B. Contributions and Innovation in Research Methods for the Study of Small Groups

My theoretical and empirical scholarship studying leadership of small groups is complemented by an interest in innovation in the methods of studying groups. This is critically important work for the small groups and top management teams field because levels of analysis problems in conducting groups research has long dogged the field, creating statistical and methodological dilemmas for scholars (e.g., Groups are formed from individuals, so should one measure the construct at the individual or group level? see Levine & Moreland, 2000). One of the key impediments to the study of group process in organizations is finding research methods suitable for studying process dynamically (Weingart, 1997). Groups researchers very often posit dynamic processes in management groups, but then most often study those processes statically with cross-sectional research methods such as single surveys or experiments (see Ancona, et. al, 2001for a full discussion). These methods are static because they assess changes in group process over time by assessing how group process evolved at the end of a group's time together rather than tracking how group process unfolds over time.

Traditional research methods used to study groups, including surveys and experiments, are at risk from at least one of two interrelated problems. The first problem concerns asking people to make retrospective judgments about group process. In general, retrospective recall of information encourages biased recall of information in support of personal implicit theories (Fiske & Taylor, 1991). More specifically, retrospective recall of group process has been shown to encourage people to combine knowledge of their groups' successes or failures with their implicit theories of the process-outcome quality relationship to report only information consistent with their implicit theory (e.g., a good outcome was the product of a reasoned debate but if the outcome had been poor the same process would have been described as dysfunctional conflict; Guzzo, et. al, 1986; Peterson, Owens, & Martorana, 1999b; Staw, 1975). Hence, retrospective recall of group process can lead the researcher to mistakenly confirm naïve implicit theories of group process and outcome, and thus prevent learning what processes actually occurred and predicted group performance.

Even when the research method does not require retrospective recall, such as behavioral coding, the method still may not provide useful information about the dynamics of group process. For example, behavioral coding schemes used to assess live group process (e.g., videotapes) typically collect information about how many times a particular event happened, but they often provide neither information about what preceded or followed the event in time sequence, nor do they provide the full context of the discussion (see Weingart, 1997). In other words, there is no information about the sequence of events or behaviors that lead to a particular outcome. As a result, the insight that can be gained from such schemes is limited. This is where the work I have been doing on research methods tries to grapple with these key problems for groups research in three ways.

1. Developing the Organizational Group Dynamics Q-Sort (O-GDQ) Technique.
2. Using the Q-Sort to Conduct Quantitative Analysis of Qualitative Data.
3. Employing Longitudinal Research Designs.

1. Developing the Organizational Group Dynamics Q-Sort (O-GDQ) Technique. The first important contribution I have made to the study of group dynamics is the development of the organizational group dynamics q-sort (Peterson, Owens, & Martorana, 1999a). The O-GDQ is a 100-item instrument designed to study group process across a wide variety of situations and using a wide variety of data sources (e.g., case studies, participant observation, etc.). Completing a q-sort involves a rater placing each of the 100 items into one of nine categories that describe the importance of the item for explaining the group. For example, item 97 asks the rater to evaluate whether "The group leader makes major efforts to persuade others to redefine their goals and priorities. versus The leader places little emphasis on persuading others." By asking the rater to evaluate which statement is most relevant for describing the group (i.e., top or bottom) and how important the statement is (i.e., extremely important, fairly important, etc.), this process permits systematic, quantitative, and reliable comparisons across observers, time, and groups to be able to make real-time assessments of groups. This aspect of the O-GDQ allows it to avoid the problem of contamination from outcome knowledge. Secondly, completing the O-GDQ requires the rater to carefully consider the importance of each item – thereby capturing some of the context lost by many behavioral coding schemes. I have used the q-sort method in a number of different papers (e.g., Peterson, 1997, Peterson, Owens, Tetlock, Fan, & Martorana, 1998, and Peterson, Smith, Martorana, & Owens, 2003).

