A Meta‐Analysis of Change in Applicants' Perceptions of Fairness

Using an event&#8208;triggered multi&#8208;stage framework, this random&#8208;effects meta&#8208;analysis examined the changes in applicants' perceptions of fairness between consecutive stages and throughout the entire personnel selection process. We integrated findings of studies with at least two measurement points, resulting in 45 effect sizes (overall N&#160;=&#160;3,038). Trajectories of perceptions of fairness decreased nonlinearly across the process, with a steeper decrease for people who held high levels of initial fairness expectations. Unjust treatment produced a decrease in perceptions of fairness from pretest to posttest and an increase from posttest to postdecision. Furthermore, the length of the time interval moderated the changes in fairness perceptions between the posttest and postdecision stage. Practical implications and an agenda for future research are discussed.

Shao, Jones, & Liao, 2014). Arguably, the personnel selection process typically consists of multiple stages in which applicants are repeatedly confronted with varying information. Consequently, their perceptions of fairness may be shaped and revised dynamically during the selection process (Breaugh, 2009;Evertz & Süß, 2017). Hausknecht, Sturmann, and Roberson (2011) demonstrated that trajectories in employees' perceptions of fairness explained additional variance in work outcomes after controlling for end-state levels of fairness. In addition, high initial levels in fairness expectations corresponded to a steeper decrease in these trajectories (Konradt, Garbers, Erdogan, & Bauer, 2016). Consequently, the first goal of the present meta-analysis is to examine the trajectories of applicants' perceptions of fairness. Accordingly, aggregating observations from individual stages in the selection process, or examining only a particular stage, give at best only a snapshot of what actually happens and may even result in relevant effects being over-or underestimated and in important insights being obscured. Describing and understanding the changes in perceptions of fairness during the selection process is crucial for developing a nuanced and detailed perspective on what applicants perceive to be fair or unfair and to develop useful advice to practitioners and organizations. Therefore, the second purpose of this meta-analysis is to examine the changes in applicants' perceptions of fairness between consecutive stages of the selection process. We also examine the impact of a particular selection practice or experience (e.g., just or unjust treatment, type of selection procedures) on applicants' perceptions. To investigate these effects, we employ a multiple-stage design (Saks, 2013), which enables us to answer questions such as "Does the length of the time interval between stages influences the change in perceptions of fairness dependent of the selection stage?". Findings on such questions not only help to achieve a better understanding of the change in perceptions of fairness, but are especially relevant from a practical point of view in order to improve the selection process (e.g., In which stages are which practices particularly relevant?).
Conclusively, this meta-analytic study is designed to systematically aggregate the results of previous longitudinal and repeated measure research to derive conclusions about that body of research and to extend previous research in the following important ways. First, in contrast to preceding work, this meta-analysis is the first to address fundamental questions about changes in perceptions of fairness. By two-stage meta-analytic structural equation modeling we are not only able to identify changes in perceptions of fairness between two consecutive stages in the selection process but also to identify the trajectory of perceptions of fairness over time, which have not been tested in any previous meta-analytical review. Finally, drawing on theories of justice (Gilliland, 1993;Jones & Skarlicki, 2013), we identify and examine factors that might influence the degree to which perceptions of fairness change, including applicants' expectations of fairness, just or unjust treatment in the selection process, the length of the time interval between stages in the selection process, the type of setting (i.e., hypothetical vs. real-world settings), and the number and type of selection procedures used. Our hypothesized model is displayed in Figure 1.
decision. Each stage is initiated by specific events, such as the organization's reaction to the applicant's application, the assessment tests, and interviews, or the communication of the hiring decision to the applicant. This structure is reflected in Barber's multi-stage framework (1998). The framework implies that, for applicants who progress beyond the first stage (i.e., the application), their perceptions of fairness develop over three stages: pretest, posttest, and postdecision. Pretest refers to the stage after the applicant has applied for a job. The posttest stage refers to the period after the applicant has undertaken a series of assessments (e.g., tests or interviews).
Postdecision (or postfeedback) refers to the period after the applicants received feedback on their application (e.g., offer or rejection).
During each stage, applicants gain new information that shapes their perceptions of how fair the selection process is, which, in turn, influences perceptions at the following stages (Ambrose & Cropanzano, 2003). Changes in perceptions of fairness are also inherent both in organizational justice theories (Jones & Skarlicki, 2013;Lind, 2001; van den Bos, Lind, & Wilke, 2001) and applicant fairness theories (Chan & Schmitt, 2004;Ployhart & Harold, 2004). Chan and Schmitt (2004) argued that applicants' perceptions of fairness change over time in consequence of the way they are treated during the application process. Similarly, applicant attribution-reaction theory (AART; Ployhart & Harold, 2004) proposes that perceptions of fairness are formed during an attributional processing sequence (i.e., situation perception, expectation, perceived satisfaction or violation, and attribution) that evolves over time. Consequently, perceptions of fairness evolve dynamically as individuals encounter new information during the selection process.

