Introduction

In a recent study presented by Mellor (2023), the chronic problem of low local union meeting attendance was viewed through a psychological lens. Through this lens, it was suggested that meetings provide employees with opportunities to experience safe environments, wherein they are encouraged to fulfill psychological needs related to work through interactions with other employees, examples of which include the need to voice and share concerns about the work environment, to participate in decisions about how work is parsed, distributed, and performed, to clarify and refine how work relationships are to be shaped and honored—and if not, how they are to be addressed and dealt with—and to be counted as a valued contributor to “our collective work effort” (see Ezorsky, 2007; Greenhouse, 2019; Kochan, 2005; McAlevey, 2020; Mellor & Holzer, 2018; Parker & Gruelle, 1999 for narrative accounts and like perspectives).

In the Mellor (2023) study, safe environments were defined in reference to a psychological construct developed by Edmondson (1999) to predict team outcomes such as “efficacy, creativity, and performance.” The construct, psychological safety, describes a shared belief by team members that meetings are safe for interpersonal risk-taking, a definition that is meant to suggest a sense of confidence held by members that the team will not embarrass, reject, or punish any member for speaking up. This shared belief is thought to emerge from the interactive experience of team members—interactions in which members encourage each other to engage in conversational turn-taking, to speak roughly in the same proportion, to express and be open to a diversity of ideas and perspectives even if they challenge team norms, and to exchange personal information intended to show others what feelings are in play. It is from these kinds of interactions that climate properties are thought to also emerge such as mutual trust and respect—properties that provide a safe environment for members to “speak out” and to explore new ideas and solutions aimed at team goals (“to learn from each other”; Edmondson, 2018).

As seen in the team literature on psychological safety, and in the widely circulated intervention studies conducted at Google, the status of safety as a predictor of team outcomes underwent a conceptual and positional shift. Conceived initially as a direct predictor of outcomes, model results consistently showed better model fit with safety positioned as an indirect predictor—that, for example, safety was linked to well-attended meetings through ratings of meeting effectiveness by team members (i.e., safety was mediated by rated effectiveness; see Edmondson & Bransby, 2023; Edmondson & Lei, 2014 for reviewed safety studies; see Duhigg, 2016; Rozovsky, 2015 for reviewed Google studies; see Gavin et al., 2008; Nembhard & Edmondson, 2006; Tucker et al., 2007; Understand team effectiveness, 2017 for mediated safety studies).

As such, when psychological safety was adapted as a predictor of local meeting attendance in the Mellor (2023) study, the hypothesized model positioned meeting effectiveness as rated by local members as a mediator. In tests of the mediation model using a broad sample of employees who attended local meetings in the last 12 months, and controlling for sample-specific member and local demographics (e.g., membership years, meeting size), and controlling for non-psychological safety at meetings (e.g., “meetings wherein members did their best to begin and end meetings on-time as scheduled”), the model was shown to fit the data, in which employees who experienced more safety at meetings were more likely to indicate meeting attendance in the next 12 months, an effect that was seen to unfold through (be mediated by) rated meeting effectiveness.

Present Study

The aim of the present study is to replicate the mediation model presented in the Mellor (2023) study with a new sample of employees who attend local meetings, and to expand the model in reference to: (a) demographic results reported in the study and (b) discussions with local officers about the demographic results. In the Mellor study, gender, minority status, and language of origin were assessed as demographics and, as non-model variables, were positioned in model tests as covariates. However, as reported in the study, tests for mean differences and corresponding zero-order correlations for model variables in relation to the demographics indicated that, on average, women employees (vs. men employees) were less likely to experience psychological safety at meetings, a difference also seen for employees for which English is a Second Language (vs. employees for which English is a First Language), with the added note that minority status and language of origin were positively correlated, in which minority employees (vs. nonminority employees) were more likely to be English as a Second Language employees. Also, as reported, on average, women employees were less likely to indicate higher meeting effectiveness and less likely to indicate meeting attendance in the next 12 months.

Reactions to Mellor (2023)

Following print publication, the Mellor (2023) study received wide circulation, with open access downloads exceeding 1500. As an online article, the study also generated a steady stream of email exchanges with labor statisticians and social science researchers who contribute studies on diversity and inclusion asking for data access, inquiring about plans for model replication, and issuing invitations to talk about study results in virtual symposia and at in-person research conferences (e.g., Levey et al., 2023). Least surprising due to the persistence of low local meeting attendance (see Mellor, 2023 for current estimates as low as 3% of eligible employees), the study also generated email exchanges and invitations with online bloggers who focus on union issues and from union members serving as local officers (representatives known as “local reps”) asking for virtual- and face-time to discuss study results, with an eye less on mediation per se and more on development and implementation of interventions to reverse low meeting attendance, especially programs that might be directed at women, minority, and English as a Second Language employees.

