Revisiting the ‘Epistemization’ of Overlaying: The computerized mapping of disease project (MOD),1965–1968

Introduction

Overlaying maps has for long been a common practice in professional geography and disciplines that use cartographic techniques. With the development of digital geographic information system (GIS) since the 1960s and the wider public accessibility of mapping applications and the use of layers as a defining software principle in the 1990s and 2000s, an even larger number of people has become accustomed to a practice of overlaying. Overlaying maps allows to superimpose different geographical datasets or distribution patterns and combines them in an integrated visual presence. Over time, the meaning of data integration and overlaying has almost become synonymous. Moreover, with the power to integrate data from different datasets into a single-screen demonstration of their relationship, map overlays represent exemplary instances of visually inferring correlations. With the contemporary interest in big data visualization and the epochal rhetoric about an age of correlation, historical examples of map-based correlation inference are regularly dug out to emphasize the innovative tradition of this technology. The 1854 Cholera map of the epidemiologist John Snow has become a perennial guest in newspaper articles about data visualization (e.g. Rogers, 2013). As this example illustrates, map overlays for correlation inference are by no means new. Yet, an important distinction and transformation lies between 19th century cartographic practice and its contemporary appropriation: the process of overlaying for inferring correlations remained implicit in the praxeological register of John Snow and his contemporaries, while it was increasingly made explicit and objectified during its translation into the protocols of computerized geographical analysis in the second half of the 20th century.

The present article will therefore return to a transformative moment in the history of overlaying at the time GIS was introduced and focus on the explication and “epistemization” ¹ of overlaying for correlation inference in one specific example. It does so by concentrating on the close-reading of a proposal for a computerized mapping system in the field of medical geography at the end of the 1960s. It needs to be noted that focusing on medical geography undoubtedly limits the scope of the analysis to follow. Readers interested in a more general historical account of geographical overlays (in the United States) might find Steinitz et al. (1976) as an interesting starting point for further inquiry. The aim of the present analysis is partially empirical, partially methodological, insofar as it presents a historical case study that has not yet received much attention in the history of disease mapping and as an attempt to think about the way a historical epistemology of overlaying could be pursued, even though this would need a much greater material base than the present article is going to discuss. Despite these shortcomings, the article seeks to trace how, within the scope of the presented case study, the practice and infrastructure of overlaying maps for correlation inference are transformed from a common professional routine to an epistemic action of great revelatory promise. If we understand practices as implicit kinds of action and infrastructures as equally implicit background operations, then the process of explicating map overlaying in the making of a computerized mapping infrastructure appears to be a worthwhile case study for a history of geomedia and geomedia practice in particular.

The second concession that needs to be made is that by focusing on the practice of overlaying maps, this article does not tackle other aspects commonly associated with the term geomedia, such as the interplay between online and offline spaces (McQuire, 2016), between mobile devices, locative technologies, and mediated localities (Thielmann, 2010). Instead, I approach the study of geomedia with the proposal by Fast et al. (2018: 4) to understand geomedia as a “concept that captures the fundamental role of media in organizing and giving meaning to processes and activities in space” and which have often revolved around particular media objects such as “maps, compasses and newspapers” without being reducible to them. Moreover, the emphasis of this article rests on the reflexivity of geomedia and how geomedia practice informs epistemic practice. Against the background of this reflexive understanding of geomedia practice, map overlays are approached as a key technical affordance of GIS that affects how we perceive, imagine, and reason about spatial relationships in the world, and which in turn has infused overlays with the epistemic promise to make connections intelligible or operable.

The article is organized as follows: In the first part, I will briefly present existing theorizations of overlaying and propose a heuristic path for the analysis that follows thereafter. In the second part, the article’s central case study is presented, a close-reading of the project report of the Computerized Mapping of Disease (MOD) project that was undertaken on behalf of the US Armed Forces Institute of Pathology from 1965 to 1968. After situating the project in developments of medical geography and GIS in the United States around the same time, the analysis focuses on the different ways the practice of overlaying was positioned within the overall architecture of the project as a key epistemic practice for correlation inference. Finally, in the third part, I will interpret the main findings of the analysis as pointing toward a shift in the meaning of overlaying that still merits further analysis: a transformation from a routine procedure toward a conceptual schema for reflecting about the general epistemic status of correlations and as part of a new apparatus of implication.

