Numerically Aided Phenomenology: Procedures for Investigating Categories of Experience

1.

Introduction

2.

Sources of Imprecision in Investigative Praxis

2.1

Imprecision in the quantification of variables

2.2

Imprecision in qualitative judgments of kind

3.

Numerically Aided Phenomenology

3.1

Thematizing description

3.2

Similarities, classes, and distinctive constituents

4.

A Demonstration

4.1

Methods

4.2

Results

5.

Conclusions

Acknowledgements

Notes

References

Author

Citation

Appendix: Texts Analyzed

Download:

141 kb

Discuss this Article:

Discussion Board

1.

Introduction

Although numeric algorithms are sometimes considered incompatible with the distinctive objectives of qualitative studies, quite frequently quantitative and qualitative methods are presented as complementary modes of inquiry. One rationale for their complementarity is that qualitative methods enable the discovery and analytic articulation of previously unobserved phenomena. And, sometimes—but not always—the qualitative methods that provide these discoveries facilitate development of the quantitative measures that are traditionally deemed a hallmark of scientific endeavours (cf. LOOS 1995). Although this rationale for complementarity gives credit to the intellectual accomplishments evident in efforts to disembed subtle phenomena from the flux of human affairs, it would be inappropriate to conclude that qualitative methods necessarily lead to quantitative refinement; quite often qualitative methods—and the connoisseurship that supports them—constitute the very best science available (POLANYI 1958). However, when investigators allow that occasionally qualitative methods lead to quantitative refinement, we regard this position as defensible within a perspective that gives both quantitative and qualitative inquiry a fruitful location within the human sciences. [1]

Another rationale for complementarity supports methodological triangulation between quantitative and qualitative research strategies. From this perspective, the trustworthiness (nee validity; LINCOLN & GUBA 1985) of findings in qualitative studies partly depends upon their convergence with results obtained using quantitative methods that are not subject to the same sources of imprecision (MORSE 1991). The emphasis on convergence (and, when appropriate, divergence) of qualitative and quantitative findings gives epistemic priority to evidential coherence—even though the relevant coherence criteria remain markedly less concrete than the decision rules articulated for psychometric construct validation (CRONBACH 1971). However, it should be remembered that coherence criteria play an important role within qualitative methods per se and that, in many circumstances, purely qualitative coherence is just as (or even more) important than qualitative-quantitative convergences.¹⁾ So, provided that deference is not automatically given to quantitative "precision," we also regard triangulation between qualitative and quantitative methods as defensible investigative praxis. [2]

Despite their appeal, the preceding forms of qualitative-quantitative complementarity beg questions that are less frequently considered and potentially more basic. Is there an unacknowledged form of quantification in what are usually considered qualitative methods? Does this unacknowledged form of quantification play an important role in the type of analytic induction that is pivotal in qualitative thinking? Is there a place—or even a need—for numeric algorithms at the inductive core of qualitative methods? We think the answer to each of these questions is, "yes." Our goal here is to clarify this conclusion and then to demonstrate that selected numeric algorithms offer significant advantages in phenomenological forms of qualitative research by adding precision to what it means to think "qualitatively." Specifically, we will introduce a form of phenomenological investigation that inductively brings categories of experience to greater distinctiveness, coherence, and richness through the quantitative systematization of categorical thought. [3]

2.

Sources of Imprecision in Investigative Praxis

2.1

Imprecision in the quantification of variables

Primarily because of its association with achievements in the physical sciences, quantified measurement seems a step toward enhanced precision. But, precision, as understood here, means more than reliability and validity; it also requires appropriately complex representation of the target construct. In phenomenological terms, precision refers to the distinctiveness that fosters reliability, the coherence that assures validity, and the richness that is appropriate to the targeted phenomenon. First, distinctiveness is the extent to which a phenomenon is discriminable from others. Judgments about distinctiveness require more than explicit (e.g., operational) definitions. They require the capacity to anticipate attributes that remain implicit in even the most explicitly conceived phenomenon and, on the basis of those implicit meanings, to consistently verify that phenomenon's presence or absence. Second, coherence is the extent to which judgments about the attribute structure of a particular phenomenon are congruent. Short of logical entailment but beyond associative contingency, judgments about coherence require consideration of both the explicit and implicit meanings of the attribute structure they describe. Third, richness is the extent to which judgments about a phenomenon capture its complexity and intricacy. Richness entails full differentiation of a phenomenon's attributes, identification of its attribute structure, and appreciation of its structural incongruities. [4]

Implicit in this tripartite conception of precision is a critique of indices in quantitative human sciences research. Indices, in contrast to direct measurement, often entrench imprecision within measurement procedures. For example, suppose we were interested in student use of library books. Direct measurement might combine the number of times library books are borrowed and the length of the borrowing period. In contrast, an index of student use of library books might be the number of times library books are borrowed, independently of how long they were borrowed. Although this index might suffice within the pragmatics of some research designs, it fails to grasp the complexity of the targeted construct. Quantitative measurement in human sciences research regularly lacks precision because, as in this simple example, the complexity of the targeted construct is not fully represented in the methods of measurement. [5]

Quantitative methods are, in part, a response to the demand for greater precision in human sciences research. However, quantification often fails in this objective because it does not meet the demand that measurement capture the rich complexity of the targeted construct. To appreciate this argument, it will be useful to examine more concretely how the promise of quantification can, in practice, betray the investigator's commitment to precision. Consider first how a dichotomous, nominal scale is implied in an affirmative response to the open-ended question, "Are you sad?" "I am sad" implicitly is contrasted with its negation, "I am not sad." And yet, this simple affirmation remains ambiguous; it does not reveal how sad the person is. In these circumstances, the introduction of an ordinal scale (0= "I am not at all sad," 1= "I am slightly sad," 2= "I am very sad," 3= "I am extremely sad") seems like discriminatory refinement of the type that can be properly compared with the physicist's instrument-guided measurement of spatial or temporal continua. Although we may have some qualms about whether "slightly," "very," or "extremely" mean the same to everyone, we are sufficiently assured of the conventionality of these meanings that this quantifying procedure plausibly enhances precision in judgments about a person's sadness. [6]