| Trajectory of applicants' perceptions of fairness
A first approach to capturing change in applicants' perceptions of fairness is to examine the latent trajectory (i.e., average growth trajectory or pattern of change over time) over the entire selection process (Curran, Obeidat, & Losardo, 2010). A longitudinal study with real job applicants has demonstrated that the latent trajectory of perceptions of fairness shows a nonlinear decline over the entire selection process (Konradt et al., 2016). Their findings have recently been replicated in a randomized controlled trial with a simulated selection context (Butucescu & Iliescu, 2018). They argued that selection processes are stressful events and thus lead to negative emotionality and feelings of stress, which trigger both a self-serving bias and a negative appraisal of the selection process relating to fairness perception.
One possible explanation of the declining fairness trajectory might be that selection procedures are stressful events that involve a high degree of uncertainty and can evoke negative emotions such as insecurity and anxiety (e.g., van den Bos & Lind, 2002). Such events can thus be a stress factor for individuals and can, in extreme cases, be a threat to their health (Sinclair, 2014). Since applicants have invested considerable resources in their application and usually regard the outcome (i.e., getting a job offer) as important but they do not receive feedback until the end, their impressions and negative emotions may tend to remain quite strong. These negative emotions may, in turn, lead to what is called mood congruency memory bias.
Here applicants will recall previous experiences that feel to be similar to their current mood and thus perceive their experience as less positive and less fair (Loeffler, Myrtek, & Peper, 2013).
Furthermore, if we look at models relating to episodic memory, studies have shown that emotional experiences are more likely to be remembered than neutral stimuli (e.g., Quevedo et al., 2003). Models of context-specific memory suggest that people in a certain mood tend to improve their coding or retrieval of stimuli that have the same affective valence, that is, information that is congruent with their current mood is retrieved more readily than information that is incongruent (Bower, 1987). Accordingly, negative information is more likely to be perceived than positive information.
Another reason for an overall decline might be that most candidates, which move through the process, receive a rejection, which evokes negative emotions (cf. Gilliland, Groth, Baker IV, Polly, & Langdon, 2001) as they are associated with disappointment and can violate one's self-view (Schinkel, Van Dierendonck, & Anderson, 2004). According to the self-threat model of procedural justice, applicants might perceive a rejection as a threat to their ego or self-concept (Lilly & Wipawayangkool, 2018). This can lead to defensive behavior in order to restore their self-esteem (Roese & Olson, 2007). Applicants tend to search for a possible explanation for a rejection that shifts the blame for their own situation onto something other than themselves. If applicants have been treated unfairly or rejected, they search for inconsistencies, biased judgments, and other signs of injustice in the selection process and blame the organization for the outcome (Lilly & Wipawayangkool, 2018). Consistent with theory and prior research we, therefore, suggest that:

Hypothesis 1a The trajectory of applicants' perceptions of fairness will
show a decline over the entire selection process.
Despite this, applicants' expectations of fairness (i.e., anticipatory fairness or initial level of perceptions of fairness)-which have been defined as "probabilistic beliefs about the characteristics of the forthcoming selection procedure" (Schreurs, Derous, Proost, Notelaers, & De Witte, 2008, p. 170)-correspond to the decline in their perceptions of fairness (Celani, Deutsch-Salamon, & Singh, 2008;Jones & Skarlicki, 2013). Expectations of fairness serve as a frame of reference, with the applicants' previous experience providing a reference point (Jones & Skarlicki, 2013). Thus, applicants' perceptions of an event are formed relative to a pretest anchor in terms of fairness expectations (Gilliland, 2008); their expectations about an upcoming situation will thus influence how they perceive a situation by guiding the type of information they pay attention to and the attributions they make (Ployhart & Harold, 2004;Schreurs et al., 2008). In situations that involve a high degree of uncertainty (e.g., selection process), applicants engage in attributional processes to identify causes of negative experiences. Applicants will search Electronic copy available at: https://ssrn.com/abstract=3734965 for any explanation that puts blame on something other than themselves (e.g., the test was not job-related) (Lilly & Wipawayangkool, 2018). As a result, they focus on those aspects that do not meet their expectations. Even small deviations from expectations have a significant and negative influence on their fairness perception.
The importance of realistic expectations is also evident from similar contexts: A realistic job preview influences applicants' expectations.
When they become an employee of the organization, these (un)met expectations, in turn, influence their performance, satisfaction, commitment, perceptions of honesty, acceptance, attraction, and turnover intentions (for an overview, see Cable & Yu, 2013). Therefore, (un)met expectations can affect how applicants perceive the fairness of an organization. Consistent with this assumption, previous work has demonstrated that elevated expectations of fairness are linked to a steeper decline in perceptions of fairness over time (Butucescu & Iliescu, 2018;Konradt et al., 2016). Consequently, we hypothesize a negative relationship between applicants' expectations of fairness (i.e., intercept) and the decline in perceptions of fairness over time (i.e., slope): Hypothesis 1b Applicants whose expectations of fairness are initially high (i.e., intercept of trajectories) will show a steeper decline (i.e., slope of trajectories) in their perceptions of fairness throughout the entire selection process.

