ArticlePDF Available

Predictors of turnover amongst volunteers: A systematic review and meta‐analysis

August 2023
Journal of Organizational Behavior

DOI:10.1002/job.2729

License
CC BY 4.0

Authors:

Vivien Forner

The University of Sydney

Djurre Holtrop

Tilburg University

Gavin R. Slemp

University of Melbourne

Show all 8 authorsHide

Volunteers represent a global workforce equivalent to 61 million full‐time workers. A significant decline in volunteering has highlighted the urgency to better understand and address turnover amongst volunteers. To address this, we conducted a systematic review and meta‐analysis of turnover amongst volunteers. We also examined whether staying or leaving has different predictors. The meta‐analysis integrated and synthesized 117 studies, encompassing 1104 effect sizes across 55 335 volunteer workers, to identify and quantify relationships between turnover and the broad range of variables that have been examined in the volunteer work domain. Amongst the strongest predictors of volunteer turnover were attitudinal variables, in particular, job satisfaction (ρ = −.58), affective commitment (ρ = −.58), engagement (ρ = −.54) and organizational commitment (ρ = −.54). Contextual variables that showed the largest effects included communication (ρ = .62), organizational support (ρ = −.61) and the quality of the relationship between volunteers and their leader (leader‐member exchange, ρ = −.55). We synthesize our findings into an integrative framework delineating the predictors of volunteer turnover. In doing so, we extend turnover research to consider non‐remunerated work contexts and provide a basis for developing turnover theory that is responsive to the unique experience of volunteers.

Content uploaded by Vivien Forner

Content may be subject to copyright.

RESEARCH ARTICLE

Stemming the Tide: An Expanded Focus on Employee Turnover

Predictors of turnover amongst volunteers: A systematic

review and meta-analysis

Vivien W. Forner

1,2

| Djurre Holtrop

3,4

| Edwin J. Boezeman

Gavin R. Slemp

| Magdalena Kotek

3,7

| Darja Kragt

| Mina Askovic

Anya Johnson

University of Sydney Business School, The

University of Sydney, Sydney, Australia

University of Wollongong, Wollongong,

Australia

Department of Social Psychology, Tilburg

University, Tilburg, The Netherlands

Faculty of Business and Law, Future of Work

Institute, Curtin University, Perth, Australia

Section of Social, Economic and

Organisational Psychology, Faculty of Social

and Behavioral Sciences - Institute of

Psychology, Leiden University, Leiden, The

Netherlands

Centre for Wellbeing Science, Faculty of

Education, The University of Melbourne,

Melbourne, Australia

Luxembourg Institute of Socio-Economic

Research (LISER), Esch-Belval, Luxembourg

School of Psychological Sciences, University

of Western Australia, Perth, Australia

Correspondence

Vivien W. Forner, University of Sydney

Business School, The University of Sydney,

Darlington NSW 2006, Australia.

Email: vivien.forner@sydney.edu.au

Summary

Volunteers represent a global workforce equivalent to 61 million full-time workers. A

significant decline in volunteering has highlighted the urgency to better understand and

address turnover amongst volunteers. To address this, we conducted a systematic

review and meta-analysis of turnover amongst volunteers. We also examined whether

staying or leaving has different predictors. The meta-analysis integrated and synthesized

117 studies, encompassing 1104 effect sizes across 55 335 volunteer workers, to iden-

tify and quantify relationships between turnover and the broad range of variables that

have been examined in the volunteer work domain. Amongst the strongest predictors

of volunteer turnover were attitudinal variables, in particular, job satisfaction (ρ=.58),

affective commitment (ρ=.58), engagement (ρ=.54) and organizational commit-

ment (ρ=.54). Contextual variables that showed the largest effects included commu-

nication (ρ=.62), organizational support (ρ=.61) and the quality of the relationship

between volunteers and their leader (leader-member exchange, ρ=.55). We synthe-

size our findings into an integrative framework delineating the predictors of volunteer

turnover. In doing so, we extend turnover research to consider non-remunerated work

contexts and provide a basis for developing turnover theory that is responsive to the

unique experience of volunteers.

KEYWORDS

meta-analysis, retention, turnover, volunteer

1|INTRODUCTION

Volunteers are the backbone of society (Salamon et al., 2018); they

underpin our essential services, they are a community's social capital

and they are what makes our societies resilient in times of crisis. Every

day around the globe, millions of people take action on issues that

matter to them, volunteering with organizations to provide access to

food, protect wildlife, educate young people, guide spectators and

athletes and provide essential safety and emergency response ser-

vices. Worldwide, 862.4 million people volunteer each month, repre-

senting an important, sizable global workforce equivalent to 61 million

full-time workers (UNV, 2021).

The global volunteering sector is currently experiencing a signifi-

cant and unprecedented decline in volunteer numbers. Over a 4-year

Received: 15 April 2022 Revised: 3 January 2023 Accepted: 7 June 2023

DOI: 10.1002/job.2729

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,

provided the original work is properly cited.

J Organ Behav. 2023;1–25. wileyonlinelibrary.com/journal/job 1

period, from 2018 to 2021, almost half (44%) the global volunteer

workforce stopped volunteering. This is a loss equivalent to 48 million

full-time workers (UNV, 2018,2021). This devastating decline in

volunteering during the COVID pandemic added to already deteriorat-

ing volunteer participation rates around the world. Indeed, countries

such as Australia, the United Kingdom and the United States have

been reporting, with increasing concern, ongoing decline in volunteer

numbers. This great (volunteer) resignation has serious consequences

for volunteer involving organizations, threatening their sustainability

and the invaluable societal services they offer. With declining access

to new workers, it is critical that volunteer involving organizations can

effectively retain their existing volunteer workforce and help support

ongoing active participation in volunteering activities (Forner, 2019;

Luksyte et al., 2021).

Achieving a clearer understanding of the turnover process within

volunteer involving organizations will be critical to address declining vol-

unteer participation. Volunteers can withhold their services more readily

than paid workers, in part because volunteers are not dependent on the

position for a salary or contractually bound to the organization. Conse-

quently, understanding factors that influence ongoing participation in

volunteering activities have been important areas of scientific inquiry,

with both practical and societal significance. There is abundant research

on turnover phenomena within the context of volunteering (Locke

et al., 2003). However, this literature remains fragmented, spanning

many different domains (Kragt & Holtrop, 2019), largely disconnected

from the organizational behaviour literature and in need of synthesis.

Our aim is to integrate this body of literature to find out what sustains

volunteering, one of the most vital theoretical and practical questions in

the space of volunteer research (Snyder & Omoto, 2008). For this pur-

pose, we conducted a systematic review and meta-analysis examining

the drivers of turnover amongst volunteers to provide a framework that

will guide future research efforts and assist organizations in managing

an invaluable volunteer workforce.

1.1 |Volunteers: A vital and poorly understood

workforce

Volunteers represent a unique occupational group that is conceptually

distinct from employees. We define volunteers as members of the com-

munity who willingly give their time to engage in freely chosen and

deliberate helping activities without remuneration. Volunteering can be

informal, occurring directly between individuals, or formal, taking place

through organizations or associations (UNV, 2021). Our focus is the lat-

ter, organization-based volunteering, where volunteer workers under-

take their activities within formal organized structures, although they

are free from obligation to do so (Bidee et al., 2013). Volunteering often

requires effortful and responsible activities that far exceed mere hobby-

ism. For example, Australia relies on more than 195 000 volunteer fire

fighters who train on a weekly basis and provide emergency response

at all hours of the day and night (Cull, 2020). In the care sector, volun-

teer hospice workers carry the heavy burden of supporting people in

their final days so that they can depart this world in a comfortable

environment (Brown, 2011). Moreover, the largest educational organi-

zation in the world, Scouts, is run entirely by volunteer leaders who

spend, on average, 7.5 h per week training and educating youth mem-

bers (Gagne et al., 2018). All these volunteering roles are highly involved

and require as much skill and training as many paid positions, meaning

that the loss of volunteer workers, in addition to societal costs, has

severe financial consequences for organizations.

Prior research has differentiated volunteer workers from paid

employees, revealing key differences in their motivation (Pearce,

1983), personality (Elshaug & Metzer, 2001; Riggs & Kaess, 1955) and

work-related attitudes (Catano et al., 2001; Pearce, 1993; van Vuuren

et al., 2008). When compared to paid employees, volunteers experi-

ence their role in unique ways and show different responses to their

organizational context (Fallon & Rice, 2015) and leadership behaviours

(Catano et al., 2001). Importantly, studies have revealed key differ-

ences in the antecedents of turnover intentions amongst volunteers

compared to employees (Boezeman & Ellemers, 2009; Fallon & Rice,

2015). For example, social aspects of the role (Boezeman & Ellemers,

2009) and receiving support and recognition (Fallon & Rice, 2015)

were found to be stronger predictors of intention to stay amongst vol-

unteers, relative to paid employees.

These differences between volunteers and employees highlight

the importance of explicitly studying the predictors of turnover

amongst volunteers, rather than generalizing findings from studies of

paid employees. Because existing theories and our current under-

standing of turnover reflect an accumulation of research conducted

with paid employees, we cannot assume that they are transferable to

the volunteer domain. While the antecedents and processes involved

in turnover of employees have been extensively addressed

(e.g., Griffeth et al., 2000; Price, 2001; Rubenstein et al., 2018), the

growing body of literature examining turnover in volunteer occupa-

tions is yet to be synthesized. To-date, there are no prior reviews or

meta-analyses summarizing the mounting research into turnover

within volunteer involving organizations. Our research addresses this

gap and provides the first systematic review of volunteer turnover

research and meta-analysis to identify the strongest predictors of

turnover amongst volunteers.

1.2 |Conceptualizing and measuring turnover of

volunteers

A conceptualization, model or theory on turnover, that is responsive

to and reflects the unique volunteer context, has yet to be established.

