In defense of OTHER
I was trying to get the registration stickers for my car. I should’ve gotten them, but I hadn’t. Instead, I’d gotten three different letters from the DMV informing me that my registration was paid but not complete. The automated system wouldn’t send the stickers. No one knew why; certainly not me, since I had paid the registration far in advance and gotten the required smog test, and not the woman at the DMV who answered after I sacrificed three hours on hold in the traditional supplication.
“I see that you paid, and I see your vehicle. There’s just a code associated with your vehicle,” was what she told me. I had been working on a puzzle during my hours on hold. This is a waiting activity I recommend. I couldn’t trust the automated call-back queue after they hadn’t actually called me back. So instead I had been hanging onto the line and reducing the waiting music to a tolerable level and finishing puzzles.
“Can you help me understand what that means?” I asked. We were two people running scripts at each other. Hers from whatever gods helm the DMV and mine from qualitative interviewing and time spent embedded on engineering teams.
After a long pause, the woman on the phone said: “I have no idea." She sounded sincere and human, which is to say, she sounded pissed. Script broken, contact established. “They don’t give us anything but the code. All I can do is look up what the code stands for. This one stands for OTHER.”
OTHER is a category paradox, a response to say response not covered.
Confession: as a researcher I have written many OTHER categories into existence at the point of data collection. I know, I know. I know it's unsatisfying. Usually these are surveys. Sometimes they are my own labelling of a scenario, or scoring of a performance. Sometimes they are bins into which I shunt a hundred thousand or so datapoints. Response not covered. Interesting but unfortunately not the point of this study. Someone decided not to instrument whatever this was so I can’t make a prediction of it. I don't trust it. I was surprised.
I know also that I choose to go into this world of measurement trade-offs. Research is always a sandbox, always a construction of a limited field. There are a lot of possible things you could store in a spice cupboard but it’s understandable your labels would say pepper and not dragons. Yet there are also good reasons to open a paradox, even a tiny one, even in your sandbox. Sometimes you have limited fields in the free version of the survey software your lab lets you use. Sometimes you only have a few labels that you feel confident spelling, defining, or throwing out to your participants, and you want to give them the chance to be generative participants, not just reactive participants. Sometimes you don’t trust the people you’re working with to protect the complexity of your participants and you write OTHER for things you already know, or live (and you hope that it doesn’t hurt people who know, or live, like you).
We--data people, researchers, some random engineer at the DMV doing her best--create standardized responses because we want to reify patterns. This requires cutting things up and picking dividing lines, even when we know real things might bleed over each other in the margins. We accept a lot of penalties when we define measurements: the flattening of vast spectra of experience into homogenous ranks and order numbers, the “best-of” choice that we know every reader has to make, the extraordinarily profound limits of our own imaginations. Still it’s a tool. We live for patterns. We live because of best-of choices, sometimes.
And then sometimes we use OTHER. OTHER is a pattern breaker. OTHER is an admittance of the missing. Writing OTHER as a response option, every quantitative researcher knows, takes away some power from your analysis and shifts it into the unknown investment that is personalized accuracy. Stakeholders don’t like this. Stakeholders don’t often hire you for this.
But OTHER can be beautiful, a validation, a pilot test in miniature, a rummage bin for people's unwrangled self-words, their non-standardized representations.
The woman at the DMV gave me the phone number of the Meta DMV, DMV Plus, the place where codes go to die. My problem had become some kind of higher order problem, and I needed a translator. I opened up another puzzle.
Meta DMV had the kind of hours that speak to organizations under meta stressors. You could only reach them during a lunch hour window on a particular weekday. There was no callback waiting queue.
I called and called and called. I logged in and out of the DMV website. The OTHER code was not visible on my customer UI. It was invisible, backend, logged somewhere. I had already paid my registration so I couldn’t pay it again (I was about ready to). I didn't want to drive my car with an expired (??) registration so OTHER cost me experience as well as time.
OTHER can be insulting. OTHER can be painful. I have no desire to pretend it’s not, least of all when otherization in the sociological sense is itself a decision with cascading permutations. Those included and excluded from “default” definitional measurement have always known this. This exclusion shows up in our data science, and in our standard methodologies, and is all the worse for being invisible to so many who create those methodologies.