Viewing local reps as key informants on the problem of low meeting attendance, we held neutral-site discussions with six reps (three women, three men, two of whom were “non-native” English speakers). Queued by their discussion points, we asked about their slanted interest in the meeting attendance of women, minority, and English as a Second Language employees, and why their points on intervention were pitched toward these employees. Herewith, is a composite of the answers we received with quotations paraphrased and edited for clarity, answers that summarily represent our derived understanding of why meeting attendance as viewed in the Mellor (2023) study had garnered their attention and why the demographic results in particular had augmented their interest.

Input From Local Reps

What we were led to understand . . . is that the business of local reps is to represent the voice of members about all aspects of work to employer representatives (supervisors, managers, “heads”). As a conduit of this voice, reps explained that a certain kind of credibility is required, the premise being that reception rests on the extent to which the voice is recognized as a collective voice. Without the credibility of a collective voice, the voice may be heard as the voice of the few (“there are always disgruntled employees”) rather than the voice of the many (“the working backbone of the company”). Assumed is that employer representatives focus on the power of the many, in which monitored as a “tell” is attendance at local meetings, where it is correctly assumed that the collective voice is forged. In terms of the power of the collective, if few attend, the voice is perceived as weak and perhaps negligible; if many attend, the voice is perceived as neither weak nor negligible. According to reps, especially monitored are identifiable groups of employees who may not attend meetings, the counter-premise being that they amount to an assumed voice of disinterest or dissention, as for example, “yes, we might expect that point of view from ‘the kind of members’ who frequent your meetings, but what about the [fill-in-the-blank group members] who do not?”.

As for their interest in the reported demographic results, the local reps indicated that they bear on easily identifiable groups of employees who, based on their gender, minority status, or language of origin, can be set off against each other at the cost of attending meetings, and in support of the counter-premise, that the collective voice forged at meetings is negligible. With surprising candor, reps conveyed to us that their efforts to “regroup” fractionalized groups of employees who tend not to attend meetings or, when attending, to encourage their participation, had yielded mixed results. When asked why, the consensus view centered on “invasive cultural norms” about how men and women, nonminority and minority peoples, and peoples who are non-native English speakers and native English speakers are expected to interact—norms that overreach into the work environment and, when trying to encourage employees who feel “marginalized” to meet as a collective and to interact freely, can seem “impossible to eradicate.”

When we probed local reps further about why they thought that the idea of psychological safety might be well-suited for interventions to reverse low attendance, we received one connected response from reps, viz., “it fits in with what we’re trying to do in our local,” “it applies to all of our members in a very important way,” and “it gets right at the heart of what we try to do at every meeting—and that is to get everyone’s input on the table no matter who or what or what else is going on.”

With our eye on how the mediation model might be expanded to include the interactive effects of gender, minority status, and language of origin, we also asked the local reps to draw on their experience with women, minority, and English as a Second Language employees who tended not to attend meetings, and where, with respects to these employees, relationships in the model might be especially weak or especially strong (i.e., whether direct effects might be better conceived as conditional effects). The answers we received were sharply uniform, with one rep saying, “look, it’s simple—if people feel that they’re a part of what’s being said, they’re going to rate meetings as productive, and if they feel that they’re excluded from what’s being said, they’re going to rate meetings as anything but productive, and asking them to come to meetings or asking them to speak up is just a big waste of time.”

Model and Hypotheses

Given the reaction to the mediation model in reference to the chronic problem of low local meeting attendance, it seems obvious that model replication with a new sample of employees who attend meetings is warranted. Also, given the interest in the reported demographic results associated with the model with intervention in mind, it seems equally obvious that an expanded model is warranted with gender, minority status, and language of origin positioned as moderators.

Thus, we constructed a hypothesized model that expanded the mediation model (see Fig. 1), in which we advanced model-based hypotheses, noting that the first hypothesis (Hypothesis 1) is a replication hypothesis.

  • Hypothesis 1: In reference to local union meetings, employees who experience more psychological safety at meetings are more likely to indicate meeting attendance in the next 12 months, but the relationship between psychological safety and meeting attendance is expected to unfold through (is mediated by) meeting effectiveness, such that safety is associated with higher rated meeting effectiveness, which in turn is associated with more attendance.