However, such enhancement is often illusory. Consider, for example, quantification of the extent to which mentation reported upon awakening from sleep is "dreamlike." As with sadness, a dichotomous nominal scale is implied in an affirmative response to the open-ended question, "Were you experiencing dreamlike imagery just before you awakened?" "I was experiencing dreamlike imagery" implicitly is contrasted with its negation. And, as with sadness, the introduction of an ordinal scale might seem a welcome advance in precision (0= "I was not experiencing dreamlike imagery," 1= "I was experiencing slightly dreamlike imagery," 2= "I was experiencing very dreamlike imagery," 3= "I was experiencing extremely dreamlike imagery"). However, elaboration of the original nominal scale is problematic in this case because, perhaps more obviously than with "sadness," the targeted construct "dreamlike imagery" is subtle and multifaceted. The conventionality of its meaning—and hence the meaning of every point on the scale—is not by any means assured, and neither then is precise measurement. [7]

Thurstone scaling (THURSTONE & CHAVE 1929) is one attempt to ensure the conventionality of meaning in scale items. The first step in these procedures is to present an open-ended question to a large number of people (e.g., "What was going through your mind just before you awakened?"). Then, based on the level of the targeted construct expressed in each response (e.g., "dreamlikeness"), each statement is assigned a value by a panel of judges. Typically, each judge independently sorts these statements into ten numbered piles, attempting to maintain equal intervals between each pile. Those statements with high inter-judge reliability are selected as options for the scale, assuring that each item on the scale is conventionally understood as representing a particular level of the targeted construct. When the completed scale is administered, respondents check the statement that most clearly describes their experience (e.g., "I experienced imagery in the form of an emotional and bizarre story.) [8]

Imprecision in scales developed using this procedure resides partly in those responses to the original open-ended question that are "unreliably" sorted. Even ignoring unique responses, statements that recur (if not verbatim, then at least in the theme-repeating form familiar to qualitative researchers) but that are not reliably sorted according to level, are removed from the final scale and, hence, from the measurement procedures. To this extent, the resulting scale becomes an index and not a direct measure of the targeted construct. Moreover, among the individuals responding to the completed scale, those whose experience is thematically captured by the recurrent but not reliably sorted statements will be required to align their ratings with the scale's reliably sorted statements. To that extent, their scale responses distort their self-perceptions and constitute a critical source of imprecision in quantitative measurement. One goal of qualitative methods is to recover and relocate those recurrent but not reliably sorted statements—and restore conceptual precision to investigative efforts. [9]
 

2.1.1 An example of quantitative imprecision

An actual example may help to consolidate this argument. Although not developed using Thurstone scaling, an instrument frequently used to judge the dreamlike quality of sleep mentation is the Dreamlike Fantasy Scale (FOULKES 1970). To develop this scale, FOULKES combined several dream attributes within a single continuum. The resulting scale, designed for use after systematic awakenings from different stages of sleep, introduces the following 8-point continuum:

1. Mind was blank

2. Experienced something but forgot what it was

3. Conceptual (no imagery), everyday content

4. Conceptual (no imagery), bizarre content

5. Non-hallucinatory (sensory) imagery, everyday content

6. Non-hallucinatory (sensory) imagery, bizarre content

7. Hallucinatory (sensory) imagery, everyday content

8. Hallucinatory (sensory) imagery, bizarre content [10]

Two features of this scale underline the extent to which it is an imprecise measure of the dreamlike fantasy continuum. First, points 3-8 on the scale reflect interactive conjunctions of sensory imagery, hallucinatory quality, and bizarreness. For example, conceptual (no imagery) mentation cannot be hallucinatory, whereas sensory imagery can be either hallucinatory or non-hallucinatory. Thus, unlike the sadness continuum (see above), points along the dreamlike fantasy dimension do not tidily manifest "more" or "less" of some single attribute. In fact, as FOULKES acknowledged, such conjunctions suggest that this instrument might be better understood as an array of nominal categories rather than as an ordinal scale.²⁾ Second, these interactive conjunctions are not arbitrary, but rather reflect the investigator's considerable experience with responses to the open-ended question, "What was going through your mind just before you awakened?" However, these conjunctions are not exhaustive. Some recurrent forms of response to the open-ended question are difficult to locate on this continuum. For example, dreams in which a traumatic event (e.g., a combat episode) is vividly re-experienced are not "bizarre," and yet such dreams would be rated as less dreamlike than those containing a minor incongruity (e.g., a figure that combines the features of two different people). Thus, despite FOULKES' (1985) argument that there is "general agreement" about "what dreams are" (p.18), the re-experiencing dream is a recurrent response that is not (to use a Thurstone scaling analogy) among those "reliably" sorted. In this way it conceals imprecision in ways that betray its scientific objectives. [11]

We believe that the type of imprecision found in the Dreamlike Fantasy Scale is representative, rather than rare, among instruments developed for measurement in the human sciences. Reference to the Thurstone scaling procedures demonstrates that assessment of the precision gained—or lost—through quantitative measurement depends upon familiarity with the full array of responses to an original open-ended question. To be precise, an instrument must exhaust the complexity of that array. If recurrent responses are omitted, the instrument's faithfulness to the targeted construct is diminished. This argument does not preclude scale refinements that more and more adequately address such complexity. But determining how adequately any scale addresses such complexity requires systematic consideration of the recurrent expressions presented in response to the original open-ended question. In effect, qualitative investigation is logically anterior to precise "scale" development—and it plays a critical role in the assessment of that precision. [12]

