Affect and attitudes in the gig economy: a meta-analysis of Amazon’s Mechanical Turk workers Open Access

Amazon’s Mechanical Turk (MTurk) platform is a freelance website that connects independent gig workers with requestors. MTurkers, as they are referred to throughout the paper, sign up on the platform and get matched with requestor projects that suit their interest and expertise. MTurk is regularly used as a data source in organizational behavior/human resources (OB/HR) research, from pilot testing newly developed scales to examining theoretical models in quantitative research and experimental studies (Mason & Suri, 2012). Aguinis, Villamor and Ramani (2021) note that between the years 2005 (when MTurk was launched) and 2020, 510 empirical papers published in management journals had used MTurk as their sample.

With studies increasingly using the MTurk platform as their data source, social sciences scholars have cautioned about data quality and validity issues when using MTurk samples (Thomas & Clifford, 2017). MTurk is known to have high-frequency participants with irregular employment status mostly from non-US populations (Harms & DeSimone, 2015), who use server farms and virtual private servers to mask their true location data to gain access to surveys that are intended for US participants only (DeCelles, Sonenshein, & King, 2020). This has led to a substantial number of participants misrepresenting theoretically relevant characteristics (like demographics and employment status) to meet eligibility criteria explicit in behavioral studies (Chandler & Paolacci, 2017). When using MTurk or other online data samples, Thomas & Clifford (2017) caution organizational researchers to use robust data cleaning procedures and subtle methods of conducting data quality checks like adding screening questions in their surveys. Others like Kennedy et al. (2020) advise that rather than directly relying on internet protocol (IP) addresses, researchers use latitude and longitude coordinates provided by the survey software to identify fraudulent participants. Even with an increasing presence of MTurk samples in behavioral management studies and such potential data issues, a deeper understanding of MTurkers as actual workers is largely lacking in management literature.

Meta-analyses have investigated potential differences in work attitudes for traditional employee samples and unconventional workforce samples like contingent workers (e.g. Wilkin, 2013) and online panel data (e.g. Walter, Seibert, Goering, & O’Boyle, 2019). Since scholars have recognized the importance of studying digital gig workers (Wu & Huang, 2024), we contend that examining the experiences of MTurkers is important. They are part of the burgeoning nontraditional gig workforce. Neither the MTurk platform (and its parent company Amazon Web Services) nor the gig requestor is contractually obligated to an MTurker (Bruckner, Frenzel, & Veit, 2020) beyond ensuring payment for a completed gig. Studies like ours that focus on the emotions and attitudes of these workers could provide insights into how workers keep the gig economy running efficiently even in the absence of gig employer support. Examining how MTurkers’ (positive and negative) affect impacts their (positive and negative) work attitudes (job satisfaction, organizational commitment and turnover intentions) could be a small step in understanding this gig workforce better.

In the present study, we extend previous meta-analytic reviews of affect and job attitudes, such as the one conducted by Thoresen, Kaplan, Barsky, Warren, and De Chermont (2003), by narrowing the focus to a specific type of worker sample, i.e. MTurker workers. We aim to meta-analyze the relationship between (positive and negative) affect and work attitudes (job satisfaction, organizational commitment and turnover intentions) specifically for MTurkers and examine the magnitudes and directions of these affect–attitude relationships for this worker sample. A second aim of this study is to observe and compare the MTurker affect–attitude correlations found in the current study with the affect–attitude correlations reported in prior studies for other types of samples. We did not conduct separate analyses to make this comparison, rather we gathered relevant published meta-analyses papers of conventional OB/HR samples and published papers that reported the concerned affect–attitude correlations for nonconventional samples like online panel data samples and made observational comparisons of our study findings with findings in those papers. As an additional step, we also compared our results with the effect sizes found in the cloud based MetaBUS platform, which is a resource for exploring meta-regressions of OB/HR variables (Bosco, Uggerslev, & Steel, 2017).