| Changes in perceptions of fairness between consecutive stages
Besides examining the trajectory of perceptions of fairness over the entire selection process, we also study the change in perceptions between consecutive stages. Fairness theories suggest that perceptions of fairness are "affected by expectations about future treatment and memories of past treatment" (Fortin et al., 2016, p. 7).
Consequently, the stages in the selection process are interdependent and can affect each other. Findings on such interdependencies are particularly relevant from a practical point of view. Many selection processes consist of sequential hurdles that applicants need to pass. It is, therefore, important for organizations to improve the stages but also to improve the transitions between them so that applicants do not withdraw their application during the process.
In terms of the pretest stage, "applicants learn about the job and organization, prepare application materials, and interact with company representatives for the first time" (Hausknecht et al., 2004, p. 13). Through this initial contact (e.g., via email or telephone), an applicant will form certain expectations regarding the organization and upcoming selection process. As already indicated, those expectations will, in turn, affect that individual's perception of an event's fairness at the posttest stage (Proudfoot & Lind, 2015). Moreover, we argue that fairness judgments in the pretest stage not only affect judgments in the posttest stage but will also decline over these two consecutive stages. As argued previously, applying for a job and participating in assessments involves a high degree of strain and uncertainty for applicants, who do not know whether they will receive a job offer or not (van den Bos & . The more time, resources, and hope applicants have invested in applications, the more they will fear failure. Furthermore, the applicant attribution-reaction theory (AART; Ployhart & Harold, 2004) poses that fairness perceptions are shaped by attributional processes. In the posttest stage, applicants engage in attributional processes to identify causes of, for example, their performance in a test, and they will search for any explanation that puts blame on something other than themselves (Lilly & Wipawayangkool, 2018). As a result, applicants mainly focus on those aspects of the assessment phase that do not meet their expectations. Even small deviations from expectations thus have a significant and negative influence on the perception of fairness in the posttest stage. In line with the mood congruency effect, applicants will then retrieve experiences that fit with their current mood and will perceive the process as more unfair (Loeffler et al., 2013). Based on this literature, we put forward the following hypothesis: Hypothesis 2a Applicants' perceptions of fairness of the personnel selection process will decrease between the pretest and posttest stages.
Between the posttest and postdecision stage, we assume there will also be a decline in applicants' perceptions of fairness. The outcome of the application process (i.e., the hiring decision) itself influences applicants' perceptions of the procedure: Those who receive a job offer will perceive the process more favorably than those who get a rejection (Gilliland, 1994;Ryan, Sacco, Mcfarland, & Kriska, 2000). Since most applicants will be rejected, negative emotions (e.g., disappointment, anger, frustration) in the postdecision stage are thus more likely (cf. Gilliland et al., 2001). The self-threat model of procedural justice suggests that applicants who are rejected become defensive and make external attributions by blaming the organization for their situation (Lilly & Wipawayangkool, 2018). In this stage, there should accordingly be a clear decrease in applicants' perceptions of fairness. In line with theory, we consequently predict the following change will occur: Hypothesis 2b Applicants' perceptions of fairness of the personnel selection process will decrease between the posttest and postdecision stage.

| Moderators of the changes between consecutive stages in perceptions of fairness
Drawing on theories of fairness, we have also explored the variation in these changes between stages in perceptions of fairness by identifying potential moderators. The first moderator, which forms part of all fairness theories, is the just or unjust treatment to which applicants were exposed.
this, unjust treatments are likely to have a stronger negative effect on perceptions of fairness, whereas just treatments will only confirm the applicant's expectations (Jones & Skarlicki, 2013). This is in accordance with the principle of "bad is stronger than good" (Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001), where bad events have a more lasting and intense impact than good events. Furthermore, in uncertainty management theory (Lind & van den Bos 2002;van den Bos & Lind 2002) it is argued that fairness matters more when people are uncertain about something. This is especially the case in a selection process, where applicants are uncertain about the outcome, which is of direct relevance for their career. Lind and van den Bos (2002, p. 196) argued that unjust treatment "under conditions of uncertainty gives the uncertainty a particularly sinister complexion and makes people even more uneasy". Thus, unjust treatment will itself produce feelings of uncertainty. Conversely, just treatment helps applicants overcome their feelings of personal uncertainty and insecurity by giving them confidence that they will achieve a good outcome or that rejection will provoke fewer negative emotions. This is in line with Mullen's (2007) suggestion that people only think about justice when they experience negative emotions. Consequently, applicants who are treated unjustly will focus much more on justice issues and are, therefore, more sensitive to any violation of justice. These arguments lead us to the following hypothesis: Hypothesis 3 Applicants' perceptions of fairness will decrease more between the pretest and posttest stage (a) and between the posttest to postdecision stage (b) when they are treated unjustly as opposed to justly.

