Be Careful How You Treat Your Coworkers: The Reciprocal Relationship between Ethnic Outgroup Coworkers’ Reactions to Voice and Ethnic Majority Employees’ Attitudes regarding Immigrant Entitlements

Abstract We study the reciprocal relationship between interethnic interactions among coworkers and native (Dutch) employees’ attitudes regarding immigrant entitlements. Building on contact theory, we hypothesize that voice support by ethnic outgroup coworkers leads to more favorable, while voice suppression leads to less favorable attitudes regarding immigrant entitlements. Furthermore, we examine potential reciprocal effects. The hypotheses are tested using a three-wave panel survey of native Dutch respondents. Findings indicate a negative effect of voice suppression by ethnic outgroup coworkers on attitudes regarding immigrant entitlements, implying that workplace interethnic contact can shape political attitudes. Moreover, findings indicate that the less favorable native employees’ attitudes regarding immigrant entitlements are, the more likely they are to subsequently experience suppression by ethnic outgroup coworkers.


Introduction
Throughout the end of the 20th and beginning of the 21st century, Western European societies underwent rapid changes in ethnic composition due to increased globalization and migration. Alongside these developments, the relevance of studying ethno-cultural diversity and related social issues has been steadily increasing. According to Hollifield (1997, p. 30), "few issues have had a greater impact on the politics and society of contemporary Western Europe than immigration." Therefore, majority groups' attitudes regarding immigrant entitlements (and related political decisions) currently represent a crucial political issue for Western European democracies (Thomsen, 2012).
The ever-more diverse Dutch society is no exception in this sense (Brouwer & Boros, 2010;Schaafsma, 2008). In fact, the Netherlands is known as an example of a multicultural society, with almost one in five inhabitants coming from an immigrant background (Indexmundi, 2021). However, since the early 1990s, the issue of immigration has become more salient and the social climate has started to shift from a preference for multiculturalism to a penchant for monoculturalism, as reflected in the rise of anti-immigration parties (Van Heerden et al., 2014). Thus, the Netherlands serves as a quintessential social context for studying factors that affect ethnic majority group members' support for ethnic minority rights. In this paper, we approach the study of majority groups' support for immigrant entitlements within the context of workplace interactions.
The workplace is considered by some to be a small-scale democracy, where employees develop political skills and attitudes that they later apply outside of the workplace (Greenberg et al., 1996). Therefore, ethnically diverse workplaces provide an important social context for forming attitudes regarding immigrant entitlements. The goal of this paper is to investigate relationships between ethnic majority employees' attitudes about immigrant entitlements and their contact (i.e. social interactions) with ethnic outgroup coworkers at the workplace. Specifically, the social interactions we examine are support and suppression by ethnic outgroup coworkers in reaction to native Dutch employees' voices. We focus on voice support and suppression by ethnic outgroup coworkers because these interactions can be consequential in the workplace context, and as such are likely to be salient in the minds of native Dutch employees who are on the receiving end of support or suppression. Thus, we study how ethnic outgroup coworkers react (support and suppression) to Dutch employees' expressions of workplace discontent (voice), and how that is related to native Dutch employees' attitudes regarding immigrant entitlements.
The research question concerns the relationship between native Dutch employees' experiences of workplace voice support and suppression by ethnic outgroup coworkers on the one hand and their attitudes regarding immigrant entitlements on the other. To illustrate the relevance of this research question, we reflect on a real-life example from the Netherlands: in 2017, the employees of the distribution centers of the Dutch supermarket chain Jumbo went on strike to voice their discontent with the working conditions. In response to that, the chain flew in Polish flex workers to cover the shifts, and effectively act as strike breakers ( Van der Velden, 2017). Several potential effects are of interest in this anecdote: how does strike-breaking by Polish workers, which can be considered suppression of voice by ethnic outgroup coworkers, affect native Dutch employees' attitudes regarding immigrant entitlements? Conversely, how may Dutch employees' attitudes regarding immigrant entitlements affect the likelihood that their Polish coworkers will show solidarity with their workplace voice (i.e. support or suppress them)? We predict that native Dutch employees' experiences of voice suppression by ethnic outgroup coworkers result in the Dutch employees' less favorable attitudes toward immigrant entitlements, while them experiencing voice support by ethnic outgroup coworkers results in more favorable attitudes regarding immigrant entitlements. At the same time, we expect that the more favorable attitudes on immigrant entitlements ethnic majority employees have, the more likely they are to receive support from their ethnic outgroup coworkers, while the less favorable attitudes on immigrant entitlements they have, the more likely they are to receive suppression from their ethnic outgroup coworkers.
As a theoretical framework, this paper utilizes contact theory literature, which studies the antecedents, characteristics, and consequences of social interactions between members of differing groups. This study contributes to the contact literature in four important ways. The first contribution lies in studying workplace interethnic contact, meaning social interactions that occur between members of different ethnic groups at the workplace. Interethnic contact has been relatively understudied within the workplace context even though it provides the benefit of being a "no-choice" context (Pagotto et al., 2010, p. 318). Namely, while self-selection represents a potential bias in most contact research, this is less of an issue in the workplace context, as interethnic contact in the workplace is not necessarily voluntary (Kokkonen et al., 2015;Savelkoul et al., 2015).
The second contribution of this study lies in simultaneously studying positive and negative workplace contact, in the form of voice support and suppression. Accounting for both positive and negative contact is important because negative contact has been relatively understudied, creating a positivity bias in the literature (Graf & Paolini, 2016;. Thirdly, this study predicts a substantial outcome: attitudes regarding immigrant entitlements. While there is plenty of evidence that intergroup contact reduces prejudice 1 (Pettigrew & Tropp, 2006), less is known about how contact affects more substantial outcomes, such as political attitudes and voting (Thomsen, 2012).
The fourth and final contribution of the present work lies in the use of a three-wave longitudinal design, thereby reducing the risks of the common drawbacks of cross-sectional research designs. This is especially important in the context of contact theory research, given that much of the controversy on the supportive evidence for this theory points to problems with self-selection bias (cf. Manevska et al., 2018).