The term “affect” is used to denote a general category and subcategories of all feeling states that include emotions, moods and sentiments (Frijda, 1994). Positive and negative affectivity are primarily distinguished based on their valence (the degree to which affect is pleasant or unpleasant) and whether they refer to a disposition or response state (trait- and state-level). Specifically, forms of positive affect and negative affect can be framed according to the circumplex model, which views affect in terms of its positive or negative valence and its perceived arousal (Russell, 2003). Therefore, to conceptualize affect within the valence-arousal dimensions of the circumplex model, extant literature has profiled variables that adequately capture positive feeling states under positive affect and variables that capture negative feeling states under negative affect. For example, Wright & Staw (1999) and Lyubomirsky, King, & Diener (2005) used well-being as a positive descriptor of affect because well-being captures the hedonic side of affect; Cohen, Tyrrell, & Smith (1993) used a negative affect scale to capture psychological stress. Following this line of research, the current meta-analysis conceptualized well-being as a form of positive affect and stress as a form of negative affect. The complete lists of studies, sample sizes, independent/dependent variables and their correlations are provided in Table 1. (Note that some papers used distinct MTurk samples of different sizes for their various studies).

Affect-related positive or negative events in the workplace are associated with emotional responses in individuals, which influence their affective and attitudinal outcomes. Therefore, state-level affective reactions in employees generated by such events would likely impact attitudes like job satisfaction (Ashkanasy & Humphrey, 2011). It is expected that individuals with high positive affect experience higher affective commitment (Cropanzano, James, & Konovsky, 1993). For turnover intentions, because high-negative affect employees have a general inclination toward discontentment and tendency to experience negative moods, they are more likely to experience turnover intentions than high-positive affect employees (Judge, 1993).

State positive affect positively correlates with job satisfaction and organizational commitment and negatively correlates with turnover intentions, and state negative affect negatively associates with job satisfaction and organizational commitment and positively with turnover intentions (Thoresen et al., 2003).

Furthermore, individual studies that used various types of conventional samples have established relationships of positive/negative affect with work attitudes. As examples, positive affect and state job satisfaction were positively correlated for university employees (Judge & Ilies, 2004) and a sample of business owners (Delgado‐García, Rodríguez‐Escudero, & Martín‐Cruz, 2012); Positive affectivity was positively correlated with affective commitment and negatively correlated with turnover intentions for automobile industry workers (Lam & Liu, 2014); In Cropanzano et al. (1993) study, positive affect had a positive correlation with organizational commitment and negative correlation with turnover intentions for an employee sample, negative affect was negatively correlated with organizational commitment and positively correlated with turnover intentions.

We expect these relationships to be consistent for MTurkers. That said, MTurkers are part of the evolving digital gig economy, which is yet to be well-researched in OB/HR literature (Kuhn & Galloway, 2019). The experiences of gig economy workers may be akin to contingent workers, whose attitudinal and behavioral outcomes are known to be different due to their varying employment relationships (Wilkin, 2013). Furthermore, MTurkers have been used mainly as data samples in OB/HR research and have not been studied separately to understand their unique attitudinal experiences (Brawley & Pury, 2016). For these reasons, we are unable to hypothesize whether these affect–attitude relationships would be stronger or weaker for MTurkers:

As we meta-analyze MTurk studies for affectivity and work attitudes, we also examine whether certain design-related characteristics in selected studies, like MTurker sample age, work tenure and gender, moderate the affect–attitude relationship. Such demographic characteristics could have significant relationships with work attitudes and have been used in previous meta-analyses of contingent workers (Wilkin, 2013). For brevity and exploratory purposes, we only hypothesize age moderation for negative affect on a positive work attitude (i.e. job satisfaction) and age moderation for negative affect on a negative work attitude (i.e. turnover intentions).