Toward a historical epistemology of overlaying

Geomedia researchers have already shown some interest in the concept of layers (Döring and Thielmann, 2009; Manovich and Thielmann, 2009). According to Manovich, layers are part of the “cultural DNA” of contemporary media culture, and even though services such as Google Earth combine different cultural forms, the layer concept is considered particularly important among them (Manovich and Thielmann, 2009: 387). Layers are interpreted as indicators of a more general shift from two-dimensional “sites”—a website or any other planar combination of text, images, and so on—to a “three-dimensional platform as a space for the integration of different media” (Manovich and Thielmann, 2009: 390, my emphasis). What geomedia research has not investigated is the epistemological reflexivity of this increased habitual use of layers. That is to say that authors such as Manovich poignantly identified the importance of the layer in contemporary software (see also Manovich, 1999: 88), but they did not reflexively ask how this might have affected the conceptualization of knowledge practices.

What could be a way forward in this direction? A first attempt in the direction of a historical epistemology of layering could be to investigate for what epistemic tasks layers and overlaying have been mobilized and addressed as worthwhile. This would risk, however, turning historical epistemology into a rather ideational history of concepts against which more “technical materialist” approaches in historical epistemology (Hacking, 2002: 44), models of distributed cognition in cognitive anthropology (Hutchins, 1996), and praxeological angles in science studies (Knorr Cetina, 1999; Pickering, 1992) have provided alternative perspectives. Beyond the focus on the tasks, for which the practice of overlaying has been deployed, one would also need to inquire into what implements an epistemic task, from algorithms and routines to artifacts and inscription devices. For the level of routines it becomes interesting to ask, then, with which other practices overlaying was compared or competed, and how it obtained the status of a legitimate epistemic resource within this assemblage of action. That is not to put overlaying in stark opposition to other related activities; as has been claimed for other practices, they cannot be reduced to single acts but often come as “practice bundles” (Schatzki, 2001). The question is then how overlaying is centered in this network of practices and obtains the status of an anchoring practice (Swidler, 2001) or coordinating schema.

One action scheme that is of particular relevance in the praxeological context of overlaying is data integration. Harvey (1997), for example, has subsumed the epistemic dimension of overlaying to the practice of data integration and in relation to the geographer’s understanding of place and representational accuracy. He claimed that GIS overlaying is bound up with a mechanistic conception of integration and reduces the holistic conception of place with its various interlinking parts to integratable systemic sets, represented as themes and layers. Moreover, it was emphasized that the digitization of maps involves uncertainty and inaccuracy which is rendered invisible in the seemingly smooth integration through overlaying. In Harvey’s account, data integration is addressed as a normative ideal in geographical practice, which already existed before GIS and one that computer-aided overlaying can only insufficiently realize. As far as the empirical analysis of historical documents is concerned, however, the term “integration” might not have been used by the authors studied. Nevertheless, one can be sensitized by the conceptual bracket of integration but still pay attention to the particular articulations of integration in the field.

In addition to the previous line of theorization, which highlights the association between overlaying and other practices or related processes, two further conceptual angles will be relevant in the following. The first angle takes the materiality of the surface seriously, and emphasizes the perceptual dimension of the epistemic practice of overlaying. For overlaying to be a trustworthy and even habitual technique of correlation inference, the embodied sense of an above and below surface in the practice of overlaying likely contributes to this epistemic end. It makes the epistemic practice also “perceptually warranted” (Burge, 2003, 526–537). The second angle picks up the insight from science studies that epistemic practices are also objectual practices. Both Rheinberger’s (1998) and Knorr Cetina’s (2001) discussion of epistemic objects and experimental systems reminds us that epistemic objects lure and orient the work of scientists or scientific collectives but that they are also constitutively precarious. For an epistemic object to be productive, it remains to be in the making and somewhat unfinished to trigger further investigation. A map overlay, or the viewer’s concept of a correlation, is surely different from an epistemic object such as a subatomic particle in physics but the analogy is nevertheless helpful to be reminded that an overlay can hold the luring promise of being a new informational entity on its own; something that has not materialized before and which realizes a connection such as a correlation that is epistemically novel and relevant to a knowledge worker.