| Time span between stages in the personnel selection process
In line with the logic of change, the length of the time interval between consecutive stages in the selection process might predict the magnitude of the changes in applicants' perceptions of fairness. When two points in time are far apart, people might have changed their mental representations of fairness through causal attributions and sense-making processes (Jones & Skarlicki, 2013;Ployhart & Harold, 2004). Additionally, previous research (Paterson, Green, & Cary, 2002) indicated a low test-retest reliability for perceptions of fairness and the authors suggested three different causes for this: Individuals changed their norms of fairness over time, used different reference points, or were prone to memory biases. Rosy retrospection bias is another possible memory bias that might play a role in perceptions of fairness over time. That is, people tend to remember events in the past more positively than they actually felt at the time (Mitchell, Thompson, Peterson, & Cronk, 1997). Applied to our context, applicants tend to remember something that happened in the past as being less unfair the longer ago the event took place. As such, the following hypotheses are offered: Hypothesis 4 When the intervals between (a) pretest and posttest and (b) posttest and postdecision are longer, applicants' perceptions of fairness will decrease less than when the intervals are shorter.

| Hypothetical versus real-world settings
In their meta-analysis, Hausknecht and colleagues (2004) assume that effect sizes drawn from real-world settings will differ from those found in hypothetical studies that simulate selection settings. In real-world settings, job applicants are taking part in an actual selection process. In hypothetical settings, individuals are taking on the role of an applicant or base their judgments on descriptions of different procedures. Hausknecht et al. (2004) reported that average correlations with perceptions of fairness are stronger in hypothetical settings than in real-world settings. They argued that one reason for the difference in effect sizes between real-world and hypothetical settings might be that, in hypothetical settings, applicants are confronted with scenarios that trigger perceptions that are more to do with aspirations or wish fulfillment. In addition, these applicants may behave less strategically, because complaints of unfairness will have no consequences in terms of the hiring decision. The authors also noted that a methodological bias could be an explanation for the moderating effect of the type of setting: Applicants in real-world settings may represent a restricted range of applicants because of self-selection. The restriction, in turn, could mitigate the observed effects. This led us to hypothesize that:

| Number and type of selection procedures
Selection procedures used in a selection process vary in number and/or type. Referring to the number of procedures used, applicants are more strained with an increasing number of procedures (cf. Ryan & Ployhart, 2000). They could thus perceive the process as less fair.
In contrast, the more procedures are used, the more valid and thus fairer the selection process is perceived. Research has for example demonstrated that the validity of a selection process increases if a multimethodological approach is used (see Schmidt & Hunter, 1998, for details).
Referring to the type of procedures, the trimodal selection approach distinguishes between the trait (e.g., personality tests, cognitive ability tests), biographical (e.g., interviews), and simulation (e.g., role-play exercise, work sample) approach (Deters, 2017). Hausknecht and colleagues (2004) observed the highest favorability among applicants for interviews (biographical approach) and work samples (simulation approach). Cognitive ability and personality tests (trait approach) only came fifth and sixth in the ranking, respectively. This corroborates more recently meta-analytical evidence that selection procedures could be clustered into three distinct groups: most preferred (e.g., interviews, work samples), favorably evaluated (e.g., cognitive ability and personality tests), and least preferred (e.g., honesty tests, personal contacts) (Anderson et al., 2010).
However, previous research did not examine whether different approaches (e.g., biographical and trait approach) or their combination influences applicants' perceptions of fairness. Similarly, the number of procedures should be considered, because the number increases with a combination of procedures. What is not yet clear, therefore, is the impact of the number and type of procedures on the change between the various stages in applicants' perceptions of fairness. We thus pose the following research question: Research Question: Do the number and type of selection procedures moderate the changes in perceptions of fairness between consecutive stages in the selection process?

| Inclusion criteria
Studies had to meet the following criteria to be included in the metaanalysis: (a) they deal with justice/fairness and examine one or more justice/fairness variables (e.g., procedural justice); (b) they have a repeated measure design and consider at least two of the three stages in the selection process (pretest, posttest, postdecision); (c) the samples had to consist of real applicants or participants assuming the role of applicants; (d) and they had to provide sufficient statistical information (i.e., correlations, sample size, mean values) to calculate effect sizes.