With the average MTurker being under 35 years (Michel et al., 2018), and with negative affect decreasing as we age (Charles, Reynolds, & Gatz, 2001), it could be expected that younger MTurkers would exhibit stronger negative affect-job satisfaction and negative affect-turnover intentions associations. Older adults have shown significantly less variability in negative affect than young adults (Röcke, Li, & Smith, 2009). Older adults score lower in both frequency and intensity of negative affect. For example, in a longitudinal study with 7,000+ individuals, Windsor and Anstey (2010) found lower and more stable levels of negative affect with advancing age (over 40 years).

Most gig work, including MTurk, is low paid, temporary, offers no conventional employment benefits like health insurance or retirement and is considered precarious work. The gig economy is overrepresented by vulnerable working groups like the younger generation (Gen Z and Millennials) with low financial stability (Norwani, Ismail, Nasir, Yusof, & Jamaluddin, 2022). We contend that if youngsters experience higher frequency and intensity of negative affect, and knowing that negative affect impacts one’s job satisfaction, then it is likely that the negative affect-job satisfaction correlation would be stronger for younger MTurkers. Furthermore, an increasing number of young employees contemplate job-hopping or leaving their current job because they value attributes like pay level, job variety and personal growth in a job (Fan & DeVaro, 2019), which gigs like MTurk work do not offer. Hence, we anticipate the negative affect-turnover intentions correlation to be stronger for younger MTurkers:

To maximize the results of the intended search and to find all available studies on various topics in OB, HR and industrial organizational (I-O) psychology that used MTurk as a data sample, we adopted the following procedures. The start date for the search was set to the year 2005, as Amazon’s MTurk platform was launched then (Newman, Bavik, Mount, & Shao, 2021). The keywords, Mechanical Turk and MTurk, were set to be searched throughout an article including the Abstract. The following terms were searched in the Abstracts and body of the papers for the independent variables – positive affect, well-being, negative affect and stress and for the dependent variables – job satisfaction, organizational commitment, affective commitment, commitment, turnover intentions, quit and leave. The manual key word searches performed were consistent with how prior meta-analyses on work attitudes have conducted word searches (e.g. Miao, Humphrey, & Qian, 2017).

These electronic databases were searched – ABI/Inform, EBSCOhost, Google Scholar, JSTOR, ProQuest Dissertations and Theses and PsycNET (i.e. PsycInfo and PsycArticles). In addition, using the site Eigenfactor.org, we determined the top journals in business management, based on their Eigenfactor (EF) score and Article Influence (AI) score, and arrived at the following 16 journals. Academy of Management Journal (AMJ), Academy of Management Perspectives (AMP), Administrative Science Quarterly (ASQ), Human Resource Management Journal (HRMJ), Journal of Applied Psychology (JAP), Journal of Business and Psychology (JBP), Journal of Business Ethics (JBE), Journal of Management (JOM), Journal of Managerial Psychology (JMP), Journal of Management Studies (JMS), Journal of Occupational and Organizational Psychology (JOOP), Journal of Organizational Behavior (JOB), Organizational Behavior and Human Decision Processes (OBHDP), Journal of Vocational Behavior (JVB), Leadership Quarterly and Personnel Psychology (PPsych). We then individually searched the website of each journal to find relevant papers.

Next, following the procedures in previous meta-analytic papers in OB/HR (e.g. Miao et al., 2017), we searched the websites of two prominent academic conferences in the management field, namely the Academy of Management (AOM) and the Society for Industrial and Organizational Psychology (SIOP). Subsequently, authors of four papers from SIOP and one paper from AOM conference presentations were emailed to provide us with correlation matrices. Finally, to gather data where MTurk was used as the sample in unpublished studies, dissertations and theses, we reached out to researchers through the mailing lists and discussion boards of AOM for the OB and HR divisions.