“Mapping correlations” might appear as a fairly common saying to contemporary readers. Despite this familiarity, the different operational parts of this epistemic process are rarely made explicit, apart from stressing the distinction between correlational and causational judgments. At most, the hint is made that correlation mapping entails presuppositions as to the factors that are chosen to be potentially related. In other words, that it entails hypotheses and that correlation mapping is rather a form of hypothesis testing than an epistemic revelation. Or, even more than that, that correlation mapping is the visualization of a correlational argument and rather a rhetorical device than an experimental tool. On such readings, the novelty aspect of correlation mapping is deconstructed by stressing the human agent’s pre-knowledge that went into making or interpreting it. For a historical epistemology of overlaying, I believe it is necessary that one retains some sense for the wonder and curiosity for the epistemic process of correlation mapping but nevertheless exposes how it is a structured process built on certain material and practical preconditions. For the latter, it is helpful to stress that the aim and ultimate epistemic process of correlation mapping is correlation inference and that inferences entail a certain organization of premises and conclusions, presume the commitment to engage in the play of giving and demanding reasons (Brandom, 1998), and would not make sense without a corresponding process of implication. In a similar way, creating a system for mapping correlations can be regarded as an organized process with premises and conclusion as well as specific medial affordances of implication. To implicate in writing or speech is not the same as to implicate through images, maps, or diagrams. Approached from this perspective, computerized overlaying for correlation inference carries something specific in the way it affords for implications, and which might be found in the way it brings to presence the synthesis of non-intelligible data points.

In the following, these different heuristic aspects will be put to work in the analysis of a particular case study from the history of disease mapping, and in whose context the practice of overlaying was explicitly transformed into a process of great epistemological significance. I will focus on the research proposal for a computerized mapping system for disease mapping during the 1960s and concentrate on the research project’s final report. It is necessary to emphasize the hybrid text genre of the document between report and manual. Because the report proposes a novel technical system, its tone is not just descriptive but equally prescriptive and suggestive in that it highlights not only the promises but also the imagined effort of building the new system. In order to analyze and extract different dimensions of the meaning of overlaying, the analysis will follow different reading strategies mirroring the ideas for a historical epistemology of overlaying mentioned above: one strategy is to focus on the formulation of tasks and the distribution of (epistemic) subprocesses; another strategy concentrates on epistemically reflexive articulations of the authors themselves such as the conceptual distinction between knowns and unknowns or the intuitiveness of map reading. Before commencing with the analysis, I will give a brief introduction to the case study’s historical context.

Case study: computerized mapping of disease (1965–1968)

GIS and computer mapping have reshaped professional geography since the mid-1960s (Chrisman, 2004; Foresman, 1998). The Canadian Geographic Information System is considered the forerunner of GIS by using a computer mapping system developed by a company called Spartan Air Services as well as IBM between 1962 and 1967, becoming operational in 1971 (Tomlinson, 1998). In parallel, from 1965 onward, the Harvard Laboratory for Computer Graphics and Spatial Analysis developed different software packages for calculating and visualizing geographical data, most famously SYMAP and ODYSSEY (Chrisman, 2004). The first idea for the development of the computer mapping package SYMAP originated already in 1963 and its development was headed by the architect Howard T. Fisher (Sheehan, 1979). Eventually, the work of the Harvard Laboratory led to the commercialization of GIS software in the 1980s, the foundation of the company Environmental Systems Research Institute (ESRI) and its standard software ArcGIS.

Important early case studies for the development and popularization of computer-assisted overlaying in the context of the Harvard lab and ESRI were the Delmarva project by landscape architect Carl Steinitz starting in 1967 (Chrisman, 2004: 4), and a commissioned study by ESRI in the Zulia region of Venezuela during the 1970s (Dangermond, 1979). ² Moreover, particular software packages and modules played an important role in algorithmically objectifying the process of overlaying: SYMAP and ODYSSEY included particular subprocess for overlaying, as well as many other software companies and research projects during the 1970s had proposed computerized mapping procedures that would include overlaying algorithms in one way or another. In this context, the concept of overlay would either refer to the process of polygon overlay or to grid cell overlay. Looking back at the Zulia study, Dangermond (1979) concluded that especially grid cell overlay is well suited for advising planning decision, for example, in identifying suitable sites for development or potentially hazardous ones for environmental monitoring, by superimposing different forms of land use. Beyond the effortful transformation of overlay routines into algorithmic blackboxes, these early days of GIS and computerized map production have also been characterized by a discourse about the advantages of computerized mapping. Most commentators highlighted the advantage of rapidly producing maps and the possibility for cost-effective trial and error (e.g. Armstrong, 1972; Steinitz et al., 1976). Moreover, some emphasized that computerized overlaying put forward something genuinely novel and in distinction to what human assessment alone could have achieved (e.g. Dangermond, 1979: 58).