  • Hypothesis 2: The mediation as stated in Hypothesis 1 (psychological safety→meeting effectiveness→meeting attendance) is expected to be conditional on gender, minority status, and language of origin, such that the relationship between psychological safety and meeting effectiveness is stronger for women employees (vs. men employees), for minority employees (vs. nonminority employees), and for employees for which English is a Second Language (vs. employees for which English is a First Language).

  • Hypothesis 3: The mediation as stated in Hypothesis 1, is expected to show moderated mediation, such that the relationship between psychological safety and meeting attendance in the next 12 months through meeting effectiveness is stronger for women employees, for minority employees, and employees for which English is a Second Language.

Fig. 1
figure 1

Hypothesized model

Full size image

Method

Procedure

Anonymous survey data were collected from American employees beginning in the first week of September 2022 and ending in the last week of April 2023. Survey sites were community gatherings and public transportation areas (e.g., seasonal farmers’ markets, church-sponsored bazaars, municipal airport wait areas, commuter rail and bus stations).Footnote 1

With permission obtained at each site, the researchers circulated flyers with the following information:

“Can you volunteer to take this survey? You can if you are employed in the United States and not a full-time student. The survey is anonymous—no names. The survey takes less than 10 minutes to complete. The survey cannot be mailed. $5 is given for taking the survey. Please ask the researcher for a survey.”

Employees who responded to the flyer were given a pencil, a no-name informed consent form, a survey, and an unmarked envelope. The researchers collected surveys in sealed envelopes, paid participants, and conducted onsite debriefing.Footnote 2

Sampling

To ensure that sampling resulted in data appropriate to test the hypothesized model, the survey was embedded with eligibility items. We excluded surveys from the sample in which responses suggested: (a) noncurrent union membership and (b) nonattendance of at least one regular scheduled local meeting in-person or online in the last 12 months. An additional check for careless responses resulted in excluded surveys if responses indicated the same scale anchor for long strings of consecutive items (e.g., a page of items).

One-hundred and thirty surveys with no missing data were counted as eligible. Surveys were collected from employees working in 22 U.S. States (California, Colorado, Connecticut, Florida, Georgia, Kansas, Kentucky, Louisiana, Maryland, Massachusetts, Michigan, New Jersey, New York, North Carolina, Ohio, Oklahoma, Pennsylvania, Rhode Island, Tennessee, Vermont, Virginia, Washington), and the District of Columbia. Employees held memberships in 22 unions and 64 locals. All locals were affiliated with one of three national labor coalitions (the American Federation of Labor and Congress of Industrial Organizations, the Change to Win Federation, the Strategic Organizing Center).

MeasuresFootnote 3

Demographics

Assessed demographics included age (indicated in years), gender (coded as either 0 (man; write-in responses: hombre, male, man, nonm) or 1 (woman; write-in responses: fanm, female, mujer, woman)); minority status (coded as either 0 (nonminority; write-in responses: Caucasian, European, White) or 1 (minority; write-in responses: African, Asian, Black, El Salvadorian, Filipino, Haitian, Hispanic, Indian, Latino, Puerto Rican, Spanish); language of origin, in reference to “your first language”, response options: “English” or “English is my second language” (coded as either 0 (English as a First Language) or 1 (English as a Second Language)); socioeconomic status, in reference to “education level (highest degree), contribution to family income, and occupational job status” (numeric response options: 1 (lower class), 2 (lower middle class), 3 (middle class), 4 (upper middle class), 5 (upper class); and employment status, in reference to “number of hours you work per week” (coded as either 0 (part-time; < 35 h) or 1 (full-time; ≥ 35 h)).

Ninety-seven percent of employees were age 25 years or older (the median age was 46; the range in years was 19 to 76). Fifty-five percent were women employees. Forty-four percent identified themselves as a minority employee. Fourteen percent identified themselves as an English as a Second Language employee. Fifty-five percent identified themselves as middle class or lower (the median class was middle class; no one identified themselves as upper class). Eighty-eight percent were full-time employees.

Assessed demographics specific to locals included length of membership (indicated in years); officer status in reference to “currently serving as a local union officer”, response options: “no, member only” or “yes, this is my union title” (coded as either 0 (member) or 1 (officer)); meeting size (response options: 1 (less than 25 members), 2 (somewhere between 25 and 50 members), 3 (somewhere between 50 and 100 members), 4 (over 100 members); and meeting attendance in the last 12 months (calculated as the percent of regular scheduled meetings attended in-person or online).Footnote 4

The average length of membership was 13.42 years (the median length was 11 years; the range in years was 1 to 46). Twenty-nine percent identified themselves as officers. The median meeting size was somewhere between 25 and 50 members. The average meeting attendance in the last 12 months was 43% (the median attendance was 36%; the range was 8% to 100%).