Theory and hypotheses
Effect of workplace contact on ethnic prejudice According to the results of 73 studies on workplace contact included in the meta-analysis by Pettigrew and Tropp (2006), workplace intergroup contact has a negative effect on prejudice. However, even though the frequency of workplace interethnic contact seems to have an overall diminishing effect on ethnic prejudice, these relationships are not always straight forward. Some findings suggest that increasing diversity (and thereby potentially increasing contact frequency) could in fact increase prejudice depending on contact characteristics other than frequency (Pettigrew et al., 2010;Vezzali & Capozza, 2011).
In particular, contact quality has been recognized as an important characteristic of contact that determines the effect of contact frequency (Tropp et al., 2017). The affective quality of contact is more important than its frequency in determining contact outcomes (Bornman, 2016;Brouwer & Boros, 2010;Miller et al., 2004;Pettigrew & Tropp, 2008). Contact quality describes to what extent contact is subjectively experienced as positive or negative (Lolliot et al, 2015). Research comparing the effects of positive and negative contact suggests that negative contact might increase prejudice to a greater extent than positive contact decreases prejudice (Barlow et al., 2012;Paolini et al., 2014). Positive intergroup contact merely seems to be more common, which explains the finding that contact (of unspecified quality) is generally linked to less prejudice (Graf et al., 2014).
Since the workplace context can facilitate both positive and negative contact, accounting for contact quality is especially important in this case. On the one hand, the workplace provides opportunities for positive intergroup contact, because work often requires cooperation and shared goals. Furthermore, work teams tend to emphasize the importance of shared group identity and interdependence (cf. De Souza Briggs, 2007;Goldschmidt et al., 2017). From this perspective, the workplace can be seen as providing some of the key conditions described by Allport (1979Allport ( [1954), and hence as conducive to the formation of positive interethnic contact (Kokkonen et al., 2015). On the other hand, workplaces are also often competitive and hierarchical, which may create the conditions for negative contact, as interactions characterized by competition and unequal status are more likely to be experienced as unpleasant. Indeed, from a group conflict perspective, competition over scarce goods is identified as a central element of interethnic conflict (e.g., Olzak, 1992). As such, it is also plausible that the workplace is conducive to negative interethnic contact. Next, we consider specific examples of positive and negative workplace contact, namely support and suppression of voice by ethnic outgroup coworkers.

Voice support and suppression by ethnic outgroup coworkers as positive and negative contact
While several large-scale empirical studies on workplace contact exist (Eisnecker, 2019;Escandell & Ceobanu, 2009;Freitag & Kijewski, 2017;Klein et al., 2019;Kokkonen et al., 2014Kokkonen et al., , 2015Sønderskov & Thomsen, 2015;Thomsen, 2012), to our knowledge, only  studied positive and negative workplace intergroup contact. They find positive interethnic workplace contact to be associated with positive attitudes toward ethnic outgroups, while negative interethnic workplace contact is associated with negative attitudes toward ethnic outgroups. Due to the cross-sectional nature of the  study, reversed causality cannot be ruled out-perhaps having positive attitudes toward outgroups causes more frequent positive contact experiences, and negative attitudes cause more frequent negative contact experiences. Therefore, the present study examines the potential bidirectional causal effects between positive and negative interethnic contact in the workplace context and attitudes on immigrant entitlements, by testing two sets of opposing (but complementary) causal hypotheses. To study these effects, we focus on the suppression and support of workplace voice by ethnic minority coworkers as examples of negative and positive intergroup contact, respectively.
According to Stanojevic et al. (2020), workplace voice is defined as "any activity of individual employees, groups of employees or their representatives, aimed at improving either personal work conditions or the work conditions of an entire group." Due to the hierarchical and potentially confrontational nature of voicing discontent to a supervisor, coworkers' reactions to voice are important social cues that signal solidarity or a lack thereof. As such, coworkers' reactions are of 2-fold importance: first, because coworkers' reactions affect employees' subjective experiences of workplace voice and perceptions of organizational climate; and second, because coworkers' reactions can affect the outcome of the voiced issue by influencing the probability of it being solved. As such, voice suppression and support by coworkers are likely to be an impactful form of workplace contact.
Coworkers can react to workplace voice in two distinct ways-by supporting employees who have voiced discontent (by providing emotional support or help with goal attainment) or by suppressing them (by discouraging them or obstructing goal attainment). 2 When these reactions to voice come from ethnic outgroup coworkers, they may color the perception of these coworkers. According to Pettigrew (2008), being helped by a "foreigner" can be regarded as positive interethnic contact, and being pestered by a "foreigner" can be viewed as negative interethnic contact. Thus, receiving voice support (help) from ethnic outgroup coworkers is likely to be experienced as positive interethnic contact, while receiving suppression (pestering) from ethnic outgroup coworkers is likely to be experienced as negative interethnic contact.