Turnover has traditionally been defined as an employee's voluntary

severance of his or her current employment ties (Mobley, 1982;

Mobley et al., 1979; Price, 1977). Traditional definitions of employee

turnover do not transfer well to volunteerism, in part because

volunteer workers rarely quit or resign formally but, instead, simply

stop showing up (Jamison, 2003). An alternative, more appropriate,

way to think about turnover in volunteering is to consider and

incorporate notions of volitional active participation. We therefore

propose the following definition: turnover in volunteer organizations

2FORNER ET AL.

10991379, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/job.2729 by National Health And Medical Research Council, Wiley Online Library on [19/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

occurs when a volunteer volitionally withdraws his or her participation

with the current organization. In contrast, volunteering is sustained

when a person continually chooses to be present at volunteer

activities and willingly donates their time to the organization and its

cause. In this sense, turnover is conceptualized as a unitary construct,

with decisions and attitudes about maintaining active participation

(staying) versus decisions and attitudes towards withdrawing partici-

pation (leaving) as opposite ends of the same construct. Yet, while

staying and leaving are often considered to be different ends of the

same continuum, they may nevertheless have different predictors

(Hom et al., 2017). This notion is in line with the Two-Factor

Motivation-Hygiene Theory of Herzberg et al. (1957), which suggests

that specific factors may evoke positive attitudes while other factors

may evoke negative attitudes (i.e., for individuals, specific factors may

only evoke satisfaction while other factors may only cause dissatisfac-

tion). Investigating these differences is important to understanding the

pathways to turnover. To our knowledge, the notion that attitudes

towards leaving and staying may have different predictors has not

been verified through research synthesis. Our systematic review and

meta-analysis contributes to addressing this gap in the literature by

separately investigating volunteers' attitudes towards leaving and

staying.

Various indicators have been used to measure turnover and study

turnover phenomena in organizations. Turnover indicators include

turnover behaviours, traditionally measured using organizational data,

and workers' intentions and attitudes towards staying or leaving. Pre-

vious meta-analyses of turnover in paid work have tended to use

behavioural indicators of turnover from personnel records (Griffeth

et al., 2000). However, collecting this type of data within the volun-

teering sector is challenging as there are significant difficulties captur-

ing and maintaining reliable data about volunteer numbers (Salamon

et al., 2018). One of the main reasons for this is, unlike paid work,

there is no employee contractual agreement or formal severance

between a volunteer worker and the organization (Cnaan & Cascio,

1998). When volunteers decide to leave the organization, it is uncom-

mon for them to formally resign or quit, they simply stop turning

up. Consequently, organizations do not have a record of when volun-

teers leave, and the inaccuracy of organizational data about turnover

and volunteer workforce numbers has been highlighted as a major

source of concern (e.g., Auditor General, 2014).

Turnover intentions are another indicator of turnover that has

also been included in prior research and meta-analysis on turnover

(Park & Min, 2020; Zimmerman & Darnold, 2009). In the paid work

context, there is evidence that this is amongst the strongest predic-

tors of turnover behaviours (Holtom et al., 2008; Lee et al., 2017) and

intention to leave or stay, explaining 31% of turnover behaviour

(ρ=.56) (Rubenstein et al., 2018). For volunteers, we argue that this

association may be stronger. Turnover intentions are much more likely

to have behavioural consequences in the volunteer context where

there are fewer structural constraints, for example, pay and health

insurance, keeping people there. Reflecting these challenges and argu-

ments, the majority of the empirical research into the turnover of vol-

unteers use ‘turnover intentions’as their primary outcome variable.

Only some researchers have been able to operationalize and examine

actual volunteer turnover behaviours, for example, longitudinal stud-

ies that include participants self-reports of whether they are still vol-

unteers for an organization (e.g., Okun et al., 2016) and studies that

include observations of whether volunteers actually show up at the

organization for the volunteer work (e.g., Harrison, 1995). While these

examples of research moving beyond attitudes are encouraging, they

are still very scarce in the literature. Accordingly, in the current meta-

analysis, we consider both volunteers' intentions to stay or leave and

volunteers' actual turnover behaviours as different indicators repre-

senting the broad domain of ‘turnover’for volunteers. We posit that

this broader conceptualization of turnover is important in the rela-

tively nascent and under-developed volunteering literature because it

provides an opportunity to examine the predictors that may help us

understand what factors contribute to turnover and the different

pathways through which they have their effect.

1.3 |Creating a framework for the predictors of

turnover amongst volunteers

What makes volunteers want to stay with their organizations has

been an issue of great interest to academics, government policy

makers and the boards and directors of volunteer involving organiza-

tions. Researchers have approached this question from many different

perspectives, reflected in the extensive and broad range of psycholog-

ical and contextual variables that have been measured alongside turn-

over in the volunteer literature (Kragt & Holtrop, 2019; Wilson, 2000).

Indeed, the eligible studies in our systematic review included over

350 different predictors. Researchers have also utilized a wide range

of diverse theoretical lenses to study turnover phenomena in volun-

teer organizations, for example, theory of planned behavior (TPB)

(Bang et al., 2014; Lee et al., 2014), motivation theories (self-

determination theory; Wu et al., 2016; volunteer functions; Van

Vianen et al., 2008) and leadership theories (LMX; Ang et al., 2017).

To move this field forward, we propose an integrative framework that

utilizes the rich variations that characterize this field of research. By

bringing together studies that cut across a broad range of variables

and theoretical perspectives, we provide researchers with a compre-

hensive integrative picture of turnover phenomenon in the volunteer

work context.

To connect and organize the vast body of literature, all the predic-

tors of turnover amongst volunteers were categorized according to

the taxonomy presented in Table 1. The taxonomy, which is hierarchi-

cal in nature, integrates existing typologies in applied psychology

research (Bosco et al., 2015; Cascio & Aguinis, 2008) and conceptual

categorizations used in previous meta-analysis of turnover (Holtom

et al., 2008; Rubenstein et al., 2018). For example, the taxonomy we

use to categorize predictors of turnover amongst volunteers con-

verges with a summary model of the predictors of turnover by Holtom

et al. (2008), where the authors mapped out how the predictors of

employee turnover from various past and recent theoretical perspec-

tives can be conceptually integrated. This same model informed the

FORNER ET AL.3

structure of a seminal meta-analysis on employee turnover

(Rubenstein et al., 2018).

In total, our taxonomy arranges variables into five first level cate-

gories: (1) characteristics of the volunteer job, (2) characteristics of

the volunteer organization, (3) volunteer contributions, (4) characteris-

tics of the individual volunteer and (5) motivation and job attitudes.

These first level categories are further refined into second and third

levels, increasing in specificity. For example, characteristics of the vol-

unteer role (level 1) branch into six second level categories (i.e., job

demands, leadership, social connections, agency, mastery and stimu-

lating), and these are further refined at a third level. These finer cate-

gorizations of variables are informed by existing theoretical

perspectives from the organizational behaviour and psychology litera-

tures (e.g., Clary & Snyder, 1999; Deci & Ryan, 2008; Demerouti et al.,

2001; Maslach & Leiter, 2017; Meyer et al., 2002; Parker et al., 2017).

First, a large portion of research has focused on how job charac-

teristics or aspects of the work context affect turnover amongst vol-

unteers. These have examined job demands (e.g., Huynh et al., 2012),

psychological need satisfaction (Forner, 2019) and different aspects of

work design, such as stimulating and challenging activities (Elstad,

2003), providing volunteers with agency in their role or task that are

significant (Millette & Gagné, 2008)—when it is clear the volunteer job

contributes positively to the lives of others. Others have focused on

relational aspects of the role, such as leadership (e.g., Forner, 2019;

Henderson & Sowa, 2018), opportunities for social interactions and

the quality of social relations with co-volunteers (e.g., Nencini et al.,

2016).

Second, a number of studies have investigated if turnover

amongst volunteers may be influenced by organizational characteris-

tics. Examples include organizational climate and culture (e.g., Cho

et al., 2020), reward systems and organizational justice (e.g., Rice &

Fallon, 2011) and different human resource management practices,

such as orientation and training (e.g., Wu et al., 2019) or communicat-

ing clearly and adequately to volunteers (Lo Presti, 2013).

Third, some studies have also investigated volunteer contribu-

tions. For instance, how time spent on volunteering and tenure are

related to turnover intentions (e.g., Giel & Breuer, 2020). Others have

looked at how performance (Rogalsky et al., 2016) or financial giving

and donations (e.g., Wisner et al., 2005) are associated with turnover.

Fourth, researchers posit that different individual characteristics

of volunteers may affect turnover (e.g., Kragt & Holtrop, 2019).

Amongst those studied are demographic characteristics (e.g., gender

type, age, type of religion, marital status, level of education and socio-

economic status; e.g., Wilson & Musick, 1997), personality type

(e.g., altruism, egoism; Mykletun & Himanen, 2016) and personal cir-

cumstances (e.g., family conflict; Cowlishaw et al., 2014).

TABLE 1 Hierarchical variable taxonomy used to classify 1104 correlates of turnover reported in studies of volunteer workers.

Level 1 Level 2 Level 3

Characteristics of the

volunteer job

Job demands •Time-related pressures, general job demands

Relational: social connections •Connectedness, group integration, organizational support, social

interactions, supportive peer relationships

Relational: leadership •Leader autonomy support, leader-member exchange (LMX), supervisor

support, leadership behaviours (other)

Work design: agency •Empowerment, voice, autonomous work

Work design: mastery •Appreciation/recognition, feedback, role ambiguity, mastery (other)

Work design: stimulating •Contributing productively, task significance, stimulating work (other)

Basic psychological need satisfaction •Autonomy, competence, relatedness

Characteristics of the

volunteer organization

HRM practices •Communication, learning and development, psychological contract,

support from paid staff

Organizational culture and climate and

other characteristics.

