Linkography

Design protocols are generated and parsed into segments like any other protocols. Unlike some other protocols, they are often accompanied by sketches that are used primarily to interpret verbalizations. As we have seen, parsing may be time-based (e.g., three minutes of verbal output), or semantically based (e.g., one sentence), or they may be based on units of content that comprise one sentence, part of a sentence, or more than one sentence; in teamwork, turn taking is also a common parsing principle. In the content-based category we find definitions of analysis units such as step, action, movement, or move. What “move” connotes is akin to its meaning in chess: a step in the process that changes the situation. There is some agreement on the use of this term (e.g., Dillon 2010; Goldschmidt 1995; Suwa et al. 1998), although this definition does not match Schön’s use of the term “design move” (Schön 1992). In his interpretation a move is an act, rather close to a decision, made between acts of seeing; in this view the design process is composed of cycles of seeing, moving, and seeing. Here a more general view of a design move is upheld, whereby a design move is a step, an act, an operation, that transforms the design situation somewhat relative to the state it was in before that move (Goldschmidt 1995). Seeing can therefore be a design move, if we can capture it through the designer’s verbalization in think-aloud documentation. Moves are normally small steps, discernible from their contents, and with some training it is not difficult to reach agreement and consistency in parsing a protocol into moves and deciding which utterances are meaningless and should not be designated as moves (“yeah,” “OK,” “emm,” and so on); such utterances are removed from the analysis. David Botta and Robert Woodbury (2013) point out that the term “conversational move,” coined by Paul Grice (1975), carries an evocative similarity to the term design move, as a design protocol can be seen as a record of a conversation among designers or of a designer with himself or herself.

Several studies concluded that the average duration of a move in a design session is about 7 seconds (Baya 1996; Baya and Leifer 1966; Goldschmidt 2012; Kan and Gero 2008; Andrew Milne, personal communications ²). Moves generated in teamwork are shorter. However, if we use the concept of move as the basic unit of analysis and not a time-based unit, we should bear in mind that moves vary in length and duration.

The following example is a short vignette from a protocol derived from a recording of an architect named Martin, who was thinking aloud while working on the design of a small library (see the appendix). It is parsed into three moves, which in this case correspond to sentences:

We start creating a hierarchy: the large trees, the parking lots, the pedestrians, an entry axis.

I would then look for a direct relationship between entrance and exterior, because here, the real edge is not this [edge of building], for me it’s that [edge of site].

I would try to have an important element; would therefore make the axis I mentioned before, this one [points to sketch].

Moves are not the smallest possible units of analysis. A move may be parsed into its constituent arguments, which are the smallest semantic units that hold a comprehensible concept. For example, the third move in Martin’s utterance above is composed of two arguments. The first argument states that the designer would like to introduce an important element, without specifying what kind of element that could be; the second argument specifies the important element in terms of an architectural feature, in this case an axis:

For certain purposes it is of interest to analyze protocols using arguments as the basic units. This was done to investigate the relationship between embodiment and rationale in design thinking (Goldschmidt 2001, 2012), where embodiment denotes physical properties of a design element or component and rationale the reason for selecting or desiring certain properties. These two modes of reasoning are the scheme of categories by which the arguments are encoded. However, more often we want to concentrate on larger units: the move, or, in some cases, derived units such as ideas or decisions that are more complete and are stand-alone units.

The difficulty in establishing appropriate universal schemes of categories for protocol analysis in design, which was discussed in chapter 2, is not surprising. Design tasks differ greatly in nature and scope, resulting in diverse foci, and accordingly designers must attend to changing aspects of the design task. As a result, attention may be directed to form or to function, to the whole or the details, to technological issues, to ergonomic considerations or aesthetic values, and so on. Within a single task, too, different phases may require different approaches. Personal dispositions of designers also affect what they choose to tend to and determine their priorities. In addition, the purpose of each study of the process of design is different. Consequently, it is hardly possible to develop a single scheme of analysis categories that would be useful across the board for protocol studies in design. When I first began to use protocol analysis, I experimented with schemes of categories but was disappointed with the results. Not only did no assurance surface that the categories were appropriate, but the significance of the findings (which indicated what kind of design activities were carried out, what the relationships among them were, and when they occurred) was not satisfactory. Findings of this sort are, of course, important for some purposes, especially when they are used to compare performances of different populations—for example, experts vs. novices—on the same task. Comparisons are of the essence in protocol analysis of any kind. However, categorizing design activities was not helpful when the goal was to fathom thinking processes.