Effect of support and suppression by ethnic outgroup coworkers on native employees' attitudes regarding immigrant entitlements
To understand how support and suppression of workplace voice by ethnic outgroup coworkers might affect attitudes toward immigrant entitlements, we must consider the process of social categorization. Although the mechanisms of social categorization are not empirically tested in the current study, they are considered the underlying theoretical mechanism bridging the effects of ethnic outgroup support and suppression on attitudes regarding immigrant entitlements. Humans are able to automatically (without conscious effort or control) sort people into meaningful categories (Brewer, 1996). As a consequence of such categorization of self and others into groups, people spontaneously distinguish between the group containing oneself-the in-group, and other groups-the outgroups . The capacity for social categorization is functional, as it facilitates managing unpredictable social situations (Liberman et al., 2017). Although crucial, this ability comes with some perceptual biases. Because people naturally strive to maintain a positive self-image, a positive in-group bias occurs . This positive in-group bias is not necessarily paired with negative out-group bias and prejudice, but does provide a fertile ground for it (Gaertner et al., 1993). As positive in-group biases guide perception, beliefs and behaviors, they solidify negative affect toward outgroup members. This way, social categorization can form and maintain prejudice, including ethnic prejudice in the workplace (James et al., 1994). It might therefore come as no surprise that ethnic minorities often suffer from ethnic prejudice at work (Emerson & Murphy, 2014;Nunez-Smith et al., 2009;Van Laer & Janssens, 2011).
However, social categorization is dynamic-people are able to reorganize their own existing social categories, change the categorization criteria, or abandon certain social categories all together (Dovidio et al., 2009). One such dynamic cognitive mechanism of social categorization often mentioned as underlying the effect of intergroup contact on intergroup attitudes is recategorization. Recategorization refers to redrawing social category boundaries in such a way that former out-group members are included in a new, superordinate in-group (Brown & Hewstone, 2005). According to the common in-group identity model, recategorization mediates the effect between contact and its emotional, cognitive, and behavioral consequences (Gaertner et al., 1993(Gaertner et al., , 1996. Several affective reactions are likely to occur following recategorization (Dovidio et al., 2004). Studying such affective responses, most prominently intergroup empathy and anxiety, is becoming central in contact research (Brown & Hewstone, 2005). Intergroup empathy refers to the ability to feel the emotion experienced by an out-group member (Vanman, 2016). Intergroup anxiety is an uncomfortable emotion that stems from the expectation of negative consequences of intergroup contact (Stephan & Stephan, 1985). Apart from cross-sectional research that identifies the affective responses of intergroup empathy and anxiety as mediators between contact and intergroup attitudes (Pettigrew, 1997;Pettigrew & Tropp, 2008), this causal chain is also supported by longitudinal research findings (Vezzali & Capozza, 2011).
A study by Capozza et al. (2010) conducted in an Italian workplace further elaborates on the steps of the causal chain between intergroup contact and prejudice. This study reports that contact-induced a common in-group identity (recategorization), which increased empathy and decreased anxiety toward known out-group members. Finally, increased empathy and decreased anxiety toward known outgroup members in the workplace are generalized to unknown members of the same outgroup. Moreover, other studies found that contact also reduces prejudice toward out-groups not represented through contact, i.e. out-groups one did not have an experience of contact with (Pettigrew, 2009;Pettigrew & Tropp, 2006). Relying on these findings, we predict the effect of ethnic outgroup coworkers' support and suppression on attitudes regarding immigrant entitlements.
Following the common in-group identity model developed by Gaertner et al. (2000), we assume that the described instances of workplace voice support and suppression by ethnic outgroup coworkers affect the process of social categorization by providing new associations with ethnic outgroup members. In cases of support from ethnic outgroup coworkers, recategorization is likely to happen, as support signals cooperation and solidarity, and increases the likelihood of attaining a common group goal. Therefore, it is plausible that supported native employees would shift their social cognitive scheme from "us (the Dutch) and them (the 'foreigners')" to "we (the workers)." This is followed by increased intergroup empathy and decreased intergroup anxiety toward those ethnic outgroup members in the workplace, which generalizes to increased empathy and decreased anxiety toward members of ethnic outgroups outside of the workplace. Therefore, native employees experiencing voice support from ethnic outgroup coworkers are better able to understand the perspectives of ethnic outgroups and to expect fewer negative consequences of future contact with them, causing them to adopt more favorable attitudes regarding immigrant entitlements.
Conversely, suppression by ethnic outgroup coworkers may hinder the goal attainment of employees voicing an issue. For native Dutch employees who voiced, this is an unwanted outcome. Therefore, it is adaptive for them to take note of the differing group membership of coworkers who suppress their voice, to adapt to future interactions with members of the outgroup. Thus, voice suppression could in this case make the differing ethnic group membership even more salient. This reduces intergroup empathy and induces anxiety toward ethnic outgroup members performing suppression, which generalizes to reduced empathy and increased anxiety toward ethnic outgroup members in general, finally creating less favorable attitudes regarding immigrant entitlements. From these theoretical assumptions, the following hypotheses are tested (as shown in Figure 1): H1a: Experiencing support by ethnic outgroup coworkers increases native Dutch employees' subsequent support for immigrant entitlements.
H1b: Experiencing suppression by ethnic outgroup coworkers decreases native Dutch employees' subsequent support for immigrant entitlements.

