I’ll keep discussing the quantification of all things social in this essay. I’ll also stay on the same book as in my last three essays, covering parts of the recently published ‘Reterritorializing Linguistic Landscapes: Questioning Boundaries and Opening Spaces’. This time I’m focusing on a book chapter titled ‘The Quality of Quantity’ by Kate Lyons.
In summary, this book chapter deals with what you might come to expect just by looking at the chapter title. We are still dealing with the issue of what we get out of qualitative and quantitative approaches. Lyons (31) notes that what became known as linguistic landscape studies (LLS) largely shifted from quantitative methods to qualitative methods when it gained more and more traction among scholars. In short, the statistical approaches were largely replaced by ethnographic approaches, which seek to “position optimal analyses of a LL within careful consideration of the sociohistorical context(s) in which a sign may occur and/or bring about, as well as highlight the fluidity of interpretation researchers must allow in their assessments of signs’ significance”, as she (31) goes on to point out. In other words, the shift was motivated by thinking in terms of processes rather than objects, contextualization rather than context, which meant that relying “on the establishment of discrete categories” was deemed problematic, as clarified by her (31). Lyons (31) dedicates this chapter to going against the grain, illustrating the usefulness of quantitative methods, which, I approve of because it’s basically heresy to those who matter, no, sorry, to those who think they matter. This does not, of course, mean that she is against qualitative approaches, but rather that she “advocate[s] the use of statistics not for validation or irrevocable proof but as a potential complement” to them, to teasing out “subtle patterns”, as she (31) points out herself. She looks into utilizing inferential statistics and exemplifies its use with a study (which I’m sure you can check out yourself, so I won’t cover that part in detail).
Lyons (32) states that there are two key issues that one needs to come to terms with: to count or not to count and if one opts to count, what should one count? Some opt to count, whereas others opt to not count. If you ask me, there is no right or wrong answer to this. It depends on how you opt to approach this issue and what it is that you are trying to accomplish by opting to count or not to count.
Summarizing what I’ve stated in the previous three essays, if we think in terms of singularities and multiplicities, or haecceities, then counting is counter-productive because no matter what you do, say or write, you’ll end up expressing something that is subtractive. In less abstract terms, experience can never be put into words. If it is put into words, it’s never the experience in itself. No matter how many words one uses, one will always fail to match the richness of the experience in itself. The point is that we can’t subsequently add up what we have subtracted in hopes of piecing it back together. Once you subtract something, you’re already counting, you are already doing something quantitative. It’s going to be reductive, no matter what. Does this mean that it’s then just pointless to not count, because in terms of research it’s more or less impossible? No. I actually think this way and promote thinking this way. It’s more of a way of life than anything else, if you ask me. You can, of course, express something that is not a reduction of experience, a mere futile attempt at resemblance, but that’s art for you. Art is great, don’t get me wrong, but it has nothing to do with this.
To get back to the question, to count or not to count, I think there’s nothing wrong with counting, inasmuch as what one counts is thought of as partial objects that come together as components of this and/or that metastable arrangement. One is to be thought of as many; the whole that the many form is only whole in the sense that it’s one, which, turn can be thought of as being one of the many of another whole, and so on, and so on. What’s particularly important to realize here is that the whole is never something original that has been fragmented. The goal is therefore not to piece together the fragments in hopes of reconstructing something that had some ‘original’ form. The ones or wholes, what we think of as objects are just effects, partial objects of other partial objects that are also partial objects of other partial objects, machines within machines that are also machines within machines. What you see is what you get. Counting ones is thus a mere device that can help us understand how those partial objects come operate in unison. Yes, this process is always going to be reductive, no matter what you do, but, as I pointed out, the task is never to complete a puzzle or to mend a broken vase, or the like. What we are dealing with is no longer a closed set, but open ended. Nothing ever begins, nothing ever ends. We are always in the middle of things.
Anyway, what I just went on about is, more or less, what Lyons is saying in this book chapter. The point of using statistics is not to find all the pieces of a puzzle in hopes of completing it, once and for all. Instead, the point is to illustrate how things are interdependent, that is to say defined not in isolation, but in connection to everything else that they are connected (which are, of course, connected to everything else in the same way, inasmuch as they are, of course). The strength of quantitative approaches is that they can help us identify these co-occurances, how things are related to one another, how they appear to be contingent on the presence of something else, whatever that might be in this and/or that context.
