A conceptually clustered matrix has its rows and columns arranged to bring together major roles, research subtopics, variables, concepts, and/or themes together for at-a-glance summative documentation and analysis. Deciding what composes a row and column heading can happen in two ways: (1) deductively, the analyst may have some a priori ideas about key concepts, themes, or theories that will be explored in a study or, (2) inductively, during early analysis, you may find that participants are giving very similar or vastly different responses to questions or that unexpected variables, concepts, and themes are emerging. The basic principle is conceptual or thematic documentation of data in matrix cells, which may or may not be accompanied by researcher-assigned evaluative descriptors (see Displays 7.4 and 7.5).
Applications
Many studies are designed to answer a lengthy string of research questions. As a result, doing a separate analysis and case report section for each research question is likely to tire out and confuse both the analyst and the reader. One solution is to cluster several research questions so that meaning can be generated more easily. Having all of the data in one readily surveyable place helps you move quickly and legitimately to a boiled-down matrix by making sure that all the data fit into a reasonable scheme and that any evaluations or ratings you make are well-founded.
Display 7.4
Conceptually Clustered Matrix: Motives and Attitudes (Format)
Source: Miles, M. B., & Huberman, A. M. (1994). Qualitative data analysis: An expanded sourcebook (2nd ed.). Thousand Oaks, CA: Sage Publications.
Conceptually clustered matrices are most helpful when some clear concepts or themes have emerged from the initial analysis. They also can be used with less complex cases, such as individuals or small groups.
Example
In our (Miles and Huberman) school improvement study, we had a general question about users’
and administrators’ motives for adopting a new educational practice, and a more specific question
about whether these motives were career centered (e.g., whether participants thought they could get a promotion or a transfer out of the project). So here we had an a priori idea of a possible relationship between two concepts. Then, during data collection, we saw some inkling of a relationship between the motives questions and two others: (1) a centrality question (whether the innovation loomed larger than other tasks in the daily life of a user) and (2) an attitude question (whether the participant liked the new practice when first introduced to it). We wondered whether a relationship existed between people’s motives and their initial attitudes toward the practice.
The best way to find out would be to cluster the responses to these questions. Not only is there a relationship to probe, but there is also a general theme (initial attitudes) and a possibility of handling three research questions and their concepts at the same time.
The conceptually clustered matrix is a format that
• displays all of the relevant responses of all key participants;
• allows an initial comparison between responses and between participants;
• lets us see how the data can be analyzed further (e.g., repartitioned or clustered);
• for multicase studies, lends itself easily to cross-case analysis and will not have to be redone;
and
• for multicase studies, provides some preliminary standardization—a set of content-analytic themes that all case analysts will be using.
When you are handling several conceptually or thematically related research questions together, a likely start-up format is a simple participant-by-variable matrix, as shown in Display 7.4. Thus, we have on one page a format that includes all respondents and all responses to the four research questions (i.e., the concepts of interest in this study). Note that we have set up comparisons between different kinds of participants (users and administrators), so it is role ordered as well as conceptually ordered. The format also calls for some preliminary sorting or scaling of the responses:
types of motive, career relevant or not, degree of centrality, and valence of initial attitudes.
Next, we go back to coded segments of data keyed to the research questions and their suggested concepts. The analyst notes down the Motives given by or attributed to a participant and then tries to put a label on the motive. One participant, for example, gave several motives: She heard how good the new practice was (social influence), her principal was “really sold on it” and “wanted it in”
(pressure), most other teachers were using it or planned to—“It’s what’s coming” (conformity), and using the new practice was an occasion to “keep growing” (self-improvement). At this stage, it is best to leave the start-up labels as they are, without trying to regroup them into fewer headings that cover all participants; this practice gives you more degrees of freedom while still providing a preliminary shaping of the data.
Turning to Career Relevance, the second concept, the analyst summarizes in a phrase or sentence the relevance of adopting the practice for each participant. The next task is to look for evidence of the Centrality of this new practice for people and what their Initial Attitudes seemed to be. For these two columns, the analyst assigns a general rating, backing it with specific quotes. When these data are entered in the matrix, we get something like Display 7.5.