Still at the beginning of these early developments of GIS between the mid-1960s and mid-1970s, a few pioneer studies were undertaken in the field of medical geography about computerized disease maps in the United Kingdom and the United States. A first handbook article that sought to overview these early developments and present the new possibilities of digital computing and new printing techniques in medical geography appeared in 1972 (Armstrong, 1972). In Britain, the medical geographer G. Melvyn Howe used computers for the production of a national atlas of disease mortality (Howe and Phillips, 1983; Koch, 2005: 229). In the United States, the US Armed Forces Institute of Pathology together with an association of different university departments received funding from the Department of Defense to develop a system for the “Computerized Mapping of Disease (MOD)” (Hopps et al., 1968). The research ran from 1965 to 1968, when it was eventually discontinued and a report made public that summarized its findings.

In the analysis that follows, I will focus on the final report of the above-mentioned MOD project. The project was probably the first of its kind in the United States and it aimed at developing a whole system for data input, storage, processing, and output with particular emphasis on map production. Other and parallel attempts at computerized mapping system were mentioned in passing such as Fisher’s SYMAP program. In addition, the authors situated their endeavor in continuity with developments in US medical geography in the decades prior to its foundation, emphasizing their “disease-ecological” perspective (cf. Mayer, 1982). Representative for the disease-ecological understanding of medical geography at the time was the author Jacques May, who at the beginning of the 1950s had proposed a system for studying disease-ecological correlations by the use of multi-factor complexes, using in tandem the tabulation of these complexes and their demonstration in maps (May, 1950, 1952; cf. Brown and Moon, 2004). In some sense, May’s work can be seen as an intellectual and diagrammatic precursor to the MOD project. In the spirit of Jacques May, the authors of the MOD project situated their research in the tradition of disease ecology and the report did not tire of stating that “it is the ecology of an area which determines what diseases might become serious problems” (Hopps et al., 1968: 9-2). Consequently, the authors considered it paramount to combine data from both environmental and medical records. Even though they stated that this data was available in abundance, what they saw lacking was a system that would bring the different sources together and thereby inform disease-ecological analysis.

Overlaying between embodied technique and protocols of data integration

In accordance with the different reading strategies mentioned above, I will concentrate first on the main tasks that the MOD report defined and for which overlaying was proposed as a relevant epistemic process. However, in order not to artificially isolate the system from contextual epistemological expectation, the task-process relationship needs to be leveled against the general attitude toward computerized mapping at the time

At the time of interest, computer mapping did not necessarily have a unified enthusiastic reception among medical geographers. According to Armstrong (1972: 69), there had been reservations to the extent that “most cartographers prefer[ed] not to call the computer graphic a map.” Early critics such as Armstrong and Howe (1970) themselves voiced their reservations when it came to the quality of the map and its underlying data, or to the large effort of preprocessing the data. Computer mapping was considered useful where extensive data analysis was required (e.g. for the mapping of aggregates and ratios) or where maps had to be regularly (re)produced with the same underlying base data (e.g. for surveillance). But where more substantial work of data preprocessing was necessary or where manual mapping techniques could not be easily automatized (e.g. in the case of isopleths), the computer mapping infrastructure supposedly lost its initial promise. In a similar vein, the authors of the MOD project report carefully balanced and justified which tasks and processes were distributed across human and machine in which way.

In the preface to the MOD report, its main author Howard Hopps declared that “the objective of the MOD system is to provide a means whereby the Disease Panorama can be quickly and effectively presented in map form” and “to develop a system whereby available data could be used most effectively to gain new insights into the multifactorial causes of disease” (Hopps et al., 1968: ii/v). Later in the report, the authors further distinguished between a first objective of building a system for “preprocessing,” “storing,” “manipulating,” and “instructing the plotting” of data, a second one “to produce meaningful maps (and other graphical displays) that show the distribution of disease(s),” and finally a more broader “effort to illuminate the geographic pathology of disease” (1-5–1-7). As far as analysis was concerned, the authors also maintained that “the computer system itself cannot perform an analysis of the maps that it produces” but that the “visual pattern recognition will continue to be the major method of analyses in most instances” (my emphasis). While visual pattern identification was described in a way that it fully delegated the analysis work to human agency, some calculating subtasks were discussed as being more efficiently handled by the machine. These considerations of the line between human and machine work also found its expression in terminology. While Hopps et al. used the term “synthesis” solely to refer to the machine processes of calculation, for the final process of “making correlations,” the authors steered the distribution between machine and human agency again toward the side of the “user”:

In the MOD computerized system it is the user, not the system, who makes correlations between the raw data and output map, evaluating the various factors which make the map look as it does. The computer system will not perform analysis of the maps produced nor will it make judgments; it will merely manipulate (according to rigidly defined algorithms) extracted, formatted data (from that pool of data which was previously put into the system) and output these manipulated data in the form of maps or other reports, in the manner specified. (Hopps et al., 1968: 8-9)

As these brief examples illustrate, the MOD report cannot simply be read as an allocation of tasks but it also articulated a way of navigating boundaries between human and machine agency, presenting the reader with the possibilities that s/he as “user” would have and mitigating any expectable skepticism toward the proposal of a computerized system. This should be kept in mind when turning toward the presentation of overlaying in the report. The structure of agentive ascriptions and terminological divisions provided important praxeological coordinates for a process such as overlaying to obtain its meaning. Technically, synthesis and overlaying had in common that they worked on the integration of different data points. But only overlaying added an intuitive familiarity with the materiality of the planar map and it was introduced in conjunction with the human analytic capacity for visual pattern recognition both at the very beginning and end of the report:

Overlaying and visual pattern comparing is a very powerful process because it permits human detection of relationships so complex that standard mathematical methods may be unable to detect them. ( Hopps et al., 1968 : ii / 8-9)

For the designers of the MOD system, the prospective users of the MOD would bring along an embodied knowhow of how to use maps even beyond the professional field of cartography. The authors of the report stressed how intuitive the understanding of maps has become when first introducing maps as ideal output: They claimed that “maps have a unique advantage over most other forms of graphical display” because of the “extensive and continual usage of map forms, beginning in early childhood, [which] has conditioned most (educated) people to an intuitive understanding of maps” (Hopps et al., 1968: 1–9). It is likely that overlaying translated between this presumed familiarity with maps and the epistemic promise of machinic synthesis and the user’s “making” of correlations.

Technically, there were essentially two ways in which the authors envisioned a computer-aided overlay to materialize. Either through overprinting on a single sheet or by printing separate intermediate but aligned maps on transparent sheets, which would then be overlaid manually. In the case of overprinting, the computer outputted two distinct map values onto magnetic tape; the printer would re-print along the same coordinates, and the overlay materialized. The SYMAP program by Howard Fisher had equally used overprinting on line printers for the display of multiple map values. However, the authors of the MOD report did not make any reference to other uses of overprinting and computerized overlaying at the time.

Instead, and as apparent in the last indented quote above, Hopps et al. associated their conception of overlaying with the practice of (visual pattern) comparison. They mentioned overlaying and comparing as the epistemically most productive techniques of medical geography for arriving at judgments about disease correlations, and as something that purely mathematical calculations would not be able to obtain. These techniques would make it possible, the authors contended, to move beyond the correlation between disease and only one variable toward a multifactorial analysis.

Logically, a map represents only one dependent variable (or one disease-environmental factor), but more than one variable can be represented either by a series of maps [. . .] or by overprinting the mapped patterns of several such variables onto the same base sheet. (Hopps et al., 1968: 3–15)

Comparison and overlaying were joined in offering the combinatory or calculative space for analysis.

Establishing an association between overlaying and comparison is a peculiar feature of the MOD report. The aim of the practice of map comparison is the diagrammatic representation of distribution patterns, the visual identification of differences and similarities among them, and subsequently the formulation of inferences about causal or determinate relationships or temporal changes. Overlaying collapses the process of visual comparison between two images into a single display. It is in a sense more insisting by using the iconic power to show and present without negation (Mersch, 2011, cf. Alloa, 2019), not even by contrast that the comparison across parallel screens still provides.