Effect of native employees' attitudes regarding immigrant entitlements on their likelihood of experiencing coworker support and suppression by ethnic outgroup coworkers
Of course, the causal effect of the opposite direction is also plausible and should not be neglected, especially since the self-selection of participants into situations of contact is a well-known caveat of contact research (Smith, 1994;Wilson, 1996). The workplace context leaves less room for selection into contact situations than, for example, the context of the neighborhood or voluntary organizations. This is because the choice of the workplace is limited, and the information about who one might have contact with at a potential workplace is often unavailable when looking for a job. However, it is still possible that self-selection happens, albeit on a more subtle level than explicitly choosing a workplace with more or less opportunities for contact. This type of selfselection may relate to the quality of contact, rather than its quantity (which is, in the context of the workplace, less affected by personal factors as it is not completely voluntary). Namely, it is possible that (unconscious) selfselection affects workplace interactions, whereby native employees' existing attitudes toward ethnic outgroups make them more or less likely to experience certain kinds of interactions (positive or negative contact) with ethnic outgroup coworkers. We assume this happens because native employees' behaviors toward ethnic outgroup coworkers are likely to reflect their attitudes regarding immigrant entitlements and attitudes regarding ethnic outgroups in general.
Despite not always being perfectly concordant, attitudes are known to affect behavior (Ajzen, 1989;Ajzen & Fishbein, 1977;Homer & Kahle, 1988). Therefore, it is plausible that attitudes on immigrant entitlements, together with related attitudes (i.e. beliefs about immigrants themselves or about ethnic outgroups in general) affect employees' behavior at the workplace. Specifically, these attitudes may govern the way that native Dutch employees interact with ethnic outgroup members in everyday social interactions at work. In this case, native Dutch employees with favorable attitudes toward immigrant entitlements may exhibit behavioral expressions of those attitudes in their interactions with ethnic outgroup coworkers by, for example, being more friendly or expressing empathy. On the other hand, Dutch employees with less favorable attitudes toward immigrant entitlements may behave more reserved in relation to ethnic outgroup coworkers. These, albeit perhaps subtle, behavioral variations may be perceived by ethnic outgroup coworkers and, consciously or subconsciously, become a guideline for how they in turn interact with the native employees in the future. Therefore, when a native Dutch employee experiences a workplace issue and attempts to voice discontent, ethnic outgroup coworkers may rely, among other things, on past experiences with them to decide whether to support or suppress their voice. Thus, as shown in Figure 1, we hypothesize: H2a: The more favorable attitudes regarding immigrant entitlements native employees hold, the greater their likelihood of subsequently experiencing support from ethnic outgroup coworkers.
H2b: The less favorable attitudes regarding immigrant entitlements native employees hold, the greater their likelihood of subsequently experiencing suppression by ethnic outgroup coworkers.

Data
The data used for this study were gathered using a panel survey (Akkerman et al., 2017(Akkerman et al., , 2019, consisting of three waves of data collection about one year apart. The sampling was conducted by a professional survey company Kantar Public, starting from a panel of 135,000 respondents in the Netherlands (TNS NIPObase). From this panel, 12,013 participants were approached, out of whom 7,599 filled out the survey, amounting to a 64% response rate. Respondents were selected to ensure the representativeness of the Dutch labor force, aged between 15 and 67. Of the 7,599 respondents participating in the first round of data collection, 6,008 took the survey in the second, while 4,855 took the survey in the third wave of data collection. This amounts to a panel attrition rate of 21% in the first, and 36% in the second wave. Additional constraints were placed when selecting the sample analyzed due to choices necessary for valid support and suppression measures, as explained in the following section.