The feeling that coding-based protocol analysis was too case specific, and the uncertainty about what could be learned from these analyses that would have some general validity, led to the question whether design protocols could not be analyzed in a different way. Our interest was clearly in design cognition and not in kinds of design acts, which were taken to be contingent on the task, the phase, and the designer’s priorities. Experience as a designer and as a teacher in dozens of design studios of varying standing led to the conclusion that the most crucial thing in a design process is the solidification of a major idea, or combination of ideas, that could bring together all the major aspects the design had to respond to. In other words, synthesis, or integration, was the major goal of conceptual design. As Christopher Alexander pointed out so eloquently, even the simplest of design tasks is complex and necessitates an integration of responses to many requirements and desires, which may even be conflicting (for example, advanced technology and high-quality materials versus low cost). Alexander (1964) talked about a “good fit” that must be achieved among design components by resolving “misfits.” John Archea (1987) used the metaphor of “puzzle making” to describe how architects achieve a unified whole by fitting discrete pieces together. Arriving at a state of good fit is no trivial matter; indeed, it has long been considered a mystery, magic, a manifestation of creativity that cannot (and some thought should not) be explained. Alexander, Archea, and some other researchers thought otherwise, of course; so did some of the psychologists who began to unveil creative processes.

On the view that design processes are not a matter of magic and that the thought processes that underlie them must be explicable in terms of thinking in general, a way was sought to analyze design protocols that would reveal something about the thinking involved, regardless of the specific design domain or task. It was clear that the analysis would have to be based on very small segments that mapped well onto design moves. This happened at precisely the time when the design research community began to direct its attention to design thinking, and psychological concepts gained a foothold in the explorations that researchers began to undertake. The concomitance of two modes of thinking and reasoning is one of these concepts.

Several psychologists subscribe to the view that we use two systems of reasoning—indeed two modes of thought—in everyday life, and that the balance between them is particularly pertinent to the understanding of creative thought. Different authors have used different terms to describe the two systems. Steven Sloman (1996) described an associative, similarity- based system versus a symbolic, rule-based one. The associative system makes use of visual representations when they are relevant, and design is a case in point; the rule-based system specifies a rationale. In chess, for example, there is evidence that pattern recognition and rational “forward search” are deeply entangled (Linhares et al. 2012); this may suggest a similar entanglement of systems in other instances of problem-solving thinking, such as designing. Liane Gabora (2010, 2) talked about associative thought and analytic thought; the former tends to be intuitive and “conducive to unearthing remote or subtle associations between items that share features or are correlated but not necessarily causally related.” “This,” Gabora continued, “may lead to a promising idea or solution, although perhaps in a vague, unpolished form.” In contrast, analytic thought is rule based and convergent, and is “conducive to analyzing relationships of cause and effect between items already believed to be related” (ibid., 3). Recently Daniel Kahneman, whose interest in decision making is well known, published a book (Thinking, Fast and Slow) in which he argued that fast thinking is mostly intuitive and based on memory and emotion, whereas slow thinking is rational and entails calculating consequences. All these descriptions may be seen as roughly corresponding to divergent thought and convergent thought. These terms have been used by creativity researchers, with a pronounced emphasis on divergent thought, which was seen as the hallmark of creative thinking (see, e.g., Finke et al. 1992; Mednick 1962). Divergent thinking is defined as “thinking that moves away in diverging directions so as to involve a variety of aspects and which sometimes leads to novel ideas and solutions; associated with creativity.” Convergent thinking is demarcated as “thinking that brings together information focused on solving a problem (especially solving problems that have a single correct solution).”³ There is much more literature on divergent thinking than on convergent thinking.

Today there is evidence that creative thinking involves both divergent and convergent thought. Gabora (2010), who gave a neurological account of creative thought in terms of memory activation, suggested that divergent thought is associated with defocused attention and convergent thought is related to focused attention, and that these two types of attention give rise to different memory activation patterns (that is, different patterns of connections among neurons in the brain). Gabora asserted that creative thinking requires the flexibility to shift between the two modes of thinking. In her words, “creativity involves the ability to either shrink or expand the field of attention, and thereby match where one’s mode of thought lies on the spectrum from associative to analytic according to the situation one is in” (ibid., 3). Min Basadur (1995) used the words “ideation” and “evaluation” to describe the scope of the spectrum to which Gabora referred, and related them to divergent and convergent thinking. These researchers emphasized that the two modes of thought complement one another, and that shifts between them are frequent. According to Paul Howard-Jones and Steve Murray (2003), in creative thought these modes of thought serve to ensure originality (divergent thought) and appropriateness (convergent thought).

Design must perforce be seen as a creative activity, in that one brings into being a representation of an entity that does not yet exist. Design thinking and reasoning is therefore a typical case of creative thinking, and we can now talk about the synthesis that is to be achieved during the early phase of the design process as a series of cycles of divergent and convergent thinking (or any of the related terms used in the relevant literature) in which ideation and evaluation follow each other in frequent proximity, pertaining to embodiment and rationale. This notion is in line with the agreement that design is not a linear process, which may in itself imply more than one mode of thinking. We shall revisit the important concept of divergent and convergent thinking and reasoning in chapter 6, which focuses on design creativity.