I have faced down a lot of surveys where OTHER is the only label that fits my response and it does not feel like it will ever help: for example, at many “women in tech” events that might discuss sexism without ever imagining the ways in which gender-expansive people experience different weaponizations. The push for data-driven rarely extends to being measurement-driven. Being data-driven is usually about the point of analysis, not the point of collection. This is why they ask so much about coding languages and so little about causal inference in most data-related job interviews.
I’ve lived through a thousand mundane moments of grief about losing rich measurement working in this field. Data cleaning courses seem to spend a lot of time on dumping, not expanding. The fact that quantitative work sees success as majoritizing patterns means that what we dump in OTHER itself is rarely random and often systematic. Plus feature reduction is efficient and efficiency is, of course, profit. Being called inefficient as an applied researcher can feel like a death knell.
So it’s not a fix, using OTHER. I think it can be like cracking open a window, though. There is always a tension between exploratory research and confirmatory research. There is always a struggle to own the definitions and not have them applied to you. It is difficult to attain the power and the agency to create labels. Sometimes what you can do is sneak in an OTHER. Sometimes it’s breathing room against all of the frustration, a tiny little moment where you say: hey. YOU tell me.
I finished two puzzles in the course of never getting anyone at the Meta DMV to answer my calls. Several weeks later, I got a very tiny reimbursement check. Mystery thickened! Then I got a letter with my sticker that said: your process is complete.
(Ok, I never found out what the code meant. But here is my best guess: I had moved, but I had paid a registration fee with my car still listed to an old address without realizing it, maybe even while my address change was still in process. My registration had therefore been overcharged by the value of approximately three dollars and fifteen cents. This shunted my registration away from normal DMV processing, and into some baroque and time-dilated reality of reimbursement. They probably couldn’t send the sticker before they’d processed the reimbursement. Edge case, conditionalities I couldn't see, OTHER. Or maybe it was pandemic vibes hitting the DMV, I don't know. Everything has been an OTHER since March 2020 hasn't it.
I also got a ticket while my car was parked next to my house waiting for its sticker! So going nowhere didn't save me from the OTHER. I explained that I had paid registration: OTHER was not an acceptable supplication to that set of gods, who had codes of their own, the ticket was dismissed but I had to pay some kind of service fee and reduced ticket for existing while coming to their attention. Category? Probably OTHER)
I’m not telling this story to complain about it--it was kinda stupid to experience, but hardly heinous--but because I keep thinking about it. The invisibility of what that OTHER was supposed to stand for. The impossibility of solving a situation once the definitions were rendered unreachable. That woman on the phone who was allowed to read the labels but not see what they meant. Sometimes when I encounter an OTHER category in my data I play that excellent Sesame Street song in my head, one of these things is not like the other.
This trivial DMV story is an anti-example. I thought about writing down OTHER stories where the label worked--qualitative insights from open text survey responses that drove decision making and changed what we worked on next, or ways that I’ve used mixed-methods research to try to co-create data with participants in the first place. But this story just kept sticking in my head. We’re surrounded by OTHER everyday. We’re dealing with outliers by not dealing with them. We’re deciding that our categorizations must work because we have a code for them not-working, and therefore our sorting of things-into-categories is one hundred percent effective.
Despite this some of us become OTHER spotters, likely to volunteer for the labor of reading the text. I guess I'm one of those. In the classic Sesame Street song you have thirty seconds to identify which thing is not like the others and I think that mounting tension is very accurate. But I like OTHER. It can be a terrible solution but not having that strange category feels like it could be worse. Pretending that our responses are exhaustive when they are not. Worshipping at the church of standard limited options. Pre-selecting only for participant reaction, not participant generation. OTHER is dodgy, a standardization-breaker, but also a space for imagination. I like to think that it lets us admit that we have not drawn a boundary around the entire world just because we wrote out five possible answers.
I try not to create invisibilities masked by OTHER. I still like writing those small OTHER options whenever I can. I know, inefficiency, etc. I know that harried quantitative researchers or dismissive data scientists will probably ifelse(x = OTHER, delete) or whatever. But once in a while it must get through. Once in a while someone must read it, and consider the existence of a category they never imagined.