•Organizational climate, reputation, rewards/incentives

Volunteer contributions Positive behaviours •Performance, tenure, volunteer participation (time spent), financial

giving/donations

Characteristics of the

individual volunteer

Demographics •Age, education

Personal circumstances •Time commitments to ‘other work’, work–home conflict

Motivation and job attitudes Commitment •Organizational commitment, affective commitment

Identification •Fit, role identity

Motivation •Altruistic motivation, autonomous motivation, controlled motivation

Volunteer functions (motives/reasons) •Care, enhancement, protective, social, understanding, values

Favorable job attitudes •Engagement, enjoyment, job satisfaction, self-efficacy

Theory of planned behavior •Attitude, perceived behavioral control, subjective norms.

Ill-being •Burnout, depression, stress and anxiety, health and ill-being (other)

Well-being •Happiness

4FORNER ET AL.

Lastly, many researchers have investigated volunteerism through

a motivational and attitudinal lens. These have identified motivational

mechanisms involved in volunteering, including intrinsic motivation

(e.g., Boezeman & Ellemers, 2009; Millette & Gagné, 2008) and per-

sonal reasons for volunteering—the latter often referred to as volun-

teer functions (Clary et al., 1998) or the affordances volunteering

provides. For example, ‘wanting to make new friends through volun-

teering’describes a social function, and ‘doing activities that offer

opportunities for learning new things’describes an understanding

function. Further, researchers have investigated a range of broad job

attitudes commonly studied in the paid employment context such as

job satisfaction, engagement and wellbeing (e.g., burnout, stress and

happiness), organizational commitment and organizational identifica-

tion. In addition, perceived ability to perform the job, perceived fit

with the organization and trust, for instance, have been investigated

as related to turnover amongst volunteers (e.g., Miller et al., 1990;

Okun et al., 2016; Valéau et al., 2013; Van Vianen et al., 2008).

1.4 |Exploring moderators of turnover for

volunteers

In addition to examining the key predictors, we also conduct explor-

atory moderation analysis to examine the influence of demographic

factors: gender, age and the type of volunteering engagement (epi-

sodic vs. long term) on the strength or direction of the relationships

between turnover and its predictors.

Age and gender are two demographic moderators identified by

earlier meta-analysis of employee turnover (e.g., Griffeth et al., 2000)

that may also influence the nature of the predictor–turnover relation-

ships amongst volunteer workers. The moderating effect of age may

be even more important in the volunteering context because people

continue volunteer work through retirement and into old age, which

results in a broader age range and greater generational differences

across volunteer workers relative to employees. Prior studies have

shown that older and younger volunteers have different motivations

for volunteering and find different aspects of the volunteering experi-

ence rewarding (Okun & Schultz, 2003; Tiraieyari et al., 2019) that

require distinct retention strategies (Hopkins & Dowell, 2022). For

example, Hopkins and Dowell (2022) found that volunteering helped

retired older people feel needed, and these older people valued the

social aspects of the volunteer experience, such as making friends. In

contrast, younger people were more likely to volunteer as a means of

skill development and placed greater importance on personal achieve-

ment and broadening their experience through volunteering.

Gender makes a difference to the type and amount of volunteer-

ing people do (Wilson, 2000), and as with employee turnover phe-

nomenon, there are likely to be gender differences that influence the

predictor–turnover relationship amongst volunteers. Einolf (2010), for

example, found females were more motivated to help others, contrib-

uted more volunteering hours and scored higher on traits that

typically predict proactive helping behaviours, such as agreeableness

and moral obligation and prosocial role identity, relative to men.

Tiraieyari et al. (2019) found that female volunteers have stronger

affiliation needs than male volunteers that increase their likelihood of

‘staying’in their volunteer position. It is likely that such differences

between male and female volunteers may reduce the influence of

some predictors, such as motivation and attitude variables for exam-

ple, on turnover for organizations that have a predominance of female

volunteers. Gender and age are often used as controls in volunteering

research, and this meta-analysis is an opportunity to determine

whether these controls are justified and if substantive differences in

the associations between predictors and turnover outcomes exist for

these demographic factors.

There may also be differences in retention depending on the

nature or context of the volunteer role. In this study, we investigate

the moderating influence of the type of volunteering engagement,

classified as either episodic or traditional. In traditional volunteering,

individuals participate in regular ongoing volunteer activities with the

same organization over a long term, such as emergency service volun-

teers (Henderson & Sowa, 2018; Lewig et al., 2007; Rice & Fallon,

2011). In contrast, episodic volunteering engagements are character-

ized by short distinct periods or one-off events, such as volunteering

at a music festival (Bachman et al., 2016) or a sporting event (Bang

et al., 2019; Rogalsky et al., 2016). These different types of volunteer-

ing may be important as for example, Chen and Yu (2014) found that

long-term volunteers are more susceptible to burnout and turnover

compared to short-term episodic volunteers. It may be that some

aspects of the volunteering experience are more salient for continua-

tion amongst traditional volunteers' who spend more time (number of

hours and frequency) in their organization relative to episodic volun-

teers. Most volunteering research is conducted within a specific con-

text, making the outcomes difficult to generalize across all contexts.

This meta-analysis provides a valuable opportunity to better under-

stand whether the observed associations are generalizable to most

volunteering contexts, be they short-term engagements or traditional

volunteering roles.

1.5 |Aims of the research

This systematic review and meta-analysis sets out to answer the

question: What are the strongest predictors of turnover for volun-

teers? The objective is to synthesize and integrate prior studies that

have examined factors that impact turnover in volunteer organiza-

tions, to provide a foundation for further theory building. The

research aims to identify and quantify the relationships between

turnover and a broad range of psychological and contextual vari-

ables that have been tested in the volunteer organizational litera-

ture. To this end, we employ an empirical meta-analytic approach,

whereby we systematically review the volunteer turnover literature

to identify all the predictors used in past research, categorize them

and report the meta-analysis findings to generate novel theoretical

insights into volunteer turnover.

To quantify the predictors of volunteer turnover, we present evi-

dence of the most common correlated variables with estimates of the

FORNER ET AL.5

size of the corrected correlations. To investigate whether the predic-

tors of volunteers' intentions to remain (stay) and intentions to quit

(leave) are similar or different, we calculate and present effect sizes

separately for stay, leave and overall turnover. Along with this, we

also aim to examine the extent to which moderating factors (age, gen-

der and type of volunteer engagement) influence the strength of

these relationships. We synthesize our meta-analytic findings into an

integrative framework delineating the predictors of volunteer reten-

tion and turnover. In doing so, we extend turnover research to con-

sider unique effects in non-remunerated work contexts and provide a

basis for developing a theory of turnover that is transferable and

responsive to the unique experience of volunteer workers.

2|METHOD

2.1 |Search strategy

We conducted a systematic search of the literature to identify studies

that have measured indicators of turnover, operationalized as turn-

over intentions, continuation intentions or behaviours, in adult volun-

teer workers within an organizational context. To identify records for

inclusion in the meta-analysis, we used two approaches. First, we

searched electronic databases for relevant published studies, which

included PsycINFO, Scopus and Web of Science Core Collection,

including Sciences Expanded (1965+), Social Sciences (1965+), Arts &

Humanities (1975+), Conference Proceedings (1991+) and Emerging

Sources Citation (2005+), as this combination provided optimal cover-

age of relevant research. To develop the most relevant search terms,

three of the authors first conducted a scoping review to test combina-

tions of preliminary terms that were later iteratively refined for both

relevance and coverage in the final search. The final agreed search

query consisted of a combination of keywords across two categories:

(1) volunteering (volunteer* OR unpaid OR voluntary work) AND

(2) retention criteria (retention OR turnover OR quit). Searches were

conducted using all keywords across both word sets, using the

Boolean ‘OR’operator to separate words within sets and the ‘AND’

operator to combine sets. This ensured that only records with at least

one word from each category would be captured. We also refined the

search by excluding ‘unpaid leave’as they led to abundant irrelevant

hits. We imposed no date restraints on the search, but because the

term ‘volunteer’frequently appears in abstracts referring to voluntary

participants in clinical medical research, we further limited the search

to the Psychology, Social Sciences, Humanities and Business litera-

tures. To ensure this did not exclude potentially relevant studies, we

conducted a subsequent search using the string (volunteer* OR

unpaid OR ‘voluntary work’) AND (workplace OR organization OR

work) across the excluded subject areas. Second, to supplement our

electronic search and to curate relevant records from the ‘gray litera-

ture’(Kepes et al., 2012), we made calls for unpublished data on the

Academy of Management OB division ListServs and conducted an

additional literature search on ProQuest Dissertations to identify

unpublished doctoral dissertations and theses.

2.2 |Inclusion criteria and eligibility screening

Our screening process was managed through the digital systematic

review platform ‘Covidence’and is presented visually in Figure 1.As

shown, our search identified a total of 4741 records, of which 1391

were duplicates that were removed. An additional 106 unpublished

samples were provided from our call for unpublished data (32) and

ProQuest dissertation search (74), leaving a total of 3456 unique

records for screening. All 3456 records were assessed against our eli-

gibility criteria (described below) that were established a priori in our

systematic review protocol. At the beginning of the screening task,

three coders individually screened a random selection of 300 records.

Results and disagreements were discussed and resolved between the

coders and used to develop and refine a codebook to guide and

improve coding accuracy to increase reproducibility (Belur et al.,

2018). After screening the title and abstracts of the 3456 records,

2891 hits were excluded due to obvious irrelevancy. We then

assessed the full text of the remaining 565 records for eligibility. All

authors and four research assistants contributed to full text screening

and subsequent coding of the studies.

Studies were to be included in the systematic review if they satis-

fied four criteria: (1) The sample were adult volunteer workers exam-

ined in an organizational context; (2) the study measured turnover;

operationalized as turnover intentions, continuation intentions or

turnover behaviours; (3) the authors reported a zero-order correlation

coefficient between turnover and another variable that could be

empirically assessed as an antecedent or outcome; (4) the source was

a primary empirical quantitative study. Studies were excluded if they

used non-volunteer worker samples, such as studies examining inten-

tions to volunteer in the future, clinical samples, students volunteering

as research participants, donors (e.g., blood donors and financial

donors) and children aged <18 years. Essays, book reviews, theoretical

papers, literature reviews and editorials were also excluded. The

search and eligibility assessment yielded 117 independent studies

(N=55 335), reporting 1104 distinct correlations between turnover

criteria and another variable for inclusion in the present study.