We can now go back to design moves, which are obviously not generated in isolation but which relate to one another in various ways. It is proposed that revealing the links among design moves is the key to understanding the dual modes of design thinking and arrival at a design synthesis. This is the case because “effective reasoning in a creative endeavor must perforce aim at first mining and then relating to one another the many items of data that are relevant to the task” (Goldschmidt and Tatsa 2005, 595).

Design moves are brief acts of thinking, lasting around seven seconds. They are generated sequentially in time. Taken together, they express the design thinking process. Since moves are not generated as autonomous entities, they form continuums of various lengths in which they are interrelated, or linked. The pattern of links is neither known in advance nor fixed in any way, but it can be established empirically for each sequence of moves. How can we determine links among moves?

Links are based on the contents of moves. Deciding whether two moves are linked is done by using common sense under the condition of good acquaintance with the discipline and with the design episode in question. Acquaintance is essential to ensure that the person who judges whether or not a link exists understands precisely what the designer was doing, or thinking, despite difficulties that can arise in deciphering think-aloud documentation, for example, repetitions, jargon, incomplete sentences, unclear speech, and use of “this,” “that,” “here,” “there,” and similar words when referring to representations such as sketches or other objects that were present in the work environment. Video recordings and copies of sketches help in interpreting such utterances but do not always aid in making sense of incomplete or unclear sentences. However, a sufficiently experienced observer in the relevant field of design who followed the design session carefully can, upon reading the (parsed) protocol several times, use what may be called “educated common sense” to determine whether a link exists between two moves. In a perfect study, three judges are employed and the decision “link” or “no link” is determined by the majority: consensus or two votes for one of the two options.⁴ No encoding of moves is required; we want only to establish the existence of links, or the lack thereof.

How is a parsed protocol analyzed for links among moves? It is done systematically, by asking “Is there a link?” for every pair of moves in a sequence. First we number the moves sequentially. Then, starting with move 2, we test whether it has a link to move 1. Next we go to move 3 and test whether it has a link to move 2 and whether it has a link to move 1. For move n, we have to ask this question n – 1 times for possible links between move n and all preceding moves, namely 1, 2, 3, . . . , n – 1. For n moves, we must perform this test n(n – 1)/2 times in order to include every pair of moves in the sequence. Figure 3.1 illustrates the process of testing for links in a sequence of five moves; the number of tests is 5(5 – 1)/2 = 10.

Figure 3.1

Testing for links among five moves. The test is performed ten times.

The tests described above were performed directionally—that is, we tested for links between every move and each of the preceding moves. The direction is backward in terms of a linear sequence in time. Any links we establish this way are therefore called backlinks. The symbol < denotes backlinking. However, once we have established a backlink, say between move 2 and move 1, we may also refer to a forward-bound link, or a forelink, between move 1 and move 2. Forelinking is denoted by the symbol >. We say “forward” because this is a link between a move (move 1) and an anterior move (move 2). Forelinks are therefore virtual and can be established only after the fact. Every link is one move’s backlink and the other move’s forelink. Therefore, the number of backlinks and forelinks is equal, and if we add up all backlinks and forelinks we obtain twice the total number of actual links. Figure 3.2 shows a link between moves 1 and 2; the same link is move 2’s backlink and move 1’s forelink.

Figure 3.2

Link between moves 1 and 2. (a) Move 2 has a backlink to move 1. (b) Move 1 has a forelink to move 2.

In a sequence that contains more than two moves (as practically all design episodes do), we get a network of links that are represented as nodes. Most network depictions or graphs of processes represent the entities (moves, in this case) as nodes and the links as connecting lines. Here we wish to concentrate on links; they are our variable, and therefore they have become nodes in a graph that portrays the network of links among moves. The representation is therefore called a linkograph. Figure 3.3 shows a linkograph in which five moves are interconnected through six links. In this linkograph, move 1 has one forelink (to move 3). Move 2 has no backlink but has two forelinks (to moves 4 and 5). Move 3 has one backlink (to move 1) and two forelinks (to moves 4 and 5). Move 4 has two backlinks (to moves 2 and 3) and one forelink (to move 5). Move 5 has three backlinks (to moves 2, 3, and 4). A move’s backlinks are “strung” along grid lines that originate at the moves and turn diagonally leftward to meet the linked move. Respectively, forelinks are lined along grid lines that originate at the moves and turn diagonally rightward, again to meet the linked move. Obviously, the first move can never have backlinks and the last move can never have forelinks.

Figure 3.3

Linkograph with five moves and six links. Nodes are links; lines are network grid lines only.