| Data collection
The first search procedure step yielded a total of 1,092 records Another eleven records were identified by looking up articles that cited Jones and Skarlicki (2013) or Bauer et al. (1998); three unpublished records were also included in the analysis. A total of 56 records were assessed for eligibility.
Two independent judges read the articles in full and excluded a further 38 articles. Reasons for exclusion at this stage were that the studies were not original articles, but were reviews, commentaries, and so forth (15); that they dealt with justice in a broader sense (8); that they did not use a repeated measures design (7); that they dealt with perceptions of fairness among employees in organizations, not specifically during the selection process (7); or that the paradigm was not suitable (1).
Additionally, we received a response from the NIP, which led us to include two additional articles and one unpublished data set that fitted our criteria for inclusion (six data sets in total). Another study published shortly after the completion of the literature search was also included (two data sets). Via the various platforms indicated above and using the overview table from McCarthy et al. (2017), we found four further studies to include in the meta-analysis (Butucescu & Iliescu, 2018;Horvath, Ryan, & Stierwalt, 2000;Oostrom, Bos-Broekema, Serlie, Born, & van der Molen, 2012;Wiechmann & Ryan, 2003). Furthermore, of these studies, twelve studies were excluded prior to analyses for different reasons: First, four were excluded because no correlations or no pre and postmeans were reported (e.g., Schinkel, van Vianen, & Ryan, 2016 (Study 4); Ployhart & Ryan, 1998;Truxillo, Bauer, & Sanchez, 2001), thus rendering quantitative analysis regarding our hypothesis impossible, or because we could find no current contact information for the author (Carless, 2003 (2008) because its retrospective design did not meet the inclusion criteria: Participants were only asked about their perceptions of fairness if they were already or soon to be employed. Due to cognitive bias, this design is not compatible with that of other studies in which applicants were asked about their perceptions of fairness immediately after each selection stage. Finally, we had to exclude six studies because they did not consider the relationships between fairness variables at different points in time (e.g., Chapman & Webster, 2006).

| Coding of study characteristics
For each study, relevant information was coded, including the publication year, sample size, and sample characteristics (age, gender, and the type of job being applied for). Authors were asked to complete basic statistics when they were missing. To compute the effect sizes, we used standardized mean changes (see "Data Analysis"). For studies that considered more than one of our dependent variables, or more than one type of a particular dependent variable, we looked at the size of each effect. We did not devote any attention to identifying possible dependencies between the effect sizes within studies, thus underestimating the variance of the composite effect size, giving us a smaller confidence interval (Cheung, 2014).
To test the potential moderators, we coded the treatment, type of setting (hypothetical or real-world setting), time intervals (in days) between measurement points, and the number and type of selection procedures. The treatment of applicants during the selection process (before the decision is communicated to the applicant) was coded as "just" (1) and "unjust" (0). For hypothetical studies, we coded the described treatment. For real-world settings, a description of the treatment was missing in most cases. In those cases, the treatment was coded as "just" because in all these studies the mean value of the perception of fairness was above the scale mean value.
The type of setting (real-world vs. hypothetical settings) was also dummy coded as 0 and 1, respectively. 2 We also coded the num-

| Data analysis
We calculated the meta-analytic effect sizes and effects of the moderators using the 'metafor' package (Viechtbauer, 2010) in R (R Core Team, 2017). We used a random-effects meta-analysis based on Schmidt and Hunter, in line with our assumption that the true effect is not the same in all studies, since the contexts will vary (cf. Schmidt, Oh, & Hayes, 2009;Borenstein, Hedges, Higgins, & Rothstein, 2009). The calculated summary effect is, therefore, the estimate of the mean of the true effects. In accordance with the procedures suggested by Schmidt and Hunter (2014), the estimated effect sizes were adjusted based on the sample sizes and also corrected for the reliability of the measures. As a substitute for any missing reliability coefficient, we used the mean of the reliabilities of the respective study (Schmidt & Hunter, 2014). 3 We calculated d and the respective confidence intervals in the form of a change score effect size, using the standardized mean change with change score standardization (SMCC; Morris & DeShon, 2002). 95% confidence intervals have to exclude zero in order to conclude that the effect size is different from zero. We also calculated the I 2 statistic, which estimates the proportion of variability that is due to heterogeneity as opposed to chance (Higgins, Thompson, Deeks, & Altman, 2003).
Moderator analyses were performed using a factorial approach (also referred to as subgroup analysis, and helpful in interpreting categorial moderators): For each category of a categorical predictor, a subgroup was specified, and the effect size was calculated. To assess whether our point estimates were affected by publication bias, we conducted a fail-safe N analysis (Rosenthal, 1979) and tested its robustness by testing whether the obtained value was larger than 5 x k + 10 (cf. Rosenthal, 1991).
To analyze the LGC models, we used a two-step procedure. As a first step, we compared the AIC (Akaike information criterion) and BIC with fixed loadings for the slope at T1 (fixed to 0) and T3 (fixed to 1) (cf. Chan & Schmitt, 2000). Lower AIC and BIC values indicate a better fit. As a second step, we interpreted the root mean square error of approximation (RMSEA) (Browne & Cudeck, 1989), comparative fit index (CFI) (Bentler, 1990), and Tucker-Lewis index (TLI) (Bentler & Bonett, 1980). An acceptable model fit was indicated by CFI and TLI values greater than or equal to 0.95 and by an RMSEA of up to .10 (Browne & Cudeck, 1989).