To estimate the representativeness of the sample with respect to the 2022 population of American union employees, the 2023 January issue of the U.S. Bureau of Labor Statistics was consulted (Union Affiliation, 2023). In doing so, we compared the percentages in the sample for age group, gender, minority status, and employment status with reported national percentages. The results indicated that employees 25 years or older, women employees, and minority employees were oversampled by 11% or less, + 0.0270, + 0.0940, + 0.1102 respectively.Footnote 5 The results also indicated that full-time employees were undersampled by 4%, -0.0357.

Psychological Safety at Local Meetings

To assess psychological safety at local meetings, we adapted seven items from the Psychological Safety Scale (Edmondson, 1999).

The items were prefaced with the statement:

“We are interested in a frank and accurate description of how your local meetings were run based on your experience—not hearsay from others—strictly and exclusively based on your experience.”

The statement was followed with a response instruction (“Check (√) one blank”) and an item stem:

“In reference to local meetings I have attended in the last 12 months, the following describe what I experienced at these meetings . . .”

An example item is:

“. . . meetings wherein members were at no risk of embarrassing themselves even when they couldn't always express themselves clearly.”

Response options were “yes” or “no”.

A principal components analysis was performed on the psychological safety items. The analysis produced one eigenvalue greater than 1.00, λ = 3.257, % of variance explained = 57.522, item loadings ≥ 0.365.

The Cronbach’s α for the items was 0.79.

As a result of these analyses, responses were averaged, yielding continuous psychological safety scale scores from 0 (experienced less) to 1.00 (experienced more).

Non-psychological Safety at Local Meetings

To assess non-psychological safety at local meetings, we adapted five items from nonunion design meeting scales, taxonomies, and commentaries (e.g., Hoffman, 2018; Rogelberg, 2019).

The items were interspersed randomly with the psychological safety items, following the same preface statement, response instruction, and item stem.

An example item is:

“. . . meetings wherein members did their best to stick to items on the agenda.”

Response options were “yes” or “no”.

A principal components analysis was performed on the non-psychological safety items. The analysis produced one eigenvalue greater than 1.00, λ = 2.582, % of variance explained = 42.634, item loadings ≥ 0.350.

The Cronbach’s α for the items was 0.67.

As a result of these analyses, responses were averaged, yielding continuous non-psychological safety scale scores from 0 (experienced less) to 1.00 (experienced more).

Meeting Effectiveness

To assess meeting effectiveness, we adapted two items from the Global Scale of Local Union Effectiveness (Hammer & Wazeter, 1993).

Prefaced with a response instruction (“Circle one number”), the items followed the psychological safety and non-psychological safety items.

An example item is:

“Based on how your local meetings were run, how effective do you think they were in doing the business your local needed to do?”

Response options were based on Likert scaling with 1 to 7 anchors (Not very effective to Very effective).

A principal components analysis was performed on the meeting effectiveness items. The analysis produced one eigenvalue greater than 1.00, λ = 1.812, % of variance explained = 90.588, item loading = 0.952.

The Cronbach’s α for the items was 0.91.

As a result of these analyses, responses were averaged, yielding continuous meeting effectiveness scale scores from 1 (lower effectiveness) to 7 (higher effectiveness).

Local Meeting Attendance in the Next 12 Months

To assess local meeting attendance in the next 12 months, we adapted one item from other items used to indicate attendance (e.g., Flood, 1993; Lund & Taylor, 2010).

The item followed the meeting effectiveness items:

“Based on how your local meetings were run, how likely are you to attend meetings in the next 12 months?”

Response options were based on Likert scaling with 1 to 7 anchors (Not very likely to Very likely), yielding continuous meeting attendance in the next 12 months item scores from 1 (less likely) to 7 (more likely).

Controls

To control for sample-specific statistical associations between the demographics and psychological safety at meetings, psychological safety scale scores were regressed onto the demographics (exempting hypothesized moderators), and in the same analysis, onto non-psychological safety scale scores.

From this analysis, the unstandardized residual psychological safety scores were used to represent psychological safety in model tests.

In the same vein, significant zero-order correlations between the demographics and meeting effectiveness, and between meeting attendance in the next 12 months, were used to select covariates (see Table 1; see Sauer et al., 2013 for a discussion on covariate selection). As such, in separate analyses, meeting effectiveness scores were regressed onto socioeconomic status, and meeting attendance was regressed onto socioeconomic status, officer status, and meeting attendance in the last 12 months.