2.3 |Coding procedure

The records were coded using a systematic coding template. Database

fields in the coding template included (a) the turnover variable (objec-

tive turnover data, intention to stay, intention to leave, sustained

volunteering), (b) whether the variable was a stay or leave indicator,

between each turnover variable and each correlate variable, (e) the

reliability of turnover intention measures, (f) the reliability of the cor-

relate variable, (g) the sample size pertaining to each correlation coef-

ficient, (h) the country in which each study took place, (i) the context

(e.g., type of volunteer activity), (j) the type of volunteering (episodic

event, traditional, tourism), (k) the percentage of the sample who were

female, (l) mean age, (o) publication status and (p) study design. A ran-

dom subset of 224 effect sizes, representing 21% of our data, were

6FORNER ET AL.

independently re-coded by the fifth author to establish coding accu-

racy via interrater agreement. Overall interrater agreement was 99%.

2.4 |Quality assessment

Studies that met the inclusion criteria were critically evaluated to

determine the quality of the research. Quality was judged by the

respective coders individually, using the Martin's (2017) critical

appraisal checklist for analytical cross sectional studies, a tool

designed for quality appraisals in systematic reviews. Guided by a

series of checklist questions, for example, ‘Were the study subjects

and the setting described in detail?’and ‘Were the outcomes mea-

sured in a valid and reliable way?’, coders appraised the methodologi-

cal quality of each study to determine the extent to which the study

has addressed the possibility of bias in its design, conduct and analy-

sis. Based on these assessments, the coders provided an overall

appraisal of the quality of the paper: (a) good (meets all or most of the

criteria); (b) lower quality (misses critical elements). Overall, the

research methodologies used in this literature had substantive limita-

tions, an observation also made in other reviews of the volunteer liter-

ature (e.g., Kragt & Holtrop, 2019). There are also a number of

challenges associated with conducting quality assessments for sys-

tematic reviews of management research (Tranfield et al., 2003).

Employing a rigorous quality assessment in order to exclude lower

quality studies so that only the ‘best evidence’is used would result in

exclusion of most of the available evidence in this literature and would

bias the conclusions of the meta‐analysis (see Schmidt & Hunter, 2015).

Thus, a decision was made to use the overall quality rating as a mod-

erator and estimate differences in effects of studies assessed to be

stronger methodologically (good quality) compared to studies that

used weaker methodologies (lower quality) (Stone et al., 2019).

2.5 |Taxonomy

The 1104 variables from the correlation tables from each primary

study were coded using the taxonomy proposed in the introduction.

Details of all extracted variables, with accompanying values, and how

they were categorized are available on this project's Open Science

Framework page (https://osf.io/w54e2/). After extracting variables

from the correlation tables, we followed an approach similar to Bosco

et al. (2015) and Aguinis et al. (2009) to further develop and refine our

taxonomy. This involved several rounds of error checking, refinement

and discussion between first, second, third and eighth authors. Ini-

tially, the first and third authors examined all 1104 extracted variables

and grouped them at the finest level, creating the narrow third level

categories of the taxonomy. For example, all variables that measured

affective commitment were grouped together and assigned to the

level 3 category ‘affective commitment’. The categorizations were

checked and discussed amongst the author team. Narrow categories

(level 3) were subsequently grouped into intermediate (level 2) and

higher order (level 1) categories, reflecting theoretical perspectives

(e.g., Clary & Snyder, 1999; Deci & Ryan, 2008; Demerouti et al.,

2001; Maslach & Leiter, 2017; Meyer et al., 2002; Parker et al., 2017)

identified by previous reviews as prevalent in volunteering research

FIGURE 1 PRISMA diagram showing

literature search and inclusion process.

FORNER ET AL.7

(e.g., Kragt & Holtrop, 2019; Wilson, 2000,2012). As a last step, we

compared our initial taxonomy to integrative models from the

employee turnover literature (Holtom et al., 2008; Rubenstein et al.,

2018) and reconsidered some hierarchical classifications. In doing so,

we maximize synergies between our review of the volunteer turnover

literature to prior synthesis of the employee turnover literature—

helping to bridge the two bodies of research. The final taxonomy

arranges the variables in a hierarchical fashion, with each predictor

coded at a high level (level 1), intermediate (level 2) and narrow level

(level 3), increasing in specificity.

2.6 |Statistical method

We used the Schmidt and Hunter (2015) psychometric approach for

meta-analysis in the current study. All analyses were conducted in R

(version 3.5.0), using the R-Studio interface (version 1.1.453). The

majority of analyses were performed using the ‘psychmeta’package

(Dahlke & Wiernik, 2019), except our publication bias analyses (dis-

cussed shortly) that were completed using MetaSen (Field et al.,

2021). Thus, we calculated a sample-size weighted mean observed

correlation between each predictor variable and each turnover crite-

rion and then disattenuated the correlations for measurement by

using the reliability information that was reported in the included

studies. We specifically used reliabilities to construct artefact distribu-

tions for each variable and thus ran our meta-analysis using the arte-

fact distributions method. All analyses were conducted using the

unbiased sample variance estimator available in psychmeta, which

leads to more accurate results with conservative confidence intervals

(CIs), particularly when the analysis involves a small number of

studies.

Following previous meta-analyses that have examined turnover

(e.g., Rubenstein et al., 2018), we corrected some relationships from

studies that reported point-biserial correlations between a predictor

and a dichotomous criterion that reflected objective staying versus

leaving behaviour. Such correlations are biassed downwards due to

restricted variance (Kemery et al., 1988), particularly when they are

not normally distributed, often attributable to an uneven split of

stayers and leavers in the sample. In the few instances where this

occurred (n=3), we used the procedures described by Schmidt and

Hunter (2015) to disattenuate these correlations, which required cod-

ing the proportion of stayers and leavers from the sample and using

this information in the correction formula.

The psychometric meta-analysis approach (Schmidt & Hunter,

2015) is based on the random effect model, which does not assume

homogeneity of effect parameters (see Schmidt et al., 2009 for a com-

parison with other meta-analysis approaches). Instead, it estimates

mean effect sizes and true (non-artifactual) variability of effect sizes

across studies. It is well established that random effect meta-analyses

lead to more generalizable population effect size estimates than fixed

effect models, the assumptions of which inhibit generalizability and

are generally inappropriate for applied settings (Hunter & Schmidt,

2000). Random effect meta-analyses also yield more conservative CIs

(Field, 2003; Hunter & Schmidt, 2000; Kisamore & Brannick, 2008)

and, as such, are less likely to overestimate precision. In order to pre-

serve statistical independence of effect sizes contributing to each

meta-analytic association, where studies reported multiple correla-

tions of the same relationship from a single sample, we computed

composite correlations and composite reliabilities (see Schmidt &

Hunter, 2015, pp. 441–447 for a detailed discussion).

We used the 95% CI to evaluate the precision of each meta-

analytic association (ρ). If the CI around ρencompassed 0, we

assumed the association between the two constructs was not signifi-

cant. To assess the magnitude of each association, we used the

benchmarks reported in Bosco et al. (2015), which arose from a data-

base of almost 150 000 research findings and thus have greater valid-

ity than Cohen (1988) benchmarks. According to Bosco et al. (2015),

correlations of .07, .16 and .32 reflect the 25th, 50th and 75th per-

centiles, respectively, which we used as the lower-bound thresholds

for weak, moderate and strong effect size magnitudes in the current

study.

To assess heterogeneity, we used two approaches. First, we used

, which serves as an indicator of cross-study heterogeneity; larger

values indicate greater dispersion in effect size distributions. We also

used the 80% credibility interval (CV), which serves as an indicator of

heterogeneity distributed around each effect size. The CV is inter-

preted such that 80% of the distribution of true-score correlations lie

within this range (Whitener, 1990). To assess moderators, we used

two approaches. Where sufficient studies were available, we used cat-

egorical moderator analyses, which involved conducting a series of

subgroup meta-analyses across different levels of the moderator

(e.g., traditional volunteering vs. episodic event volunteering). If only

continuous data were available (e.g., age), we used meta-regression to

examine whether the moderator was related to study-level effect

sizes.

We calculated an effect whenever there were at least three stud-

ies between each predictor and each criterion. To summarize our main

effect and moderator analyses, we use 10 pieces of information:

(a) k=number of studies used to calculate each effect,

(b) N=combined sample size, (c) r=the ‘bare bones’meta-analytic

correlation before artefact corrections are applied (Schmidt & Hunter,

2015), (d) SD

=the observed standard deviation, (e) SD

res

=residual

standard deviation, (f) ρ=correlation corrected for sampling and

measurement error (i.e., the true-score correlation), (g) SD

=the stan-

dard deviation of ρ, (h) SD

=the observed standard deviation of

artefact corrected correlations, (i) 95% CI =95% CI for true-score

correlations, (j) 80% CV =80% credibility interval around each

effect size.

3|RESULTS

The results of our meta-analysis are summarized in Table 2. For each

variable presented in the table, we calculated separate effects for

staying intentions (i.e., stay) and leaving intentions/behaviour

(i.e., leave). The table also presents a pooled turnover estimate

8FORNER ET AL.

TABLE 2 Meta-analysed correlations between predictors and turnover criteria in volunteer workers.