| Latent trajectory
Hypothesis 1a  supported. In support of Hypothesis 1b, the decrease in perceptions of fairness was steeper for applicants whose initial expectations of fairness were higher than for those whose expectations were lower.

| Moderators of change
To determine whether the analysis of moderators of changes between consecutive stages was reasonable, we first assessed the level of effect size heterogeneity, using the I 2 heterogeneity index.

| 375
KONRADT eT Al. Note: k = number of effect sizes; N total = the sample sizes of each study were aggregated; I 2 = total heterogeneity/total variability. For effects in bold print, the confidence interval excludes zero.

TA B L E 2 Analysis of moderators on changes in applicants' perceptions of fairness between consecutive stages of the selection process
Abbreviations: B, biographical approach; CI, confidence interval; H, hypothesis; RQ, research question; S, simulation approach; SMCC, standardized mean change with change score standardization; T, trait approach. a For hypothetical studies we have set the time interval to 0 days when they were completed on the same day.
Electronic copy available at: https://ssrn.com/abstract=3734965 Our analyses showed that more than 93% of the total variability among effect sizes was caused by true heterogeneity between studies. These results consistently indicate a high level of effect size heterogeneity across the studies in our sample (cf. Higgins & Thompson, 2002), suggesting that moderator analysis was warranted. The results of the moderator analyses are presented in Table 2.
Hypothesis 3a predicted that the decline in applicants' perceptions of fairness from pretest to posttest would be stronger when they were treated unjustly than when they were treated justly.  Table 2).
For the length of the time interval between two consecutive stages (Hypotheses 4a and 4b), the analysis did not reveal a significant change in applicants' perceptions of fairness between the pre and posttest (0 or 14 days; see Table 2). In accordance with Hypothesis 4b, however, there was a stronger decrease for smaller  Table 2).
Hypothesis 5  The research question examined the moderating role of number and type of selection procedures on applicants' perceptions of fairness. As shown in Table 2, there were almost no significant effects.

Significant changes in perceptions between the pretest and posttest
were found for studies that used two compared to one, three or more selection procedures (SMCC = −.72, CI [−1.12 | −.31], k = 10). In addition, a significant moderation effect was found for those studies who pursued the trait + biographical approach (SMCC = −.71, CI [−1.07 | −.35], k = 9). However, the 95% CIs of the subgroups overlapped, which restricts this finding (see Table 2).

| Supplementary analyses
Supplementary analyses were conducted to determine whether the effects varied depending on the type of source, as 15% (