The juxtaposition of comparison across multiple images on the one hand, and generating a single integrated image space through overlaying on the other hand, also played a didactic role in the way Hopps et al. attempted to demonstrate their case for the advantages of overlaying. After presenting the most common map styles that the MOD system would aim at reproducing (dots, shades, isolines), the authors demonstrated a number of examples where environmental and disease records could beneficially be combined in one map. These examples were presented by a series of miniature images, beginning with maps for each variable and ending at a combined map that integrated the different variables. The final integrative image was either an overlay map, which was produced through overprinting the different component maps (Figure 1), or a map showing synthetic values that were already computed by the machine on the basis of integrating environmental and disease data before printing (Figure 2). The suggested promise of the examples was that the MOD was able to make visible an association between environment and disease that would otherwise be hard to recognize from comparison across parallel screens alone.

OPEN IN VIEWER

Figure 1.

Exemplary presentation of map layers (A, B, C) and the resulting composite map (D) in the MOD project report (Hopps et al., 1968: 8-14).

OPEN IN VIEWER

Figure 2.

Exemplary presentation of map overlays, “showing distribution of temperature [A], rainfall [B], and schistosomiasis data [C] in eastern Brazil.” Map D displays the overlaying of the previous maps. Maps E and F show data points and contours “that would have been retrieved and output by MOD system in response to query asking for the combined factors,” in this case the distribution of schistosomiasis where the July temperature is lower than 7o° F and annual rainfall greater than 50 inches (Hopps et al., 1968: 8–17).

Overlay and correlations

The formulation of inferences about causal or determinate relationships can be confirmatory as in the case of already known factors, or it can be revelatory as in the case of unknown factors whose causal implication is inferred. On this spectrum, the MOD system is steered toward revelation by the authors. They themselves introduce the distinction between known and unknown factors when first making their case for the map as ideal output format (ranking before written reports, for example):

We believe that a mechanism/system which can produce many kinds of map-patterns quickly, in response to specific query, will offer two very important advantages: First, such a mechanism will make it possible to have current information about the distribution of specific diseases and the distribution of known causally related agents or conditions. Second, the repaid availability of a large number and wide variety of disease-environmental maps will give the observer an opportunity to compare location patterns of unknown but possibly related ecological factors and, in this way, help him to identify causal relationships that might otherwise have escaped notice. (Hopps et al., 1968, 1-9, original emphasis)

The division between known and unknown factors is not only an abstract remark about the confirmative or revelatory gesture of “drawing connections” and “inferring correlations,” but it rhetorically opens an imaginative space of virtual combinations that can be exploited by the MOD system. The MOD system promises that unknown factors can be mapped with less effort and different combinations can be tested repeatedly; and it makes evident the operative division between database and overlaid map output, between retrieving or selecting on the one hand and showing on the other hand. With the possibility of unknown factors in mind, the authors of the MOD system imagined and promoted a vast database of factors from which one can select at will and discard if shown unrelated by visual proof.

In addition, for Hopps et al. the epistemic promise of the MOD system seemed to be closely tied to the two-dimensionality of the medium of the map, to the embodied capacity to see correlations or associations in the cartographic plane, and to be able to virtually extend this plane in the vertical dimension. This vertical virtuality worked in a positive and negative way. The associational plane of the map could be extended vertically through overlaying, but it could also be vertically undercut by invisible relationships below the map, as Hopps et al. showed in a diagram of possible relationships (Figure 3). The promise of overlaying turned the map not only into an epistemic medium that makes something visible and known to a human operator, it also provided a model for conceptualizing the existence of correlations. The concept of the layer with an apparent above and virtual below became reflexively applied to think about and illustrate a general problem of knowledge.

OPEN IN VIEWER

Figure 3.

Conceptual diagram of different types of possible relationships between disease and environmental factors (Hopps et al., 1968: 8–8).

The image in Figure 3 makes a simple division between above surface data—which can be easily observed and in which case a relationship between two types of data is perceptually suggestive—and below surface factors that are beyond the analyst’s range of vision. On the one hand, the image seems to suggest that to relate two factors based on what is visible on the surface of a map carries the risk of ignoring all the interfering factors that are not visible in this specific case, or the risk of stipulating a determinate relationship even though it could be accidental. The authors thus remind that mere contiguity relations in a representational space can indicate relationships of various kinds in reality: causal (direct, indirect), associations, incomplete, and accidental.

