A theme in my blogs is how the “structure” of data – rather than just the “content” – affects what that data can say and is capable of doing. In particular, I suggest that certain structures tend to reinforce certain contents; this means that a structural imposition can have an effect similar to a contextual imposition. Structure is an interesting conversation because, at first glance, arguably data has none. Structure is something that is attributed to physical things. But data normally has no physical existence – at least not in a tangible sense except maybe in relation to storage devices. Yet I am certain that almost everyone who has programmed in an object-oriented language has had to deal with multifarious data structures. The question really is “how” the structure of data can influence the use and interpretation of the data. I actually go a step further in this blog by already assuming that most readers will probably go along with me; for me, the question is how to implement the concept of “accommodation” in data in order to overcome the constrains and impositions inherent in structural expression.
Even a series of numbers can have structure: 1) it is possible to express the data in a particular order; 2) there is a high likelihood of periodicity or regular intervals; 3) the unitary basis is probably homogenous. Not all readers will recognize the latter point as structural. If I say that all of the units of something being counted can be expressed as oranges, chestnuts, or pistachios, the units are actually structural in a physical sense. Still, for those that make use of data as primitives (such as a series of numbers), the issue of structure might seem unimportant or superfluous. Primitive data structures go well with a “commodification” regime. I personally am not surprised that primitive data has followed growth in capitalism. Nonetheless, as a general observation, I would say that there is hardly anything structural about a primitive body of data. Lack of structure invites certain types of analysis (externally-defined and quantitative) while discouraging the use of more abstract objects (such as the internal narrative); the latter seems less likely without the support of structure. I don’t believe object-oriented programming was around when Karl Marx was alive; but I believe he would have recognized the usefulness of being able to retain constructs in order to address disassociative concerns such as “alienation” among workers and in the workplace.
I have chosen to formulate my structural preamble in a completely non-numerical manner; this “non-numerical” approach is possible in relation to structural manifestations. Often when major structures are being constructed in cities, there is considerable debate over long-term development. Because the structures will persist for many decades, they will tend to reinforce the intended use. Not only this, but they also entrench the intended “manner of use,” creating beneficial bias among users. For example, if a city is constructed without considering wheelchair accessibility, the city forms a “physical” bias against certain people with disabilities, the elderly, those that have injuries, those that are pregnant or pushing strollers, emergency responders, and shoppers that buy and carry lots of merchandise. At this moment in my life and perhaps for some time into the future, I am experiencing difficulties going long distances on foot; this actually affects what I do and where I choose to go. There are different forms of structures that affect personal abilities. As I mentioned earlier, I will be covering “data structures.”
Can a data structure behave like a physical structure – to create disablement affecting certain segments of the population? Imagine a “victim impact statement” in court proceedings being reduced to a tabulation of costs, which I believe some would argue represents an accurate depiction of “impact” to the victim. I read some insurance agreements containing lengthy tables indicating the maximum payout for different forms of dismemberment. Suffice it to say, it is possible to express data in a manner that sets boundaries on the impact to the victim. For a master craftsman, loss of her hands might mean permanent impairment of livelihood. But there are more quotidian examples. Consider a client contacting a warranty department for product support. The agent instinctively attempts to pigeonhole the call. Is the client complaining about fire damage, rust, discolouration, chipping, denting, flaking, scratches, cracks, missing parts, wrong trim, right trim wrong colour, warping, misalignment, broken welding? The client says, “My daughter touched the surface of this product, and she badly burnt her fingertips!” Unfortunately, there isn’t a pigeon hole for this complaint. So the data-system cannot directly express the context of the call. The data becomes insulated from reality in a structural sense. Somebody had created a persistent bias – having the effect of concealing a certain aspect of reality – denying some individuals and their problems of proper recognition.
The specific example involves a relational database where there might be a limited number of fields; where the fields must be predefined; where there might be no basis to invoke a particular tabulation such as physical injury. There is a question as to how often and reliably the designer of a data system can anticipate “every” important development. The risk of exclusion and structural insulation seems quite high. If the inability of the designer is due to negligence or incompetence, then this is certainly one angle or route that might be addressed. A concern that I have formed recently relates to “database dependency.” Data is becoming the watershed for decisions. At some point in the future, decision-makers might, rather than accept blame entirely, declare the data scientist blameworthy for his or her failure to provide relevant and timely data.
In the image above, to the left, I show how disablement affects a person with mobility impairment physically. Oftentimes people focus on the physical impairment of the person as the source of difficulty. But I suggest – and this concept is well-documented in disability literature – the impairment is due to the disabling environment itself. Because so much of the human environment is regulated and controlled by data, certainly the “content” of that data can affect expression within the system. On the other hand, when discussing the “context” of the data, it is necessary to consider the retention, configuration, setting, prioritization, and ability to give notification – its structure (image to the right). In this way, the implementation of code can radically affect the ultimate usefulness of data. Maintaining crime statistics to follow crime does much to perpetuate it; for the mobilization of resources is driven by disconnection and hindsight.
The principle of “accommodation” using data structure is the same as accommodation for people with disabilities – e.g. in relation to physical and workflow structures. I believe that at the design level, it is generally unrealistic to expect a data system to capture every important aspect of business phenomena. It is necessary to make use of devices such as internal narrative during the data-gathering process. I was recently hospitalized for an extended period of time. About a week after I returned home, I received a survey in the mail asking for my feedback. “Well, I met a French-speaking patient. Yet nobody on staff could speak French. The hospital should have at least one nurse who can speak our second official language,” I was going to say. Of course this comment doesn’t fit any of the checkboxes on the questionnaire. “A nurse tried to wake up a patient who just had a surgical procedure done to his heart. According to the patient, he almost had a heart attack,” I thought I should mention. Alas there isn’t a checkbox for this either. It is necessary to incorporate what the patient considers important – if at all possible, in the manner the patient wishes to express the details – the “internal narrative.” My biggest issue with any institutional service probably relates to tolerance for constructive criticism. I am certain some people feel there is never room for improvement. There is no positive place to go from that starting point.