There have been high levels of both scholarly and practitioner interest in this work. I have been asked to conduct three half-day workshops at schools around the world (most recently at INSEAD). Managers have also expressed interest in my methodological work as a method for assessing the culture of top management teams. I have, for example, been collaborating with a consulting firm working with all state sector schools in the United Kingdom to assess school leadership teams. More recently I have also been collaborating with Brandes Institute and Watson Wyatt in using the O-GDQ for assessing the culture of top management teams in the investment management industry. To date we have q-sorts completed from 46 chief executive officers about their teams, and comparison sorts completed by corporate analysts at Watson Wyatt. These top management team data generated a great deal of interest at the Brandes client meeting in October 2004, but also promise to be extremely interesting to the scholarly community as I plan to use them to test my recent theoretical thinking on the central role of legitimacy for teams (Peterson, Ronson, Anand, & Matthews, 2004).
back to top

2. Using the Q-Sort to Conduct Quantitative Analysis of Qualitative Data. To study group processes dynamically, some scholars have turned to case study or ethnographic methods. The principle advantage to these methodologies are their great descriptive richness and sensitivity to change over time. The detailed and nuanced storytelling quality of these research strategies gives them both intuitive and persuasive appeal. Compared to most quantitative studies the "thick" descriptions of group process provide a more thorough understanding of the dynamic processes in groups over time as well as context in which each group is embedded.

The classic problem with the case study approach from a social science perspective, of course, is the lack of generalizability. The problem of generalizability comes from the unique language and emphasis of each researcher. Although the use of unique language usually is where special insight into group functioning is created, it is also the cause of difficulty in assessing agreement between researchers. It is often exceedingly difficult to assess the level of agreement between two experts who research the same group. Scholar A may emphasize a group's camaraderie and esprit de corps, whereas Scholar B emphasizes its sense of purpose in vanquishing opponents. Here, the reader might be left with the impression that these two scholars disagree about the dynamics of the group when they are in fundamental agreement about a complex reality. Thus, understanding the differences between multiple researchers' descriptions of the same often-studied group is quite difficult (e.g., the many academic assessments of the group dynamics in the Space Shuttle Challenger Launch Team), but the task of identifying themes for theory-building becomes nearly impossible when trying to make comparisons across different groups studied by different scholars (e.g., many scholars writing about many different top management teams). There is, in short, no good systematic way of combining many case studies together to come to reliable conclusions. Without a coherent understanding of where there is underlying consensus, it is impossible to build a cumulative theory of group dynamics.

This is where my application of the O-GDQ to the study of groups with good quality historical or case studies written has allowed the O-GDQ to operate as a bridge methodology to translate qualitative into quantitative data for more rigorous comparison and analysis. In general, the O-GDQ embraces many of the strengths of the case study and ethnographic approach by providing a reasonably comprehensive array of items assessing the details of interaction among group members, but also has the rigor of a quantitative approach because the 100 items of the instrument are standardized, rank-ordered, and placed in one of nine categories for statistical comparison. These quantitative data can then be used to pinpoint specific differences between groups over time, in different contexts, or between multiple observers of the same group at the same point. For example, I used the O-GDQ to track the increasing rigidity, corruption, and risk aversion in the top management team at Chrysler under Lee Iacocca's leadership from 1980-1990 (Peterson, Owens, & Martorana, 1999a).

Given that the O-GDQ has the ability to bridge qualitative into quantitative data, the method has also made many historical groups accessible to many other scholars that have previously been off-limits either because the group does not generally grant access (i.e., current top management teams), or because the group no longer exists (i.e., learning from past teams where there is written history but participants are no longer available). This has, I believe, made the O-GDQ a significant addition to the methodological toolkit for groups scholars because it also makes data available to a wide scholarly audience rather than only to a few with privileged access. This allowed me, for example, to analyze contemporary accounts of top management teams from the 1970s through the 1990s to assess the causes and consequences of their failures (Peterson, Owens, & Martorana, 1999b).
back to top

3. Employing Longitudinal Research Designs. I have also developed an interest in analysing problems in longitudinal research designs. Beyond those already described in the q-sort work, I have also tracked groups across time with multiple surveys in order to study how conflict in groups begins and evolves over time (e.g., Behfar, Peterson, Mannix, & Trochim, 2004; Peterson & Behfar, 2003). Here along with Ph.D. student Kristin Behfar, we have had a number of important breakthroughs finding, for example, that relationship (affective) conflict is generally a product of poor group performance rather than a cause of poor performance. In other words, we found that early relationship conflict did not predict later performance, but early performance measures predicted subsequent reports of relationship conflict. This stands current research thinking on its head, having historically suggested that relationship conflict predicts poor group performance.

The second major insight we have found from a longitudinal research design explores the link between conflict resolution tactics and group performance in teams over time (Behfar, Peterson, Mannix, & Trochim, 2004). We found that conflict resolution tactics are an essential source of later group structure – predicting both group process and performance. Our results suggest a group adaptive structuration theory (GAST) approach to predicting team performance. In other words, early conflict resolution statements set the boundaries and rules for later group discussions. This suggests that the long-term success or failure of a team is largely set in how they resolve their first major conflict(s).
back to top