2.2

Imprecision in qualitative judgments of kind

We suggested earlier that precision refers to the distinctiveness that fosters reliability, the coherence that assures validity, and the richness that is appropriate to a targeted construct. Implicit in this conception of precision is a critique of the failure in qualitative research to address issues pertaining to the "extent" to which a category of experience is manifest in an experiential narrative. The failure to address these issues entrenches a source of imprecision within qualitative research, a source that is complementary to the sources of imprecision in quantitative studies. To justify this conclusion, we will consider just one form of qualitative study: phenomenology. The reason for this focus is that phenomenologically oriented investigators explicitly aspire toward distinctiveness, coherence, and richness in the articulation of their targeted constructs. How effectively do they attain their objectives? [13]

Phenomenologically grounded qualitative methods address questions of the kind: What are the attributes A₁...A_N such that, when they are jointly present as aspects of O, O is an X? For example, what are the attributes of an experience such that, when they are expressed in an experiential narrative, that narrative is an expression of "dreaming"? Usually the phenomenon of interest is tacitly known. For example, it is assumed that the investigator can recognize when—but perhaps not how—a narrative describes an experience of "dreaming." With that confidence, the phenomenologically oriented investigator examines a set of narratives to identify the attributes that are invariably present in the experience of "dreaming." [14]

Implementation of this strategy begins with a form of open-ended questioning comparable to that required in Thurstone's scaling procedures: "What was going through your mind just before you awakened?" And, similarly, phenomenological investigators act as judges who review the resulting array of responses and sort them. However, rather than sorting each response according to the level at which a targeted construct is manifest (e.g., "dreamlikeness"), judges sort them according to kind. For example, in one study, GIORGI (1985) asked participants to describe an experience of learning. Through detailed analysis of their responses, he identified two different kinds of learning and articulated the attributes that were characteristic of each. By identifying kinds of experience but ignoring levels of their manifestation, phenomenological investigations such as GIORGI's avoid quantification. Moreover, by addressing questions of kind, these studies preclude the forms of imprecision that often corrupt efforts toward quantification. That is, recurrent forms of response that may be difficult to sort by level (e.g., dreams in which a traumatic event is realistically re-experienced) are instead understood by phenomenologically oriented investigators to constitute a qualitatively different kind of response. [15]

And yet, the phenomenological gain in precision may obscure a complementary problem: in the human sciences, questions of kind cannot be asked without also entertaining questions of level. The justification for this seemingly impertinent assertion becomes evident when phenomenological methods are considered more closely. Categories of experience can be identified independently of the levels of their manifestation only when investigators are concerned with what HUSSERL (1913/1931) called "exact essences," e.g., the essence of geometric forms. In the articulation of exact essences, variations in the phenomenon of interest may even be pursued in imagination; there is no need to examine their occurrence empirically. In the examination of a concrete instance of a geometric form such as a triangle, it is possible to imagine that a single discrete attribute of that phenomenon (e.g., its colour or its number of sides) is changed. After varying this single attribute in imagination, judging whether the variation remains an instance of the phenomenon (as in the case of varying a triangle's colour) or whether it does not (as in the case of varying its number of sides) reveals whether that varied aspect is essential for the phenomenon to be the kind of object that it is, i.e., a triangle. [16]

The manner in which imaginative variation has been advocated by some phenomenologists (cf. GIORGI 1985; 1997) resembles its use in the articulation of exact essences. In the confidence that the investigator can recognize when a narrative describes a particular kind of experience, it seems possible to examine narrative variations in imagination and to determine whether particular attributes distinguish that experience. By systematically imagining possible modifications of concretely given narratives, the investigator presumably can uncover the essence of the phenomenon. However, this methodological proposal is put in jeopardy when, as is almost invariably the case in the human sciences, investigation moves beyond the realm of exact essences. HUSSERL (1913/1931) contrasted exact essences, such as the essence of "triangle," with what he called "morphological essences". To appreciate this distinction, consider the kind of intentional object that those who know trees call a "red oak." Red oaks usually appear as trees with leaves having spiny pointed lobes; in contrast, white oaks usually appear as trees with leaves having rounded lobes. We emphasize "usually." The bluejack oak, for example, is a type of red oak that, unlike its red oak companions, has leaves with rounded lobes. Nonetheless it is considered a red oak because it possesses many of the other attributes of members of the red oak family, such as their characteristically bitter acorns, the hairy surfaces inside their acorns, and so on. In these circumstances, varying a concrete representation of a red oak tree in imagination so that it has rounded lobes cannot help to determine whether spiny pointed lobes are essential for a red oak tree to be the kind that it is because some red oaks do not have leaves with spiny pointed lobes. Within this deciduous domain, using the imaginative variation of single discrete attributes to identify essences simply (and in principle) does not work. [17]

As MERLEAU-PONTY (1962) pointed out, considerations such as these prompted the later HUSSERL (1938/1970) to more tightly circumscribe the role of imaginative variation. Imaginative variation became the means by which the inductive examination of numerous actual instances of a phenomenon progressively leads to articulation of their idealizing limits, a process based upon a kind of "interpolation" across concrete instances. In our deciduous example, such interpolation enables the articulation of a prototype, i.e., the ideal, prototypic limits of the category "red oak tree." This formulation captures the structure of our experience with red oak trees—and of most phenomena within the human sciences as well. It rescues the notion of essential structure from the strictures of exact essences and places it in the service of morphological essences, a shift that is seldom acknowledged in critiques of phenomenology's "essentialism." Equally important, it suggests the need to incorporate consideration of the "degree" to which an instance "fits" a prototype within phenomenologically oriented qualitative studies. [18]