Predictor and turnover criterion Effect size kN r SD

res

ρSD

95% CI 80% CV

Characteristics of the volunteer job

Job demands Strong 35 5638 .29 .18 .16 .35 .21 .19 [.27, .42] [.10, .60]

Job demands (general)

Stay — —— ——————— —

Leave Strong 31 3706 .35 .17 .15 .40 .20 .17 [.33, .48] [.18, .63]

Turnover Strong 33 4592 .30 .20 .18 .35 .23 .21 [.27, .43] [.08, .62]

Time related pressures

Stay — —— ——————— —

Leave — —— ——————— —

Turnover Moderate 3 1605 .27 .05 .01 .31 .05 .02 [.18, .44] [.28, .34]

Relational: Social connections Strong 62 21 674 .35 .19 .18 .43 .24 .23 [.49, .37] [.73, .14]

Connectedness

Stay Strong 5 1178 .39 .08 .06 .50 .11 .08 [.37, .63] [.38, .62]

Leave ns 4 5057 .25 .20 .20 .31 .26 .25 [.72, .09] [.73, .10]

Turnover Strong 9 6235 .27 .18 .18 .34 .23 .22 [.52, .17] [.66, .03]

Group integration

Stay Strong 5 1531 .33 .14 .13 .39 .17 .16 [.18, .61] [.15, .63]

Leave — —— ——————— —

Turnover Strong 6 1648 .32 .14 .12 .38 .16 .15 [.55, .21] [.60, .16]

Organizational support

Stay Strong 6 3340 .41 .09 .08 .54 .12 .10 [.41, .66] [.39, .69]

Leave Strong 3 2172 .57 .20 .20 .72 .26 .25 [1.00, .08] [1.00, .24]

Turnover Strong 9 5512 .47 .15 .15 .61 .20 .19 [.76, .46] [.87, .35]

Supportive peer relationships

Stay Strong 10 4697 .38 .18 .18 .49 .24 .23 [.32, .66] [.17, .81]

Leave Strong 30 3041 .33 .11 .06 .38 .12 .07 [.43, .33] [.47, .29]

Turnover Strong 40 7738 .36 .16 .14 .44 .19 .17 [.50, .38] [.66, .22]

Social interactions

Stay Strong 4 2169 .24 .06 .04 .32 .08 .05 [.19, .44] [.23, .40]

Leave — —— ——————— —

Turnover — —— ——————— —

Relational: Leadership Strong 51 13 520 .34 .14 .12 .39 .16 .14 [.44, .35] [.58, .21]

Leader autonomy-support

Stay — —— ——————— —

Leave Moderate 3 3527 .25 .04 .04 .28 .05 .04 [.41, .15] [.36, .21]

Turnover — —— ——————— —

Leader–member exchange

Stay Strong 4 1384 .48 .23 .22 .57 .27 .26 [.14, .99] [.14, 1.00]

Leave — —— ——————— —

Turnover Strong 5 1803 .46 .20 .19 .55 .24 .23 [.85, .25] [.91, .19]

Leadership behaviours

Stay Moderate 5 1548 .22 .04 .00 .25 .05 .00 [.19, .31] [.25, .25]

Leave — —— ——————— —

Turnover Moderate 6 1783 .21 .04 .00 .27 .05 .00 [.32, .21] [.27, .27]

Supervisor support

Stay Strong 6 1211 .34 .09 .06 .42 .11 .07 [.31, .54] [.32, .53]

(Continues)

FORNER ET AL.9

TABLE 2 (Continued)

Predictor and turnover criterion Effect size kN r SD

res

ρSD

95% CI 80% CV

Leave Strong 30 4935 .40 .12 .10 .45 .13 .11 [.50, .40] [.59, .31]

Turnover Strong 36 6146 .39 .11 .09 .44 .13 .11 [.49, .40] [.58, .31]

Work design: Agency Strong 42 7020 .33 .18 .17 .41 .23 .21 [.48, .34] [.68, .14]

Empowerment

Stay ns 4 2271 .16 .10 .09 .22 .14 .12 [.00, .43] [.02, .42]

Leave — —— ——————— —

Turnover — —— ——————— —

Voice

Stay — —— ——————— —

Leave Strong 30 2780 .36 .24 .22 .41 .27 .25 [.51, .31] [.74, .08]

Turnover — —— ——————— —

Autonomy

Stay Strong 7 3186 .35 .09 .06 .46 .11 .09 [.36, .57] [.34, .59]

Leave ns 3 561 .21 .21 .20 .28 .29 .27 [.99, .43] [.79, .22]

Turnover Strong 10 3747 .33 .12 .10 .44 .16 .14 [.55, .33] [.63, .25]

Work design: Mastery Strong 46 18 157 .29 .17 .16 .35 .20 .19 [.41, .29] [.60, .10]

Appreciation/recognition

Stay Strong 7 9704 .29 .12 .11 .35 .14 .14 [.22, .49] [.16, .55]

Leave Strong 30 4935 .42 .13 .11 .49 .14 .12 [.54, .43] [.65, .33]

Turnover Strong 37 14 639 .34 .13 .12 .40 .16 .14 [.45, .35] [.59, .21]

Feedback

Stay Moderate 3 1065 .26 .03 .00 .31 .03 .00 [.23, .40] [.31, .31]

Leave — —— ——————— —

Turnover Moderate 4 1189 .23 .11 .09 .29 .13 .11 [.50, .08] [.47, .11]

Role ambiguity

Stay — —— ——————— —

Leave Moderate 31 3840 .22 .27 .25 .25 .31 .29 [.37, .14] [.64, .13]

Turnover Moderate 32 4168 .18 .29 .28 .21 .34 .32 [.33, .09] [.63, .22]

Work design: Mastery (other)

Stay ns 3 1118 .27 .16 .15 .34 .20 .19 [.16, .83] [.02, .70]

Leave Moderate 29 2629 .23 .15 .11 .27 .17 .13 [.34, .20] [.44, .10]

Turnover Moderate 32 3747 .24 .15 .12 .29 .17 .14 [.35, .22] [.47, .11]

Work design: Stimulating Strong 44 8877 .31 .19 .18 .38 .24 .22 [.45, .31] [.67, .09]

Contributing productively

Stay — —— ——————— —

Leave Strong 29 2629 .42 .13 .10 .47 .15 .11 [.52, .41] [.61, .32]

Turnover Strong 30 2781 .40 .15 .12 .45 .16 .13 [.51, .39] [.62, .28]

Task significance

Stay Strong 5 2961 .30 .13 .12 .39 .17 .16 [.18, .60] [.15, .63]

Leave Strong 6 2299 .32 .16 .15 .44 .21 .20 [.66, .21] [.73, .14]

Turnover Strong 11 5260 .31 .14 .13 .41 .18 .17 [.53, .29] [.64, .18]

Stimulating work

Stay ns 4 1111 .13 .29 .29 .16 .35 .34 [.40, .71] [.40, .71]

Leave — —— ——————— —

Turnover ns 5 1235 .12 .27 .26 .15 .33 .32 [.55, .26] [.63, .34]

10 FORNER ET AL.

TABLE 2 (Continued)

Predictor and turnover criterion Effect size kN r SD

res

ρSD

95% CI 80% CV

Basic psychological need

satisfaction

Strong 5 4815 .37 .10 .09 .44 .12 .11 [.59, .30] [.61, .27]

Autonomy

Stay — —— ——————— —

Leave — —— ——————— —

Turnover Strong 3 3620 .32 .10 .10 .43 .13 .13 [.76, .10] [.67, .19]

Competence

Stay — —— ——————— —

Leave — —— ——————— —

Turnover Moderate 4 3799 .18 .08 .07 .23 .10 .10 [.40, .07] [.39, .08]

Relatedness

Stay — —— ——————— —

Leave — —— ——————— —

Turnover Strong 3 3620 .30 .09 .09 .38 .11 .11 [.66, .10] [.58, .18]

Characteristics of the volunteer organization

HRM practices Strong 50 15 627 .37 .16 .15 .45 .19 .18 [.50, .39] [.67, .22]

Communication

Stay Strong 5 2039 .44 .11 .09 .62 .16 .13 [.42, .82] [.42, .82]

Leave — —— ——————— —

Turnover — —— ——————— —

Learning and development

Stay Strong 11 7611 .38 .20 .19 .46 .24 .23 [.30, .62] [.15, .78]

Leave — —— ——————— —

Turnover Strong 13 8917 .38 .18 .18 .46 .22 .22 [.59, .32] [.75, .16]

Psychological contract

Stay Strong 5 2023 .47 .11 .10 .54 .12 .11 [.39, .70] [.37, .72]

Leave — —— ——————— —

Turnover — —— ——————— —

Support from paid staff

Stay — —— ——————— —

Leave Strong 26 2439 .45 .16 .13 .50 .17 .15 [.57, .43] [.69, .31]

Turnover — —— ——————— —

Organizational culture climate

and other characteristics

Moderate 8 6019 .27 .07 .05 .31 .08 .06 [.37, .24] [.39, .22]

Organizational climate

Stay — —— ——————— —

Leave — —— ——————— —

Turnover ns 4 4616 .22 .15 .14 .25 .17 .16 [.52, .01] [.52, .01]

Reputation

Stay — —— ——————— —

Leave — —— ——————— —

Turnover Strong 4 1403 .31 .10 .09 .41 .13 .12 [.62, .20] [.60, .22]

Reward/incentives

Stay Moderate 7 7511 .20 .09 .08 .27 .12 .11 [.17, .38] [.12, .43]

Leave — —— ——————— —

Turnover Moderate 8 7746 .21 .09 .08 .28 .12 .11 [.38, .18] [.43, .13]

(Continues)

FORNER ET AL.11

TABLE 2 (Continued)

Predictor and turnover criterion Effect size kN r SD

res

ρSD

95% CI 80% CV

Volunteer contributions

Positive behaviours Moderate 31 16 621 .19 .19 .19 .25 .26 .26 [.35, .16] [.59, .08]

Financial giving/donations

Stay ns 4 2535 .22 .33 .33 .28 .42 .42 [.39, .95] [.41, .96]

Leave — —— ——————— —

Turnover — —— ——————— —

Performance

Stay Weak 3 6507 .12 .02 .00 .15 .03 .00 [.08, .22] [.15, .15]

Leave — —— ——————— —

Turnover Moderate 4 6631 .12 .03 .02 .16 .04 .02 [.22, .09] [.20, .12]

Tenure

Stay ns 11 2804 .04 .14 .13 .04 .16 .14 [.06, .15] [.15, .24]