To get back on track here, Lyons (32) argues that the answers to her own questions depends on what the researcher seeks to accomplish. The hypotheses and the research questions should help the researcher to pick the suitable course of action. I agree with her problem oriented take on this issue. For example, if you deal with singularities such as the fall, as in all bodies fall, it’s beside the point to quantify it, as pointed out by Gilles Deleuze (‘U’) in ‘Gilles Deleuze from A to Z’. Summarizing Deleuze (‘U’), grabbing objects and letting go of them, one after another, proves just about nothing. Reproductions of the fall don’t really tell us about the fall itself. You can test something, let’s say, a hundred times, to prove that it is the case, but someone can always still doubt that and ask you to test it one more time, only to ask you to do it just one more time, and so on, and so on. Qualitatively, it’s the fall itself that is interesting, not that it appears to hold if we keep letting go of objects. To give you more examples, qualitative study focuses on the conditions of apparition, what is it that might explain why we distinguish leaflets from pamphlets or green from blue? It’s beside the point to examine a certain number of leaflets and pamphlets when we try to figure out why we’ve come to distinguish them. Sure, we may have to involve actual leaflets and pamphlets but it’s not a numbers game. We can also look at a certain number of instances that involve the said colors, but, again, that’s beside the point. But if we are dealing with something quantifiable, like how many legionnaires there were in the Roman legion at a certain time, then, yeah, quantify away.
Lyons (33) moves on to deal with the second issue, what should one count if one opts to count. I won’t get into detail here as I sort of covered this issue in the previous essay,. Anyway, in summary, it’s best to address this issue through one’s hypotheses and research questions, what it is that one wants to find out through quantification.
After addressing the two key issues, Lyons (34-38) explains the benefits of applying two different types of models in quantitative linguistic landscape studies, generalized linear regression models (GLM) and generalized additive models (GAM). I leave it to you to read the specifics on these four pages. Which one is better then? She (36, 38) addresses this question by noting that no statistical model or approach is inherently better than another one, albeit she found GAM to work better than GLM for her data because of the complexity involved and because it allows her to address spatial categories (latitude and longitude). Again, the fit really depends on what one seeks to find out and what kind of data one is working on, as she (38) points out. Usually different models, approaches and visual illustrations have their pros and cons. Some work great with certain kinds of data, whereas others not so. For example, as pointed out by her (35), ideally one does not resort to binaries, such as the indicating the frequency of one language vs. ‘other languages’, but if the number of cases for those other languages is negligible, the selected model may not be the best fit for that purpose. Moreover, presenting something in a certain way tends to result in ignoring or undermining other aspects. Tables are great in the sense that you get to see the counts, as well as the math involved, but it’s only likely that you need to be familiar with the terms used, how it all works, to make sense of it all, which is not so great for someone who is unfamiliar with it all but wants to understand what’s the deal. In other words, tables aren’t great because they aren’t that intuitive. That’s why the graphs tend to work much better. They allow you to get what’s important at a glance, as done by her (45-46) in this study. That said, graphs tend to be limited by what you can present on a page. Firstly, a page forces you to reduce what you have to two dimensions. In her case, it’d be great to see the presence of languages examined in relation to latitude and longitude, but presenting more than two dimensions in two dimensions just won’t do. Sure, you could do that on a map, but then it might risk obscuring what she manages to present in her graphs. Anyway, the two dimensions are constrained by the actual physical dimensions of the page itself. Of course, this is a problem that has to do with academic publishing that, for some reason, still relies on articles and books. Secondly, even if you could present more dimensions, you can’t just cram more and more in the same graph because while it does take into account more, it tends to make it harder to comprehend what’s at stake at a glance, as I pointed out. That’s counter-productive because the main benefit of graphic illustrations is making the work easier to comprehend.