By formally describing this state-of-affairs, the means by which we can more precisely address questions of "degree" become clear. Within some domains, categories of experience will have the following structure (cf. BECKNER 1959):

1. Each instance of the category has a large but unspecified number of attributes;

2. Each attribute in that array is an attribute of many instances of the category; and

3. No attribute in that array is an attribute of every instance of the category. [19]

By virtue of the third criterion, which defeats imaginative variation in its commonly advocated form, no attribute is strictly invariant. By virtue of the second criterion, each attribute is more-or-less invariant—and extrapolation from these more-or-less invariant attributes identifies the ideal, prototypic limits of the category, i.e., its morphological essence. By virtue of the first criterion, the comparison of entities across a large number of their attributes becomes the mode of access to these categories, as well as to their prototypes. Within such comparative thought, HUSSERL argues, we necessarily rely on "degrees" of similarity, primarily because the "extent" of similarity between two or more wholes depends upon how many parts of each are, in themselves, similar (HUSSERL 1948/1973; §45, p.193).³⁾ By implication, we propose, HUSSERL is acknowledging that there is inherent quantification of "extent" and "degree" within similarity judgments—and hence within the determination of morphological essences. [20]

Categories defined in the preceding way have been called polythetic classes (SNEATH & SOKAL 1973). Techniques of numerical analysis exist for the identification of such classes and of their more-or-less invariant attributes. Briefly, if members of a set of phenomena are examined for the presence or absence of an array of attributes, functions such as a correlation or distance coefficient may be used to express the degree of similarity between any two members of the set. A number of cluster analytic algorithms can then be used to classify together members of the set that have a certain degree of mutual similarity. These steps may be used to form classes whose more-or-less distinctive attributes can then be identified. And, extrapolation from these distinctive attributes enables articulation of the ideal, prototypic limits of the class. Finally, correlation or distance coefficients can be used to articulate the "degree" of similarity between any instance of the class and its ideal prototype. [21]

The conception of prototype used here must be distinguished from the conception of "typicality" that is often invoked in quantitative human sciences research. The calculation of an average as a measure of central tendency—and the determination of deviations from that average—is common statistical procedure. For example, the four values 0, 1, 1, and 1 have a mean of .75, and the deviations from that mean are -.75, +.25, +.25, and +.25. What deviates minimally from .75 is commonly considered "typical." In contrast, the values in that same array could be compared with the highest value, and then the deviations would be -1, 0, 0, and 0. The ideal, prototypic limits that constitute the HUSSERLian morphological essence are analogous to the use of the extremes (in this case, 1) of the array to identify "typicality." Moreover, this procedure is extended across a number of attributes that are more-or-less invariant for the class. So, if members of a polythetic class had the following profiles of 10 attributes (where 0 = attribute absent and 1 = attribute present):
 

Class Member #1

1, 1, 0, 0, 0, 1, 0, 1, 1, 0

Class Member #2     

0, 1, 1, 0, 1, 0, 1, 1, 1, 1

Class Member #3

1, 0, 1, 0, 1, 0, 0, 1, 0, 1

Class Member #4

1, 1, 1, 0, 1, 0, 0, 0, 1, 1
 

The ideal prototype would be:
 

Ideal Prototype

1, 1, 1, 0, 1, 0, 0, 1, 1, 1 [22]
 

Note that, because this is a polythetic class, no actual profile is identical to the ideal prototype. Moreover, by determining the proportion of coinciding 0's and 1's, the profile of each member of the class can be characterized according to the "degree" to which it resembles the ideal prototype: Class Member #1 (.6), Class Member #2 (.8), Class Member #3 (.8), and Class Member #4 (.9). Due to variation in attribute frequency (e.g., .00 for the fourth attribute and .25 for the sixth), correlations between profiles provide a more exact indication of the "degree" of similarity to the ideal prototype: Class Member #1 (r = .273), Class Member #2 (r = .575), Class Member #3 (r = .636), and Class Member #4 (r = .779).⁴⁾ [23]

Within the domain of experiential narratives, this strategy—embedded within an approach that is, in general, derivative from HUSSERL's (1938/1970) examination of the life world—defines numerically aided phenomenology (KUIKEN, SCHOPFLOCHER, & WILD 1989). We will now proceed to describe and demonstrate this method. [24]

3.

Numerically Aided Phenomenology

The objective of numerically aided phenomenology is to bring to distinctiveness and coherence the full complexity of different categories of lived experience. With all the refinement and nuance that language will allow, and with the support of selected numeric algorithms, the aim is to articulate the (morphological) essences of these categories of experience. In what follows, we present the sequence of steps that defines these methods. [25]

3.1

Thematizing description

A consistent feature of phenomenological methods, including numerically aided ones, is the identification and explication of recurrent themes within a set of experiential narratives. In our approach, recurrent expressions must be evident within an analogous component of the genre that characterizes the chosen set of narratives. Sometimes, for example, the genre is temporally linear narrative, such as in the recollection of a dream. Sometimes the genre has a hierarchical structure, such as when a general open-ended question is followed by a series of specific questions that invite elaboration of particular aspects of the response to the open-ended item. Although narrative genre varies from study to study, the general principle remains: the recurrent themes must be expressed in analogous components of the given genre. [26]

For example, in CARRARE's (1989) study of tragic life events, each person described the moment during which realization of the tragic turn occurred. One person, describing that aspect of his experience of a pogrom in Poland, said:

A woman, a neighbor, was crying, crying wounded, calling my mother by name ... A shudder went through me from this constant moaning of this woman in the street ... (p.119). [27]

Another person, describing an analogous aspect of her experience of the death of her brother, said:

All of a sudden, this shrieking, animal-like sound came out of my father. It was the most painful noise I think I have ever heard ... (p.119). [28]