Leave ns 4 969 .05 .23 .22 .07 .35 .34 [.64, .49] [.63, .48]

Turnover ns 15 3773 .04 .16 .15 .06 .22 .21 [.18, .07] [.33, .22]

Volunteer participation time

Stay ns 4 768 .09 .09 .05 .10 .10 .05 [.06, .25] [.01, .18]

Leave Strong 7 3047 .22 .21 .21 .33 .32 .31 [.62, .04] [.77, .12]

Turnover Moderate 11 3815 .19 .19 .19 .28 .29 .28 [.48, .09] [.66, .09]

Characteristics of the individual volunteer

Demographics — —— ——————— —

Age

Stay ns 12 3665 .06 .13 .12 .07 .14 .13 [.02, .16] [.11, .24]

Leave ns 12 4293 .08 .15 .14 .09 .17 .16 [.21, .02] [.32, .13]

Turnover Weak 24 7958 .07 .14 .12 .08 .16 .14 [.15, .02] [.27, .11]

Education

Stay ns 5 1425 .03 .25 .24 .04 .30 .29 [.33, .41] [.41, .49]

Leave ns 4 1180 .07 .15 .14 .08 .18 .17 [.36, .21] [.35, .19]

Turnover ns 9 2605 .05 .20 .19 .06 .25 .24 [.25, .13] [.39, .27]

Personal circumstances ns 12 5199 .19 .17 .16 .24 .21 .20 [.11, .38] [.03, .52]

Time commitments to

other work

Stay — —— ——————— —

Leave ns 5 2038 .03 .07 .04 .03 .08 .05 [.13, .07] [.11, .05]

Turnover ns 6 2209 .02 .07 .05 .02 .09 .06 [.11, .07] [.11, .07]

Work–home conflict

Stay Moderate 3 1373 .22 .07 .05 .30 .09 .07 [.53, .07] [.43, .16]

Leave Strong 3 1839 .31 .07 .06 .38 .09 .08 [.16, .60] [.23, .52]

Turnover Strong 6 3212 .27 .08 .07 .34 .10 .08 [.24, .44] [.22, .46]

Motivation and job attitudes

Commitment Strong 57 12 346 .43 .22 .21 .54 .27 .26 [.61, .47] [.87, .20]

Organizational commitment

Stay Strong 17 5869 .45 .22 .21 .56 .27 .26 [.42, .70] [.21, .91]

Leave Strong 37 5061 .41 .15 .12 .51 .19 .16 [.57, .45] [.71, .31]

Turnover Strong 54 10 930 .43 .19 .17 .54 .23 .22 [.60, .47] [.82, .26]

Affective commitment

Stay Strong 5 1878 .45 .29 .29 .55 .36 .36 [.10, 1.00] [.00, 1.00]

Leave ns 3 722 .41 .21 .20 .62 .32 .30 [1.00, .17] [1.00, .05]

Turnover Strong 8 2600 .44 .26 .25 .58 .34 .33 [.86, .29] [1.00, .11]

12 FORNER ET AL.

TABLE 2 (Continued)

Predictor and turnover criterion Effect size kN r SD

res

ρSD

95% CI 80% CV

Normative commitment

Stay — —— ——————— —

Leave — —— ——————— —

Turnover ns 4 1362 .20 .32 .31 .29 .46 .45 [1.00, .45] [1.00, .45]

Continuance commitment

Stay — —— ——————— —

Leave ns 3 1294 .38 .23 .23 .56 .34 .33 [1.00, .28] [1.00, .07]

Turnover — —— ——————— —

Identification Strong 10 3181 .32 .08 .06 .40 .10 .08 [.47, .32] [.50, .29]

Fit

Stay — —— ——————— —

Leave — —— ——————— —

Turnover Strong 4 1234 .28 .11 .09 .35 .14 .12 [.57, .13] [.54, .16]

Role identity

Stay Strong 4 1063 .36 .06 .02 .43 .07 .03 [.32, .54] [.39, .47]

Leave — —— ——————— —

Turnover Strong 6 2154 .36 .08 .06 .43 .09 .07 [.53, .34] [.54, .32]

Motivation ns 14 9505 .09 .33 .33 .11 .40 .40 [.12, .34] [.43, .64]

Altruistic motivation

Stay Moderate 8 7558 .22 .23 .22 .26 .27 .27 [.04, .49] [.11, .64]

Leave — —— ——————— —

Turnover — —— ——————— —

Autonomous motivation

Stay Strong 5 1892 .37 .08 .07 .46 .11 .09 [.33, .59] [.33, .60]

Leave ns 3 561 .33 .31 .30 .47 .44 .42 [1.00, .61] [1.00, .33]

Turnover Strong 8 2453 .36 .15 .14 .46 .19 .18 [.62, .31] [.71, .21]

Controlled motivation

Stay — —— ——————— —

Leave — —— ——————— —

Turnover ns 4 876 .12 .10 .07 .15 .13 .09 [.05, .36] [.00, .31]

Volunteer functions (reasons) Strong 23 21 721 .28 .21 .20 .35 .26 .25 [.46, .24] [.68, .02]

Motive: Career

Stay ns 9 9352 .13 .16 .16 .15 .20 .20 [.00, .31] [.12, .43]

Leave — —— ——————— —

Turnover Weak 11 9668 .12 .16 .16 .15 .20 .20 [.28, .01] [.42, .12]

Motive: Enhancement

Stay ns 5 1698 .26 .23 .22 .30 .26 .26 [.03, .63] [.09, .69]

Leave — —— ——————— —

Turnover ns 6 1856 .23 .23 .22 .28 .27 .26 [.56, .01] [.67, .11]

Motive: Protective

Stay ns 4 1784 .18 .19 .18 .22 .22 .21 [.14, .57] [.14, .57]

Leave — —— ——————— —

Turnover ns 5 1942 .17 .18 .17 .20 .22 .21 [.47, .07] [.52, .12]

Motive: Social

Stay Moderate 8 7108 .18 .14 .14 .25 .19 .19 [.08, .41] [.02, .51]

Leave — —— ——————— —

Turnover Moderate 9 7266 .18 .14 .14 .24 .19 .19 [.39, .09] [.50, .02]

(Continues)

FORNER ET AL.13

TABLE 2 (Continued)

Predictor and turnover criterion Effect size kN r SD

res

ρSD

95% CI 80% CV

Motive: Understanding

Stay ns 5 2310 .25 .20 .20 .30 .24 .23 [.00, .59] [.05, .65]

Leave — —— ——————— —

Turnover Moderate 7 2626 .21 .22 .21 .26 .26 .25 [.50, .01] [.62, .11]

Motive: Values

Stay Strong 8 7609 .25 .08 .07 .33 .10 .09 [.24, .41] [.20, .46]

Leave — —— ——————— —

Turnover Strong 9 7767 .25 .09 .08 .32 .12 .11 [.41, .23] [.47, .17]

Favourable job attitudes Strong 90 36 779 .49 .18 .18 .58 .22 .21 [.63, .54] [.86, .31]

Engagement

Stay Strong 6 1630 .47 .09 .07 .58 .11 .09 [.47, .69] [.45, .71]

Leave Strong 33 4957 .46 .11 .09 .53 .13 .10 [.58, .49] [.67, .40]

Turnover Strong 39 6587 .46 .10 .08 .54 .12 .09 [.58, .50] [.66, .42]

Enjoyment

Stay — —— ——————— —

Leave ns 3 875 .32 .28 .27 .37 .32 .32 [1.00, .43] [.96, .23]

Turnover ns 4 1231 .37 .24 .23 .40 .26 .25 [.82, .01] [.82, .01]

Job satisfaction

Stay Strong 38 22 018 .51 .20 .19 .61 .23 .23 [.53, .69] [.31, .91]

Leave Strong 41 10 900 .44 .16 .15 .52 .20 .19 [.59, .46] [.77, .28]

Turnover Strong 79 32 918 .49 .19 .18 .58 .23 .22 [.63, .53] [.86, .30]

Self-efficacy

Stay — —— ——————— —

Leave — —— ——————— —

Turnover ns 4 1667 .14 .24 .24 .16 .28 .27 [.61, .28] [.61, .29]

Theory of planned behavior ns 3 656 .37 .20 .19 .44 .24 .23 [1.00, .16] [.88, .00]

Attitude

Stay ns 3 656 .40 .30 .29 .47 .35 .35 [.41, 1.00] [.19, 1.00]

Leave — —— ——————— —

Turnover — —— ——————— —

Perceived behavioral control

Stay ns 3 656 .51 .44 .43 .61 .52 .51 [.68, 1.00] [.36, 1.00]

Leave — —— ——————— —

Turnover — —— ——————— —

Subjective norms

Stay ns 3 656 .36 .21 .20 .45 .27 .26 [.21, 1.00] [.03, .93]

Leave — —— ——————— —

Turnover — —— ——————— —

Ill-being Strong 45 11 492 .33 .17 .16 .40 .21 .20 [.34, .46] [.14, .66]

Burnout

Stay Strong 4 1668 .28 .14 .13 .35 .18 .16 [.63, .07] [.62, .09]

Leave Strong 35 7530 .40 .16 .15 .48 .19 .18 [.42, .55] [.25, .71]

Turnover Strong 38 9145 .38 .16 .15 .46 .19 .18 [.40, .52] [.23, .69]

Depression

Stay — —— ——————— —

Leave — —— ——————— —

Turnover ns 3 1578 .22 .10 .09 .27 .13 .12 [.05, .59] [.05, .49]

14 FORNER ET AL.

(i.e., turnover) where we combined stay/leave indicators in each anal-

ysis and reversed the sign of the correlation for staying intentions so

that the associations are consistently in one direction and more easily

interpreted. In addition to reporting omnibus meta-analytic results, we

present findings from a range of exploratory moderator analyses that

investigate the moderating effects of categorical and continuous vari-

ables, using sub-group and meta-regression approaches, respectively.