The following criteria were applied for a study to be included in the current meta-analysis. First, only quantitative studies were considered and other inductive, experimental and qualitative studies that did not report correlations among study variables were excluded. Second, studies should have reported MTurk data independently, without combining with other online panel samples. Third, studies that did not specifically capture work attitudes but measured other facets of satisfaction or commitment (such as pay/career/family/life satisfaction and commitment toward identity) were excluded. Fourth, studies that analyzed both state and trait affect as independent variables were included because although the two are theoretically distinguishable, they are most often captured by the Positive and Negative Affect Schedule (PANAS) scale (Watson, Clark, & Tellegen, 1988) and yield very similar results. Fifth, papers that captured and analyzed any form of stress (e.g. job stress) were included. The majority of samples were from the USA.

Upon completion of the search process, 26 papers and 31 unique samples were discovered (sample count includes both positive/negative affect studies). The papers are included in the References section indicated by an asterisk (*) and the samples are listed in Table 1. The overall sample size for positive affect studies was 9,950 MTurkers and that for negative affect studies was 14,772 MTurkers.

For the independent variables, well-being was coded under positive affect and stress was coded under negative affect. The coding procedure for the dependent variables (namely job satisfaction, organizational commitment and turnover intentions) was straightforward and involved no judgment calls to be made by the authors.

Coding procedures for the moderator variable were as follows: We coded MTurker age as (35 years and younger = 0 or older than 35 years = 1). There were two reasons for splitting the moderator sample at this age data point. First, this age split is consistent with the dichotomization used in the meta-analysis of contingent workers (see Wilkin, 2013). Second, the median age in our current sample is close to 35 years and hence, dichotomizing the sample as under/over 35 years split our sample somewhat evenly to be able to perform a subgroup analysis. Specifically, for age moderation on negative affect-job satisfaction (H3a sample), the median age was 35.8 years (mean age = 36.8 years) and categorizing age as under/over 35 resulted in an even split of eight and nine papers, respectively, for the sample of 17 papers. For age moderation on negative affect-turnover intentions (H3b sample), the median age was 33.6 years (mean age = 35.4 years) and even though dichotomizing the sample as under/over 35 years yielded a somewhat uneven split of 10 and six papers, respectively, for the sample size of 16, both subgroups had sufficient power to warrant an analysis.

We used the meta-analytic procedure described in Hunter & Schmidt (1990) to analyze the effect sizes of affect–attitude relationships. We used the inter-item reliabilities for the independent variables and dependent variables to correct any measurement error reported correlations. For the studies where independent or dependent variable reliabilities were not reported, we imputed the missing reliability by using the mean of the reliabilities of that variable reported in other studies included in our meta-analysis. For single item measures (with no reliability measure), we corrected the reported correlation without using the reliability for that variable as that reliability was not reported. Instead, we calculated the average reliability measure (i.e. Cronbach’s alpha) of that variable, and consistent with Hunter & Schmidt (1990) recommendation, we imported that average into the studies that had a single item measure for that variable. In the current sample, there were two studies (namely, Cohen, Panter, & Turan, 2013 and Tucker & Jimmieson, 2017), which used a single item measure for turnover intentions and one study (namely, Reif et al., 2022), which used a single item measure for negative affect. In addition, one study (namely, Leavitt, Barnes, Watkins, & Wagner, 2019), used more than one measure for the same main variable (i.e. state job and person job satisfaction) and hence, reported more than one effect size. Following Hunter & Schmidt (1990) recommendation, we calculated the composite correlation using the corrected correlation for this study.

Following these steps allowed us to correct for measurement error for each reported correlation and subsequently, calculate the corrected sample-size-weighted mean correlation (ρ) and sample-size-weighted standard deviation of corrected mean correlations (SDρ). To test if the corrected mean correlations for the hypothesized relationships were statistically significant, we computed the lower and upper bounds of the corrected 95% confidence interval.

Subgroup analysis, which tests the between-group statistical significance of the differences in effect sizes of two or more groups (Borenstein, Hedges, Higgins, & Rothstein, 2021), was conducted to check the effects of the moderator. First, we dichotomized the (age) moderator using the categorization discussed previously, which generated two subsamples for each analysis. Next, consistent with previous meta-analyses in OB/HR and following Hunter & Schmidt (1990) approach, we used a z-test to check for statistical significance of the between-group difference in mean corrected correlations.