Measures
To assess attitudes regarding immigrant entitlements, respondents were asked whether they agree or disagree with the following statements: People from the following groups should be able to … 1. Come and live in the Netherlands 2. Vote in local elections 3. Vote in national elections 4. Become a candidate in an election 5. Be eligible for the same labor rights as native Dutch citizens 6. Be eligible for the same social welfare benefits as native Dutch citizens The respondents were asked to express their (dis)agreement with the above statements for three ethnic out-groups: Eastern European, Turkish/Moroccan, and other non-Western. Attitudes regarding immigrant entitlements for these three out-groups make up three manifest variables. They are operationalized as a sum score of binary responses on the six items presented above. The total score for each manifest variable thus ranges from zero to six with higher scores reflecting more support for immigrant entitlements (i.e., more favorable attitudes regarding immigrant entitlements). We assume these three manifest variables covering attitudes regarding immigrant entitlements of (all three possible) groups of non-Western immigrants make up a latent construct-immigrant entitlements of non-Western immigrants-which is to be empirically tested in the results section.
To assess support and suppression by ethnic outgroup coworkers, respondents were first asked if they experienced a workplace issue and voiced it. If they answered these questions affirmatively, they were then asked whether they agree or disagree with the following statements: Support At the first data collection point (T1), respondents were asked to recall whether they had experienced the above-listed suppression/support instances over the past three years, while at the second and third data collection points (T2, T3), they were asked to recall whether they had experienced such instances of suppression/support since filling out the previous questionnaire (over the preceding year).
In contact research, presenting participants with such concrete examples of positive or negative incidents with ethnic outgroup members is referred to as the specific incidents approach to assessing contact valence (Laurence & Bentley, 2018;. This approach has several advantages over the more common overall-valence approach, where participants are asked to retrospectively assess the valence of contact. First, the specific incidents approach allows for the fact that a person may experience both positive and negative contact. Second, participants are not forced to assign either positive or negative valence to contact experiences they might have experienced as neutral. Finally, the specific incidents approach is preferred with regard to social desirability issues, as participants might be inclined to report interethnic contact experiences more positively if asked to assess overall valence. Thus, in the present study, we presented respondents with specific incidents of ethnic outgroup coworkers' support and suppression that represent positive and negative interethnic contact, respectively. Additionally, respondents were given the chance to describe their own experiences of support or suppression when not covered by the items (as shown above), which were later coded as positive or negative by the researchers.
The presented support and suppression statements referred to coworkers of four ethnic groups, namely Dutch, Eastern European, Turkish/Moroccan, and other non-Dutch. The statements were presented separately for these four groups, and only to participants who have previously indicated that they work with coworkers of the respective ethnicities. Of these groups, the Eastern European and Turkish Moroccan groups can be regarded as ethnic outgroups of native Dutch people. 3 Ethnic outgroup support and suppression were operationalized as binary variables indicating whether or not the participant experienced any type of support or suppression from coworkers of either of these ethnic outgroups. 4 Thus, respondents experiencing any form of support from Eastern European and/or Turkish/Moroccan coworkers would have a score of one for ethnic outgroup support and a score of zero when they have not experienced any type of support from either outgroup. Similarly, respondents who have experienced any form of suppression from Eastern European and/or Turkish/ Moroccan coworkers would have a score of one on ethnic outgroup suppression and a score of zero if they have not experienced any form of suppression from either outgroup.
To compute a valid measure of suppression/support, additional constraints were imposed. Specifically, participants who did not have an issue and did not voice are coded as missing (rather than zero) on suppression and support measures. This is important because the tendency to experience issues at the workplace and voice them might also be correlated with immigrant entitlements (for example, people with a less agreeable personality might be more prone to perceive issues, voice them, and also to have less favorable attitudes on immigrant entitlements). Therefore, including participants who did not have an issue or did not voice it in the support and suppression measures would produce conflated effects, which would be impossible to distinguish from the effects of support and suppression.
Additionally, only native Dutch participants who work (or have worked over the previous three years) in interethnic environments where they might have experienced interethnic contact with Eastern European or Turkish/Moroccan coworkers were included in the analyses. This was an important selection to make because mere exposure to people of different ethnic backgrounds can affect prejudice  even without having experienced support or suppression. This way, it is ensured that the measures capture only suppression and support, rather than the propensity to have an issue and voice it, or exposure to ethnic outgroups at the workplace. This selection resulted in a restricted sample to be analyzed but was necessary for the validity of measures in this longitudinal design.

Data analysis
Cross-lagged panel model (CLPM) analysis using lavaan package in R was used to test the hypotheses (Rosseel, 2012). The diagonally weighted least squares (DWLS) method was used for path estimation. The DWLS method does not assume multivariate normality and is thus the appropriate estimation method here since the independent variables are binary, thereby disenabling multivariate normality. First, a three-wave CLPM 5 (N ¼ 685) is estimated. However, due to the combination of panel attrition and list-wise deletion, this model is tested on a rather particular sample, namely employees who experienced a workplace issue and voiced it in three successive waves of data collection. This might be a selection consisting of workers who are particularly expressive about workrelated problems or work in specific problematic workplaces. Therefore, we additionally assess two twowave models, which enable studying more inclusive samples: a first-to-second wave model (N ¼ 1,294) and a second-to-third wave model (N ¼ 1,008).

Results
The results section is organized as follows. First, we present descriptive statistics (Table 1) and correlations ( Table 2) between manifest variables used in the models. Second, a confirmatory factor analysis is performed to test whether the three manifest variables (immigrant entitlements of Eastern European, Turkish/Moroccan, and other non-Western immigrants) indicate a latent dependent variable (immigrant entitlements of non-Western immigrants). Finally, a three-wave CLPM is estimated (Figure 3), as well as two two-wave models (Figures 4 and 5). Table 1 reveal that support from ethnic outgroup coworkers is more common than suppression from ethnic outgroup coworkers, which is in line with Graf et al. (2014) findings that positive intergroup contact occurs more often than its negative form. Furthermore, the correlation matrix presented in Table 2 shows indications of negative correlations between suppression and immigrant entitlements, and positive correlations between support and immigrant entitlements, which provides a further indication to explore the empirical tenability of our hypotheses.

