1
What's Past Is Prologue:
Exploring a Biodata Approach to Team Selection
Michael J. Stevens
University of Missouri - St. Louis
8001 Natural Bridge Rd.
St. Louis, MO 63121
(314) 516-6297
Robert G. Jones
Donald L. Fischer
Southwest Missouri State University
Department of Psychology
901 South National Avenue
Springfield, MO 65804
(417) 836-4790
Address correspondence to:
Michael J. Stevens
University of Missouri - St. Louis
College of Business Administration, SSB 487
8001 Natural Bridge Rd.
St. Louis, MO 63121
(314) 516-6297
What's Past Is Prologue:
Exploring a Biodata Approach to Team Selection
Abstract
One of the original key assumptions underlying biodata inventories is that knowledge about past behavior can help us predict future behavior (Owens, 1976). This assumption should hold true for self-directed teamwork environments just as readily as it does in other areas of work. Consequently, a 64-item Teamwork Biodata Inventory (TBI) was developed to test this assumption, and was evaluated on field subjects for its validity in making teamwork staffing and selection decisions. Both team and individual performance were both predicted by some, but separately predicted by different TBI scales.
Introduction
Staffing teams requires consideration of both individual effectiveness regarding core task proficiencies, as well as "mix" variables associated with team coordination (Klimoski & Jones, 1995; Stevens & Campion, 1994; Jones, Stevens & Fischer, 2002). Although there is a growing body of research that addresses these two issues separately, there is little work which has simultaneously considered both team and individual effectiveness prediction together. This is an important issues since both team-level and task-specific components of performance may influence effectiveness of work teams and its members (Bannick, Salas & Prince, 1997). Further, while the current literature has looked at individual member characteristics and team composition as they affect processes and outcomes, most of this work done to date has been done in laboratories (Levine & Moreland, 1990), while team composition is generally treated using demographic variables of team members, rather than experience and background variables (see Barrick, Stewart, Neubert, & Mount, 1998 for a notable exception). By taking into consideration important background experiences of the individual team members, this study attempts to extend these previous research inquiries.
Simultaneous examination of both predictors of individual effectiveness within teams and of biodata predictors of team performance has not been accomplished. Still, theoretical work on relationships between individual member characteristics, team composition factors, and their combined relationships with processes and performance in group situations provides some useful directions for dealing with these questions. In particular, several theorists (Heslin, 1964; Jackson, May, & Whitney, 1995; Levine & Moreland, 1990) have suggested that team performance can be understood as a function of both member characteristics and the relative mix of these characteristics. Evidence has supported the idea, at the group level, that heterogeneity (usually defined in terms of demographics rather than experiential variables) facilitates performance in creative problem solving tasks and reduces performance in structured tasks (Heslin, 1964; Jackson, et al., 1995). Similarly, team processes may be facilitated by members’ general sense that the team is effective (Guzzo, 1986; Hyatt & Ruddy, 1997; Sniezek, 1992), though there is less evidence concerning this notion. Heterogeneity may also reduce group cohesiveness, though the exact nature of this relationship appears to be contingent on which aspect of cohesiveness is considered (Mullen & Copper, 1994) and what sort of task the group performs (Saavedra, Earley, & Van Dyne, 1993).
Empirically, and without reference to process and context moderation, there is some evidence of composition being related to team outcomes. For example, in two field studies, elevation and dispersion of several big five personality characteristics related to team process and performance (Wagner, Neuman, & Christiansen, 1996; Barrick et al., 1998). Similarly, combined member ability has been associated with performance (Colarelli & Boos, 1992; Tziner & Eden, 1985; Wright, McMahan, Smart, & McCormick, 1995) though the nature of this set of relationships is controversial (Hill, 1982; Watson, Michaelson, & Sharp, 1991; Brannick et al., 1998). Finally, Campion Medsker and Higgs (1993) found significant correlations with performance only for some composition factors. These mixed results of course suggest a need to evaluate moderation of composition effects on team outcomes, and return us to the problem of particularistic selection.
The interaction of individual performance with team contextual factors in predicting performance is also a potentially important issue we examine in this study. Exactly which individual characteristics may be related to which team outcomes has been less clearly delineated at the individual level (Barry & Stewart, 1997). Some have suggested that individual personality (e.g. Schlenker, Weigold & Hallam, 1990), interpersonal skills (e.g. Campion & Stevens, 1994), and preference for working in teams (Campion et al, 1993; Eby & Dobbins, 1995) may predict individual performance in teams, as well as team performance as a whole. However, the questions of which team processes interact with these individual level differences remains a matter for conjecture, since methods for statistical evaluation of interactive relationships between individual and group levels of analysis do not yet exist. However, individual performance within a team context can be assessed using traditional analytic tools.