Overall results of the meta-analyses that tested the relationships among positive/negative affect and three work attitudes as well as the moderating effect of age are provided in Table 2.

The relationships between positive affect and (a) job satisfaction was positive and statistically significant (k = 7, n = 4,389, ρ = 0.6694), (b) organizational commitment was positive and statistically significant (k = 6, n = 2,290, ρ = 0.4802) and (c) turnover intentions was negative and statistically significant (k = 7, n = 3,271, ρ = −0.4319) because the corrected 95% confidence intervals spanned from 0.6693 to 0.6696 for job satisfaction, 0.4797 to 0.4807 for organizational commitment and −0.4323 to −0.4315 for turnover intentions and no interval included zero. Therefore, H1a–H1c were supported. The relationship between negative affect and (a) job satisfaction was negative and statistically significant (k = 17, n = 6,137, ρ = −0.3742), (b) organizational commitment was negative and statistically significant (k = 7, n = 2,875, ρ = −0.2948) and (c) turnover intentions was positive and statistically significant (k = 16, n = 5,760, ρ = 0.2157) because the corrected 95% confidence intervals spanned from −0.3744 to −0.3740 for job satisfaction, −0.2954 to −0.2942 for organizational commitment, and 0.2154–0.2160 for turnover intentions, and no interval included zero, hence, providing support for H2a–H2c. Effect sizes are considered statistically significant when corrected 95% confidence intervals exclude zero. We also note that the narrow confidence intervals recorded for these relationships indicate a highly precise estimate (Tan & Tan, 2010).

To test the moderating effect on the affect–attitude relationships, we conducted exploratory analyses using a demographic variable that is frequently reported in studies that use MTurk samples, i.e. age. As an exploratory analysis, the decision to test age moderation for negative affect instead of positive affect was based purely on our current paper sample. A count from Table 1 will reveal that there are 17 and 16 samples, respectively, for the negative affect ^[1] job satisfaction relationship and negative affect-turnover intentions relationship as opposed to just seven samples each for positive affect ^[2] job satisfaction and for positive affect-turnover intentions. We, therefore, selected relationships that had a large enough sample size to warrant sufficient power. For the dependent variables, we decided to explore a positive work attitude (i.e. job satisfaction) and a negative work attitude (i.e. turnover intentions).

Contrary to the hypothesized relationship in H3a, the analysis revealed that studies that contained mostly older MTurkers over 35 years (k = 9, n = 3,362, ρ = −0.3803) reported a stronger negative correlation (z = −5.6677, p < 0.001) than younger MTurkers under 35 years (k = 8, n = 2,775, ρ −0.3668) for negative affect-job satisfaction, thereby rendering H3a unsupported. H3b, on the other hand, was supported because studies with mostly younger MTurkers (k = 10, n = 3,560, ρ = 0.3928) reported a stronger positive correlation (z = −55.6923, p < 0.001) than studies with mostly older MTurkers (k = 6, n = 2,200, ρ = −0.0708) for negative affect-turnover intentions.

This research study aimed to attain two goals – first, it documented the effect sizes of affect and work attitudes in MTurks samples. We compiled relevant research papers and theses that used MTurk samples and conducted a standard meta-analysis of 31 different papers that captured the correlations for the concerned affect–attitude relationships. Second, this study attempts to compare the affect–attitude correlations of MTurk samples with non-MTurk samples. This is accomplished in the following section, wherein we searched for relevant papers published prior to the year 2005, i.e. before the MTurk platform was launched. Making such a comparison allows us to address the differences and similarities between MTurk contra non-MTurk samples.

We made two types of comparisons to attain the paper’s second goal. First, we compared our MTurk sample results with conventional samples used in social sciences research and second, we compared our MTurk sample results with panel data samples. The comparisons are reported in Table 3 and explained ahead.