Table 1 Zero-order correlations (rs), means, and standard deviations
Full size table

From these analyses, the unstandardized residual scores for meeting effectiveness and for meeting attendance were used in model tests (see Fig. 2 for the tested model with covariates indicated in grayscale).

Fig. 2
figure 2

Tested model

Full size image

Results

Raw score zero-order correlations, means (Ms), and standard deviations (SDs) for all study variables are presented in Table 1.

Descriptive Tests

To discern significant mean differences in reference to demographics and model variables, we performed t-tests, using median splits for non-dichotomous variables to construct high-low subgroups.

On psychological safety at meetings, on average, women employees (vs. men employees) were less likely to experience safety at meetings, a difference also seen for minority employees (vs. nonminority employees), and for employees for which English is a Second Language (vs. employees for which English is a First Language), ts(128) ≤ -2.393, ps < 0.05. In contrast, on average, employees with high socioeconomic status (vs. employees with low socioeconomic status) were more likely to experience safety at meetings, t(128) = 2.953, p < 0.01.

On meeting effectiveness, on average, employees for which English is a Second Language were less likely to rate meetings as effective, t(128) = -2.665, p < 0.01. In contrast, on average, employees with high socioeconomic status were more likely to rate meetings as effective, t(128) = 2.568, p < 0.05.

On meeting attendance in the next 12 months, on average, women employees, minority employees, and employees for which English is a Second Language were less likely to indicate attendance, ts(128) ≤ -2.152, ps < 0.05. In contrast, on average, officers (vs. member only) were more likely to indicate attendance, a difference also seen for employees who attended a higher percent of meetings in the last 12 months, ts(128) ≥ 2.636, ps < 0.01.Footnote 6

Preliminary Model Tests

The zero-order correlation between psychological safety at meetings and meeting attendance in the next 12 months was positive and significant, r = 0.54, p < 0.01, as were correlations between psychological safety and meeting effectiveness, r = 0.66, p < 0.01, and between meeting effectiveness and meeting attendance in the next 12 months, r = 0.63, p < 0.01 (see Table 1).

Also, the zero-order correlations between hypothesized moderators (gender, minority status, language of origin) and psychological safety, meeting effectiveness, and meeting attendance in the next 12 months were negative and significant, rs ≤ -0.18, ps < 0.05; the range of rs was -0.18 to -0.48 (see Table 1).

Model Tests

To test the hypothesized model, we ran a series of regression-based analyses selected from PROCESS 4.0 written for SPSS by Hayes (2022). In each regression analysis, we used 10000 bootstrap (Boot) samples to generate Boot 95% confidence intervals (CIs) for direct and indirect effects.Footnote 7

Mediational Analysis (no moderation)

We first tested a model with no moderation that estimated direct paths to and from model variables in each instance, viz., the direct effect with meeting effectiveness as outcome (psychological safety→meeting effectiveness), and the direct effects with meeting attendance as outcome (psychological safety→meeting attendance, meeting effectiveness→meeting attendance). The analysis also included a test for the hypothesized indirect effect of psychological safety on meeting attendance through meeting effectiveness (psychological safety→meeting effectiveness→meeting attendance).Footnote 8

In the analysis, the direct effect of psychological safety on meeting effectiveness and the direct effect of meeting effectiveness on meeting attendance were significant (the CIs for the effects did not include zero; see Table 2). In contrast, the direct effect of psychological safety on meeting attendance was nonsignificant (the CI for the effect did include zero). The test for the hypothesized indirect effect indicated that the effect of psychological safety on meeting attendance through meeting effectiveness was significant, indirect effect = 1.8871, Boot SE = 0.3998, Boot CI [1.1251, 2.6854].

Table 2 Regression results: mediation analysis without moderator
Full size table

Moderated Mediation Analyses

We next tested the model with moderation that estimated the conditional path in the model from psychological safety to meeting effectiveness in reference to either gender, minority status, or language of origin as moderator, viz., the conditional effect of psychological safety by gender with meeting effectiveness as outcome (psychological safety by gender→meeting effectiveness), the conditional effect of psychological safety by minority status with meeting effectiveness as outcome (psychological safety by minority status→meeting effectiveness), and the conditional effect of psychological safety by language of origin with meeting effectiveness as outcome (psychological safety by language of origin→meeting effectiveness). The analyses (conducted separately) also included a test for the hypothesized conditional indirect effect of psychological safety on meeting attendance through meeting effectiveness (e.g., psychological safety by gender→meeting effectiveness→meeting attendance).