We conclude this section by reporting the results of analyses for pub-

lication bias to gauge the robustness of our findings, as proposed by

the APA's Meta-Analysis Reporting Standards (Appelbaum et al.,

2018) and best practice recommendations (Field et al., 2021).

3.1 |Predictors of retention amongst volunteers

3.1.1 | Motivation and job attitudes

Within the broad category of Favourable Job Attitudes, we identified

engagement and job satisfaction as the strongest predictors of reten-

tion. The results in Table 2indicate strong meta-analytic associations

between engagement and stay (ρ=.58, CI =[.47, .69]), leave

(ρ=.53, CI =[.58, .49]) and turnover (ρ=.54, CI =[.58,

.50]) and strong estimated average effect sizes between job satisfac-

tion and stay (ρ=.61, CI =[.53, .69]), leave (ρ=.52, CI =[.59,

.46]) and turnover (ρ=.58, CI =[.63, .53]). Other indicators of

job attitudes such as enjoyment and self-efficacy did not reach signifi-

cance. The level two category of Commitment features organizational,

affective, normative and continuance commitment. While we found

strong meta-analytic associations with affective commitment (stay:

ρ=.55, CI =[.10, 1.00], turnover: ρ=.58, CI =[.86, .29]) and

organizational commitment (stay: ρ=.56, CI =[.42, .70], leave:

ρ=.51, CI =[.57, .45], turnover ρ=.54, CI =[.60, .47])

effects for the remaining relationships were not significant. Notably,

some associations within this category featured large standard devia-

tions of residuals as well as wide CVs that imply potential moderating

effects impacting these associations (affective commitment and leave:

=.30, CV =[1.00, -.05]; normative commitment and turnover:

=.45, CV =[1.00, .45]).

We also found strong meta-analytic associations within the cate-

gory Identification. Specifically, we found strong estimated average

effect sizes for fit (turnover: ρ=.35, CI =[.57, .13]) and role

identity (stay: ρ=.43, CI =[.32, .54], turnover: ρ=.43, CI =[.53,

.34]). Similarly, results in Table 2concerning Motivation showed

strong estimated average effect sizes for autonomous motivation

(stay: ρ=.46, CI =[.33, .59], turnover: ρ=.46, CI =[.62, .31])

and moderate associations for altruistic motivation (stay: ρ=.26,

CI =[.04, .49]) The associations with controlled motivation included

CIs encompassing zero, as did the associations between autonomous

motivation and leaving, which additionally featured high levels of het-

erogeneity (SD

=.42, CV =[1.00, .33]). The majority of the associ-

ations with Volunteer Functions featured CIs encompassing zero, with

the exception of associations with the volunteer functions of career,

social, understanding and values motives. While effects of value

motives showed strong meta-analytic associations (stay: ρ=.33,

CI =[.24, .41], turnover: ρ=.32, CI =[.41, .23]), we found mod-

erate estimated average effect sizes for social motives (stay: ρ=.25,

CI =[.08, .41], turnover: ρ=.24, CI =[.39, .09]) and under-

standing motives (turnover: ρ=.26, CI =[.50, .01]). The results

showed small meta-analytic associations between career motives and

turnover (ρ=.15, CI =[.28, .01]). Additionally, associations with

TABLE 2 (Continued)

Predictor and turnover criterion Effect size kN r SD

res

ρSD

95% CI 80% CV

Poor health and ill-being

Stay ns 4 1457 .09 .13 .12 .12 .17 .15 [.38, .15] [.37, .13]

Leave ns 3 4086 .13 .05 .05 .19 .08 .07 [.00, .38] [.07, .32]

Turnover Moderate 7 5543 .12 .08 .07 .16 .10 .09 [.07, .26] [.03, .30]

Stress/anxiety

Stay — —— ——————— —

Leave — —— ——————— —

Turnover ns 3 1692 .14 .08 .07 .18 .11 .09 [.09, .44] [.01, .35]

Well-being ns 4 2605 .18 .18 .18 .24 .24 .24 [.63, .14] [.63, .14]

Happiness

Stay — —— ——————— —

Leave — —— ——————— —

Turnover ns 4 2605 .18 .18 .18 .24 .24 .24 [.63, .14] [.63, .14]

Note:‘Turnover’refers to the aggregation of stay and leave data into a combined association, with stay reversed to reflect its opposite.

Abbreviations: ρ, mean true-score correlation; CI, confidence interval around ρ; CV, credibility interval around ρ;k, number of studies contributing to meta-analytic

associations; N, combined sample size; ns, confidence interval includes zero (non-significant); r, mean observed correlation; SDres, residual standard deviation of

correlations after accounting for sampling error and measurement error; SD

, observed standard deviation of corrected correlations (r); SD

, observed standard

deviation of correlations; SD

, residual standard deviation of ρ.

FORNER ET AL.15

three indicators derived from the TPB (i.e., Attitude, Perceived Behav-

ioral Control and Subjective Norms) were not significant and have

large standard deviations of residuals as well as wide CVs (attitude:

=.35, CV =[-.19, 1.00], perceived behavioural control:

=.51, CV =[-.36, 1.00]).

Finally, within the Ill- and Well-being predictors, we found strong

meta-analytic effects for burnout (stay: ρ=.35, CI =[.63, .07],

leave: ρ=.48, CI =[.42, .55], turnover: ρ=.46, CI =[.40, .52]) and

moderate effects for poor health and other ill-being (turnover:

ρ=.16, CI =[.07, .26]). Other associations such as depression,

stress/anxiety and happiness all featured CIs that encompassed zero,

so these results were not significant.

3.1.2 | Characteristics of the volunteer organization

With respect to the characteristics of the volunteer organization, we

investigated HRM Practices, Organizational Culture and Other Char-

acteristics of the Organization. We consistently found strong esti-

mated average effect sizes for the HRM Practices of communication

(ρ=.62, CI =[.42, .82]), learning and development (L&D) (stay:

ρ=.46, CI =[.30, .62], turnover: ρ=.46, CI =[.59, .32]), psy-

chological contract (stay: ρ=.54 CI =[.39, .70]) and support from

paid staff (leave: ρ=.50, CI =[.57, .43]). Within Organizational

Culture, associations with organizational climate featured CIs encom-

passing zero. We found a strong meta-analytic association between

reputation and turnover (ρ=.41, CI =[.62, .20]) and moderate

effect sizes for rewards and incentives (stay: ρ=.27, CI =[.17, .38],

turnover: ρ=.28, CI =[.38, .18]).

3.1.3 | Characteristics of the volunteer job

Within the level two category of Job Demands, we analysed the

effects of (general) job demands and time-related pressure. There are

strong effects for (general) job demands (leave: ρ=.40, CI =[.33,

.48], turnover: ρ=.35, CI =[.27, .43]) and a moderate meta-analytic

relationship between time-related pressure and turnover (ρ=.31,

CI =[.18, .44]). Turning to Basic Psychological Need Satisfaction, the

findings revealed strong meta-analytic correlations between auton-

omy and turnover (ρ=.43, CI =[.76, .10]) and relatedness and

turnover (ρ=.38, CI =[.66, .10]), as well as a moderate effect

size between competence and turnover (ρ=.23, CI =[.40, .07]).

Within the Relational category, we analysed variables grouped as

Social Connection, such as connectedness, group integration, organiza-

tional support, supportive peer relationships and social interactions.

The meta analytic effects for these predictors were consistently

strong, with the weakest effects still showing moderately strong rela-

tions, such as social interactions and stay (ρ=.32, CI =[.19, .44]),

while the strongest effect was between organizational support and

turnover (ρ=.61, CI =[.76, .46]). Similarly, Leadership, also

within the Relational category, also included moderate to strong meta-

analytic effects, with the strongest predictors being leader-member

exchange (stay: ρ=.57, CI =[.14, .99], turnover: ρ=.55, CI =

[.85, .25]) and supervisor support (stay: ρ=.42, CI =[.31, .54],

leave: ρ=.45, CI =[.50, .40], turnover: ρ=.44, CI =[.49,

.40]). Leader autonomy support (leave: ρ=.28, CI =[.41, .15])

and leadership behaviours (stay: ρ=.25, CI =[.19, .31], turnover:

ρ=.27, CI =[.32, .21]) had moderate associations with volun-

teer turnover.

There are moderate to strong meta-analytic effect sizes for the cat-

egories within the broader Work Design grouping. With regard to Agency,

voice appeared to have a strong impact on leave indicators (ρ=.41

(CI =[.51, .31]), as did autonomy on stay indicators (ρ=.46, CI =

[.36, .57]) and on the consolidated turnover measure (ρ=.44, CI =

[.55, .33]). Empowerment included CIs encompassing zero. The Work

Design predictor within Mastery shows strong meta-analytic associa-

tions, specifically appreciation and recognition (stay: ρ=.35, CI =[.22,

.49], leave: ρ=.49, CI =[.54, .43], turnover: ρ=.40, CI =

[.45, .35]), whereas we found moderate average effects for feedback

(stay: ρ=.31, CI =[.23, .40], turnover: ρ=.29, CI =[.50, .08]),

role ambiguity (leave: ρ=.25, CI =[.37, .14], turnover: ρ=.21,

CI =[.33, .09]) and other mastery predictors (leave: ρ=.27, CI =

[.34, .20], turnover: ρ=.29, CI =[.35, .22]). Similarly, we found

that Stimulating Work category predictors strongly impact volunteer turn-

over. We found a strong, positive meta-analytic relationship between

task significance and stay indicators (ρ=.39, CI =[.18, .60]) and strong

negative estimated average effect sizes between task significance and

turnover (leave: ρ=.44, CI =[.66, .21], turnover: ρ=.41, CI =

[.53, .29]) as well as between contributing productively and turnover

(leave: ρ=.47, CI =[.52, .41], turnover: ρ=.45, CI =[.51,

.39]).