To ensure that we were comparing our results with conventional samples that did not contain any MTurker data, we searched for meta-analytic reviews prior to the year 2005 both in extant literature and in a research database. As seen in Table 3, the current meta-analysis had higher effect sizes than Connolly & Viswesvaran’s (2000) meta-analysis for both positive affectivity and job satisfaction (0.6694 vs 0.49) as well as for negative affectivity and job satisfaction (−0.3742 vs −0.33). Similarly, our meta-analysis had higher effect sizes than Thoresen et al. (2003) meta-analysis for all relationships, with the exception of negative affect-turnover intentions wherein our study had a slightly lower effect size (0.2157 vs 0.28).

Next, we also compared effect sizes (prior to 2005) obtained from the metaBUS ^[4] database for positive affect-job satisfaction and negative affect-job satisfaction. Again, our correlations were stronger than those found in MetaBUS for positive affect (0.6694 vs 0.376) and negative affect relationships (−0.3742 vs −0.212).

For the second comparison – of our MTurk sample results with panel data samples, we relied on Walter et al. (2019) who captured the effect size estimates for online panel data (OPD) (like Qualtrics and Zoomerang). Our MTurk sample’s effect sizes were higher than the effect sizes of Walter et al. (2019) MTurk samples for all reported relationships, with the exception of negative affect-turnover intentions wherein the current study had a slightly lower effect size [0.2157 (current MTurk) vs 0.34 (Walter et al.’s MTurk) vs 0.42 (Walter et al.’s OPD)]. This is similar to what was observed when comparing our study with Thoresen et al. (2003).

The results of the current meta-analysis for MTurkers are consistent with those found for traditional workers and non-MTurk online data panels. We note two important observations from our analysis and the comparison of MTurk samples with non-MTurk samples.

First, the affect–attitude correlations for each work attitude were stronger for positive affect (ρ_JS = 0.6694; ρ_OC = 0.4802; ρ_TI = −0.4319) than for negative affect (ρ_JS = −0.3742; ρ_OC = −0.2948; ρ_TI = 0.2157). This could indicate the presence of a “halo effect” among MTurk participants, wherein their overarching impressions of positive concepts like positive affect and positive work attitudes are more strongly linked than negative concepts. MTurkers with positive affect may be subconsciously evaluating their work experiences, either with the MTurk platform or at their other jobs, more favorably, i.e. scoring their satisfaction and commitment high even if aspects of their job have been neutral. These high scores collectively could have led to inflated correlations in the positive affect-work attitude associations when meta-analyzed. We see this as a potential contributor to the low data quality issues with MTurk samples. Hence, a contribution of this meta-analysis lies in identifying the presence of a halo effect among MTurkers. On the other hand, the negative affect-work attitude associations may be weaker due to a “horn effect”, wherein MTurkers’ satisfaction, commitment or turnover intentions were not overly influenced by their negative affect.

Second, our MTurk sample’s affect–attitude correlations were stronger than the affect–attitude correlations reported in prior studies and meta-analyses of non-MTurk samples (i.e. Connolly & Viswesvaran’s (2000); Thoresen et al. (2003); metaBUS Database; and Walter et al. (2019) panel data sample of Qualtrics and Zoomerang workers). Our MTurk sample also had stronger affect–attitude correlations than Walter et al. (2019) MTurker sample. Such inflated effect sizes could be another indicator of MTurk samples’ poor data quality. Hence, a second contribution of this meta-analysis lies in cautioning researchers recruiting MTurk survey participants of potentially finding inflated effect sizes in their study results. In Eyal, David, Andrew, Zak, & Ekaterina (2021) study, while Prolific and CloudResearch samples showed high data quality on participant attention, comprehension, and honesty, MTurk samples revealed alarmingly low data quality on these factors even with data quality filters in place. Interestingly, the lowest data quality came from MTurk participants who reported MTurk as their main source of income but only spent a few hours on it per week. Kennedy et al. (2020) noted that such problems with MTurk data began surfacing in 2018, before which MTurk was both the most popular platform for survey recruitment for social science research and until then, had been providing high quality data.