Display 7.5 contains about as many data as a qualitative analyst can handle and a reader can follow. The analyst has ordered participants according to their time of implementation (Early Users, Second Generation, and Recent Users) and their roles (Users and Administrators) and, within the group of users, has included a Nonuser to set up an illustrative contrast between motives for adopting and motives for refusing the new practice.
For cell entries, the analyst reduced the coded chunks to four kinds of entries: (1) labels (e.g., self-improvement), (2) quotations, (3) short summary phrases, and (4) ratings (none/some, low/high, and favorable/unfavorable). The labels and ratings set up comparisons between participants and, if needed, between cases. The quotations supply some grounded meaning for the material; they put
some flesh on the rating or label and can be extracted easily for use in the analytic text.
The summary phrases explain or qualify a rating, usually where there are no quotations (as in the Career Relevance column). In general, it’s a good idea to add a short quote or explanatory phrase beside a label or scale; otherwise, the analyst is tempted to work with general categories that lump together responses that really mean different things (as seen in the “high” responses in the Centrality column). If lumping does happen and you are puzzled about something, the qualifying words are easily at hand for quick reference.
It’s important to hold on to the common set of categories, scales, and ratings for each case—even if the empirical fit is poor in one or another of these columns—until the full set of cases can be analyzed.
Analysis
Reading across the rows gives the analyst a thumbnail profile of each participant and provides an initial test of the relationship between responses to the different questions (tactic: noting relations between variables). For example, L. Bayeis does have career-relevant motives, sees the practice as very important, and is initially favorable. But R. Quint’s entries do not follow that pattern or a contrasting one. We have to look at more rows.
Reading down the columns uses the tactic of making comparisons between the Motives of different users and administrators, as well as comparisons between these groups. It also enables similar comparisons between responses to the Career Relevance, Centrality, and Initial Attitudes data.
A scan down the columns of Display 7.5 provides both information and leads for follow-up
A scan down the columns of Display 7.5 provides both information and leads for follow-up analyses. The tactic of making contrasts/comparisons leads to conclusions. For example, there is some career relevance in adoption for users but practically none for administrators. Centrality is high
—almost overwhelming—for users but less so for administrators. Users are less favorable initially than administrators.
Looking across rows, we can use the tactic of noting relations between variables and see that for two of three career-motivated users a relationship exists among the variables: High centrality and favorable attitudes are also present. But the opposite pattern (low career relevance, low centrality, and neutral/unfavorable attitudes) does not apply. In fact, it looks as if some people who are neutral would have been favorable were they not so apprehensive about doing well (tactic: finding intervening variables).
In sum, a conceptually clustered matrix brings together key data from key participants into a single matrix. The goal is to summarize how things stand with regard to selected variables, concepts, or themes of interest. Avoid using more than five related research questions for a conceptually clustered matrix, otherwise the mind will boggle. There will be too many data to see inclusively at one time and too much time spent manipulating blocks of data to find clusters and interrelationships.
Notes
Conceptually clustered matrices need not be organized by persons or roles, as in Display 7.5.
More general concepts and themes can be the ordering principle in the rows as well as in the columns. For example, rows can consist of cells broken into Types of Problems , with columns divided into various Forms of Coping Strategies. Less emphasis is placed on specific cases and people and more on the conceptual and thematic matters of the study.
Folk Taxonomy
[hierarchical] lists of different things that are classified together under a domain word by members of a microculture on the basis of some shared certain attributes. (pp. 44–45)Spradley (1979) further defines a folk taxonomy as “a set of categories organized on the basis of a single semantic relationship.” The taxonomy “shows the relationships among all the folk terms in a domain” (p. 137). A verbatim data record to extract folk terms is necessary for constructing a taxonomy. But when no specific folk terms are generated by participants, the researcher develops his or her own—called analytic terms.