The descriptive indicators presented in
Before fitting a structural model, it is necessary to assess the fit of the measurement model using confirmatory factor analysis. As shown in Table 2, correlations between attitudes regarding immigrant entitlements of different out-groups at the same time point are high (r>.75), thereby providing grounds for testing whether they form a latent construct. For the measurement model, which consists of the immigrant entitlements latent variable measured at three time points, the MLR estimator (maximum likelihood estimator with robust standard errors and scaled test statistics) was used. Using a robust maximum likelihood estimator, in this case, is recommended since the manifest variables are continuous, but not necessarily normally distributed. The factor loadings of the measurement model are presented in Figure 2, and the fit indices are provided in Table 3.
The measurement model is calculated on the restricted sample of valid cases in all three waves so all three latent variables could be fitted in the same measurement model. As shown in Table 3, fit indices indicate a good fit: SRMR and RMSEA are lower than .08, and CFI and TLI are above .95 (fit index threshold guidelines according to Kline, 1998). Moreover, reliability analysis showed that the Cronbach's Alpha for this scale over three waves amounts to .97, .97, and .93, respectively, which indicates that the items form a reliable scale. Therefore, these latent variables can be used in the panel models. Figure 3 shows the three-wave CLPM effect estimates. The key paths for assessing causal hypotheses in cross-lagged panel models are lagged effects (effects over time). These paths are marked blue (for H1a and H1b) and red (for H2a and H2b). The paths between the same variables in consecutive waves (marked grey) signify autocorrelation. Autocorrelation serves as control to partial out the constant sources of variability between waves (potentially coming from sources, such as age, gender, education levels, and personality traits). However, these autocorrelation paths are not key for the interpretation of results, and thus for simplicity, their estimates are not shown in the path diagrams. This is the case with cross-sectional paths within waves as well (also marked grey). However, within the current study design (due to retrospective measures of support and suppression) they are informative, and as such their estimates are shown in the two-wave models.
The fit indices shown in Table 4 indicate a good fit. Therefore, we proceed with interpreting the path coefficients. We only present coefficients of which the analyses indicate that these are likely to differ from zero. Figure 3 shows the results for the three-wave model. In this model, we find indications for the hypothesized negative effect of suppression by ethnic outgroup coworkers on subsequently measured immigrant entitlements (H1b). However, this is only true for the effect of suppression in the first wave on support for immigrant entitlements in the second wave (b ¼ À.07, b ¼ À.85, SE ¼ .36). Furthermore, the threewave model provides indications for the hypothesized negative effect of favorable attitudes regarding immigrant entitlements on the likelihood of subsequently experiencing suppression by ethnic outgroup coworkers (H2b). However, this was only found between wave 2 and wave 3 (b ¼ À.11, b ¼ À.01, SE ¼ .00).
Turning to the two-wave models, which are less restrictive in terms of sample composition, we find similar indications for the hypothesized effects. Figures 4 and 5 show that there are no indications for the effect of support on immigrant entitlements (H1a), nor for the reversed effects (H2a), which is in line with the three-wave model. Like in the threewave model, we only find indications for the negative effect of suppression on immigrant entitlements (H1b) between wave 1 and wave 2 (b ¼ À.04, b ¼ À.20, SE ¼ .24). Furthermore, we find indications for the negative effect of support for immigrant entitlements on subsequent suppression (H2b) for both two-wave models, namely between T1 and T2  (b ¼ À.04, b ¼ À.01, SE ¼ .00; Figure 4), as well as T2 and T3 (b ¼ À.07, b ¼ À.01, SE ¼ .00; Figure 5). Thus, in the more inclusive sample, we do find indications for such effects, which were not found in the three-wave model on the more restricted sample. 6 Summarizing the results from the three-wave model, which is tested on a more restrictive and   specific sample (respondents who had an issue and voiced it in all three waves) with the two-wave analyses on broader samples, the overall pattern that arises is 2-fold: on one hand, the analysis finds indications for both the negative effect of suppression by ethnic outgroup coworkers on subsequent immigrant entitlements-H1b (although partially-only between the first and second wave), 7 and the negative effect of immigrant entitlements on subsequently experiencing suppression by ethnic outgroup coworkers-H2b. On the other, when it comes to the relationship between support by ethnic outgroups and attitudes regarding immigrant entitlements, the models do not provide indications for such effects.

General discussion
The strongest (largest effect size) and most consistent (found across all waves) support was found for the negative effect of natives' attitudes regarding immigrant entitlements on their likelihood of experiencing suppression by ethnic outgroup coworkers (H2b). Next, the findings partially (between T1 and T2) support the hypothesis about the negative effect of suppression by ethnic outgroup coworkers on immigrant entitlements (H1b). This partial support of H1b is most likely due to differences in time intervals that suppression measures encompass in T1 compared to T2. 8 When it comes to voice support, the results do not seem to support the hypotheses-there were no indications for the positive effect of support from ethnic outgroup coworkers on immigrant entitlements (H1a), nor for the positive effect of attitudes regarding immigrant entitlements on the likelihood of experiencing support by ethnic outgroup coworkers (H2a). A stronger effect of negative than positive experiences on human perception, also known as negativity bias, is typical across the board (Baumeister et al., 2001). This is reflected in, for example, loss aversion, a psychological phenomenon that describes peoples' tendencies to prioritize minimizing losses rather than maximizing gains (Kahneman & Tversky, 1979). This tendency is attributed to negative experiences being more directly related to survival, than positive ones (Haselton & Nettle, 2006;Krueger & Funder, 2004). These insights can help explain the stronger effect of negative compared to positive intergroup contact on prejudice, as found in the previously mentioned study by Barlow et al. (2012). Barlow et al. (2012) find that negative contact increases prejudice more than positive contact decreases prejudice, although positive contact happens more frequently than negative contact. Similarly, our results indicate that suppression has a greater negative effect on immigrant entitlements than support has a positive effect on immigrant entitlements.
Not only that, but the reciprocal effect seems to reflect negativity bias as well. The more unfavorable attitudes native employees held regarding immigrant entitlements, the more likely they were to experience suppression by outgroup coworkers. This suggests that native employees' preexisting attitudes toward ethnic outgroups, captured through their attitudes toward immigrant entitlements, affect the interactions native employees' have with ethnic outgroup members. However, natives were not more likely to experience support from ethnic outgroup coworkers if they held more favorable attitudes regarding immigrant entitlements. This pattern can also be explained by a negativity bias, as natives' favorable and unfavorable attitudes on immigrant entitlements can result in outcomes of different weight for the ethnic outgroup coworkers (i.e. discrimination affecting ones' working conditions more than a friendly chat). Another explanation for this could be that dominant ethnic groups' behavioral expressions of favorable attitudes regarding ethnic outgroups are sometimes perceived as performative by the ethnic outgroup members (Crittle, 2017), which would prevent the underlying mechanism assumed for the effect of attitudes regarding immigrant entitlements on the likelihood of experiencing voice support by ethnic outgroup coworkers.
Next, we turn to compare the effect of suppression by ethnic outgroup coworkers on attitudes regarding immigrant entitlements (H1b) and the effect of attitudes regarding immigrant entitlements on the likelihood of experiencing suppression by ethnic outgroup coworkers (H2b). While H1b is only supported between the first and second waves, H2b is supported both between the first and second, as well as between the second and third waves. Considering factors that might have affected effect sizes provides insight into the differential support for H1b and H2b. One such factor is the way suppression was measured, which is as an incidence. Arguably, one single experience can have a limited effect on subsequent attitudes, leading up to partial confirmation of H1b; while attitudes may have a greater effect on social interactions, due to their quality of continuity and possible multiple instances of attitude expression. This would explain why the effect of attitudes on experiencing suppression (H2b) is more robust than the effect of having experienced suppression on attitudes. Future research could benefit from employing a definition of voice support and suppression that captures degrees thereof (by assessing, for example, frequency to differentiate between one or several experiences, or the number of coworkers involved), instead of a dichotomous measure used in the present study. At the same time, although the effects found are of relatively small magnitude, they are particularly indicative because the experience of voice suppression captured in this research may be just one of the representations of negative interethnic contact at the workplace, thereby making the present findings potentially a fragment of a larger reciprocal process.
Although comparison of CLPM effects is sometimes used to determine causal predominance (whereby the stronger effect would be considered as the first one in the causal chain), this approach is widely criticized and CLPMs are considered better suited for exploratory purposes than definite causal conclusions (Allen, 2017). This leads to the main limitation of the present study, namely the inability of CLPM to pinpoint causal effects with certainty, due to conflating between and within subject variability. This is the reason an extension called RI-CLPM has recently been introduced as the preferred choice (Hamaker et al., 2015). The RI-CLPM extracts between subject variability by introducing random intercepts, therefore leaving only the within subject variability under cross lagged paths, which would make these estimates more valid indicators of causality. However, the data used in this study does not lend itself to RI-CLPM analysis. A possible explanation is apparent from the descriptive statistics in Table 1-namely, since very few people experienced voice suppression (dichotomously measured), variables measuring suppression in T1, T2, and T3 have a very small variance, especially relative to immigrant entitlements, measured as continuous variables with multiple indicators. Therefore, the reason RI-CLPM could not be computed in this case is likely because there was not enough variability in the suppression variables to successfully extract random intercepts and estimate cross-lagged paths. Thus, although this study does not yield the definitive answer to the "which came first, the chicken or the egg" question, it does show that natives' attitudes regarding immigrant entitlements and ethnic outgroup coworkers responses to their workplace voice are reciprocally related.