In reference to gender as moderator, the direct effect of psychological safety on meeting effectiveness was significant, but the direct effect of gender was not (the CI for the latter effect included zero; see Table 3). Also, the expected conditional effect of psychological safety by gender on meeting effectiveness was nonsignificant (the CI for the effect included zero), as was the expected conditional indirect effect of psychological safety by gender on meeting attendance through meeting effectiveness, index of moderated mediation = 0.2994, Boot SE = 0.4335, Boot CI [-0.5254, 1.2046].Footnote 9

Table 3 Regression results: moderated mediation analysis with gender, minority status, or language of origin as moderator
Full size table

In reference to minority status as moderator, the direct effect of psychological safety on meeting effectiveness was significant, but the direct effect of minority status was not (the CI for the latter effect included zero; see Table 3). However, the expected conditional effect of psychological safety by minority status on meeting effectiveness was significant, unstandardized coefficient (B) = 2.0387, Boot SE = 0.6814, Boot CI [0.6901, 3.3872].

Simple slopes for the conditional effect with minority status as moderator are presented in Fig. 3. For minority employees, the effect of psychological safety on meeting effectiveness is positive and significantly different from zero, B = 3.9867, Boot SE = 0.4551, Boot CI [3.0860, 4.8874]. For nonminority employees, the effect of psychological safety on meeting effectiveness is also positive and significantly different from zero, B = 1.9480, Boot SE = 0.5072, Boot CI [0.9444, 2.917]; however, in reference to the significant conditional effect, the effect for nonminority employees is less significantly different from zero than the effect for minority employees (i.e., the simple slopes are significantly different from each other).

Fig. 3
figure 3

Simple slopes for psychological safety at local meetings and meeting effectiveness in relation to minority status

Full size image

As expected, the test for the hypothesized conditional indirect effect of psychological safety by minority status on meeting attendance through meeting effectiveness was significant, index of moderated mediation = 1.2668, Boot SE = 0.4462, Boot CI [0.4423, 2.2087].

In reference to language of origin as moderator, the direct effect of psychological safety on meeting effectiveness was significant, but the direct effect of language of origin was not (the CI for the latter effect included zero; see Table 3). However, the expected conditional effect of psychological safety by language of origin on meeting effectiveness was significant, B = 1.8289, Boot SE = 0.8976, Boot CI [0.0527, 3.6052].

Simple slopes for the conditional effect with language of origin as moderator are presented in Fig. 4. For English as a Second Language employees, the effect of psychological safety on meeting effectiveness is positive and significantly different from zero, B = 4.3035, Boot SE = 0.8095, Boot CI [2.7016, 5.9055]. For English as a First Language employees, the effect of psychological safety on meeting effectiveness is also positive and significantly different from zero, B = 2.4746, Boot SE = 0.3877, Boot CI [1.7073, 3.2419]; however, in reference to the significant conditional effect, the effect for English as a First Language employees is less significantly different from zero than the effect for English as a Second Language employees (i.e., the simple slopes are significantly different from each other).

Fig. 4
figure 4

Simple slopes for psychological safety at local meetings and meeting effectiveness in relation to language of origin

Full size image

As expected, the test for the hypothesized conditional indirect effect of psychological safety by language of origin on meeting attendance through meeting effectiveness was significant, index of moderated mediation = 1.1365, Boot SE = 0.4033, Boot CI [0.3541, 1.9365].Footnote 10

Summary of Model Tests

In relation to the hypothesized model, the model test results provide support for mediation. Consistent with Hypothesis 1—the replication hypothesis— employees who experience more psychological safety at meetings are more likely to indicate meeting attendance in the next 12 months, but the path from psychological safety unfolds through meeting effectiveness, such that more psychological safety is associated with higher rated meeting effectiveness, which in turn is associated with more attendance.

Also, in relation to the hypothesized model, the model test results provide support for conditional effects. Consistent with Hypothesis 2, the positive relationship between psychological safety at meetings and meeting effectiveness is stronger for minority employees and for employees for which English is a Second Language.

Moreover, in relation to the hypothesized model, the model test results provide support for moderated mediation. Consistent with Hypothesis 3, the positive relationship between psychological safety and meeting attendance in the next 12 months through meeting effectiveness is stronger for minority employees and for employees for which English is a Second Language.

Discussion

What Then?