Structural accommodation in the physical world allows people with impairments to overcome barriers. If we consider these dynamics conceptually, the predefined behaviours and activities presumed by the designers can, on the absence of accommodation, lead to the imposition of limitations characterizing the lived experience of people with disability. We make it possible for people to overcome these barriers through deliberate redesign; as a result there is greater personal autonomy. People get to write their own stories rather than being forced to live the stories written by others. Similarly, although many data structures can be primitive, it should be possible to accommodate the needs of others by incorporating their own personal perspectives and insights. So important is structural accommodation in relation to data, I believe that it should be mandated by law and tested to ensure efficacy. Moreover, data scientists should indeed be responsible for the quality of data collected in an organization; this creates a need for accommodation when dealing with data structures.
Structural Accommodation from the Perspective of the Service Provider
Invariably the context of disability is taken in relation to vulnerable individuals. There might be a presumption of a power imbalance between individuals and organizations: e.g. between an employers and employees; producers and clients. Such an imbalance conforms perhaps to more traditional lines of thought suggestive of deep power conflicts. However, it should be evident that organizations these days routinely struggle to survive; they might suffer liability; or be held accountable for poor management, lack of accountability, defective products, or faulty service. It has been my contention that the disassociative disease pervasive among organizations is likewise “structural” – as opposed to attitudinal or ideological. The implication is that a business having certain structural attributes involving its data might become “impaired”: it is possible for disability to arise from phenomena involving organizations rather than just individuals. I have access to historical management records that I believe reasonably support my position; my research remains unpublished due to lack of academic opportunity.
Take for the sake of argument a company that maintains fabulous sales records but nothing else. The company can monitor and track patterns in its sales exceptionally well. Let us assume also that the data system cannot go beyond this type or nature of data perhaps due to logistic constraints. It should be evident on a transactional basis, the organization has almost no information about the story behind each sale; this might lead a data scientist to observe that the story is the sale. The patterns tell the story, as accountants might say. If sales decline 50 percent the following year, what does this mean? It means actually that the data scientist – as I stated earlier – doesn’t have the foggiest idea what the story is behind sales. If the broader pattern conforms to some kind of mass behaviour, then this is wonderful of course until it stops working. Peering into the design of the analytics, the data scientist responsible might not care about what triggers sales; he or she makes no provisions to obtain and retain that data.
I accept that it can be quite challenging for, say, a cashier to obtain data from somebody about to make a purchase. Let’s say that a retailer wishes to display on a map everything that I wish to purchase during a particular visit both to expedite my search and also ensure that the items are available. Before I go to the store, I provide a list of items that I wish to purchase. For the items not in the store but which I wish to purchase, I provide a reason. I remember once I was looking for a computer bag at Wal-Mart. How can a store sell computers but have no bags, I wondered. I bought the bag from some other place. A few months later, bags appeared at Wal-Mart. But that just isn’t fast enough, right. So there can be a “back to school” or “guy has a new job” template. One wonders what the map-path might be for the “I’m feeling hungry” template. What I am doing here is moving the data away from the transaction and pushing it towards the client narrative. I do not tell the client what her or his story is. My objective is to enable fulfillment. Enablement requires more complex data forms. A sale is something in the past where information can be stored on a simple table or tape. For narrative-driven data, it is necessary to have data objects for the person, situation, circumstances, and expectations.
There is expression “by” the self. Then there is expression “of” the self. There is expression for the sake of “inclusion.” Then there is expression for the purpose of “participation.” If a person is asked whether she likes the flavour of a particular cola, her response is a matter of inclusion by the self. It can register as a tick mark on a table. If her involvement in the process were limited to delivering a response, on a fundamental level she has been deprived of her rights to actual expression. For all that she is has been reduced to serve not just a corporate need but to remain within the predefined boundaries of quasi-intellectuals in the marketing department. We therefore see that within this constrained paradigm, there would be no place for structural accommodation. To invoke structural accommodation is to care enough to take into account how the client actually feels – and to convey this essence through the data. “Did you like the flavour of the pop – yes or no? Does it taste better than this other brand – yes or no?” If we return to the pop-tasting example, the company isn’t concerned about the person but rather the aggregate. On an individual basis, there have always been many people who dislike excessive sweetness in pop. These are the people who have stopped buying the products. The marketing experts can have meetings over cocktails while nibbling on shrimp – talking amongst themselves. Or they can collect real client data.
Having the opportunity of expression at the level “of” the self rather than “by” the self is made possible through the structure of the data. I would therefore argue that the ability to participate in society as a human is enabled or disabled by data structures. A person doesn’t even have to be human to be quantified and handled like cattle. I suppose centuries ago there were beasts of labour such as mules and oxen; these days there are robots. Many aspects of the conversation remain much the same. People are expected to give up certain needs and desires in order to function properly in a work environment. They continue to express “by” the self – the narrative belonging to the organization. But they seem much less likely to express “of” the self – the narrative belonging to the individual. I think rather mysteriously, these dynamics now extend to clients whose involvement in an organization might be to make a purchase or buy products elsewhere. The dynamics can be explained if an organization has systematically and pathologically encouraged structural primitives to form the basis of decisions. An organization might treat its employees and clients in much the same way because its intellectual capital is not meant to be sensitive to the reality “of” the self. Structural accommodation gives voice to individuals and restores their place in organizational development.