Team member performance. We measured team member performance using four items from Welbourne et al.’s (1998) measure of task
178 PERSONNEL PSYCHOLOGY
performance. The team leader rated each focal team member’s contri- bution to the performance of the team on a 7-point scale ranging from 1 = needs much improvement to 7 = excellent including “quantity of work output” and “quality of work output.”
Team performance. We measured team performance using six items from Kirkman and Rosen (1999) designed to assess key team performance indicators. Team leaders rated the extent to which they disagreed or agreed that their team was effective in terms of each performance indicator, including: “Meets or exceeds its goals,” and “Completes its tasks on time.”
Control variables. Several variables of theoretical relevance to the dependent variables were explored as potential controls at both the indi- vidual (gender, age, team tenure, and number of other teams on which the team member simultaneously participated) and team level (task in- terdependence, team size, and extent of face-to-face interaction). Follow- ing recommendations and past research (e.g., Bono & McNamara, 2011; Erdogan & Bauer, 2009; Hu & Liden, 2013; Seibert, Kraimer, Holtom, & Pierotti, 2013; Shoss, Eisenberger, Restubog, & Zagenczyk, 2013), in order to preserve degrees of freedom, only those that were significantly correlated with any of the study variables were carried forward in the analysis. At the individual level, this led us to include team member vir- tual teamwork experience as a control in the analysis, because there is research to show that experience can positively impact job performance (McDaniel, Schmidt, & Hunter, 1988). The focal team member reported on different aspects of his/her virtual teamwork experience on a 7-point Likert scale ranging from 1 = no previous experience to 7 = significant amount of experience. At the team level, the type of team (i.e., global com- modity team or process improvement team) and team technology support were included as controls. For team type, commodity team was coded as 0, and process improvement team was coded as 1. Team technology support refers to the extent to which the team as a whole has adequate access to technology tools required to support distributed collaboration among team members (Kirkman, Rosen, Tesluk, & Gibson, 2006). Inad- equate technology support can be an impediment to virtual collaboration (e.g., Kirkman et al., 2006). Team technology support was measured with three items from Kirkman et al.’s (2006) measure of technology support. Because of space limitations on the team member surveys, the team leader reported on the team’s level of technology support.
Due to the nested structure of the data (i.e., individuals nested within teams), we used hierarchical linear modeling (HLM: Raudenbush &
HILL AND BARTOL 179
TABLE 1 Individual-Level Descriptive Statistics and Correlationsa
Variable M SD 1 2 3 4
1. Virtual teamwork experience 5.74 1.21 (.90) 2. Virtual teamwork situational judgment 16.39 3.31 .06 (.62) 3. Team member virtual collaboration 5.47 0.70 .20∗ .04 (.98) 4. Team member performance 5.29 1.16 .13 .18∗ .43∗∗ (.97)
Note. aReliabilities are shown on the diagonal (n = 193 individuals). ∗∗p < .01; ∗p < .05
Bryk, 2002) to test the individual-level and cross-level (Hypotheses 1) relationships in the model. HLM is a statistical approach that provides a more appropriate estimate of standard errors than other analytic meth- ods when data are nested in teams, and assumptions of independence, therefore, are not warranted. We used hierarchical ordinary least squares regression with mean-centered predictor variables to test the predicted team-level interaction effect (Hypotheses 3). Although testing the rela- tionships in the model separately can provide preliminary evidence of moderated mediation (Hypotheses 2 and 4), researchers have noted the limitations of using a stepwise approach to testing moderated mediation (Edwards & Lambert, 2007; Preacher et al., 2007). Therefore, we tested the moderated mediation effects in the model in a more integrative fashion by using the Monte Carlo method (Mackinnon, Lockwood, & Williams, 2004) to obtain estimates for the size of the indirect effects at different levels of the moderator, including confidence intervals for the indirect effect.
Table 1 and Table 2 show descriptive statistics and bivariate corre- lations for all study variables. These correlations do not account for the non-independent nature of the data at the individual level and should be viewed with caution until properly modeled using HLM.
Hypotheses Predicting Team Member Virtual Collaboration and Performance
To confirm that the use of HLM was appropriate for testing influ- ences on team member virtual collaboration and performance, we first ran an HLM null model for virtual collaboration and team member perfor- mance. The resultant ICC(1) value reflects the percent of variable variance residing between teams. If the team-level variance is significant, then the use of HLM is considered to be warranted. The ICC(1) values and
180 PERSONNEL PSYCHOLOGY
TABLE 2 Team-Level Descriptive Statistics and Correlationsa
Variable M SD 1 2 3 4 5 6
1. Team typeb 0.41 0.50 – 2. Team technology support 5.15 1.17 .02 (.63) 3. Empowering team
leadership 5.57 0.47 .22 .35 (.95)
4. Aggregate virtual collaboration
5.45 0.45 .44∗ .40∗ .53∗∗ –
5. Team performance 5.65 0.99 .29 .16 .32 .65∗∗ (.95) 6. Team geographic
dispersion 0.00 2.75 −.05 .05 −.03 −.35 −.38∗ (.80)
Note. aReliabilities are shown on the diagonal (n = 29 teams). b0 = Commodity team; 1 = Process improvement team ∗∗p < .01; ∗p < .05
associated significance tests showed significant between-group variance for team member virtual collaboration (26.05%, τ 00 = .13, p < .000) and team member performance (23.91%, τ 00 = .34, p < .001). Hence, we pro- ceeded with the use of HLM for the analyses predicting these outcomes. The results presented below include all control variables discussed ear- lier that were significantly correlated to the variables in the model (Bono & McNamara, 2011). Some of these controls were ultimately not sig- nificant predictors in the models analyzed to test the model hypotheses. We also ran all analyses without these controls included (Hu & Liden, 2013; Sluss, Ployhart, Cobb, & Ashforth, 2012), and the results remained the same.
Hypothesis 1 predicted that empowering team leadership has a cross- level moderating influence on the relationship between a team member’s VT-SJ and team member virtual collaboration. We tested this relationship by first entering the level 1 control and VT-SJ as predictors in the level 1 equation and both the level 2 controls as well as empowering leadership as level 2 predictors of the level 1 intercept. For the interaction between em- powering leadership and VT-SJ, we entered empowerin