3.1.4 | Volunteer contributions

Within the category of Volunteer Contributions, we found volunteer par-

ticipation time had a strong meta-analytic association with leave

(ρ=.33, CI =[.62, .04]) and a moderate average effect on turnover

(ρ=.28, CI =[.48, .09]). Performance only had a weak effect on

stay (ρ=.15, CI =[.08, .22]) and turnover (ρ=.16, CI =[.22, .09]).

Associations with financial giving and tenure were not significant.

3.1.5 | Characteristics of the individual volunteer

We examined volunteer Demographics but did not find associations

that were significant, with the exception of the weak meta-analytic

association between age and turnover (ρ=.08, CI =[.15, .02]).

Overall, our findings suggest that demographic variables have a negli-

gible effect on volunteer retention and turnover. Within the category

of Personal Circumstances, work–home conflict was moderately and

negatively associated with stay indicators and strongly and positively

associated with leave and turnover measures (stay: ρ=.30, CI =

[.53, .07], leave: ρ=.38, CI =[.16, .60], turnover: ρ=.34, CI =

[.24, .44]).

16 FORNER ET AL.

3.1.6 | Summary of meta-analytic results

Most of the research in this literature investigated desirable features

of volunteering, so their associations with retention (stay) and turn-

over (leave) criteria were generally positive and negative, respectively.

Most effects were in the moderate to strong range (Bosco et al.,

2015). An exception was demographic predictors, which yielded weak

to negligible relationships. Some of our observed associations showed

substantial heterogeneity, indicated by broad CV and high SD

values.

This led to some very broad 95% CIs in such instances, which meant

statistically significant results were less likely to emerge for these

associations. Indeed, an inspection of the 95% CIs (Table 2) suggests

64% (28/44) of the stay associations, 62% (18/29) of the leave associ-

ations and 74% (39/53) of the consolidated turnover associations

reached statistical significance. We turn to potential moderating fac-

tors of these associations in the following section.

3.2 |Exploratory moderator analyses

We conducted exploratory moderator analyses to investigate whether

the observed meta-analytic results were moderated by other factors.

We first did this for categorical moderators, type of volunteering (epi-

sodic and traditional) and study quality rating (good and low), testing

whether effects for sub-groups differed (Borenstein et al., 2009). We

only did this if there were at least three studies available for each level

of the moderator. Using this procedure, our results did not provide

any evidence of moderation for either study quality or type of volun-

teering, as all moderator sub-group CIs showed considerable overlap.

The second way we tested for moderation was to use meta-

regression to examine whether a continuous moderator (sample aver-

age age, percentage of the sample who were female) was related to

effect size magnitudes across studies. We concluded that moderation

was significant only when the CI of the regression coefficient did not

encompass zero. Complying with the Cochrane guidelines (Higgins &

Green, 2011), these analyses were only conducted when a minimum

of 10 effect sizes per variable were available. Using this procedure,

we found that the association between work engagement and turn-

over was moderated by the percentage of the sample who were

female: β=.004, SE =.002, CI =[.001, .007], indicating effect sizes

became stronger as the percentage of female participants decreased

in the samples. Similarly, we found that the association between job

satisfaction and turnover was moderated by sample age: β=.009,

SE =.004, CI =[.002, .017], indicating effect sizes were stronger in

younger volunteer workers.

3.3 |Publication bias

Our final step was to examine the possibility that our findings are

impacted by publication bias. We did this in several ways by perform-

ing a range of sensitivity analyses using the open-source Meta-Sen

software (see https://metasen.shinyapps.io/gen1/; Field et al., 2021).

Meta-Sen provides a range of outlier and publication bias assess-

ments, including contour-enhanced funnel plots (Peters et al., 2008),

trim and fill models (Duval & Tweedie, 2000), cumulative meta-

analyses by precision (Kepes et al., 2012) and precision-effect test-

precision effect estimate with standard error analyses (PET-PEESE;

Stanley & Doucouliagos, 2014). In accordance with recommendations

for such analyses, we used Meta-Sen only when there were at least

10 effect sizes available for a variable (Kepes et al., 2012). Thus, for

these analyses, we focused on the combined turnover estimate as it

involved higher k. All of these analyses can be found in our

supplemental file.

We first screened distributions for bias created by outliers, with

results generally suggesting outliers did not have a major influence on

the meta-analytic associations. Outliers were detected in 7/19 (37%)

of analyses performed, yet outlier-adjusted meta-analytic mean esti-

mates were often very similar in magnitude or sometimes stronger

than the unadjusted estimate. There were three exceptions to this.

Specifically, the volunteer participation time to turnover association

reduced from .14 to .10, the L&D to turnover association reduced

from .36 to .21 and the motives: career to turnover association

reduced from .13 to .05, suggesting that these associations were

likely influenced by outliers. Next, we screened output for evidence of

publication bias and small study effects. Overall, results showed some

evidence of possible upward bias for appreciation and recognition, job

demands (general), role ambiguity, burnout and task-significance, with

bias-adjusted estimates weaker than unadjusted estimates. While

these effects should be interpreted with more caution, it is important

to note that most of these distributions (i.e., job demands, apprecia-

tion and recognition, role ambiguity and burnout) were primarily com-

posed of unpublished literature. Thus, data suppression is unlikely to

be the cause of the detected bias. Interestingly, after accounting for

the effect of publication bias across most other effect size distribu-

tions, results suggested that the unadjusted estimates were underesti-

mated. That is, in these analyses, unadjusted estimates were weaker

than those adjusted for bias.

Taken together, our sensitivity analyses suggest that for most

meta-analytic associations in the current meta-analysis, publication

bias is not having a major influence on the results. Thus, we hence-

forth refer to the original meta-analytic estimates in interpreting the

results of our meta-analysis.

4|DISCUSSION

The objective of this research was to achieve a better understanding

of turnover phenomena in the volunteer work context. To this end,

we conducted the first comprehensive systematic review and meta-

analysis of the predictors of turnover for volunteers, with turnover

behaviours and intentions as indicators. Our research integrated and

synthesized 117 studies, encompassing 1104 effect sizes across

55 335 volunteer workers, to identify and quantify relationships

between turnover and the broad range of variables that have been

examined in the volunteer literature.

FORNER ET AL.17

FIGURE 2 Predictors of turnover amongst volunteer workers. Conceptual model derived from meta-analysis summarizing 1104 effect sizes across 55 335 volunteer workers. Note: This model

shows the mean true-score correlations (p) with turnover and a broad range of psychological, behavioural and contextual variables that have been tested in the volunteer organization literature.

Variables closer to centre of the figure have a stronger association with turnover, and correlations get progressively weaker towards the perimeter. Variables with weak or non-significant effects are

excluded from the model.

18 FORNER ET AL.

From our results, we have derived an integrative framework

delineating the predictors of turnover amongst volunteers, which we

present in Figure 2. The model illustrates which characteristics of

the volunteer job, characteristics of individual volunteers, motivation

and job attitudes, volunteer contributions and characteristics of the

volunteer organization are the strongest predictors of turnover in

volunteer organizations, along with estimates of the size of the cor-

rected correlation. The integrative model derived from our meta-

analysis provides a basis for theoretical development and guiding

future research. Moreover, the model offers an evidence-based

framework that can help organizations to minimize turnover rates in

their volunteer workforce.

A few variables stand out as the strongest predictors of turnover

amongst volunteers. These variables are shown in the centre of

Figure 2and include communication (ρ=.62), organizational sup-

port (ρ=.61), job satisfaction (ρ=.58), affective commitment

(ρ=.58), leader–member exchange (ρ=.55), engagement (ρ=-

.54), organizational commitment (ρ=.54), psychological contract

(ρ=.54), support from paid staff (ρ=.50), L&D (ρ=.46) and

autonomous motivation (ρ=.46). When considering the predic-

tors more broadly (level 2), favourable job attitudes (ρ=.58) and

commitment (ρ=.54) were particularly strong indicators of volun-

teers' turnover attitudes and behaviours, with a large body of

research evidence (k=90 and k=57, respectively). It is perhaps not

surprising that volunteers are more likely to stay if they are satisfied

with their role, engaged and committed to the organization. The

more interesting predictors are aspects of the volunteering experi-

ence that an organization can directly influence, such as the way vol-

unteer roles are designed, leadership, policy and practices. Of the

contextual variables examined, HRM practices (ρ=.45) and social

connections in the volunteer role (ρ=.43) had the strongest influ-

ence on turnover. People are more likely to continue volunteering if

the organization provides regular informative and supportive com-

munications and L&D opportunities. Volunteers also need roles

where they are supported by the organization, paid-staff and super-

visors and tasks that enable them to contribute productively

towards the organization's cause on issues that matter to them.

Demographic factors, age and gender had a moderating effect for

two of our strongest predictors of turnover: engagement and job sat-

isfaction. Specifically, we found the engagement–turnover relation-

ship is weaker when there is a higher proportion of female volunteers,

suggesting, work engagement is more important for turnover of male

volunteers, whereas female volunteers are more likely to stay even if

they are less engaged. We also found that the job satisfaction–

turnover relationship is weaker when there is a greater proportion of

older volunteers. This suggests that satisfaction with the volunteering

role is less important in influencing older volunteers' decisions to stay

or leave, relative to younger volunteers. Prior research has highlighted

key gender and age differences in volunteer workers (Einolf, 2010;

Hopkins & Dowell, 2022; Okun & Schultz, 2003), and our findings

suggest that these differences also impact the extent to which volun-

teers' attitudes towards their volunteer work influence their turnover

decisions.

4.1 |Neither a dominant nor a best theoretical

perspective

The results of the meta-analysis help in answering theoretical notions

and considerations of researchers. First, similar to studies of paid

employees, investigations of volunteer turnover are overwhelmingly

focused on a handful of variables, namely, job satisfaction and organi-

zational commitment (e.g., Rubenstein et al., 2018). Despite good evi-

dence that these two variables are strong predictors of turnover

intentions, many other predictors remain underexplored, albeit a

strong theoretical fit with the volunteering domain. For instance, psy-

chological need satisfaction (Deci & Ryan, 2008)

Predictors of turnover amongst volunteers: A systematic review and meta‐analysis

Abstract