| D ISCUSS I ON
In conducting this meta-analytic review, we aimed to address several limitations and research gaps in the applicant fairness literature.
First, we used studies with matched groups and repeated measures designs to address questions of causal influence, change, and psychological processes that have not been addressed meta-analytically before. By focusing exclusively on studies that used these repeated measurement designs with at least two stages in the selection process, this meta-analysis allowed us to examine how perceptions of fairness changed between the various stages. Second, we used an event-triggered multi-stage framework, in which we distinguished between pretest, posttest, and postdecision stages in the personnel selection process. Third, we combined meta-analysis with structural equation modeling, which allowed us to examine the latent trajectory of applicants' perceptions of fairness that had not been examined in earlier meta-analyses.
In general, our results suggest that to fully understand the dynamic nature of perceptions of fairness, we need to consider the influence of previous perceptions of fairness and the change in perceptions of fairness between different stages. In addition, this meta-analysis set out to identify and examine moderators that might influence the degree to which perceptions of fairness change.
Our meta-analysis indicates that there is a considerable change in perceptions of fairness during a selection process, with an overall decrease across the entire process, irrespective of how fairly applicants were treated. Our finding that the trajectory in perceptions of fairness follows a nonlinear rather than a linear pattern, mirrors previous findings (Butucescu & Iliescu, 2018;Guo, 2012;Konradt et al., 2016). Interestingly, applicants with high expectations of fairness showed a steeper decline in their perceptions of fairness than those whose expectations were initially lower. This fits with the logic that applicants mainly focus on those aspects that did not meet their | 377 KONRADT eT Al.
expectations and that even small deviations from expectations have a significant and negative influence on their perceptions of fairness.
Another central finding is that the change in perceptions of fairness between consecutive stages of the selection process is related to how fairly or unfairly applicants were treated. In the just treatment condition, there was no change between consecutive stages of the process, whereas in the unjust condition the change from pretest to posttest was initially negative, but from posttest to postdecision it was positive. As summarized by Ambrose and Cropanzano (2003), when two events are more widely spaced (Jones & Skarlicki, 2013;Ployhart & Harold, 2004).
The first meta-analytic review on applicants' perceptions of fairness by Hausknecht and colleagues (2004) found that the average effect sizes were higher for hypothetical studies than for real-world studies. Consistent with these findings, our results revealed that, in hypothetical settings, the mean effect sizes for the negative change between pretest and posttest were significant and slightly higher than in real-world studies, although the confidence intervals overlapped. The effects found in hypothetical settings should, therefore, be regarded as representing the upper bounds of what may be found, with effects tending to be generally less pronounced in real-world settings.
The ambiguous effect of the number and type of selection procedures on the change in perceptions of fairness is somewhat surprising. One explanation for this inconsistent effect might be a methodological one: The variables number and type were confounded. Arguably, the more approaches were considered in a selection process, the more procedures were used. Furthermore, this result is based on a comparatively small number of sample sizes within subgroups that were suitable for the meta-analysis (e.g., the combination of the trait, biographical, and simulation approach in the second phase of the selection process). A small sample size cannot preclude an unreliable estimation of the effect sizes and inefficient standard errors. Consequently, this finding is less likely to be precise and thus should be interpreted with caution. Another explanation for this finding is that we classified the selection procedures according to the trimodal selection approach. Thus, we distinguished between the trait, biographical, and simulation approach but not between different procedures within these groups. A modified classification of the selection procedures may have led to different results. We hope that we can encourage researchers to take up this point and consider the effects of number and type of selection procedures in selection processes in a more nuanced way, so that important implications for organizations can be derived thereof.

| Practical implications
The perceptions also vary, these variations will have an impact on the applicants' subsequent attitudes and behaviors (Gilliland & Steiner, 2012;Truxillo & Bauer, 2011). Understanding who these applicants are, and which groups of applicants are at risk of being disappointed, provides an important starting point for managers and personnel staff in terms of deciding what might be done to minimize the dysfunctional effects of applicants perceiving there to be a low level of fairness. This is important because whether applicants would recommend an organization as an employer has been found to be related to the perceptions of fairness of its selection practices (Hausknecht et al., 2004). For example, applicants who show strongly decreasing fairness perception trajectories (i.e., the high-decreaser class) are at risk of a substantial decline in perceptions of fairness over time and might thus withdraw their application, even if they prove suitable. Being able to identify these applicants very early in the selection process and provide them with Supporting Information on the organization's values during the process may help organizations to enhance selection effectiveness.
We have demonstrated in our meta-analyses that higher expectations at the beginning of the selection process lead to a greater decrease in perceptions of fairness throughout the entire process.
Drawing on the justice rules proposed by Gilliland (1993), organizations should raise realistic nonexaggerated expectations of the upcoming process in the application stage. HR managers should explain transparently what is being done and why, and provide truthful information about the procedure, the decision-making process, and the people involved in the selection process (for a meta-analysis on the effects of explanations on applicant reactions see Truxillo et al., 2009).
In terms of the treatment during the selection process, Bauer, McCarthy, Anderson, Truxillo, and Salgado (2020) have published a best practice guide, which provides specific recommendations for HR managers. Among others, HR managers should take particular care to ensure that the selection procedures are face-valid, job-related, have perceived predictive validity (e.g., develop a selection system that is based on a job analysis), and are consistently administered (e.g., provide training to interviewers). Furthermore, applicants should have the opportunity to perform well (e.g., ensure that the selection system is comprised of multiple components) and they should be treated with respect (e.g., be pleasant, do not ask inappropriate questions).
Finally, we have also demonstrated that applicants' perceptions of fairness can change between the assessment and acceptance stage. In order to avoid a decrease (and hopefully facilitate an increase) in perceptions, HR managers should provide both timely and useful feedback and a justification for the hiring decision (e.g., give as much information as possible, provide information regarding future job applications) (Bauer et al., 2020;Gilliland, 1993).