What I really like about this book chapter is that she explains how statistics work, what’s the deal, or so to speak. I especially like the part where she (35) points out that inferential statistics should not be seen as a tool that proves something to be simply true or false (with recourse to a low p-value):
“This point is of crucial importance in motivating and interpreting inferential models, particularly in the case of the field of LLs. Inferential statistics should not be viewed as a way to prove something about a landscape, but an enhance way of describing and characterizings [its] features. The benefit of these techniques over descriptive statistics is not that a configuration of variables can be deemed significant, but that certain aspects can be highlighted and noted for future surveys and that relationships between variables may be more precisely characterized.”
Thank you! To paraphrase this, using statistics, going beyond description (counts and ratios), can be used to indicate which phenomena tend to co-occur and, conversely, which do not. That’s interesting in itself already, even if it does not tell us why some phenomena seem to co-occur whereas others don’t seem to co-occur. By doing just that, indicating that, hey, these things seem to be linked, pushes us to look at those things closer, to investigate them further, which may even lead us to figure out why it is that they keep appearing together. In short, this helps us see patterns. It’s as simple as that.
She (35-36) provides an example to make this easier to comprehend. So, let’s take language as one variable and combine it with some other variables. The problem is that looking at percentages alone makes it harder to see if there’s some connection between them. This does not mean that it’s impossible to intuit that some things seem to be linked, whereas others don’t. I mean you don’t even have to look at percentages. You can see this just by looking at cross-tabulated counts. The presentation is not what matters. It’s rather that in many cases the counts and ratios aren’t clear cut and you are left wondering, going through the data, which, of course, takes time. In her (36) example, certain types of businesses seem to prefer English over Spanish, yet other types of businesses seem to prefer Spanish over English. The problem is, of course, that we don’t know any specifics about the variables, so the percentages won’t do the trick. This is actually why one should always mention the counts as well, even if one is using percentages. Firstly, it’s very important to keep the proportions in mind (especially what the total is, are we dealing with the full set of data, a subset of data or a subset of a subset of data etc.). Secondly, if you only indicate percentages, one type might contain, let’s say, 200 units that were assessed, whereas another type might contain only 12 units. The fact that is the case, that something is that skewed, is, of course, interesting in itself, but I’d hesitate to say much about those 12 units. I’d recommend investigating that aspect more, whether it’s a thing in general or just a fluke in that data. I mean, it could be that the area in question just happens to have a low number of those types, but a high number of the other type. We might able to get a lot out of the 200 units (or maybe not, it really depends), but those 12 units might mislead us to think that that type tends to have these and/or those features associated with it (which could be the case, sure, but at least I’d be hesitant to say so). Conversely, if we do not look at the numbers, we might be working with very little data, which may risk saying something like this or that co-occurs (which, again, may or may not hold).
What I also like about this book chapter is that it incorporates spatiality (latitude and longitude) in the core of its analysis, something that I haven’t seen previously in similar studies. My own work has been mainly indoors, so this wouldn’t work for me, as such, but it’s interesting nonetheless. Using GPS data is something that has been underutilized thus far, which is actually kind of surprising, considering how big a deal GIS is in geography. As I pointed out in the previous essay, we did that type of spot mapping with GPS loggers in undergraduate geography classes years ago. It’s not even that hard to do and the mapping software is much better and intuitive these days.
There still seems to be interesting chapters in the book. I might cover (some) of them, but it really depends on if there’s enough to comment on. I mean the point of me writing about them is not to replicate them, nor to review them, but rather to comment on them, to give my take on them.
References
- Deleuze, G. ([1994–1995] 2011). Gilles Deleuze from A to Z (P-A. Boutang, Dir., C. J. Stivale, Trans.). Los Angeles, CA: Semiotext(e).
- Lyons, K. (2020). The Quality of Quantity. In D. Malinowksi and S. Tufi (Eds.), Reterritorializing Linguistic Lansdscapes: Questioning Boundaries and Opening Spaces (pp. 31–55). London, United Kingdom: Bloomsbury Academic.
- Malinowski, D., and S. Tufi (Eds.) (2020). Reterritorializing Linguistic Lansdscapes: Questioning Boundaries and Opening Spaces. London, United Kingdom: Bloomsbury Academic.