give feedback on this paper on its writing/spelling grammar/study design/formatting/readability/ etc

Study 2: Inclusive Culture Reduces Dominance And Increases Relational Leadership

in Coded Speech Behavior

The goal of Study 2 was to examine the effect of culture on women's and men's leadership approach using a more behavioral measure of dominance as opposed to self-reports. Participants were thus asked to record a video of themselves as leaders in a typical workplace situation, which independent coders then rated for different types of leadership behaviors.

Adding evidence to the findings of Studies 1a and 1b, we tested whether an equity-inclusive (vs. competitive-meritocratic) culture allows women and men to use less dominant leadership behaviors when leading others. Testing the underlying mechanism, we again examined whether such a culture relates to weaker perceived masculine defaults and subsequently releases pressures to use dominant behaviors as well as women's otherwise constrained career aspirations without negatively affecting men's. Study 2 tested these ideas with preregistered hypotheses and analyses (seehttps://osf.io/96vgz/?view_only=0a5d4416257f428da7cfd41c39de8dcf for the data collection and analysis, andhttps://osf.io/8ytbj/?view_only=71a98e44e14e49b58c7c629bffbc3f8e for the behavioral coding scheme and training of coders).

New to this study, we explored the effect of the culture on additional leadership behaviors. Specifically, we examined the preregistered secondary hypotheses that an equity-inclusive (vs. competitive-meritocratic) culture motivates people to lead with more relational leadership behaviors. Such leadership behaviors, which includes leading people-oriented and power-sharing, are increasingly important in organizations as it was suggested to "have positive effects on the attitudes and (ethical) conduct of employees and ultimately even on business unit or organizational performance"(see Kalshoven et al., 2011, p. 51).

Methods

Participants

The preregistereda prioripower analysis indicated the need for at least 387 participants to have 80% power to detect small to medium effect sizes (n²_p= .020,f = 0.14) for the two main effects and interaction in an ANOVA when employing the traditional .05 alpha-criterion of statistical significance (G*Power3.1; Faul et al., 2007). We thus preregistered a sample size of 450 to account for participants who would fail our manipulation check or encounter technical problems with the video recordings. Participants were recruited through Prolific. Of the 463 participants initially recruited, we excluded (as preregistered) participants who experienced technical problems with the video recordings (e.g., no audio,n = 19), self-reported that they responded randomly during the study (n = 6), who failed our free-response manipulation check of culture (n = 15), who admitted their participation is invalid (n = 2), or who did not self-identify as either women or men (n = 4).

The final sample included 417 adult STEM employees (206 women, 211 men). Their age ranged from 18 to 72 years (M = 36.09 years,SD = 10.71). Participants most frequently reported employment in the computer sciences (45%), engineering (20%), biological sciences (9%), physical sciences (5%), mathematical/statistical sciences (4%), indicated another domain (15%), or chose not to indicate which domain (1%). The majority of 71% were White, 12% Asian, 8% African American, and 3% Hispanic. All participants were compensated US$3.75 for the online survey, and the best performing 20% received an additional US$ 6.30 bonus payment after the end of the video coding.

Design and Procedure

The experiment was again a 2 (Inclusiveness of Culture: Competitive-meritocratic vs. Equity-inclusive) 2 (Participant Gender: woman, man) between-subjects design. Participants were randomly assigned to one of the two experimental conditions.

The study was conducted online. Participants were informed that the study examined how people lead teams working remotely. After giving informed consent, each participant learned that they "are working at CCB, a large tech company" and received the same background information about the company as in Studies 1a and 1b. Participants then recorded a 30 sec video to introduce themselves to allow "folks to learn more about their colleagues, who they might not meet in person" when working remotely, used to code for baseline dominance (and prestige). As in Study 1b, participants then received an executive summary of a review of CCB's workforce policies and practices, which always detailed the same mix of gender-inclusive and non-gender related policies.

To manipulate the inclusiveness of the culture, each participants saw "short messages so you can learn more about CCB's work environment and what it's like to work there," allegedly from three current employees: Justin (from the tech development department), Heather (a software engineer), and Michael (from the business development department). These employees were cited saying that CCB values the contributions of either "those individuals who most help us succeed" (competitive-meritocratic) or "everyone equally, regardless of their gender or background" (equity-inclusive), that interactions between employees from either "different office locations are characterized by a success-orientation" (competitive-meritocratic) or "different backgrounds are characterized by respect and trust" (equity-inclusive), and that CCB's management has a clear commitment to "a strong work ethic" (competitive-meritocratic) or "an inclusive workplace" (equity-inclusive). As in the prior studies, participants then summarized their impression of the work culture in their own words (used as a manipulation check).

Following past research(Engstrom et al., under review), participants then recorded a 2-3 minute video where they explained how they would "organize the way online meetings are run to ensure they are efficient." Participants learned that "in communicating your strategy and ideas, you will want to address these questions: (1) How should the goals for each meeting be set and communicated?; (2) How do new ideas / questions get communicated and vetted?; and (3) How should meetings be organized in terms of speakers so that work gets done?" Aimed at increasing their engagement and motivation, participants were reminded that their "strategy will be evaluated by the 3 employees you just read about. The top 20% performing leaders will receive an additional 5 [US$6.40] via the Prolific bonus system."

Measures

Consistent with the measures used in the prior studies, participants then completed the questionnaires for dominant leadership approach, = .82, prestige leadership approach, = .75, perceived masculine defaults, = .82, the two ladders for constrained career aspirations, and the manipulation check item used in Study 1b.