Theoretical contribution
The findings contribute to the contact literature by reaffirming the potential for workplace contact to affect intergroup attitudes. However, contrary to the general finding that workplace intergroup contact has a beneficial effect on intergroup attitudes by reducing prejudice, this study's findings emphasize the effect contact can have on impairing intergroup attitudes. This highlights the importance of the affective quality of intergroup contact at work, and in particular the importance of negativity bias for the effects of positive and negative contact on attitudes. It reinforces the findings of Barlow et al. (2012) that although positive contact is more common, negative contact is comparatively more impactful. Future research could expand on the present study by measuring the theoretical mechanisms implied as intermediaries in this research, namely categorization and intergroup empathy and anxiety.
To our knowledge, this study is the first to use a longitudinal design to empirically support the causal claim that negative experiences of workplace contact shape employees' political attitudes regarding outgroups. Moreover, the results also lend support for the effect of attitudes regarding immigrant entitlements on the likelihood of receiving suppression from ethnic outgroup coworkers. As such, the results point to a bidirectional interplay between native employees' attitudes (likely expressed in natives' behavior toward ethnic outgroup coworkers) and ethnic outgroup coworkers reactions to their voice.

Practical implications
In line with the findings pointing to a vicious cycle of unfavorable attitudes toward immigrant entitlements (which are likely an implicit reflection of unfavorable attitudes toward ethnic outgroup members) and voice suppression by ethnic outgroup coworkers, there are several ways in which companies could benefit from these findings in order to harmonize interethnic relationships at the workplace and create a more inclusive organizational culture. We discuss practical implications in three areas of interest: during the hiring process; as part of the organizational culture; and on the level of (compliance with) government legislation.
First, for organizations that value diversity and inclusivity, it might be beneficial to make such values explicit throughout the hiring process. In addition to being vocal about these values in job advertisements and in information about the organization, organizations may choose to make the topic of diversity and inclusivity part of the assessment procedure for new employees. Similar to other assessments used to test a candidate's fit to the goals and values of the organization, organizations might assess interethnic attitudes of prospective employees, to the extent that selection of candidates on such values is legally permitted. Such assessment could best be performed in a late-stage or final-stage interview, because it represents an additional criterium of assessing candidates' fit to the organization, on top of key competences that are typically assessed earlier in the selection process.
Candidates could be asked about their previous experiences working within multicultural workspaces and generally how open they are to other cultures. Such information could be used to differentiate between candidates with similar competences, or when assembling work teams.
Second, our results speak for an open organizational culture in which speaking up about problems at work is welcomed by the employer. If employees feel their relationship with their supervisors and/or higher management will deteriorate as a consequence of addressing work-related problems, they might be less likely to speak up themselves and more likely to suppress coworkers who speak up about work-related problems. The latter may be exacerbated when ethnic boundaries are at play within an organization, in which case fear of negative consequences will likely intersect with common processes of outgroup derogation. When employers/supervisors suspect that ethnic boundary issues may be at play within the organization, organizations could organize trainings in which they promote solidarity between employees of differing ethnic backgrounds. These solidarity trainings would primarily discourage suppression of voice between ethnic outgroup coworkers; and, additionally, encourage support as appropriate (only, of course, when employees themselves feel intrinsically motivated to do so). Such trainings would have the potential to decrease suppression of voice between ethnic outgroup coworkers and therefore prevent the cycle of deteriorating interethnic relationships implied by the results of this study. After all, it has been recognized that good relationships between coworkers are beneficial for productivity and a wide array of employee and organizational-level outcomes (Baruch-Feldman et al., 2002;Chiaburu & Harrison, 2008). Thus, solidarity trainings are in the interest of the company and employees alike (regardless of the employees' ethnic background, but potentially especially useful in intercultural working environments). Similarly, to improve such organizational and employee outcomes, it may be useful to compose working teams of people who will get along, starting from the selection procedure. Third, on a higher level, the results of this study can be used in the context of (compliance with) public governance policies. Coming back to the example of Polish strike-breakers from the introduction, a government that wishes to build a tolerant multicultural society and is aware of these findings would not allow organizations to employ such tactics in particular with regard to hiring migrant workers as strike breakers, which effectively contributes to the formation of ethnic boundaries. The responsibility here is one for the employer in the first place. However, governments are responsible to ensure compliance with policies that prohibit the use of strike breakers, also (or in particular) with regard to migrant workers. In a more general sense, since governments dictate the legal frameworks in which migrant workers are embedded, governments have a responsibility to reflect upon the social consequences of work-related policies, above and beyond the (economic) interests of employers.