The phenomenological task is to identify and explicate the similarities in meaning among such recurrent expressions. A paraphrase of those similarities is then portrayed in statements such as the following:

Someone I knew cried out in agony, moving me to compelling recognition of her/his pervasive suffering. [29]

It is crucial to capture as fully as possible the shared meaning of such descriptions, drawing out the nuances of meaning that they explicitly or implicitly share. A paraphrase of that shared complexity is called a constituent (KUIKEN & WILD 1988). [30]

These procedures for identifying and explicating constituents contrast with some other approaches, such as GIORGI's (1985; 1997). First, for reasons indicated earlier, imaginative variation is not considered a vehicle for articulating the shared meaning of recurrent expressions in experiential narratives. Second, in contrast with GIORGI's suggestions, it is unnecessary to provide constituent descriptions that reflect a selected discipline, such as psychology or anthropology. The similarities in meaning among recurrent expressions may, in fact, cross these somewhat arbitrary disciplinary boundaries. Third, it is not advisable to articulate a constituent with reference to some pre-established conception of the kind of experience that is under consideration, e.g., tragedy. Numerically aided phenomenological methods enable the identification of unanticipated categories or kinds of experience. Presuming to know these categories in advance may even distort constituent descriptions. Instead, in our approach, constituent descriptions are constrained exclusively by the requirement that they emerge from the comparative effort through which are captured the similar meanings of actually recurrent expressions within the set of narratives. In short, careful comparison displaces theoretical expectations. [31]

3.2

Similarities, classes, and distinctive constituents

When a constituent has been identified, each narrative within the available set is systematically re-read to determine the presence or absence of that expressed meaning. Gradually an array of such constituents is identified, each of which is neither rare (i.e., found in less than 10% of the narratives) nor ubiquitous (i.e., found in more than 90% of the narratives).⁵⁾ Elaboration of the constituent array can never be exhaustive; more likely, the pragmatics of the investigative context will determine when the next procedural step can be taken. [32]

When a full array of constituents has been identified, similarities between each pair of narratives are determined by calculating similarity coefficients (e.g., Euclidean distances). Then cluster analysis (e.g., Ward's method) is used to group narratives according to the mutual similarities in their profiles of constituents. In general, cluster analysis (cf. BAILEY 1994) is a group of numeric algorithms that groups such profiles so that the degree of similarity between members of the same cluster is maximized and the degree of similarity between members of different clusters is minimized. Ward's method, for example, begins by treating each individual profile as a group, and then it progressively combines profiles into larger groups. At each step, the criterion for the next combination is that within-cluster variance, as measured by the sum of within-cluster deviation from cluster means, is minimized. Thus, the average distances among all members of a cluster are minimized. Ward's method is hierarchical, i.e., it allows the identification of subgroups within a cluster, and the resulting clusters are mutually exclusive (rather than overlapping). [33]

Each cluster identified in this way describes a class, and, as a result, categories of similarly lived experience are derived. The numeric algorithms used in these procedures should neither be over- nor under-estimated: their use compensates for human difficulties in assessing similarities that, in our studies, involve from 50 to 300 constituents and 30 to 75 narratives. First, these algorithms provide disciplined, pair-wise assessment of the degree of similarity between the profiles of constituents associated with each narrative expression. Second, they contribute to the disciplined formation of categories of experience, categories that might not otherwise be identified. [34]

Finally, comparison of the prevalence of each constituent across clusters enables determination of those constituents that are associated with each category. Numerical procedures can be invoked here, too, to provide a disciplined characterization of those constituents that (1) differentiate a category from at least one other category, (2) distinctively identify one particular category, or (3) uniquely identify one particular category. Because in our studies cluster analysis has typically revealed 3 to 6 distinct classes, we suggest that no fewer than 30 experiential narratives are required to identify the more-or-less characteristic constituents, i.e., the morphological essences, of these classes. [35]

4.

A Demonstration

In what follows, we will demonstrate these methods by summarizing a study of different kinds of reading experience. ROSENBLATT (1978) has suggested that variations among experiences of literary texts are comparable to variations in the performance of a musical composition. Independently of skill in reading the score, musical performances may vary dramatically as aesthetic moments. Similarly, different "performances" of literary texts may be equally competent but quite distinct aesthetically. The presence of certain experiential qualities, perhaps in a certain configuration, may justify reference to some reading experiences as aesthetic ones, not simply because they are valued but because they constitute a certain kind of experience (KUIKEN 1998). Theories that articulate such aesthetic qualities in reading events have been influential. For example, ROSENBLATT's (1978) distinction between efferent and aesthetic reading and HUNT and VIPOND's (1985; 1986) related distinctions between information-driven, story-driven, and point-driven reading continue to shape empirical studies. Although think-aloud studies have already aided articulation of certain aesthetic qualities in reader experiences (cf. MANY 1991), more precise descriptions might further that objective. It was in this spirit that we examined the range of reading experiences reported by participants in the present study. [36]

4.1

Methods

Thirty advanced undergraduate English students read Sean O'FAOLAIN's short story, "The Trout." "The Trout" is the story of Julia, a young girl, who finds a live trout in a small, water-filled hollow in her garden. After worrying about the trout's predicament and listening to various explanations of how it got there, she gets up at night to rescue it by releasing it into a river. This brief story was presented segment by segment on a computer screen in a manner that allowed self-paced reading of one story segment (roughly a single sentence) at a time. Participants were encouraged to concurrently describe any and all aspects of their reading experience: thoughts, feelings, interpretations, evaluations, memories, and so on. These instructions, and a laboratory setting that was made as private and comfortable as possible, were designed to permit expression of any understanding or reaction. Although it is impossible to eliminate implicit demands, explicitly conveyed expectations to demonstrate expertise, such as instructions to comment on style (cf. GRAVES & FREDERIKSEN 1991), were not included in our procedures. [37]