The aim of the present study was to replicate and expand a mediation model constructed to address the problem of chronic low local union meeting attendance. Based on a psychological view of local meetings as providing opportunities for employees to experience safe environments in which they are encouraged to fulfill psychological needs related to work through interactions with other employees, we found support for a hypothesized model in which the mediation was replicated. Specifically, in reference to mediation, we found support for the effect of psychological safety on meeting attendance through meeting effectiveness, in which employees who experienced more safety at meetings in the last 12 months were more likely to indicate meeting attendance in the next 12 months, an effect linked to higher rated meeting effectiveness. Moreover, we found support for the hypothesized model in which reported demographics, viz., gender, minority status, and language of origin, were positioned in the model as moderators. Specifically, in reference to conditional effects, we found support for a model path from psychological safety to meeting effectiveness, in which the path was stronger for minority employees and for employees for which English is a Second Language. Moreover, in reference to moderated mediation, we found support for the indirect effect of psychological safety on meeting attendance through meeting effectiveness, in which the effect was stronger for minority employees and for employees for which English is a Second Language.

Given the model support, brought forward are the two questions that fueled our study: (1) Why replicate the model? and (2) Why expand the model? On the first question, we were concerned that the attention garnered by the mediation model might lead some statisticians and researchers to overgeneralize the model results. We think the replication shown in the present study with a new sample of employees who attended meetings rescues the model from the prospect of overgeneralization, as well the opposite—undergeneralization—in which replication offsets the prospect that support of the model rests on a single sample, and, as such, might be suggested as a “one-off study”. With replication, the model achieves a record of fit, albeit, and notably, the samples are specific to American union employees. On the second question, we were concerned that programs of intervention for low attendance might be advanced by others for so-called marginalized employees before a proper moderated mediation model was constructed and test results were in, in which yet to be indicated is whether programs could be claimed as evidence-based.

What’s New?

Based on discussions with local reps, and in reference to our reading on meeting attendance, we soon became aware of the claim made by unions that they have long championed diversity and inclusion at work on behalf of employees, and presently are non-complacent about these issues (see Greenhouse, 2019; McAlevey, 2020; Unger, 2020 for up-to-date narratives). Also, in reference to our reading of widely circulated business articles on current human resource (HR) practices, we soon became aware that no-choice (“compliant”) intervention programs on diversity and inclusion have been launched in masse by employers beginning with “front-line” employees (see, for example, Fernandes, 2023; Hunt et al., 2018). Perhaps less circulated in public view is that the development and implementation of such programs must be first “sold” to corporate gatekeepers (e.g., board members) for approval, in which approval is partially contingent on a credible claim that programs are evidence-based, a claim that is thought to come with an assurance of legal protection. How such a claim is made credible begins with groups of in-house “HR specialists” who search, find, and extract model results on diversity and inclusion from published social science studies—studies, it is safe to say, that now also appear in masse (see Bezukova et al., 2016 for a recent review and list of studies).

The foregoing provides context for why to date there are no like intervention programs on diversity and inclusion launched in masse by unions beginning with local members in reference to the problem of meeting attendance. According to our key informants (“specialists” of another kind), yet to be shown are model results that bear on the problem in which a credible claim can be made that model results can be linked to the attendance problem, a claim that might also come with evidence-based legal protection if and when model results are used to underline a program—that is, until now.

To wit, in reference to the present study, the replicated and expanded model results provide a strong bid to develop and implement intervention programs aimed at employees who show low meeting attendance. To be extracted from the results is that attendance is linked to opportunities to experience safe environments—environments in which employees can strive to fulfill work-related psychological needs through interactions with other employees. Specifically, suggested by the results is that employees who experience more psychological safety at meetings tend to rate meetings as more effective and that more safety is associated with more meeting attendance in relation to meetings rated as more effective. Moreover, in reference to diversity and inclusion, also suggested by the results is that such relationships are stronger for minority employees and for employees for which English is a Second Language in contrast to their counterparts.

What’s Next?

In our view, stated study limitations are opportunities to pose new ideas and questions to be addressed in future studies. Foremost, in reference to the present study, is the need for further sampling of employees who attend local meetings, and in particular, sampling details about local demographics. For instance, we did not find support for gender as a moderator. In hindsight, we suspect that this moderation may be conditional on the experience of women employees at meetings in reference to the mix of women and men who attend—that the hypothesized unfolding of the model variables may relate to whether the presence of women at meetings is large or small. Unknown at this time is whether model relationships may be affected by a local demographic that may be framed as gender mix at meetings and, conceivably, may be operationalized as a proportional variable. In future studies it may be possible to estimate the demographic by asking attending employees to indicate the percent of women and men present at meetings.