| Developing an agenda for future research
About three decades after Gilliland's seminal work on fairness in personnel selection, almost five decades after one of the first studies on applicant perceptions (Alderfer & McCord, 1970), and after numerous quantitative and qualitative reviews, we must conclude that it is time for the next step. To encourage advances in both theory and practice, we suggest that the following lines of research should be considered in future studies (see also McCarthy et al., 2017).
First, future research on applicant reactions should examine applicants' perceptions of fairness over more than two points in time allowing to cover a larger part of the selection process (Chan & Schmitt, 2004;Fortin et al., 2016;McCarthy et al., 2017;Rupp et al., 2014). Consequently, future studies could help in understanding the dynamics by including data that capture the entire personnel selection process and allow the analysis of within-subject change processes such as fairness trajectories (e.g., Konradt et al., 2016) and its effects on distal outcomes (e.g., work attitudes, employee health; Hausknecht et al., 2011;Walker et al., 2013 (Silverstein, 1988).
This would allow the effects to be calculated over time (including the velocity, acceleration, and inertia in change) and examined in much greater detail; it would also allow us to study change processes during major events and transitions (Bolger, Davis, & Rafaeli, 2003;Soenen, Melkonian, & Ambrose, 2017). We also suggest that the cognitive and affective processes that shape, or are shaped by, perceptions of fairness should be regarded as dynamic processes that can follow different metrics. To capture dynamic psychological processes, our meta-analytic model used a sampling protocol based on a series of specific discrete events (Bolger et al., 2003;Wheeler & Reis, 1991). We, therefore, suggest using models that are based on a multiple-hurdle selection process.
Also, existing longitudinal studies of applicants' perceptions of fairness rely exclusively on self-reports. Although this approach offers several advantages, including making it easy to capture psychologically relevant processes, it is retrospective and thus susceptible to memory bias. For this reason, other procedures might also be useful for monitoring applicant reactions, including unobtrusive methods of capturing behavior (for a review, see Chaffin et al., 2015). Specifically, McCarthy et al. (2017) proposed using web-based services such as organizational messaging and posts on social media to obtain behavioral data. In addition, future research should gather reports from sources other than applicants themselves (for example, from parents or peers) in order to address the issue of single-source bias. As a final point, our analysis has also shown that the studies we examined paid little attention to methodological issues of incomplete data in analyses (e.g., imputation) and have paid little attention to the advantages of latent structure modeling. More recent studies have been seen to respond to this shortcoming by more measurement points and latent growth curve modeling (e.g., Konradt et al., 2016).

| Limitations of the meta-analysis
Although this meta-analysis makes an important contribution to research on applicants' perceptions of fairness, it is not without limitations. The first group of limitations relates to the generalizability of our results. Instead of including as many studies as possible, our goal in this meta-analysis was to be selective by including only those studies that used at least two measurement points and that related to the domain of personnel selection. As a consequence, this metaanalysis included a small number of studies, which may have introduced bias into the analysis. However, including these studies in the meta-analysis nevertheless gives us an initial impression of their impact. At the same time, this limitation highlights areas where future | 379 KONRADT eT Al.
research is needed and may encourage researchers to study some of the relationships that we found.
At the methodological level, several issues emerged from our analyses that also highlight areas where future research is needed. First, the results are based mainly on studies of people in a relatively early adult life stage with limited professional and life experience, which differ in terms of a stable self-definition, stable attitudes, and egocentricity (Ashraf & Merunka, 2017). Consequently, it is unclear whether the findings of our study generalize to more experienced applicants.
Second, because the results of studies in which the applicants were treated unfairly used hypothetical designs, the effects may represent the upper limit of what might be expected, since effects are generally less pronounced in real-world settings (Hausknecht et al., 2004).
Third, the studies available focused on relationships and short-term fairness dynamics. However, studies of longer-term dynamics (e.g., over several weeks or months), which address how perceptions of fairness change over time, are rare. Accordingly, to make significant progress, at least three points in time should be examined, as this would allow us to model within-subject latent trajectories instead of analyzing only mean changes over the short-and medium-term.

| CON CLUS ION
In this meta-analytic review, we have integrated previous research with more than one point of measurement theoretically and statistically and examined changes in perceptions of fairness throughout the different stages of the personnel selection process. The results indicate that applicants' perceptions of fairness decline in a nonlinear way over time.
The decrease is steeper for applicants who have higher expectations of fairness before being assessed. Furthermore, we found both negative and positive changes in fairness perceptions from one selection stage to the next when applicants were treated unjustly, whereas we found no such changes when they were treated justly. The results of this study also indicated that the longer the time interval between the posttest and postdecision stage in the selection process, the smaller the decline in applicants' perceptions of fairness. This work has gone some way toward enhancing our understanding of how and why applicants' perceptions of fairness change throughout the selection process. We hope that these findings will stimulate further longitudinal research on perceptions of fairness in personnel selection.

ACK N OWLED G M ENTS
This study was supported by a grant from the German Research Community (Deutsche Forschungsgemeinschaft; DFG) to the first author (Ko-1412/12-1) that is gratefully acknowledged. Open access funding enabled and organized by Projekt DEAL.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest. All authors have seen and approved the manuscript and warranted that the article is the authors' original work, has not received prior publication, and is not under consideration for publication elsewhere.