Readers' verbal expressions were tape-recorded and transcribed for subsequent review. Their responses to each story segment, and to groups of such segments (a story section, roughly one paragraph), were systematically compared in order to identify similarly expressed meanings. Statements with similar meanings occurring in three or more reader narratives were identified and paraphrased to reflect as much of their common meaning as possible. For example, in response to the opening lines of the story in which Julia returns to a haunting natural setting called the Dark Walk, some participants offered comments such as, "The Dark Walk seems like a scary place," or, "The Dark Walk is almost gothic." Such similarly expressed meanings were identified and paraphrased as, "The Dark Walk seems threatening." As indicated above, this aspect of our procedures ensured that constraints on the meanings paraphrased were maximally provided by the expressions of other participants, rather than by the researchers' preconceived categories. This crucial aspect of our phenomenological approach can be contrasted, for example, with studies in which participant meanings are coded according to preconceived discourse categories (cf. ANDRINGA 1990; GRAVES & FREDERIKSEN 1991). [38]

The resulting constituents were articulated at three levels of generality. This allowed empirical determination of the level of analysis that would be most fruitful, thereby avoiding the excessive generality that has diluted the findings from some think aloud studies of literary reading (cf. OLSON, DUFFY, & MACK 1984; WAERN 1980). First, at Level One, constituent descriptions captured as much of the similarity in meaning as possible (e.g., "The Dark Walk seems threatening"); at Level Two, constituent descriptions captured the nature of the interpretive acts in somewhat more general terms (e.g., "I characterize the mood of a setting [threat]"), and at Level Three, constituent descriptions were still more general (e.g., "I characterize the mood of a setting"). This procedure allowed appropriate consideration of similarities between Level One constituents either within or across segments; for example, at Level One, "The Dark Walk seems like a scary place" (in response to story segment 1) and "The well seems dangerous" (in response to segment 25) are both paraphrased as, "I characterize the mood of a setting [threat]" at Level Two. This procedure, and its extension in principle to Level Three, enabled broader consideration of similarities among reader expressions. Despite such increasing generality, the meanings paraphrased remained the expressions of the participants, rather than of any theory. Our procedures are phenomenological methodologically rather than theoretically. Categories derived from the phenomenological theories of reader response, such as those of INGARDEN or ISER, were shunned as completely as those from studies in cognitive psychology. [39]

The presence or absence of such constituents was used to create matrices that could be explored using the numerical algorithms of cluster analysis. This step allowed identification of distinct categories of reader response according to profiles of similarities among reading experiences. The unit of analysis, the experiential narrative, was the complete reading experience, i.e., the reader's commentary on all 84 story segments. The genre of the experiential narrative was determined by participants' response to each of 15 story sections and by their comments after reading the entire story. [40]

Cluster analysis (Wards method, Euclidean distances) of arrays representing 299 Level Three constituents revealed four distinct types of reading experience. The number of clusters chosen was determined using the agglomeration schedule to identify a break in slope and by assessing the interpretive coherence of the profiles of distinctive attributes. The stability of these clusters was assessed by varying constituent level, similarity coefficient (Jaccard's coefficient), and clustering algorithm (average linkage) with moderately similar results. A Level Three constituent was regarded as differentiating if the proportion of individuals expressing it within a cluster was greater than the proportion expressing it in at least one other cluster, using Fisher's LSD test (p<.05) as a guideline that takes into account both mean differences and variability. (Although cluster analysis only involved Level Three constituents, identification of differentiating constituents also was extended to the Level Two (p<.05) and Level One (p<.10) constituents that they subsumed.) A constituent was regarded as differentiating (D) if it distinguished a cluster from one or two but not all three of the other four clusters. A constituent was regarded as differentiating and distinctive (DD) if it differentiated a cluster from all of the other three clusters. A constituent was regarded as differentiating, distinctive, and unique (DDU) if it was found only in a single cluster. We also monitored what we call characteristic constituents, i.e., those that, while not differentiating, were found in more than 50% of members of a cluster. This helps to preserve the context for understanding the other constituents that identify the cluster. [41]

4.2

Results

4.2.1 Cluster one: Story rejection, impersonal memories, and unresponsiveness

One identifying feature of readers in Cluster One (see Table 1) was their negative emotional reaction to the story. They reacted negatively almost immediately, suggesting that their rejection of the story was readily evoked: in response to the first segment, they indicated dislike for an ambiguous reference to the setting, a place identified only as "G---". However, their rejection was also general and persistent: after reading the entire story, they discretely indicated that it did not "grab", "excite", or "overwhelm" them.

_________________________________________________________________________________

Story Rejection:

C

I emotionally respond to a literary device

D

I negatively emotionally respond to literary style (phrase/word choice)

D

D

The story did not grab/excite/overwhelm me

_______________________________________

Impersonal Memories:

D

The setting evokes a life-world comparison

D

The setting reminds me of a personal memory

D

D

_______________________________________

LEVEL 3 CONSTITUENT

LEVEL 2 CONSTITUENT

_______________________________________

KEY: C=Characteristic; D=Differentiating (but not distinctive); DD=Differentiating and Distinctive; DDU=Differentiating, Distinctive, and Unique

_________________________________________________________________________________

_________________________________________________________________________________

Emotional Reactivity:

DD

I emotionally respond to a character

D

I positively emotionally respond to a character

D

C/D

I emotionally respond to a literary device

C/D

I negatively emotionally respond to literary style (phrase/word)

C/DD

_______________________________________

Holistic Imagery:

D

I imagine a character and/or actions (holistically)