Also, needed in future studies is sampling of attending employees at meetings within a single union. In contrast to the present sampling of attending employees from locals affiliated with different unions, a question might be asked about how the model may fair, support-wise, in a sample in which attending employees are sampled from locals affiliated with the same union. Of particular interest might be cross-level effects in which, for example, aggregated safety data within locals may be used to examine local-level effects on individual-level attendance in the next 12 months in relation to either local-level or individual-level effectiveness ratings. Perhaps, based on presentations of the model results seen in the present study themed with a projected intervention program to reverse low attendance, it may be possible to gain the kind of “buy-in” support from a union and the local reps that would make collection of such data feasible.

In reference to a question of how the model results contribute to extant literatures in ways that might inspire researchers to conduct future studies, we think two-fold recognition is in order. First, centered on the problem of low meeting attendance, the results bear on the further adaptability and refinement of safety as a psychological construct imported from the team literature on safety, a bridging of the literature from nonunion to union environments. As seen in the present study, the safety construct can be successfully adapted from the safety literature to forecast like team outcomes in union environments, with the caveat that unlike nonunion environments, meeting attendance in union environments is voluntary and, at the same time, is arguably crucial for local success (i.e., to be heard as a strong collective voice). As for refinement, shown in reference to support for the hypothesized moderated relationships, indirect safety effects are shown to be stronger for employees with certain demographics, an emphasis that provides an added ecological focus.

Second, centered on the supported moderated relationships, the literature on diversity and inclusion is a direct beneficiary of the model results. This literature—both in terms of corporate-sponsored blogs and published academic work—has yet to fully incorporate psychological safety as a key variable in intervention programs designed to enhance employee awareness of links between diversity and participation and to suggest how organizations can better craft inclusion strategies to stabilize and improve work teams. Moreover, based on the model results of the present study, we think the indirect effects associated with safety at the individual-level can now be suggested as a viable part of intervention strategies to create opportunities for employees with diverse minority and language of origin backgrounds to feel safe at meetings, especially with respect to “speaking up” without fear of being embarrassed, rejected, or punished by others.

What Now?

In the Mellor (2023) study, an intervention was suggested to reverse low meeting attendance based on the mediation results. Included here is an expanded version of the intervention based on the moderated mediation results, with attention (indicated in italicized font) to attending minority employees and employees for which English is a Second Language.Footnote 11

It is suggested that local reps meet to construct an announcement flyer to be distributed before the next local meeting, a flyer that is printed in multiple languages that includes all languages spoken by members in the local. The flyer should indicate that the meeting is to assess how employees would like to see their meetings run with special attention to how all attending members might be encouraged to participate. Before the meeting, the reps should preselect members who speak each of the languages that are spoken by members and, importantly, and if possible, members who are also bilingual. At the meeting, these members should be asked to introduce an invited guest (an outside researcher) who is present to administer a short no-name voluntary survey to be filled out by all members, a survey printed in all languages present in the local, a survey “for everyone to indicate their experience at meetings.”

The survey should be streamlined and shortened to 14 items (the 7 psychological safety items interspersed with the 5 non-psychological safety items, plus the 2 effectiveness items). Before the survey is distributed it should be indicated by the preselected language speakers that the survey responses will be tallied and reported to all members at the next meeting, the interpretation of which will be opened to the floor for discussion, in which all attending speakers will be heard and, if need be, heard through bilingual members.

At the next meeting, the outside researcher should be absent, having submitted a brief “summary only report” to the local reps, a report that, if needed, should be edited by the reps for clarity, printed in all languages, and, if possible, delivered by hand to all attending members at the beginning of the meeting. The key to this second meeting is the creation of an all-inclusive language discussion in which emergent is the prospect of an open and informal atmosphere without regard to how much time might be taken for bilingual translations. If a third meeting appears to be in order to further discuss and interpret the results, it should be called.

As an addendum to this expanded intervention, we think that the moderation results are sufficient to suggest that the local reps should petition national union coalitions for funding to sponsor free English as a Second language classes and English-to-[fill-in-the language or languages] translation classes for reps and members to facilitate the widest possible spread of bilingualism, in which, before classes, out-sourced instructors are given the chance to learn the lexicon of the local and to be informed of the take-away purpose of the instruction—viz., to facilitate “free and open” discussions at meetings by reps and members with diverse language origins.

Final Note

On the relationships indicated in the model, we take the point of view that the model itself is a work-in-progress and we welcome instructive comments from statisticians, researchers, and interested parties to expand model relationships to ensure meaningful model fit. As for the expanded intervention—another work-in-progress—we also welcome instructive comments from local reps, members, and interested bloggers with diversity and inclusion in mind to optimize the intervention to ensure well-attended meetings.