The only relationship where the current study’s correlations were lower than that of other studies (namely Thoresen et al., 2003 and Walter et al., 2019) was that of negative affect-turnover intentions. The noninflated effect size could indicate that halo effect spillover did not occur for negative affect as it did for the positive affect relationships, meaning MTurkers’ negative affect did not overly influence self-ratings of their intentions to leave MTurk work. It is possible that environmental factors like social desirability bias and impression management play a role in MTurkers completing self-report surveys (Weigold, Weigold, Jang, & Thornton, 2022), due to which they may be less likely to publicly report quit intentions.

Beyond shedding light on the effect sizes of affect–attitude relationships for MTurkers, the current meta-analysis also helps researchers consider the importance of reconceptualizing work attitudes for the digital gig economy. As gig workers are not afforded the same privileges by gig employers as traditional employees are by their organization, we may need to rethink what concepts like a “job” and “organization”, and “commitment toward one’s organization” mean for MTurkers. Scholars have recognized that such reconceptualization of conventional constructs may be necessary. For example, in their study of MTurker job satisfaction and turnover, Brawley and Pury (2016) suggest that gig worker turnover should be defined in terms of requestor-specific turnover, wherein an MTurk worker might decide to quit working for a requestor rather than quit MTurk work entirely. Moreover, in their review exploring the nomological network of job attitude constructs, Woznyj, Banks, Whelpley, Batchelor, and Bosco (2022) found that scales for job satisfaction and job involvement do not accurately reflect the definition of the construct. They recommend future research to expand the measurement of job attitudes to improve precision. This recommendation is consistent with ours to reconceptualize work attitudes in the gig worker context. Another theoretical implication also lies in MTurker age acting as a moderating boundary condition. Following our findings that older MTurkers had a stronger negative-job satisfaction association and younger MTurkers had a stronger negative affect-turnover intentions association, it would be interesting to investigate the variations in affect–attitude associations among Gen-Z, Millennial and Gen-X MTurkers.

Furthermore, we encourage management studies that pose research questions to compare the experiences of specific online panel data samples like MTurkers, Prolific or Qualtrics workers and test previously established relationships in OB/HR literature in these new work contexts. In the current papers we analyzed, even though study subjects were MTurkers who do not work in a traditional organizational setting, no study had modified the established scales of work attitudes to better fit the MTurk context. Hence, we see opportunities for future research to modify established scales of conventional variables like organizational commitment and turnover intentions to contextualize them for digital gig workers.

Finally, we contend that a follow up research study to ours could compare MTurker affect–attitude relationship in two contexts. First, pre- and post-2018, which is when data quality issues with MTurk began emerging. And second, pre- and post-2020 pandemic, to draw implications of the potential changes in this workforce. We note the skewness in sample size of papers that were published before and after the 2020 pandemic. As seen in Table 1, 17 studies were published by 2020 and nine studies were published since 2021. This reduction in published papers that used MTurk samples, as observed during our data collection, could potentially be attributed to two factors. One, perhaps because researchers are using MTurk less postpandemic and switching to other, more reliable panel data sources like Prolific or two, it is possible that some studies since 2021 are still in the review process of publication. By conducting more such comparative studies and meta-analyses of MTurkers, we will get better clarity on whether researchers are continuing to use MTurk and whether the MTurk platform should continue to be used as a data panel. If MTurk is still a viable data source, then it is important to understand for which behavioral management concepts/relationships MTurkers may be a good fit to sample.

The current meta-analysis tested whether positive/negative affect correspondingly influenced positive/negative work attitudes in MTurkers and revealed that the affect–attitude correlations of MTurk workers are consistent with those of traditional workers. More meta-studies and reviews of MTurkers are needed to understand whether their experiences differ by demographic characteristics like worker age, gender, educational level and personality traits. It is important to test previously established relationships in OB/HR literature in new contexts, both in individual studies and meta-analytically, to advance the field.