Relational Leadership Approach. Additionally, participants indicated their use of relational leadership. After the dominance-prestige questions, they responded to eight items from the ethical leadership at work questionnaire(ELW; Kalshoven et al., 2011), with four items assessingpeople orientation (e.g., "I would pay attention to others' personal needs") and four items power sharing(e.g., "I would reconsider decisions on the basis of recommendations by those who report to me"), on 7-point scales (1 = not at all to, 3 = somewhat, 7 = very much). Given that all eight items were significantly correlated and exploratory factor analysis with oblimin rotation yielded a one-factor solution (with loadings from .60 to .71), we created one relational leadership scale, = .81. Appendix D1 in the Supplemental Material displays the verbatim study questionnaire.

Human-Coded Leadership Behavior. Unique to Study 2, three coders from a team of seven coders rated each video. All coders were blind to condition and hypotheses. Following recommendations for coding open-ended data(Goodell et al., 2016), the training of the research assistants included reading articles describing dominance, prestige, and ethical-relational leadership behavior, writing definitions of all constructs in their own words, discussing these with the first author, and coding practice videos together with the first author. Each research assistant then coded 20 baseline and 20 main videos that had already been consensus-coded by the first author and two senior research assistants. Any deviations of more than 1 point from these benchmark codes were discussed.

We coded the main videos for dominance (1=not at all dominant, 3=somewhat dominant, 5=very dominant), prestige (1=not at all prestigious, 3=somewhat prestigious, 5=very prestigious), people orientation (1=doesn't care about people, 5=very people-oriented;), and power sharing (1=completely takes power, 5=completely shares power). Appendix D2 in the Supplemental Material includes examples illustrating which statements were coded as high versus low on these leadership approaches.

The codings for all four constructs showed good or excellent inter-rater agreement (ICC_Dominance= 0.63;ICC_Prestige= 0.61;ICC_{People Orientation}= 0.76;ICC_{Power Sharing}= 0.72), according to the often-quoted guidelines for interpreting ICC by Cicchetti(1994), which suggest fair reliability for values from 0.40 to 0.59, good reliability for 0.60 to 0.74, and excellent reliability for 0.75 to 1.00. Consequently, we ran the analyses based on the three coders' ratings for each video (despite having preregistered initially to add new coders if the ICC falls below .75 for a construct). Parallel to the self-reported relational leadership, we again created one relational leadership approach scale, ICC_Relational= 82.

Using participants' short introductory videos, we coded for baseline dominance and prestige, with fair inter-rater agreement (ICC_{BaselineDominance}= .45;ICC_{BaselinePrestige}= .47;Cicchetti, 1994). Given the difficulties coding these 30 second videos, we present the analyses that controlled for these baseline values (as preregistred) but also ran an additional set of analysis that did not include this covariate, showing that all pattern of results remained the same.

Machine-Learning-Coded Speech (Exploratory).To further identify and quantify leaders' speech behavior, we applied structural topic machine learning models to their speech transcripts. This method reveals topics or themes within text data by statistically identifying words that tend to appear near versus far away from each other. This bottom-up approach is akin to exploratory factor analysis in that researchers do not specify content or topics in advance. It therefore offers an independent and complementary approach to our top-down, human-coding method by unveiling what topics emerge spontaneously as prominent themes in participants' verbal behavior.

Manipulation Check for Culture Manipulation.As in Study 1b, participants indicated "How inclusive does the environment and culture at CCB feel to you?" on a 7-point scale (1 = not at all to, 3 = somewhat, 7 = very much).

Agentic and Communal Aspects of the Culture.The last part of the survey asked about CCB's focus on different aspects of the culture to better understand which aspects of the culture may drive the effects. Participants indicated on six items how much CCB culture includes agentic aspects (e.g., "being competitive and going the extra mile") and communal aspects (e.g., "showing respect and trust"). An exploratory factor analysis with oblimin rotation yielded a two-factor solution, with the first factor reflecting communal aspects (with loadings from .85 to .94) and the other agentic aspects (with loadings from .71 to .84).

Results

Manipulation Check

The culture manipulation changed participants' perception of how inclusive CCB would be. As in Study 1b, the 2 (Inclusiveness of Culture) 2 (Participant Gender) ANOVA on how inclusive the culture feels showed the expected main effect of Condition,F(1, 415) = 89.49,p < .001,_p² = .18, indicating that participants perceived the culture to be more inclusive in the equity-inclusion (M = 6.14,SD= 1.13) than the competitive-meritocracy (M = 4.92,SD= 1.49) condition.

New to this study, results further yielded a significant main effect of Gender,F(1, 413) = 5.87,p = .016,_p² = .01, and a significant Condition Gender interaction,F(1, 413) = 9.19,p = .003,_p² = .02, indicating that women perceived the competitive-meritocratic culture to be less inclusive than men did,t(413) = -3.82, p < .001, whereas no such difference existed in the equity-inclusive condition,t(413) = 0.44, p < .001. These results suggest that gender differences exist in particular for the competitive-meritocratic culture, with mainly women perceiving it to be low inclusive. Although generative for future research, this finding did not affect the hypotheses being tested given that the manipulation was perceived as intended.

Coded and Self-Reported Leadership

The preregistered analysis plan was largely identical to the one for Studies 1a and 1b. New to Study 2, and parallel to the analysis plan of the prior studies, the ANCOVA on relational leadership included dominance as a covariate (although all results hold without the covariate in the models). The subsequent sections present the results for the different leadership behavior, first as coded by our raters and then as self-reported by participants. Results then focus on perceived masculine defaults and constrained career aspirations. Table 3 shows the intercorrelations between the variables and Table 4 the means and standard deviations.