In summary, relationships between individual member characteristics (on the one hand) and team composition and team outcomes (on the other hand) require further research. Where research has been done, it tends to have been carried out in laboratory settings with limited generalizability to the complex world of workplace teams. This study therefore uses a multi-level approach to examine relationships between individual team member background and experiences (e.g., personal motives; preferences for working teams; knowledge of and experience in team settings; age, gender, language, and ethnicity demographics), team process and context, and performance in intact teams in a field setting. In particular, we will explore the relationships (at both the individual and team levels of analysis) between predictors and effectiveness. The same individual characteristics aggregated for team level analyses will also be used to predict individual performance in the team context using multiple rating criteria. Moderating effects of process and context on team level predictor-performance relationships will also be explored.
Method
Sample and procedure
Data for this study were gathered from a large metal refinery in the Southwest U.S. in twelve, two-hour sessions during work time. Individual participants were 458 employees and their 57 managers. Employees had been organized into semi-autonomous teams for approximately six years, during which time the organization had experienced significant increases in production and profitability. Team members who reported ethnic information (67.8% of the total) were predominantly of Latino ethnicities (50.4%), with fewer European- (13.5%), African- (3.7%), and Asian-Americans (.2%). Of the 321 people reporting gender (70.3%), most were male (65.3% of total) and fewer female (5% of total). Education levels varied, with 38.9% having completed high school, 21.1% with some college, and 5.3% of the sample having completed college or education beyond college. Median age of the team members was approximately 42.
There were 57 teams, of which 56 provided more than one participant response. Teams varied in size from three to 16 members, with a mean of 8.24 members (S = 3.78). Size was used as a covariate in certain analyses. Teams performed a variety of tasks, from manual labor to technical tasks (e.g. electrical or auto maintenance), and 18 were formally classified by the plant as "self managing." All teams had team leaders (n = 57), and non-self-managing teams had team coaches as well. Only six teams had women leaders.
Measures
Teamwork Biodata Inventory. Items for the Teamwork Biodata Inventory (TBI) were written based primarily on the content domain of individual level teamwork knowledge, skills and abilities for teamwork presented by Stevens and Campion (1994). This domain consists of competencies in the areas of interpersonal effectiveness (i.e., conflict resolution, collaborative problem solving, and interpersonal communication) and self-management capabilities (i.e., goal setting and performance management, and planning and task coordination). Items were written to capture aspects of subjects' personal background and experience that would be strong indicators of successful involvement with teamwork (or at least successful involvement with the interpersonal and self-management aspects that are known to be indicative of successful teamwork).
Beyond the content domain of the Stevens and Campion framework, additional biodata items were also written based on the expectation that other content areas might also provide a valuable source of prediction for the TBI. These additional areas included such things as group attitudinal constructs, cultural values and personal preferences (or predispositions) for group work. Thus, after a review of the relevant literatures, the following specific content areas were identified as a basis for potential TBI items: self-monitoring (Gangestead & Snyder, 1985, 1991; Miller & Thayer, 1989); collectivism versus individualism (Earley, 1989; Triandis, Kurowski, & Gelfand, 1994); power distance (Hofstede, 1980a, 1980b); need for recognition and individual achievement (Triandis, 1988, 1990); need for affiliation (McClelland, 1985); sociability and extroversion (McAllister, 1996); organizational citizenship behaviors (Organ & Ryan, 1995) and current time in one’s team.
After content areas were identified, a pool of potential items was written to reflect subjects' relevant historical experiences with the constructs of interest. Pilot administration and refinement of items resulted in the final version of the TBI, consisting of 64 self-report questionnaire items (see Appendix A for sample items). After appropriate coding, items were analyzed using components analysis. Various numbers of factors were considered based on different decision criteria, and several rotation methods were applied to each. In the end, the most interpretable solution included four factors and used a promax rotation. Standardized composites were formed using items loading over .30 on a single factor. Based on item content, the TBI composites were named Social Facilitation (α = .81), Sociability (α = .69), Social Pessimism (α = .71), and Independent Achievement (α = .55).
Other Individual Differences Measures. In order to test for convergent and discriminant validities, subjects also completed measures on mental abilities and teamwork KSAs. Demographic data were also gathered for purposes of prediction. For team level analyses, variability and mean levels on composites were used. Minimum and maximum levels were also used in composition analyses. Specifically, mental abilities measures included the NCS/London House (1986) nonverbal, linguistic, and quantitative tests. The Teamwork-KSA Test (Stevens & Campion, 1994) was used to evaluate team members’ understanding of the workings of teams. Variability (homogeneity/ heterogeneity) in age and ethnicity were also used in the analyses.
Moderators. Team context and process were measured using surveys of team members. These surveys included 60 items derived from literature on team processes, including potency (7 items), work assignment equity (9 items), cohesiveness (14 items), and trust/communication (6 items). The extent and character of role definition was assessed using 16 items. All items used a 5-point ("strongly agree" to "strongly disagree") scale and referred to characteristics of the team as a whole.
Principle factor analysis of the team context and process survey yielded a two-factor solution. Squared multiple correlations were used as prior communality estimates and a discontinuity rule was used to select the number of factors (38.6 % of variance accounted for). A varimax solution showed a fairly clear delineation between process and role definition items. Two items were factorially complex (similar loadings on both factors), and were assigned to scales based on logical relationships to process versus role-definition scales. Internal consistencies for resulting composite scales were .95 (38 items) and .93 (22 items) for the process components and role-defining context components, respectively. Aggregation to the team level was also justified by generally high rwg values (M = .83 for process and M = .79 for role definition context).