On the other hand, the image also appears at the end of the whole MOD report, in a section entitled “notes to user,” which summarizes some of the functions of the system, and where the authors remind once again that it is the human analyst who “makes” a correlation. In this context, the image is not only a modest reminder of what the analyst might not know, but also a plea for the potential of the MOD system: that these factors below the surface can be integrated into a composite overlayer. Consequently, the authors highlight that “the practical limit to the number of overlays is probably quite low, for the whole purpose of this type of data processing is to simplify the situation considered so that relationships are clarified” (8-7).

Even though the term correlation is not mentioned in Figure 3, the section in which the image appears is opened with the paragraph cited above that “it is the user, not the system, who makes correlations” (8-6). However, the epistemic role of the materiality of the surface and of overprinting or overlaying remains implicit at best. Yet, Figure 3 brings some of the epistemological import of the notion of the layer to the fore, as it presents the surface as a conceptual model for theorizing the general epistemic status of (causal, associative, accidental) relationships. To turn the argument around: the point about hidden factors could have been well made without mobilizing a conception of surfaces. The surface serves as an important informational threshold and a conceptual tool for theorizing relationships: a threshold above which relations can materialize as ready at hand for the correlation seeking analyst, and below which a virtual reservoir of further relationships exist. At the same time, the image reminds of the epistemic relevance of the vertical dimension. If correlation is the epistemic object that guides the user of the MOD system, this object also has an orientation (vertical) and a point of realization (overlayer). These perceptual dimensions of the epistemic object are not separate properties of a disembodied concept of correlation, but they are part of the way correlation matters, or are assumed to matter, to the prospective user of the MOD system.

Discussion: map overlays between operative background and epistemic resource

Overlaying maps had long been common practice in several fields of geography. During the late 1960s and early 1970s, this common practice received renewed attention and reflection during the process of early GIS development. Steinitz et al.’s (1976) historical account of overlaying by the mid-1970s, based on their own experience with GIS development, testifies to this increased interest and new level of reflexivity that overlaying managed to obtain during these years. Computerized cartography had set out to automatize the manual practice of overlaying and to translate it into a programmable action unit. The proposal for the MOD system falls into the few years from the mid to late 1960s, just before overlaying algorithms were beginning to be reified and popularized. The analysis above has traced the different ways in which overlaying was articulated in this particular research context, as an early example for overlaying’s transformation during that time from a common cartographic technique, effectively operating in the background of professional routine, into an epistemic practice with outstanding creative and revelatory promise.

Within the limited scope of the MOD project report, this articulation of overlaying could be witnessed on three levels: first, overlaying obtained the status of a schema that could be equally related to an embodied experience and the machinic implementation of overprinting, while also maintaining a conceptual continuity with other epistemically worthwhile algorithmic procedures of data integration such as synthesis and interpolation. In other words, overlaying coordinated between different registers of description and within the order of the computerized system it was the final process in which previous integrative procedures converged. Second, and relatedly, the schematic function of overlaying extended beyond the coordination of different levels of description into the realm of epistemological reflexivity, insofar as it was used to describe the general epistemic status of causal associations that are hidden below a surface and which must be brought to the operator’s awareness. Third, overlaying was bundled with the practice of comparison. This imaginative association between the two practices “authorized overlaying by tradition” (van Leeuwen, 2007), and it brought the material or medial specificity of overlaying to the fore. Overlaying was not merely another kind of comparison but it replaced the horizontality of two actual surfaces laid out before the operator’s eyes with the demonstrative force of one integrated display and a virtual vertical space of possible relations.

In their historical account of the practice of comparison, Epple and Erhart (2020: 15-16) emphasized that comparison is never “neutral or innocent” but always entangled with the “perspectives of the ones who compare.” Even though one can compare “everything with everything” there is usually also a presumed “comparability” and a “tertium comparationis” from whose perspective or toward which the comparison is undertaken and it is these latter two criteria that make a practice of comparison historically specific. The same could be claimed for the practice and apparatus of data integration and overlaying. Overlaying, data integration, or comparison in the context of the MOD system were approached from a disease-ecological angle and on the back of an already existing tradition of earlier medical geographers who regarded the visual inference of correlation a primary aim of the discipline. In addition, the authors of the MOD report argued for the value of comparison and overlaying on the condition of an abundance of available, yet unorganized information. As much emphasis they put on output, they also stressed the necessity of file structure and database design. The authors made clear that data integration is an ideal which needs a lot of work and effort for making all the existing records of environmental and epidemiological data available in a machine-readable way. The epistemic promise of the overlay was the lure that could legitimize this effort, in a similar way that Rheinberger (1998) has described the attraction of epistemic objects in experimental systems, and Knorr Cetina (2001) the lack and desire relationship in “objectual” epistemic practices.

