I will eventually talk about nursing homes and older adults today, but first I wanted to talk about online dating. Wait, what? Yes, I’m recently single, but that’s not why I’m on this subject.
Bear with me. So I was watching a TED Talk video recently, plus browsing some articles about predictive analytics and predictive matching algorithms. The video was here and yes, it relates to romantic matchmaking – increasingly an important way that single adults find romantic partners in this day and age.
There are a number of dating sites out there. Many take the approach of simply having their clientele create a profile, add pictures, give them access to other profiles, and then allow people to somehow message each other for dates. This is basically the approach of sites like Blendr / Badoo, or AdultFriendFinder – you see something you like, you message them. There are other sites that cater to “special interests” – such as married people who want to cheat on their spouses (AshleyMadison.com), or for specifically finding and dating school alumni (DateMySchool.com).
I wanted to focus on the sites that do predictive analytics, or predictive matching. These are sites like EHarmony.com, which uses a personality profile matching system to find you good dating partners – Eharmony was apparently founded by a psychologist and they tend to be staffed by people schooled in statistical techniques and personality theory.
Then there’s OKCupid (my personal favorite!). Unlike EHarmony, they don’t focus exclusively on personality as part of their matching algorithm. What they do is when users log on, after they create a profile they must answer a large number (at least 75, if I remember correctly) in order to optimize the algorithm and start getting good match predictions from their proprietary system. A lot of the questions seem rather random, and relate to issues like politics, lifestyle, aesthetics, etc.
So how do they construct this algorithm? Basically they do three things. First, they ask you the questions, and then you, as the user, are required to provide your answer. Second, they ask you to rate how you’d like the other person to answer (‘answers you’d accept’) and then third, they ask you to rate the importance of the question. After tabulating your answer, this allows OKC to weight your answer and then use it to calculate a match percentage. On OKC, if you have a match of 85 percent or more with someone else, its widely considered that you should probably check them out and maybe even go see ‘em for a date (of course, there’s debate about whether this faith in these proprietary matching algorithms is misplaced or not).
In practice, the OKC algorithm operates both using theory and empiricism. To a certain degree, the OKC algorithm doesn’t appear to care what the user says when they answer – it just needs to know how each person answers a particular question and then provides them with a match percentage that’s based ideally on the desired outcome (which in OKCs case, is whether you’d disabled your profile and indicated you’ve done so because you’ve successfully found someone to be exclusive with). In other words, OKC’s software is interested in the following: based on past experience of successful matches, how likely is it that a pair of potential partners who each answer a given question in a particular way or pattern going to be a successful match?
To be fair, it appears based on the little I know about the OKC algorithm, there is some theory in how they construct their questions. But I wanted to focus on empirical test design and pure empirical matching algorithms. I often think that psychologists and social scientists spend too much time on theory and not enough time on utilizing the raw predictive power potentially found in computational mathematics.
How can this apply to nursing homes? One thing that I was thinking about is the periodic issue of poor roommate matches. Here at the VA nursing home where I work, a majority of the patients are in double rooms, some are in four-person rooms. There’s a lot of issues that go into putting roommates together in nursing homes. The ‘first cut’ issues, of course, are things like microbe compatibility (MRSA positive / negative), whether a resident needs access to wall oxygen, a bariatric room (e.g., larger toilet, larger bed, etc).
Once those issues are taken care of, then the “art” of roommate matching takes over. What this looks like, in practice, is an animated discussion which takes place primarily between nurse managers and the physicians (with me occasionally joining in). It’s clearly a fun discussion for most, because of the inexact nature of it. It goes like this:
“So we should put Mr. X in with Mr. Y. Mr. X. is quiet and so is Mr. Y.”
“But Mr. Y. is African-American. Didn’t you say Mr. X has made some racial comments in the past?”
“Yes, but I’ve seen them chatting pleasantly. They both like football too.”
“I think Mr. Z. would be a better choice.”
“But Mr. Z is a night owl, and Mr. Y likes to get to sleep by 8pm.”
Over the years, I’ve observed that when roommate matches go poorly, it can result in all sorts of untoward events. It can result in time and labor-intensive moves of patients and their belongings. It can result in “behavior problems.” It can result in fights. All of these things are negatives for patient health and well-being, and are an unnecessary drain on nursing resources.
Years ago, I witnessed how bad this problem can get. We had a resident (let’s call him Mr. Bob). He was a latino male, and was very sensitive about his racial background and very sensitive about sleights and perceived them frequently as being borne of racism (which may be based on many painful, real experiences he has had in his life). He also had issues with paranoia that were likely at least somewhat secondary to his previous cerebrovascular hemmorage, which had left him wheelchair-bound and with some cognitive impairments.
Well, we tried to get this gentleman properly matched up with roommates. His first roommate was a 90-plus year old gentleman and Air Force veteran, who was born and raised in the Bay Area but whom Mr. Bob immediately had trouble getting along with. They began sniping at each other almost immediately. We moved Mr. Bob, and moved him in with another older, white gentleman (most of our residents are Caucasians) and Marine veteran, but one whom we thought would be a good match with Mr. Bob has they both had asked to room with each other…. Guess what, it was even worse. The two eventually stopped speaking to each other, were calling the nursing station and hitting their call buttons constantly to complain about each other – it was a nightmare!
After a couple more moves, we found a roommate for Mr. Bob that seemed to work and he’s been fine now, more or less, for the last few years – but obviously it required a significant degree of trial-and-error to get the job done, a lot of nursing hours and time wasted, and along the way, lots of unneeded “behavior problems,” fights, and lost sleep of residents that may have been avoidable.
There is a good deal of theory and anecdote out there as to what makes for good roommate matches in the long term care environment. There’s a body of research on roommate matching that can be drawn from studies of undergraduates (who are often the preferred guineau pigs of psychology departments, since they are the most available), but that may have limited applicability to the geriatric, long-term-care crowd.
But the thing I’ve been struggling with is how to go about tracking this as an outcome. In the case of OKCupid, they have a great way of doing this – when people disable their accounts, they are asked as they are leaving, “why are you leaving?” Users are then able to tell them that they found someone, and then OKC asks them who. Bingo, there’s the outcome data, which can later be mined for variables to further optimize their matching algorithm. How would this get done in the nursing home environment?
Well, we could ask, I suppose – via questionnaires and the like (which of course now starts to sound like it would require formal research – given the issues with privacy and risk posed by “rating” each other). Of course, there’s the issue that sometimes, roommates (like romantic partners) may be attracted to each other as potential matches, but may in fact be terrible for each other (like the above example).
Sure, there are any number of potential issues, such as personality, politics, race, culture, medical issues, family visits, et cetera, that may make or break roommate matches. But until we start tracking this important outcome, I don’t think there will be a way to get a handle on this in the future.