Why do we stipulate forelinks, and why is it logical and beneficial to distinguish between backlinks and forelinks? Let us remember that design synthesis at the micro (cognitive) level is understood as emanating from a search that consists of cycles of acts of ideation and evaluation. These acts gradually shape an original design proposal, or solution, until it can be deemed appropriate. The design acts in question are defined as moves, and we posit that links among them enable an integrated end result, or a “good fit” of the synthesis. Are originality and appropriateness achieved in the same way? They cannot possibly be achieved in the same way. To achieve originality, the designer must propose something. To ensure appropriateness, it is necessary to evaluate a proposal by ensuring that it is consistent with previous steps taken to accommodate requirements; if it is not, there may be contradictions (misfits), which are likely to cause difficulties or even to lead to the failure of a proposed solution that does not meet the requirements. Since moves are very small steps, it is never enough to propose something; further steps must build on it (or challenge it) and develop it in order to arrive at a complete synthesis. Likewise, when we evaluate, we often match a proposal with several previously raised issues. When we propose, we look forward, and the move we make will probably link to further subsequent moves, which follow immediately thereafter or occur later in the process. When we evaluate or assess, we look backward to what has already been done to make sure that a good fit exists between the current move and previous work, and that no apparent contradictions, mismatches, or other negative consequences are evident in the design process. Looking backward creates backlinks. Looking forward creates forelinks. The two types of links have very different meanings, and therefore we distinguish between them. Such a distinction can be expressed in quantitative terms, and expressing them quantitatively is significant when we analyze and when we compare protocols of design sessions.

To demonstrate this notion, let us return to the brief vignette from Martin’s process of designing a small library that was presented earlier in this chapter. Here is the vignette again, parsed into three moves:

Move 1 We start creating a hierarchy: the large trees, the parking lots, the pedestrians, an entry axis.

Move 2 I would then look for a direct relationship between entrance and exterior, because here, the real edge is not this [edge of building], for me it’s that [edge of site].

Move 3 I would try to have an important element; would therefore make the axis I mentioned before, this one [points to sketch].

Martin had been given a shape representing a library’s “footprint” in the middle of a rectangular site. In move 1 he referred to the site, in which there were large trees and on which he had placed parking lots and determined where the entry to the library would be. Martin is an architect who often uses axes in his designs (Wrede 1986), an axis being an imaginary line that visually connects design elements. In this vignette, as a matter of habit or routine, he talked about creating an “entry axis.” Move 2 expanded on the relationship between the entrance and the exterior, and, since this move referred to the entry and its axis mentioned in move 1, a backlink is established between move 2 and move 1. In move 3 Martin returned to the entry axis of move 1 and asserted that it was an “important element.” Move 3 therefore has a backlink to move 1, but it does not relate to the relationship to the exterior and therefore it is not linked to move 2. After the fact we see that, whereas move 2 has no forelink to move 3, move 1 has forelinks to both move 2 and move 3. Move 2 asserts the position of the entry as part of a hierarchical arrangement of elements connected by an axis. Move 3 provides an explanation for the axis: it is an important element. Move 1 also connects building elements with outdoor elements, and move 2 explains that this is done because, for Martin, the extent of the design is the building together with its surroundings, ending only at the edges of the site. A linkograph depicting the network of moves and links in this short vignette is shown in figure 3.4. This example demonstrates that a good fit between a proposal and its rationale is not pulled out of the designer’s sleeve as a “fait accompli” but is built up small step by small step. Proposals are made and then assessed, or a rationale is put forth and a suitable embodiment is then proposed; the order in which these types of moves succeed one another makes no difference (Goldschmidt 2012).

Figure 3.4

Linkograph derived from a short vignette of Martin’s protocol.

As was claimed earlier, good design boasts a well-integrated solution that takes care of the many issues that must be addressed. Good and creative design also manifests a coherent, novel, comprehensive “leading idea” that elevates the design from a mere solution to a problem to a social, artistic, technological, or generally cultural statement that is appreciated by the public at large. A leading idea is not a “thing apart”; rather, all design decisions must be compatible with it. This is precisely what a design synthesis denotes. If this is indeed the case, then understanding the properties of the network of links among design moves should shed light on the thinking that brings into being well-integrated works of design—both good, creative ones and lesser ones.

Linkography thus concerns itself with links among design moves, as it is believed that this is the best way to capture the essence of design cognition and behavior. Although links are associated with the achievement of synthesis, it must be emphasized that linkography does not explain causality, as the actual contents of design moves is disregarded after links are established (contents may, of course, be visited in complimentary analyses). Linkography is flexible, since base lines for variables to be measured can be calibrated to suit the needs of a study. However, we must insist on consistency in handling the data and in performing the analyses.