Unobtrusive Measures, Content Analysis, and Systematic Observation
An unobtrusive measure reflects data that are gathered in a way that cannot be affected by the originator’s (or communicator’s) knowledge that they are being used for research purposes.
This approach typically means that the originator of the data is unaware that a study is or will be conducted with the information that becomes data for the researcher. Webb, Campbell, Schwartz, and Sechrest (1966) provided the first detailed analysis of such measures, and they indicated the advantages of such measures over the more traditional measures that rely on self-reports gathered by questionnaires and interviews. Unobtrusive data are nonreactive: Because the individuals are unaware that they are being studied, they cannot try to (a) impress the investigator by displaying positive qualities (a social desirability effect) or (b) guess the hypothesis that is being tested and act either to confirm it (the “good” participant) or to disconfirm it (the “bad” participant).Unobtrusive measures include archival data and records generated from individuals, companies, and governments; trace evidence (by accretion, such as nose prints on an exhibit’s glass enclosure or trash in a backyard, or by erosion, such as worn-down steps); and hidden observation of behavior, such as blogs monitored for conflict interactions. In his book on methods for studying conflict, Druckman (2005) indicated the benefits of this kind of measure: This type of measure “avoids subject reactivity.” It “can capture trails (or traces) not recoverable by asking questions,” and it provides “indicators of activities not known by other methods, as in archaeology” (p. 345). On the down side, Druckman indicated that there is a “lack of control over the data production” and that “interpretations [of these data] may be ambiguous, especially when other data are not available.” These data are also “subject to observer biases” (p.
345).If one studies interpersonal conflict, the problems of using reactive measures may loom large: People want to put their best face forward, and therefore their behavior in the presence of the investigator or their verbal responses to the investigator may be different from those behaviors or responses without the participant’s awareness of the investigator. Although not perfect, unobtrusive measures may be invaluable here.
Content analysis, which can fall under the rubric of unobtrusive measures depending on what content is analyzed, has been used to study conflict. Starting with Bales (1950a, 1950b), content analysis of communication has been used to examine conflict in small- group interaction, including during negotiation and mediation. Bales (1950b) assigned the communicative possibilities exhibited by interactants into 12 categories. A communication by an interactant may (1) show solidarity, (2) show tension release, (3) agree, (4) give a suggestion, (5) give an opinion, (6) give orienting information, (7) ask for orienting information, (8) ask for an opinion, (9) ask for a suggestion, (10) disagree, (11) show tension, or (12) show antagonism (p. 258). In any observed interaction, there may be conflict, as indicated by messages of type (10), (11), or (12). However, the coding system is not restricted to interactions that necessarily involve conflict. Negotiation research often looks at the unfolding of interaction, with some level of conflict inherent within the interaction because of the competing goals of the interactants.
Content analyses based on interaction coding schemes were developed from the mid-1980s to the mid-1990s, reflecting a continued interest in interaction during conflicts. Hostage negotiation, business negotiation, and marital mediation were studied using the methods of conversation analysis, which can be a form of content analysis. For example, lag-sequential analysis was used to study interaction patterns during simulated negotiation (e.g., Cai & Donohue, 1996), and phase mapping was used to study conflict phases during interaction (e.g., Holmes, 1997). These approaches should have created opportunities for conflict communication research; however, such investigations have declined over the past several years.
Indeed, interaction analysis of conflict has been largely abandoned. A notable exception to this trend was a collection of articles in a 2003 issue of the International Journal of Conflict Management (Parks, 2003). A more elaborate discussion of content analysis for studying conflict communication may be found in Druckman (2005) and Roberts (1997), and detailed guides to the general issues associated with content analysis may be found in Krippendorff (2004) and Neuendorf (2002).Interactional data typically result from content analyses based on systematic observation. Coders (judges, observers, or raters) are given coding rules and then code or rate the behaviors that are observed. If the behaviors are to be counted, then rules are needed to unitize the behaviors (i.e., determine where one behavior ends and another begins). For example, in a study regarding the interaction between hostage takers and negotiators, Taylor (2002a) analyzed transcripts of nine hostage incidents. A rhetorical structure analysis was conducted to divide each transcript into separate episodes, or dialogue movements, based on changing themes. In addition, thought units were unitized from each episode before they were coded and subjected to data analysis. Other examples of creating rules to unitize data may be found in Gordis, Margolin, and Garcia (1996) and Gordis et al. (2001), both of which examine family conflict.
A great deal of literature is available on the factors that affect the unitization of behavior as assessed by actors and observers; the process of unitization should not be thought of as without difficulties (see, e.g., Girbau, 2002; Lemus, Seibold, Flanigan, & Metzger, 2004; an elaborate discussion of this issue is found in Krippendorff, 2004). In addition to creating rules for unitization, the results of the unitization must be assessed for reliability; as Krippendorff (2004) indicated, there must be agreement not only on the total number of units but also on what the actual units are that are observed.
Typical reliability measures for this type of analysis include Guetzkow’s U, which assesses whether coders agree reliably about unitizing the content, and κ (either Cohen’s or Krippendorf’s), which assesses whether coders are using the coding scheme with reliable agreement on how the codes apply (see Folger, Hewes, & Poole, 1984; Krippendorff, 2004, for how to compute these measures).If observational data are quantitative ratings rather than frequency counts, reliability is typically assessed by intercoder reliability, most commonly in the form of one or more bivariate correlations. For scores derived from multiple coders, Cronbach’s α may also be reported. Additional examples of unitization and reliability assessment in interactional conflict data may be found in research on hostage negotiation (Donohue, Ramesh, & Borchgrevink, 1991; Donohue & Roberto, 1993, 1996; Rogan & Hammer, 1995a ; Taylor, 2002b), family conflict (Gottman, Levenson, & Wooden, 2001; Smetana, Yau, & Hanson, 1991), interethnic conflict (Collier, 1996), and third-party mediated conflict (Jones, 1988).
Research Design
Studies can be differentiated by whether they employ cross-sectional data, such as surveys at one point in time; panel data that employ at least two points in time; and time-series data, based on many points in time. These studies may be experimental or nonexperi- mental. Next, we discuss how the conception of conflict interacts with the type of design employed.