Rheinberger (1998: 287) had noted that epistemic objects function as “generators of surprise,” so that experimental systems do not come to a standstill but always produce something beyond the present state of knowledge. Building on Rheinberger, Knorr Cetina (2001: 190) has added that epistemic objects are “processes and projections rather than definite things” and that they serve as “stand-ins for a more basic lack of object.” The description of the geographic overlay in the context of the MOD report mirrors some of these aspects, because of the overlay’s status as a threshold space which concretizes associations for the viewer and simultaneously virtualizes the space beyond. As such the overlay is as much revealing as it is implicating and thereby embodies an ambivalent and intermediate epistemological status. The authors of the report articulated this status most clearly in their final conceptual diagram. It turned the MOD system into an imaginative apparatus of revelation and implication, with the overlay obtaining the status of an integrative epistemic relay in its own right.

After all, the project never saw the light of day and the research was discontinued. It was rather an assessment of possibilities with a significant amount of projection and aspiration for what a computerized system for disease mapping could be. The report’s narrative was set up as a prefigurative design brief and it rested on articulating overlaying not just as another process but as an epistemic technique. Overlaying was presented as the focal point of an organized epistemic operation and which served them as a model to be reflected back upon a general epistemology of (non)causal associations toward the very end of their report. Altogether, the MOD system articulated a material-semiotic schematization and epistemization of the practice of overlaying. On the one hand, the process of overlaying was singled out from a background of tacit routine and addressed as a valuable action unit, thus externalizing and objectifying the act. On the other hand, it was re-embedded as an abstract conceptual resource and schematic bracket for translating between manual experience and overprinting as well as making meta-conceptual claims about the correlations and the epistemic potential of the whole system. In other words, the process of schematization can be traced through the increasing schematic operability and plausibility of the process of overlaying, its application across several levels of description, and implementation.

The analysis above has focused on a single project report and the historical claims that can be derived from it are necessarily limited. But the example of the MOD project might have the potential for offering clues and reading strategies for subsequent case studies that could not themselves be part of the research that led to this article. Future work steered toward a synchronic comparison would have to include other user references and project reports of the many pioneering design proposals for computerized overlaying during the mid-1960s to mid-1970s. As said at the beginning, one way of situating these projects in relation to each other could be to focus on how the main epistemic aim of overlaying has been construed. A complementary approach, one that the present analysis wished to contribute to, would be to pay attention to the implementation of these goals and how the breaking down of the mapping process into valuable and operable subunits proposes hierarchies of action and an overall organized epistemic system. Moreover, how actions are relatively weighted, and which ones might be unmade or invisibilized instead, becomes of equal interest for a diachronic perspective of overlaying. For example, comparing map series as a way of visually understanding changes in environmental variables was already mentioned by earlier medical geographers. Overlaying, by contrast, was a common technique in medical geography as much as it was in geography in general; but, to my knowledge, it was not emphasized as strongly as a particular epistemic capacity as it was during and after its attempted automatization.

Against this background, would it be too farfetched to claim that the formalization and explication of overlaying in early GIS has contributed to a “geomediatization” (Fast et al., 2018: 6) of the epistemic process of correlation inference? It is rather impossible, however, to delimit clear lineages between the interdiscursive meaning of overlaying today and the particular field of medical geography during the 1960s and the MOD system specifically. This was also not the goal of the analysis presented in this article. Instead, I have proposed a close-reading of a historical case study of overlaying to contrast, maybe estrange, in any case make contingent the contemporary experience of ubiquitous background correlations and layers as default organizational principles of knowledge work. As far as avenues for further research is concerned, a more comprehensive study of the history of overlaying would also need to inquire into fields of practice beyond geography. For example, the conceptual model of vertical layers again gains an important role in discourses about machine learning. If layers are introduced in the design of artificial intelligence to mediate learning, to address in-between stages of bottom-up and top-down categorization, then this hints at the continued relevance of overlaying as epistemically relevant in further technological contexts.