Conclusion
Although the non-experimental nature of the current research design does not allow for unequivocal causal inferences, the findings of this study point to coworkers' reactions to voice as a potentially relevant factor in shaping political attitudes, while at the same time being determined by political attitudes in question. The results indicate that experiencing suppression by ethnic outgroup coworkers leads to native employees holding less favorable attitudes toward immigrant entitlements. At the same time, the less favorable attitudes native employees have regarding immigrant entitlements, the greater their likelihood of experiencing outgroup suppression. Conversely, when it comes to support by ethnic outgroup coworkers, no relationship with natives' attitudes regarding immigrant entitlements is found in either direction. This study is the first to examine the consequences of voice support and suppression by coworkers. The results show that negative contact with ethnic outgroup coworkers in the form of voice suppression may cause less favorable attitudes regarding immigrant entitlements. Moreover, native employees' attitudes regarding immigrant entitlements affect their likelihood of receiving voice suppression by ethnic outgroup coworkers. These findings emphasize the 2-fold importance of being "careful with how you treat your coworkers," as there seems to be a bidirectional interplay between behaviors of native majority group members and ethnic outgroup members. Notes 1. Prejudice is here "consensually defined as a negative attitude toward members of a social outgroup" (Stangor et al., 1991, p. 360). 2. Apart from direct support and suppression, coworkers can choose to ignore the voiced issue. In the current research, ignoring was shown to correspond with suppression according to emotional impact, and is thus regarded as voice suppression in this study. 3. The "other non-Dutch" category is too broad to be analyzed, as it may encompass nonwestern ethnic outgroups, as well as other western European nationals such as Germans or Belgians. The degree to which other west Europeans are perceived as ethnic outgroup members is questionable, especially compared to nonwestern ethnic outgroup members who often differ from the Dutch with regards to cultural, religious and racial characteristics. Because of that, the reports of suppression/support by the "other non-Dutch" were omitted when computing the binary variable of having experienced suppression/support by ethnic outgroup coworkers. 4. The reason for computing binary variables for support and suppression was to maximize the sample size. As further explained in the text, support and suppression variables must be restricted according to whether native participants had the chance to experience them due to working in an intercultural environment. Computing latent variables for support and suppression that take into account the origin by ethnic group would mean further restricting the sample analyzed to participants who work with both Turkish/Moroccan and Eastern European coworkers, which seems to be an unnecessarily strict selection. 5. Recently the CLPM had been criticized for conflating within and between subject variability, and an extension called RI-CLPM has been suggested in order to disentangle the two and capture within subject change (Hamaker et al., 2015). However, in this case testing RI-CLPM was not feasible due to the specificities of the covariance structure of the current data, as explained in the discussion. 6. Additionally, we briefly reflect on the cross-sectional correlations in the two-wave models. Since coworker support and suppression were assessed via self-reports, they are necessarily retrospective (meaning they occurred before respondents took the survey, and hence, before the measurement of immigrant entitlements). As such, the cross-sectional correlations could represent early-onset effects of support or suppression on immigrant entitlements. Whereas on first sight the cross-sectional correlations between support and immigrant entitlements for respectively T1 ( Figure 4) and T2 ( Figure 5) seem to indicate an effect of support on immigrant entitlements, the effect disappears once controlled for through autocorrelations (T2, Figure 4; T3, Figure 5). Therefore, the crosssectional correlations are unlikely to represent early onset effects of support on immigrant entitlements. 7. This is likely due to suppression in T1 encompassing 3 years of suppression instances (as explained in the methods section), thereby capturing enough instances to exert an effect on entitlements in T2; while suppression in T2 contains only one year worth of suppression instances, thereby having less potential to affect entitlements in T3. The ceiling effect may provide an additional explanation (at least in case of the three wave model), whereby subsequent instances of suppression cannot further affect a person's stance on immigrant entitlements when this stance approaches the extreme ends of the scale. 8. Ibid.