strip away the noise: a 3 AM call reveals truth usually veiled beneath smalltalk or shyness. It’s a fascinating privilege to hear people explain so bravely and honestly how they work, to see instantly into the clockwork of human nature.

Since February 2011, Bridge counselors have taken nearly one thousand calls [1], and for each call they summarize basic details (while preserving the caller’s anonymity) to keep counselors apprised of important developments. My job was to look at this data. My first year I did the analysis by hand; my second year I wrote a program to do the analysis automatically; and recently I created an online robotic statistician that allows non-programmers to produce reports just by clicking a button. If you are interested in setting something like this up for your own counseling center, or for another purpose, please shoot me an email (emmap1 at alumni.stanford.edu).

So do we learn from our robot? Even the basic statistics are interesting. Try to guess (answers at end):

a) What time of day people call in most frequently
b) How long calls last on average
c) The length of the longest call
d) What percentage of calls come in between midnight and 9 AM

Our statistics also allow us to answer more important questions. An obvious one is why people call: it’s a cliche that every Stanford student is happy, but it’s almost as much of a cliche that students are often secretly unhappy, and nothing makes this more obvious than the Bridge data.

Both men and women call most frequently because of relationships (not necessarily romantic ones), which account for 39% of all calls. But after that their reasons diverge. Women worry more about school: they are more likely to call about academics (29% vs. 16% for men) and stress (34% vs 21%). They are also more likely to call about eating disorders, although this topic is rare in both genders. Men are much more likely to call about gender/sexuality issues (18% vs. 4%); in my experience, this was often related to coming out of the closet.

I think people call the Bridge in part because it’s easy: you’re talking to a fellow student, not a psychiatrist, and a phone call allows anonymity as an appointment doesn’t. Substantiating this, about ¾ of the time, the counselee phones in rather than coming in person. Interestingly, however, the counselor reports feeling more helpful if the counsel occurs face-to-face. This could be because it’s easier to connect with someone, but it could also be because of population differences: people who come in person might be more comfortable talking about their feelings.

I’m often asked how you prepare to be a counselor at the Bridge, and when I tell people that you take two quarter-long courses, they often ask the obvious follow-up: does this really prepare you to counsel people? But according to the data, there aren’t any obvious gaps in the training: I broke down calls by various categories (what the call was about, whether the caller was a grad or an undergrad, etc) and all common types of calls went well.

People also often ask what the suicide calls are like. These calls are rare but obviously pretty important. Suicide callers discuss depression two and a half times as often as callers overall, but they also talk about more common issues: academics, relationships, and stress. Suicide calls are longer, lasting an hour on average (as opposed to 40 minutes for calls overall), and more likely to result in a referral to Stanford’s professional counselors.
 
This was the only time I’ve attempted to create a robotic statistician that would just report everything interesting about a dataset, and (though I’m perhaps biased) it taught me that human statisticians are still necessary. Statistics is interactive: you ask a dataset a question, get some answer, then ask a follow-up question. So you can’t just hard-code in all the interesting questions a priori, and if you could create a robot that knew enough about people to ask good follow-up questions, you could probably just create a robotic Bridge counselor (I’m working on it) and save yourself a lot of time.

And even if you did manage to create such a compassionate machine, there’d be a limit to what it could learn from this data. Ultimately, if you want to understand why these people are in pain, you need to do more than print out statistics of calls: you need to listen to them. We say the most amazing things at 3 AM, and not just when we’re calling the Bridge -- when we’re falling in love, or getting drunk, or both at once -- and these moments are as precious as they’re mathematically ineffable. Which makes the statistical lens both indispensable and incomplete: everything is mathematical, but mathematics isn’t everything.

Lately I've been thinking a lot about seeing the world through this double lens: seeing, in a spreadsheet, the life that underlies the numbers, and seeing, in a public place, the numbers that underlie the life. Which I find enriches both experiences:

Life beneath numbers: It's easy to get frustrated when you're messing with some matrix math and you forget why the matrices matter -- I think that's why so many students hate linear algebra. I’m a happier statistician when I remember that each row captures part of a person, and I’m a better statistician when I remember that the row fails to capture an enormous number of important things about that person. (If your matrices have nothing to do with human lives, and this bores you, I might suggest finding different matrices.)

Numbers beneath life: People-watching is more fun when you view everyone as a datapoint. I got a bit carried away at the farmer's market the other day and hissed at a happy couple “you are data”. When they lift the restraining order, I'll explain to them that I was trying to comfort them. Because to me it is comforting that our flirtations and day-to-day actions are not just random events that don't matter, but part of a larger pattern. We are not alone in what we do: we move along invisible strings. This was what always drew me to physics: that underlying the chaos of everyday existence were mathematical rules of astonishing beauty [2]. (This is also why I've always loved this scene from V for Vendetta).

I think the power of this idea is one reason that people are so eager to apply mathematical and technological solutions to complex problems. But I also think the Bridge is a reminder of what you can do with no math at all -- I'm proud of our robotic statistician, but the best work done at the Bridge has never been statistical. In the midst of Stanford’s high-tech campus, the Bridge is a place that believes in the comforting power of simple things. They don't offer the bouquet of drugs the chemo or psych wards can, but they have almost as many kinds of tea, which they'll serve to you in chipped mugs from the 70s that proclaim, "Need to talk? Call 24/7!" It's a place where the lights won't turn on and the doors won't lock and the stove smokes ominously when you try to boil water; where the tiny kitchen is filled with semi-edible fossils, and the bookshelves with diaries, from counselors from years before. It is quaint and historic and occasionally helpless in the face of crises that require doctors or psychologists or even the police. But when the dawn light filters into the kitchen as the counselors are making each other tea, there is no more hopeful place to be; and even in the dead of night, it usually works out. There are problems that can only be fixed, it's true, with lasers or nanotechnology; but humans have been suffering since long before such things, and seeking relief from simpler sources: a smile, hot chocolate, a hug.

Answers to quiz:
a) What time of day do people call in most frequently? 9-12 AM (25% of calls)
b) How long do calls last on average? 40 minutes
c) What is the length of the longest call? More than 5 hours
d) What percentage of calls come in between midnight and 9 AM? 31%

Notes:
[1] In fact, they've taken more than that; I haven't worked at the Bridge since August 2013, so my data is at this point out of date.
[2] This isn't, of course, particular to physics and statistics, but those are the fields I know best.