D

C/D

I construct an imaginal story feature

C/D

I imagine the setting holistically

DD

_______________________________________

Personal Memories:

C/D

Actions evoke a life-world comparison

D

Actions remind me of a personal (childhood) memory

C

The setting evokes a life-world comparison

D

The setting reminds me of a personal memory

DD

_______________________________________

LEVEL 3 CONSTITUENT

LEVEL 2 CONSTITUENT

_______________________________________

KEY: C=Characteristic; D=Differentiating (but not distinctive); DD=Differentiating and Distinctive; DDU=Differentiating, Distinctive, and Unique

_________________________________________________________________________________

_________________________________________________________________________________

Articulable Uncertainty:

DDU

I query a character's motives

DDU

I am confused about a character's motives

DDU

D

I query a character's actions

D

I am confused about a character's actions

D

_______________________________________

Articulable Uncertainty:

DDU

I query a character's motives

DDU

I am confused about a character's motives

DDU

D

I query a character's actions

D

I am confused about a character's actions

D

_______________________________________

Clarify Understanding:

D

I clarify my understanding of the story

D

I add details to clarify the action

DDU

_______________________________________

Concern with Emotions:

C/D

I characterize a character's emotions

C/D

I characterize a character's emotions (unworried)

C/D

C

I characterize a character's emotions

D

I characterize a character's emotions (sympathy)

D

_______________________________________

Concern with Motives:

C/D

I characterize a character's motives

D

D

I characterize a character's motives

DDU

I characterize a character's motives (pleasurable desire)

DDU

_______________________________________

LEVEL 3 CONSTITUENT

LEVEL 2 CONSTITUENT

_______________________________________

KEY: C=Characteristic; D=Differentiating (but not distinctive); DD=Differentiating and Distinctive; DDU=Differentiating, Distinctive, and Unique

_________________________________________________________________________________

_________________________________________________________________________________

Characterizing Setting Mood/Connotation:

CD

I characterize the mood of the setting

D

The Dark Walk seems foreboding/evil/gothic

_______________________________________

Concern with Thoughts:

C/DD

I characterize a character's thoughts

C/DD

I characterize a character's thoughts (reflection)

C/D

_______________________________________

Concern with Character:

CD

I characterize a character

CD

I characterize a character (self-centered)

D

_______________________________________

Anthropomorphosis:

C/DD

I anthropomorphize a story figure

C/DD

I anthropomorphize the trout

_______________________________________

Symbolic Interpretation:

D

I maintain the integrity of a symbolic interpretation (across at least two sections)

C/D

I interpret a story symbolically

C/D

I interpret a story element symbolically (narrow confined world)

_______________________________________

Interpretive Coherence:

DDU

I confirm/elaborate my previous characterization

C/D

I anticipate story development

_______________________________________

LEVEL 3 CONSTITUENT

LEVEL 2 CONSTITUENT

_______________________________________

KEY: C=Characteristic; D=Differentiating (but not distinctive); DD=Differentiating and Distinctive; DDU=Differentiating, Distinctive, and Unique

_________________________________________________________________________________

5.

Conclusions

The preceding discussion is intended to demonstrate that numerically aided phenomenology integrates some of the sources of precision available through quantitative methods with some of the sources of precision promised by qualitative methods. Their integration provides both rigor and subtlety. Their rigor derives from the contribution that numeric algorithms make to judgments of "extent"; their subtlety derives from careful comparisons of the meanings that are expressed in numerous experiential narratives. This quantitative-qualitative complementarity is not a marriage of convenience; it is a relationship that is motivated by the limitations inherent in "purely" qualitative and in "purely" quantitative studies within the human sciences. [68]

Acknowledgment

The research reported here was supported in part by programme grant #53-10128 from the Social Sciences and Humanities Research Council of Canada.

Notes

1) For example, qualitative methods that provide convergence across levels of analysis (e.g., interview comments descriptive of anxiety that converge with interview comments indicating avoidance of a task) may be more revealing than qualitative-quantitative convergences at the same level of analysis (e.g., interview comments descriptive of anxiety that converge with—or are largely redundant with—questionnaire responses indicating anxiety). <back>

2) We would also argue that the sadness scale conceals an array of nominal categories, but making that case here would take us too far afield. <back>

3) HUSSERL also appreciates that, in many instances, the parts that determine the similarity of two (or more) wholes are themselves similar—and not alike. This point underlines the multi-leveled character of similarity judgments. <back>

4) These procedures enable determination of the degree of similarity between actual attribute profiles and the ideal prototype. This numeric assessment of degree involves profiles of attributes rather than individual attributes. Although we appreciate the potential importance of the latter (see note 3), we have not attempted to address the analytic problems that arise from the combination of nominal and ordinal variables in estimates of profile similarity. It should be noted, however, that some available software facilitates the assessment of ordinal information during attribute identification (cf. KUCKARTZ 1995; WEITZMAN & MILES 1995). The possibility of coordinating ordinal and nominal attribute judgments deserves further consideration. <back>

5) Using this criterion to eliminate rare or ubiquitous constituents minimizes some problems that arise when these numeric methods are applied to sparse matrices, e.g., difficulties in systematically identifying the constituents that differentiate the clusters. These criteria also pragmatically constrain how exhaustive (and exhausting) the process of constituent identification will be. <back>

References

Andringa, Els (1990). Verbal data on literary understanding: A proposal for protocol analysis on two levels. Poetics, 19, 231-257.

Bailey, Kenneth D. (1994). Typologies and taxonomies: An introduction to classification techniques. Thousand Oaks, CA: Sage Publications.

Beckner, Morton. (1959). The biological way of thought. New York: Columbia University Press.

Carrare, Robert A. (1989). Psychology of tragedy: A phenomenological analysis. Journal of Phenomenological Psychology, 20, 105-129.