Team Performance. Supervisors provided ratings of team performance across thirteen dimensions using a 5-point scale ("well above average" to "well below average"). Supervisor ratings on the thirteen team performance items were also submitted to factor analysis. Following the same decision steps as in the team context and survey, supervisor ratings yielded a two factor solution (65% of variance accounted for). Although rotation left one item loading highly on both factors, it was assigned to the conceptually similar scale. The two factors corresponded with effectiveness (9 items) and safety-related (4 items) issues and were formed into composites (alpha = .92 and .84 respectively).
All supervisors familiar with a given team's performance provided ratings for the team. All teams had at least two independent performance ratings, while as many as eight teams had ratings across five supervisors. Correlations among supervisor evaluations on the two composites (effectiveness and safety) were generally high, and alphas of .82 and .56 (respectively) were derived from the eight teams that were rated by five supervisors.
Individual Team Member Performance. Supervisor, peer, and self ratings of all individual team members were collected. The evaluations asked raters to score individuals on dimensions of both taskwork (6-items) and teamwork (6-items) performance behaviors using a 5-point scale ("well above average" to "well below average"). Although reliability was not directly estimable for these ratings, correlations between different sources suggested that there was significant systematic variance in each. This will be discussed further in the results section.
Results
Team level prediction
Supervisor ratings of team effectiveness (M = 3.78) and safety (M = 4.11) and team ratings of process and role definition were significantly related to several ability-related predictor measures. Correlations among representative team ability composition variables are presented in Table 1. These indicate that supervisor ratings of team performance were significantly predicted by team levels on mental abilities measures and heterogeneity on the Teamwork-KSA measures. Higher levels of mental ability were correlated with higher supervisor ratings of team performance, whereas variability on teamwork KSAs was correlated positively with supervisor ratings. Team processes were not predicted by abilities.
Different team level patterns emerged with regard to demographic and biodata predictors. First, supervisor performance ratings did not correlate significantly with any demographic predictors. However, team process ratings did relate to average age of group members (r = .27, p < .05) and with average level of education (r = -.30, p < .05). Group size also correlated inversely with process (r = -.37, p < .01) and role definition context (r = -.35, p < .01). Second, biodata scores and variability indicators correlated significantly with supervisor ratings as well as with team process and role definition (see Table 2). These patterns showed that variability in social facilitation within teams negatively correlated with supervisor productivity ratings. Mean, minimum, and maximum levels of social facilitation correlated positively with supervisor ratings and team role definition, as well. Levels of social pessimism also related negatively to supervisor ratings and process and role definition indicators. Mean and maximum levels of sociability correlated positively with team process and role definition measures.
Stepwise regression analysis using supervisor ratings as criterion and composition effects (other than demographics) as predictors was used to select predictors for moderated regression. Team composition variables that accounted for significant unique variability in supervisor ratings included levels of non-verbal mental ability and social facilitation, as well as teamwork KSA variability. These were entered into individual regressions using interaction terms with team process and role definition ratings as moderator terms (see Table 3).
Several moderated relationships emerged and were examined using median splits on predictor variables. First, teams with lower levels of team non-verbal mental ability were much likely to have higher performance ratings in the presence of higher perceived role definition. Performance of teams with higher mean non-verbal ability was unrelated to role definition. Similarly, variability in team member knowledge about teamwork had less of a relationship with performance when role definition was higher. Team process effectiveness had a very small moderating relationship with non-verbal mental abilities; in general when processes were better, so were safety ratings.
Individual level prediction
Correlations among individual level predictors and performance ratings by self, supervisor, and peer are presented in Table 4. These indicate that supervisor ratings were better predicted by mental abilities than by biodata, overall. Peer and self ratings were predicted more evenly by both types of individual difference measure. However, item-by-item correlation of biodata dimensions showed self and peer ratings significantly predicted by biodata, but not mental ability measures. Regression of criterion ratings on biodata dimensions yielded significant unique contributions to peer ratings (individual achievement entered, R = .27, p < .01) and self ratings (individual achievement and social pessimism entered, R = .28, p < .01).
Discussion
This study examined the relationship between performance at the individual and team levels of analysis and individual background differences measured via a teamwork biodata inventory (as well as teamwork KSAs, mental abilities and demographic differences). As with other types of work environments, the previous experiences of the subjects with team-related content domain items were helpful identifying those individuals who were more effective at contributing within a team environment. However, the patterns of predictions did vary across the different performance criteria that were examined. Specifically, team heterogeneity on social facilitation and teamwork KSAs predicted team performance ratings significantly. In addition, mean team levels of individual mental abilities, social facilitation and social pessimism also related to team performance. Interestingly, teamwork KSA levels predicted supervisory ratings of individual performance, while variability in teamwork KSAs predicted team performance.