Cronbach, Lee J. (1971). Test Validation. In Robert L. Thorndike (Ed.), Educational measurement (pp.442-507). Washington, DC: American Council of Education.

Dias, Patrick & Hayhoe, Michael (1987). Developing response to poetry. Milton Keynes and Philadelphia: Open University Press.

Foulkes, David (1970). The dreamlike fantasy (Df) scale: A rating manual. Sleep Laboratory, University of Wyoming, unpublished manuscript.

Foulkes, David (1985). Dreaming: A cognitive-psychological analysis. Hillsdale, NJ: Erlbaum.

Giorgi, Amedeo (1985). Phenomenology and psychological research. Pittsburgh: Duquesne University Press.

Giorgi, Amedeo (1997). The theory, practice, and evaluation of the phenomenological method as a qualitative research procedure. Journal of Phenomenological Psychology, 28, 235-260.

Graves, Barbara & Frederiksen, Carl H. (1991). Literary expertise in the description of a fictional narrative. Poetics, 20, 1-26.

Hunt, Russell A. & Vipond, Douglas (1985). Crash-testing a transactional model of literary reading. Reader: Essays in reader-oriented theory, criticism, and pedagogy, 14, 23-39.

Hunt, Russell A. & Vipond, Douglas (1986). Evaluations in literary reading. Text, 6, 53-71.

Husserl, Edmund (1913/1931). Ideas (W.R. Boyce Gibson, trans.). London: George Allen & Unwin.

Husserl, Edmund (1938/1970). The crisis of European sciences and transcendental phenomenology (David Carr, trans.). Evanston: Northwestern University Press.

Husserl, Edmund (1948/1973). Experience and judgment (James S. Churchill & Karl Ameriks, trans.). Evanston: Northwestern University Press.

Kuckartz, Udo (1995). Case-oriented quantification. In Udo Kelle, with Gerald Prein, & Katherine Bird, (Eds.), Computer aided qualitative data analysis (pp.158-166). Thousand Oaks, CA: Sage Publications.

Kuiken, Don (1998). Understanding the depth metaphor in aesthetic experience: Pressing the limits of psychological inquiry. In William Smythe (Ed.), Toward a psychology of persons (pp.101-117). New York: Erlbaum.

Kuiken, Don, Schopflocher, Donald, & Wild, T. Cameron (1989). Numerically aided phenomenology: A demonstration. Journal of Mind and Behavior, 10, 373-392.

Kuiken, Don & Wild, T. Cameron (1988). Meaning horizon, paraphrase, and phenomenological psychology. In William J. Baker, Leendert P. Mos, Henderikus Stam, & Hans Rappard (Eds.), Recent trends in theoretical psychology (pp.189-198). Amsterdam: Springer-Verlag.

Ingarden, Roman (1973). The cognition of the literary work of art (Ruth A. Crowley & Kenneth R. Olson, trans.). Evanston: Northwestern University Press.

Ingarden, Roman (1985). Aesthetic experience and aesthetic object. In Peter J. McCormick (Ed.), Selected papers in aesthetics (pp.107-132). Washington, DC: Catholic University Press.

Lincoln, Yvonna S. & Guba, Egon G. (1985). Naturalistic inquiry. Beverly Hills: Sage Publications.

Loos, Gregory P. (1995). Foldback analysis: A method to reduce bias in behavioral health research conducted in cross-cultural settings. Qualitative Inquiry, 1, 465-480.

Many, Joyce E. (1991). The effects of stance and age level on children's literary responses. Journal of Reading Behavior, 23, 61-85.

Merleau-Ponty, Maurice (1962). The phenomenology of perception (C. Smith, trans.). London: Routledge & Kegan Paul.

Morse, Janice M. (1991). Approaches to qualitative-quantitative methodological triangulation. Nursing Research, 40, 120-123.

Olson, Gary M.; Duffy, Susan A., & Mack, Robert L. (1984). Thinking-out-loud as a method for studying real-time comprehension processes. In David E. Kieras & Marcel A. Just (Eds.), New methods in reading comprehension research (pp.253-286). Hillsdale, NJ: Erlbaum.

Polanyi, Michael (1958). Personal knowledge: Towards a post-critical philosophy. Chicago: University of Chicago Press.

Rosenblatt, Louise M. (1978). The reader, the text, the poem. Carbondale, Ill: Southern Illinois University Press.

Sikora, Shelley; Kuiken, Don & Miall, David S. (1998). Enactment versus interpretation: A phenomenological study of readers' responses to Coleridge's 'The Rime of the Ancient Mariner.' Paper presented at the Conference of the International Society for the Empirical Study of Literature, Utrecht, The Netherlands.

Sneath, Peter H.A. & Sokal, Robert R. (1973). Numerical taxonomy. San Francisco: Freeman.

Thurstone, Louis L. & Chave, Ernest J. (1929). The measurement of attitude: A psychophysical method and some experiments with a scale for measuring attitude toward the church. Chicago: University of Chicago Press.

van Manen, Max (1997). From meaning to method. Qualitative Health Research, 7, 345-369.

Waern, Yvonne (1980). Thinking aloud during reading. Scandinavian Journal of Psychology, 21, 123-132.

Weitzman, Eben A. & Miles, Matthew B. (1995). Computer programs for qualitative data analysis. Thousand Oaks, CA: Sage Publications.

Authors

Don KUIKEN, PhD, is a Professor in the Department of Psychology at the University of Alberta. He is the author of numerous publications in the areas of dreaming, psychological aesthetics, and phenomenological psychology. Correspondence may be addressed to:

Don Kuiken, PhD

Department of Psychology
University of Alberta
Edmonton, AB T6G 2E9 CANADA

E-mail: dkuiken@ualberta.ca
URL: