Power Asymmetry and Early Intervention in Divorce

Individuals going through divorce often experience an imbalance of power, and this is likely to change throughout the divorce process. In this study, we examine the relationship between perceived differences in relative power among individuals going through divorce and their subsequently reported emotions, appraisals of agreements, and third-party involvement in divorce settlement. Our main expectation was that an initially perceived disadvantage in power would in ﬂ uence subsequent stages of the divorce process, even when the perceived disadvantage reduces over time. Furthermore, we expected an empowering effect of an educational web based intervention that can reach people early in the divorce process. Using a quasi-experimental pretest – posttest design, the sample included 312 Dutch adults who visited (260) or did not visit (52) a web-based intervention and were assessed at three points in time. As expected, and despite a decrease in perceptions of power asymmetry over time, we observed enduring detrimental effects of an early power disadvantage in terms of higher emotional costs, more dissatisfaction with the process and content of the agreements, and more third-party involvement. Interestingly, those who reported power asymmetry (both as disadvantage ánd advantage) also reported more third-party lawyer and less mediator involvement. Also as expected, in this sample, those who reported a power disadvantage and used the web based intervention, reported higher power at a later stage than those who did not use the web intervention. This study points at the importance of signaling, and potentially offering a remedy for, perceived power disadvantages in the initial stages of a divorce process.

have been a central point of focus of conflict research since the 1950s (Emerson, 1962;Tedeschi et al., 1973;Thibaut & Kelley, 1959). Increasingly, asymmetric distributions of power have received scholarly attention in this field of research (e.g., Giebels et al., 1998;McAlister et al., 1986;Olekalns, 1991). One reason for this is that power asymmetry often evokes power struggles (Giebels et al., 1998;Zartman & Rubin, 2002), which is likely to result in conflict escalation and, ultimately, intractable conflict. How power and power perceptions change over time has long been acknowledged as important (Kelley & Thibaut, 1978), yet few studies have investigated the longitudinal dynamic nature of perceived power asymmetry in conflicts and negotiations (Coleman et al., 2012;Fiske & Berdahl, 2007). This is unfortunate, as many and particularly the most severe conflicts develop gradually, over multiple interactions, and over longer periods of time (Giebels et al., 2014). As such, understanding the development of power dynamics in (severe) conflicts over time can offer important insights for intervention and help to prevent further conflict escalation.
In the current research, we explore the effects of perceived power asymmetry over time, in a conflict context that involves one of the most impactful human life events: divorce (Holmes & Rahe, 1967). Given its associated emotional and financial costs, it is remarkable that there is a relative lack of research into the (power) dynamics of divorce conflicts. Most studies on divorce focus on divorce interventions (e.g., mediation, counseling; Sbarra et al., 2012;Shaw, 2010;Strouse & Roehrle, 2011) or look at what happens after the divorce has been finalized by examining adaptation processes (Hetherington, 2003;Lucas, 2005), coparental relationships (Baum, 2003;Emery & Dillon, 1994;Harman et al., 2021), and effects on children (Amato, 2001;Nielsen, 2014;Visser et al., 2017). As far as we know, no studies so far have followed individuals through their divorce process to examine the power dynamics over time.
In this article, we will focus on a longitudinal dataset we collected with individuals in the Netherlands going through divorce. We argue that and test whether perceptions of initial power asymmetry will have a profound influence on subsequent stages of the divorce process and affect emotional as well as economic divorce outcomes, even if, as we expect, the asymmetry itself is reduced over time. We postulate that particularly the initial experience of a power disadvantage (as opposed to a power advantage) will have a detrimental effect on subsequent emotions and appraisals of divorce agreements and their sustainability and will be associated with seeking more third-party involvement in the divorce process.
Assuming that the experience of a power disadvantage is an important precursor for later detrimental outcomes, interventions aimed at countering (the effects of) power asymmetry and its consequences should ideally reach people at the very start of a divorce process, even before they contact legal, social, or psychological professionals. Web-based interventions, which reach people at the point where they start googling "what to do in a divorce," are in a good position to offer such early intervention. We therefore also explore the effects of a prototypical web-based intervention that was developed in the Netherlands, 1 which targets people at the early stages of divorce.

Perceptions of Power Asymmetry in Divorce
To date, research has provided insights into power dynamics in related fields, such as during various phases of romantic relationships (Harman et al., 2019;Kim et al., 2019;Simpson et al., 2019), including relationship decline (Hatfield et al., 2008), as well as in postdivorce families (Harman et al., 2021;Ogolsky et al., 2019). However, to our knowledge, no studies have yet looked at the changing power dynamics during the (legal) dissolution of marriages or at the effects of these power dynamics on the divorce process and resulting agreement.
One of the most encompassing ways to conceptualize power dynamics in relationships is through Interdependence Theory (Kelley & Thibaut, 1978;Kelley et al., 2003;Thibaut & Kelley, 1959;Van Lange & Balliet, 2015). As the inverse of dependence, power refers to the ways in which people can affect one another's outcomes during the course of their interaction. In a divorce process, partners can affect the other's outcomes by controlling the divorce process and its resulting divorce agreement (Farrell et al., 2015;Galliher et al., 1999;Shnabel & Nadler, 2008). For example, process control in the form of agenda setting and involving third parties may affect negotiation outcomes (Magee et al., 2007;Magee & Galinsky, 2008).
Importantly, the literature indicates that it is the subjective experience of power-dependence that is particularly influential in affecting behavior, cognition, and emotions (Bacharach & Lawler, 1976;Fiske & Berdahl, 2007;Fiske & Dépret, 1996;Greer & Bendersky, 2013;Kim et al., 2005). From the perspective of an individual in a divorce process, the division of power in their divorce may either be perceived as symmetrical (an individual perceives both parties to hold either high or low levels of control) or asymmetrical. In the latter case, individuals can see themselves at a power disadvantage or power advantage. Generally, a power disadvantage harms fulfillment of a fundamental need for agency and autonomy, and thus motivates parties with a power disadvantage to pursue strategies to restore control and even the power balance (Fiske & Dépret, 1996;Kim et al., 2005;Sheldon & Gunz, 2009;SimanTov-Nachlieli et al., 2013). That is, parties can actively try to change the balance of power by acquiring and/or withholding information, or by recruiting support (Fiske & Berdahl, 2007;Kim et al., 2005;Zartman & Rubin, 2002). This need is arguably driven by the powerless party's feelings of insecurity and anxiety over the outcomes of the conflict . We argue and expect that perceived advantages or disadvantages in power are likely to decrease during the divorce process (H1).

Effects of Power Asymmetry
The negative effects of power asymmetry are well established in the conflict literature (Coleman et al., 2012;Greer & Bendersky, 2013;Lawler & Bacharach, 1987) and beyond (Anderson & Berdahl, 2002;Galinsky et al., 2006;Keltner et al., 2003). However, research has yet to determine whether and to what extent these negative effects of perceived power asymmetry occur in divorce conflicts. In the current study, we identify and examine three domains in which we particularly expect perceptions of disadvantages to have negative effects on the divorce process in terms of emotional costs (e.g., anger, anxiety, and stress), agreement appraisals, and third-party involvement.

Emotional Costs
Divorce can be associated with negative affect such as anger, rejection or humiliation, anxiety about the future, and stress related to conflict interactions. We know that a lack of power, control and autonomy is associated with a variety of negative effects (Sheldon & Gunz, 2009;Skinner, 1996) including reduced daily well-being (Reis et al., 2000) and increased negative affect (Keltner et al., 2003). In conflicts specifically, experiencing a power disadvantage is associated with heightened levels of stress (Giebels & Janssen, 2005). These effects of perceived power disadvantages are likely to be especially pronounced in a high stakes conflict (Giebels et al., 2014) with a high degree of interdependence between the parties (Coleman et al., 2012; such as divorce. However, there is nothing inevitable or necessary about the association between intense negative affect and the divorce process. Most people adapt well to divorce and come to high-quality divorce agreements, whereas other divorce processes escalate into high-conflict divorce and problematic coparenting relationships (Amato, 2010;Whiteside, 1998;Visser et al., 2017). We hypothesize that a perceived disadvantage in power could be a determining factor in this process, increasing the extent to which individuals experience intense negative emotions. A power disadvantage has also been associated with anger (Berdahl & Martorana, 2006) and humiliation (McCauley, 2017). In a divorce, such reactions of anger and humiliation may be especially pronounced in case of a one-sided decision to initiate divorce (an asymmetry in control over the preservation of the relationship; Sprecher et al., 1998), or when one parent is able to withhold contact with children (Harman et al., 2021).

Agreement Appraisals
We also examine how those who perceive a power disadvantage evaluate the (provisional) outcomes of their divorce. We consider appraisals of the agreement itself and its perceived future sustainability, as well as the process leading up to the agreement. Generally, those in a low power position are likely to be less assertive in negotiations: they make fewer first offers, are less ambitious in their offers, and present offers in a less convincing way (Galinsky et al., 2017). They are also more likely to experience uncertainty (Bollen et al., 2010). This is likely to result in unfavorable outcomes and-reinforced by an experienced lack of control over the process-negative feelings about the obtained outcomes (Bollen & Euwema, 2013;Galinsky et al., 2017;Kim et al., 2005;Thibaut & Kelley, 1959).

Third-Party Involvement
Previous research suggests that in legal conflicts, the need for third-party help is higher for individuals who perceive a power disadvantage (Van Dijk et al., 2016). If a third party is a good fit, help can be beneficial, especially to low power holders (Bryan, 1999;Shestowsky, 2020;Ting-Toomey et al., 2000). However, third-party involvement has important disadvantages. Specifically, it can contribute to less sustainable outcomes, high financial costs, longer procedures, and more destructive conflict (Pruitt & Kim, 2004;Tesler, 1999). Mediation can often contribute to a more positive conflict process but may not be a good fit when the power differences between parties are large (Beck & Frost, 2006;Beck & Sales, 2000).

Enduring Effects of Early Asymmetry
We expect that over time, perceived (dis)advantages in power will decrease. At the same time, we expect that perceptions of a power disadvantage early on in the divorce process will set the tone and process of a divorce resulting in enduring negative effects for those with a power disadvantage (Kelley & Thibaut, 1978). These enduring effects of early power perceptions could be explained by well-researched conflict dynamics. Potential explanations include early negotiation tactics, conflict escalation spirals, and third-party dynamics. Research shows that early negotiation tactics such as agenda setting and first offer making by the high-power party at the outset of the conflict can result in substantial benefits in terms of negotiation outcomes (Magee et al., 2007;Magee & Galinsky, 2008). Early escalating behaviors can have a further profound effect on the conflict and its outcomes. Once parties start exhibiting negative conflict behaviors, it is difficult to stop or even reverse a spiral of conflict escalation (Andersson & Pearson, 1999). High-power parties who are insensitive to concerns of low-power parties and the resulting anger and humiliation of lowpower parties could act as a catalyst for further negative conflict behaviors (Anderson & Berdahl, 2002;Galinsky et al., 2006;Keltner et al., 2003). In legal proceedings such as divorce, an early choice for a third party with an adversarial style can cement these negative patterns (Beck & Sales, 2000;Shaw, 2010;Tesler, 1999). This also implies that negative patterns will be difficult to reverse, even if the underlying power imbalance is remedied. Perceived early asymmetry can thus set in motion behaviors and choices that can have an enduring negative impact. We therefore expect that people who perceive a power disadvantage at the start of the process, compared with a power advantage or power symmetry, will report higher emotional costs (increased anger, humiliation, anxiety, and stress) throughout the divorce process (H2), and more negative agreement appraisals (H3). We also expect them to have recruited more third-party involvement in the medium to long term, both in negotiations (H4a) and in bringing the divorce to court (H4b).

Power Asymmetry and Early Intervention
Given the importance of the early conflict stages, early web-based interventions might be ideally placed to alter early perceptions of power and mitigate the lasting effects of an experienced power disadvantage in the early stages of a divorce process. With digital media, sophisticated interventions can reach individuals long before professional third parties see them. Web-based interventions in legal processes are quickly becoming an important part of the legal aid landscape in many countries (Smith, 2019), and their importance is arguably even larger since the start of the COVID-19 pandemic.
One such intervention is the Dutch 'Rechtwijzer' (RW), which is translated as a signpost or roadmap to justice. It is a government funded web-based advice and support website that was originally developed by the Dutch Legal Aid Council and the University of Tilburg (Raad voor Rechtsbijstand, 2019; Van der Linden et al., 2009;Van Gammeren-Zoeteweij et al., 2018;Van Veenen, 2008;Van Veenen, 2010). The Dutch initiative was unique at the time and received quite some (inter)national attention. Subsequently, similar initiatives have emerged in for example, Canada, Australia, and the United Kingdom (Smith, 2019), and there is some overlap with divorce education programs in the United States (Cronin et al., 2017). The RW intervention was built around three goals: (a) to support self-reliance/empower individuals; (b) to direct parties to appropriate help early in the process when self-reliance was not feasible; and (c) to encourage constructive conflict resolution and avoid a lengthy court process (Van der Linden et al., 2009;Van Veenen, 2008). The website could be easily found through search engines or through links from associated (government) services. Visitors to the website were first prompted to reflect on their divorce process in a short series of questions and were then presented with a custom (only showing information that is relevant to the user) and clearly structured step by step plan for their divorce (Van Veenen, 2008). The intervention used a clear design to structure the complex divorce process. RW offered tools for self-help such as a guide to discussions between divorcing partners but also highlighted access to (subsidized) third-party support (Sandefur, 2015;Shestowsky, 2017) when this was not feasible. Throughout, users were prompted to reflect on the interests of children (if relevant), of both parties, and negative conflict behaviors and emotions.
It is an important empirical question whether such an intervention is widely applicable or whether it is particularly beneficial to specific user groups. 2 If an intervention were able to improve perceptions of one's power position, we would expect this to have lasting effects on the divorce process and result in a more equal divorce agreement. Previous work has shown that feeling powerful results in acting like one has power (Galinsky et al., 2003;Galinsky et al., 2017). Hong and Van der Wijst (2013) found that priming with power reduced the power imbalance in negotiation outcomes, suggesting that altering perceptions of power of less powerful parties can increase their negotiation outcomes. Generally, legal aid practice has warned that those in a more vulnerable position are less likely to benefit from online assistance (Smith & Paterson, 2014;Smith, 2015). However, there are several reasons to expect an intervention such as RW to be especially beneficial to those who experience a power disadvantage. First, effectively structuring information-as RW does-might be especially helpful to those who perceive a power disadvantage, as their higher levels of stress might make ordering complex information more challenging to them (De Dreu & Weingart, 2003;Giebels & Janssen, 2005. 3 Second, psychological research tells us that those without power are likely to be motivated to regain control (Shnabel & Nadler, 2008;SimanTov-Nachlieli et al., 2013) will pay more attention to information related to their divorce (Fiske & Dépret, 1996; and can benefit from reducing uncertainty at the outset of a conflict (Bollen & Euwema, 2013). Third, individuals may perceive themselves at a power disadvantage because they may not be aware of potential power bases such as subsidized legal aid or may not know how to capitalize on available power bases if they lack knowledge of the legal framework of a divorce (Fiske & Dépret, 1996;Greer & Bendersky, 2013;Kim et al., 2005). We therefore expected individuals who experience a disadvantage in power to benefit most from this web-based intervention, in terms of improved perceived own power (H5).
To summarize, in this article we argue that a perceived disadvantage in power early in the divorce has a profound impact on the further course and outcomes of a divorce process (H2-H4), even if the perceived imbalance decreases over time (H1). We also examine whether an early web-based intervention during a divorce process can counter this perceived disadvantage in power (H5).

Method
We included longitudinal data from one sample of individuals (N T1 = 312) in the Netherlands who were going through divorce. Participants were users of a web-based intervention (RW) offered by the Dutch Council for Legal Aid and a control group. We report how we determined our sample size, all data exclusions, and all measures in the study.

Sampling Procedure
Participants filled out three questionnaires over the course of their divorce process. Visitors to the web-based intervention were prompted with a request for participation in our study at the start of their visit to the "divorce" section of the website. Respondents were not made aware at this point of the goal of exploring the effects of the use of the intervention. We communicated to them only the goal of understanding the process of divorce as it develops over time. Those who agreed to participate were immediately redirected and filled out the first survey, before continuing their visit to the website (intervention). We recruited participants for the control group (i.e., individuals going through divorce without using the intervention) through messages on diverse national and regional radio, both commercial and public, in national and regional newspapers, and on social media with the same introduction to the research. Data were collected between July 2013 and June 2015. Participants received a 10-euro gift card after participation.
The second survey was emailed to respondents 1 week after the first survey. Those respondents who did not (yet) fill out the second survey were reminded up to four times, after every 5-7 days. The third survey was sent 5 months after the second. Surveys were closed to responses two weeks after the last reminders. We did not do any analyses before closing the surveys.

Sample Selection
There were 534 entries on the pretest, 429 in the test group and 105 in the control group. We removed data from 34 respondents who did not meet the criteria. For details, see Figure 1. In addition, we only included relationships that had to be dissolved in court. In the Netherlands, this means legal marriage and registered partnerships when the partners have children under the age of 18. 4 We removed those with other relationship forms (including cohabitation; n = 114). 5 To ensure that vulnerable individuals received appropriate help, the RW intervention included a question that asked if there was or had been violence in the relationship. If this was the case, the website immediately redirected these users to a page with direct links to access in person help. Because these individuals no longer received the RW intervention, they no longer qualified for the RW intervention group of this study. We copied the question on violence from the intervention for the control group questionnaire to ensure selection was done on identical criteria. In total, 47 participants were removed based on this criterion.
At T 2 , respondents reported timelines of their divorce processes. We removed individuals who had not made the final decision to end their relationship at the time of the first survey (n = 15). 6 We also removed those who had started their divorce process (arrangements or negotiations) more than a year before the first survey (n = 9) to avoid overrepresentation of people with complex and escalated divorce processes in our sample.
Finally, we could check whether both partners in a couple had participated in our study as we asked participants to create an 3 Anecdotal evidence from focus groups held during the development of RW suggests that the structured overview of information is the most appreciated feature of the website. See internal documents of Dutch Legal Aid Board. 4 For registered partnerships when the partners have children under the age of 18, a co-parenting agreement has to be presented to the court for approval. 5 A total of 16% of the intervention group had a cohabitation agreement and 9% cohabitated without a formal agreement. This shows the intervention does reach these groups to a certain degree and a separate exploration for the effect of the website for this group is needed. In this study, as our control group consisted only of people who were legally married or in a registered partnership, we did not examine the cohabitation group further. 6 Twelve of these people were from the intervention group, which suggests that people (also) use the intervention to look for information on the possible consequences of a final decision to end the relationship. We included a question on the T 2 survey to check whether individuals were indeed in a divorce process at that time.
anonymous code (Kearney, 1984) based on their and their partner's birthday and the first letters of their names. We found three couples and randomly removed one person from each couple to prevent interdependence in the dataset. The resulting dataset included 312 individuals, of which 260 in the test group and 52 in the control group on the first measurement (T 1 ), 183 and 46 on the second measurement (T 2 ), and 137 and 35 on the last measurement (T 3 ) respectively 7 (Van Dijk et al., 2022).

Sample Descriptions
As sampling methods differed, we analyzed differences between the intervention and control groups. All descriptive statistics and comparisons can be found in Tables 1 and 2. There were no statistically significant differences between the intervention and control groups on age, gender, being employed or not, partner's education, income categories, owning a business, being married versus in a registered partnership, or having children under the age of 21. There was a statistically significant difference in education of the respondent, v 2 (3) = 7.91, p = .048, V = .16 (see also Table 1). For each respondent who reported their divorce timeline on T 2 (n = 210), we calculated the number of weeks between starting divorce arrangements or negotiations and the first survey on T 1 (see Table 2). In the intervention group, average time elapsed between starting divorce arrangements or negotiations and T 1 was shorter than in the control group, t(208) = 2.80, p = .006, d = .48, We therefore also controlled for time elapsed in our analysis of the effect of RW. 8

Power Asymmetry
We measured perceived power by asking to what extent respondents felt they and their (ex-)partner controlled the process and outcomes of the divorce (Shnabel & Nadler, 2008;Van Dijk et al., 2016). All four items were measured on a seven-point scale where 1 represented low control and 7 represented high control. Respondents indicated to what extent (1: not at all, 7: to a very large extent) they agreed with statements about their own and other's perceived control over the outcome as well as over the process of the divorce. Table 3 shows the items, item means, standard deviations, and correlations for these perceived power measures on T 1 , T 2 , and T 3 .
Based on the high correlations, we combined control over process with control over outcome into perceived power measures for the respondent and the partner. There were statistically significant differences between the control and intervention groups in perceptions of These numbers correspond to respondents starting each survey. A few respondents drop out during each survey. We will report ns for each variable. 8 In the Netherlands, 90% of the divorce cases in court last between 6 weeks and a year (Rechtspraak Servicecentrum, 2017). However, successful or unsuccessful negotiations will almost always have taken place before a divorce is taken to court. In the Netherlands, divorce agreements have to be ratified by a judge, but the majority of these cases are dealt with relatively quickly. This is because often, a prepared agreement proposal is submitted for ratification only. The full divorce process, starting from the (hard to pinpoint) decision to divorce can thus last much longer (Amato, 2010;Emery & Dillon, 1994;Symoens et al., 2013). 24, did not show statistically significant differences. As we set out in the Introduction, parties going through divorce can experience high or low power and can perceive the other party to hold high or low power. This means that next to a distinction between symmetry and asymmetry, we can include both high and low power symmetry (Giebels et al., 2000;Rubin & Brown, 1975;Schaerer et al., 2020;Zartman & Rubin, 2002). To identify how these patterns of the perceived balance or imbalance in power between (ex-) partners (high-high, high-low, low-low, low-high) are distributed in the current sample, we conducted SPSS two-step cluster analysis based on Log-likelihood distances (Bacher et al., 2004;Clatworthy et al., 2005;Henry et al., 2005;Marquand et al., 2016). We only used variables from T 1 in the cluster analysis, as we were interested in effects of initial asymmetry. We set the number of clusters to four. Resulting clusters were perceptions of positive asymmetry (a power advantage, n = 41), perceptions of symmetry at low to middle levels of power (n = 126), perceptions of symmetry at high levels of power (n = 98), and perceptions of negative asymmetry (a power disadvantage, n = 47). We validated this cluster solution by running two ANOVA analyses to test whether the clusters predicted different values of perceptions of the respondent's own power and Note. N = 312 (n intervention group = 260, n control group = 52). a There were nine same sex couples in the sample. b Income questions were answered by 175 respondents. Income was measured with two questions. Respondents reported what their gross yearly income was (no income, below 10,000, 10,000-20,000, 20,000-30,000, 30,000-40,000, 40,000-50,000, above 50,000). Respondents also reported whether they had income to fall back on in emergency, for example from a parent, partner, or relative. Chisquared tests were done with the original variables. Information on the categories was condensed for the  Table S1 in the online supplemental materials. There were no associations between the four clusters and the demographic variables, age, sex, having children, and owning a company, all Pearson v 2 , 4.72, all ps . .19. There was an association between the clusters and educational level, with more people with relatively lower educational levels experiencing asymmetry (both negative and positive) and more people with relatively higher educational levels experiencing high power symmetry, v 2 = 17.84, p = .04, V = .14. More people whose partner had relatively lower educational levels experienced positive asymmetry, and more people whose partner had relatively higher educational levels experienced high power symmetry, v 2 = 21.89, p = .01, V = .15.

Emotional Costs
We measured four main constructs of divorce related negative affect, two representing stress and anxiety and two representing negative emotions toward the partner. All four constructs were measured on a 7-point scale where 1 represented low negative affect and 7 represented high negative affect. We conducted scale analyses for each measurement.

Stress Surrounding Interaction With Partner
We assessed conflict related stress using four items of the questionnaire developed by Giebels and Janssen (2005). Questions were designed to test stress associated with interactions with the other party. As not everyone had spoken to their (ex-) partner about the divorce at the time of the pretest, only 256 respondents (82% of total sample) answered this question. Scree plots for each measurement showed one clear factor, which explained 75% of variance on the T 1 , and 81% on T 2 and T 3 . All factor loadings were above .75 and Cronbach's alpha was .89 on T 1 , and .92 on both T 2 and T 3 .
Three emotional cost constructs were measured by presenting emotion words to respondents and asking them to what extent they experienced these emotions in relation to their divorce (Fischer & Roseman, 2007;Fitness, 2000;Frijda et al., 1989;Lazarus, 1991;Weiner, 1985;Wetzer et al., 2007).

Divorce-Related Anxiety
Anxiety was measured using two items. Respondents reported to what extent they felt anxious and worried about the divorce process and about the time after the divorce, r(310) = .74, p , .001, r(205) = .84, p , .001, and r(156) = .72, p , .001 on T 1 , T 2 and T 3 respectively.

Anger Toward Partner
Anger toward the partner was measured with three items. Respondents indicated to what extent they felt anger, indignation, and rage toward their (ex-) partner. A scree test showed one clear factor, which explained 72%, 78%, and 76% of variance for T 1 , T 2 , and T 3 , respectively. All factor loadings were above .66 and Cronbach's as were .88, .86, and .84.

Humiliation by Partner
Using three items, respondents indicated to what extent they felt insulted, humiliated, and belittled by their (ex-) partner. Scree tests for all three measurements showed one clear factor, which explained 71%, 83%, and 81% of variance on T 1 , T 2 , and T 3 respectively. All factor loadings were above .78 and Cronbach's as were .88, .89, and .88. Table 4 shows means, standard deviations, and correlations between these four emotional costs constructs for T 1 , T 2 and T 3 . For each time of measurement, the four variables were moderately to strongly correlated. Therefore, to condense the data and reduce the number of analyses, we averaged these variables to construct emotional costs measures for T 1 -T 3 each. Exploratory factor analyses using principal axis factoring using the four affect variables supported the validity of a general emotional cost factor, which explained 62%, 63% and 63% of variance on T 1 , T 2 and T 3 respectively. All factor scores were above .50. Alphas for the emotional costs scale on T 1 , T 2 and T 3 were all .79.

Agreement Appraisals
On T 3 , respondents evaluated the process as well as the outcomes of their divorce. Final agreements had been reached in the divorce procedures of 65 (44%) respondents, and 84 (56%) respondents answered questions about provisional or temporary agreements as they stood at T 3 . On a scale from 1 to 7, respondents reported to what extent they were content with the process of their divorce, felt the (provisional) divorce agreements were sufficient and sufficiently detailed, to what extent they felt the (provisional) divorce agreements were to their advantage and to their ex-partner's advantage, to what extent they felt that they and their ex-partner were content with the (provisional) divorce agreement, and, finally, to what extent they expected the agreements to be sustainable in the future. We used the individual items in our analyses, as we were interested in comparing effects on the different aspects of the agreements by respondents. Means, standard deviations and correlations can be found in Table 5.

Third-Party Involvement
We asked respondents about third-party involvement in the preparation of the agreement as well as in court.

Third-Party Involvement in Preparation of Agreement
Respondents first reported whether they had drawn up their divorce agreement and negotiated mostly (a) with their ex-partner without the help of a professional, (b) with the help of a mediator, or (c) whether they had left most of the negotiations to their lawyers. At T 3 , 30% of respondents had negotiated mostly with their ex-partner without the help of a professional, 48% used a mediator, and 22% had left most of the negotiations to their lawyers.

Third-Party Involvement in Bringing Divorce to Court
To finalize a divorce in the Netherlands, a lawyer has to present the case to court for approval. We therefore asked respondents whether they, together with their partner, hired one lawyer who presented their settlement to court for approval only, or whether they were not able to come to a common agreement and both hired their own lawyer. 9 The majority (70%) had contracted one lawyer together with their (ex-) partner and 21% had each had their own lawyer.

Dropout
Of the total sample of 312 respondents, 164 persons completed all three surveys. Dropout was lower in the control group where 65% of respondents completed all three surveys versus 50% in the intervention group, v 2 (1, N = 312) = 4.11, p = .04, V = .11. Respondents from couples where at least one partner owned a company were more likely to drop out of the study than those who did not own a company, v 2 (1, N = 312) = 6.10, p = .01, V = .14. There were no other statistically significant differences between respondents who did and did not drop out of the study in terms of background characteristics. Additionally, dropout was comparable across the four power perception clusters, v 2 (1, N = 312) = .59, p = .90, V = .04. Thus, aside from company ownership and group membership (intervention vs. control group), there were few differences between participants who did or not complete the three surveys, suggesting that dropout was largely random. 10

Results
We first tested our hypothesis on the development of power perceptions over time. We ran a mixed effects ANOVA to test our expectation that a perceived disadvantage or advantage in power would decrease over time (H1). The dependent variable in this analysis was power perceptions and we included two fixed within subject effects: time with T 1 , T 2 , and T 3 , and partner with perception of self and perception of other (a large negative difference between perception of self and perception of other equaled a perceived disadvantage in power). 9 Divorce cases where both parties hire a lawyer can take longer and can escalate further (Ter Voert, 2009), as the case is discussed in the oppositional setting of a courtroom. In some couples, one partner had contracted a lawyer and the other had not. As self-representation is not possible in divorce cases, this could mean that partner 1 did not oppose partner 2. However, it is equally possible that partner 1 was eligible for legal aid subsidies and partner 2 or the partners combined were not. In these cases, a seemingly one-sided divorce request is in reality a divorce by mutual consent and agreement. We excluded this group when we analyzed results related to this variable. After exclusion, N was 135, with 104 respondents reporting one lawyer and 31 respondents reporting two lawyers in opposition. 10 Anonymized data relevant to this publication are available from https:// osf.io/5gzqh/?view_only=5db8ad01cf2f4fdfa76085894e27e8f4. Correlation between the main study numeric study variables can be found in Table S3 in the online supplemental materials to this article. 8 VAN DIJK, ZEBEL, AND GIEBELS This document is copyrighted by the American Psychological Association or one of its allied publishers.
We examined whether changes in power perceptions were driven by the use of the intervention (H5). Our expectation was that for individuals who perceived a disadvantage in power, the effect of using RW (compared with not using RW) on perceived own power at T 2 would be positive. We tested this expectation using a threeway interaction effect (between perceived own power, perceived other's power, and using RW) in a regression analysis with power on T 2 as dependent variable. In light of the quasi-experimental nature of this study, we controlled for the number of weeks since starting divorce arrangements, age, sex, education, partner's education, employment status, partner's employment status, having children, and owning a business.
We then tested our hypotheses on the effects of perceptions of negative asymmetry. For these analyses, we used the four power perception clusters to facilitate interpretation (positive asymmetry, low-middle power symmetry, high power symmetry, negative asymmetry). We hypothesized that individuals who saw themselves at a power disadvantage at the outset of the process, would be more likely to report higher emotional costs (H2), lower evaluations of process and outcome (agreements) (H3), and more thirdparty involvement (H4) later in their divorce. We used a mixed effects ANOVA to examine emotional costs over time (within subjects: T 1 , T 2 , and T 3 ) and per cluster (between subjects). We tested our hypotheses on the effect of perceived power (represented by the four clusters) on agreement appraisals on T 3 and third-party involvement on T 3 , using ANOVA and chi-square analyses.

Power Asymmetry Over Time
We ran a repeated-measures analysis of variance with time (T1, T 2 and T 3 ) and partner (perception of self, perception of other) as within-subjects factors. We tested our expectation that the perceived difference in power (the effect of partner) would decrease over time. There were no main effects for time, F(1.92, 266.71) = 1.00, p = .37, h 2 = .01, or for partner F(1, 139) = 2.16, p = .14, h 2 = .02, but the expected interaction effect of time and partner was statistically significant, F(1.79, 249.04) = 7.87, p , .001 h 2 = .05, confirming that respondents perceived a decline in power disparity over time. To illustrate the decline in asymmetry in more detail, We can inspect the power perceptions pattern graphically and per cluster (see Figure 2). We see that when respondents felt that power positions were asymmetrical at the start of their divorce process, they were likely to report a decrease in power differences over time, whereas respondents who reported symmetry at the start of their divorce process, continued to report symmetry.

Effect of RW on Perception of Own Power
To test hypothesis 5, that people who perceive the other party as powerful but do not feel powerful themselves, will benefit most from the intervention, we used multiple linear regression analysis. Our dependent variable was perceived own power at T 2 , one to four weeks after the use of Rechtwijzer (RW). We controlled for time elapsed since starting arrangements, the sociodemographic  This document is copyrighted by the American Psychological Association or one of its allied publishers. and divorce-related variables, as well as perceived own and perceived other's power on T 1 . In addition, we included the interaction of the pretest power measures (Own 3 Other Power) to test for effects of early asymmetry. 11 To test our hypothesis, we added RW use and the interactions between the use of RW and the perceived power variables as well as the three-way interaction term between both perceived power measures and RW use. We predicted that the use of RW would have the strongest empowering effect to those who felt relatively powerless. In other words, we expected that when perceived own power was low and perceived other's power was high (a perceived power disadvantage), the effect of RW would be strongest (H5). The model (see Table 6) explained 43% of variance on perceived own power on T 2 , F(20, 185) = 7.01, p , .001. 12 The three-way interaction was not statistically significant, b = À.19, SE = .28, t(185) = À.69, p = .49, but we did observe a marginally significant interaction effect of perceived other's power and RW on perceived own power, b = .43, SE = .23, t(185) = 1.89, p = .060. However, we note that participant numbers in the control group on T 2 were low (n = 46) and VIF and tolerance scores for the interaction terms were high, which makes it difficult to pinpoint where in the interplay of these variables the interaction effect is located.
We can illustrate our findings with Figure 3, in which we show mean perceptions of perceived own power in the intervention and control groups on T 2 across the four clusters. We see that the intervention group in the negative asymmetry cluster reports feeling more powerful than the control group. This difference does not appear on T 1 (see Figure 4).

Emotional Costs
Consistent with H2, and throughout the process, those who perceived negative asymmetry on T 1, reported statistically significantly higher emotional costs than respondents in the other three clusters did (see Figure 5, and Table S2 in the online supplemental materials). A mixed design ANOVA showed that the expected between subjects effect of cluster on emotional costs was statistically significant, F(3, 142) = 5.34, p = .002, h 2 = .10. In addition, these analyses indicated a statistically significant effect of time, indicating that emotional costs decreased, F(1.91, 270.83) = Own and Other's Power Over Time per Cluster Note. Participant numbers on T 1 , T 2 , T 3 respectively: Positive asymmetry: n = 41, 29, 21, Low-middle power symmetry: n = 126, 85, 63, High power symmetry: n = 98, 64, 44, Negative asymmetry: n = 47, 31, 23. Means and standard deviations can be found in Table S1 in the online supplemental materials. 11 Using both original measures as well as the interaction term is preferable to using the difference score as it allows us to examine the effect of asymmetry, without losing information on total perceived power in the dyad (Johns, 1981;Van Dijk et al., 2016). 12 In a regression analysis which did not include the effect of RW, 17.72, p , .001, 13 h 2 = .11. Finally, there was no interaction effect between time and the power perception clusters, F(5.72, 270.83) = 1.27, p = .27, h 2 = .03, indicating that early effects of asymmetry persisted. We explored these statistically significant main effects in pairwise comparisons with Bonferroni corrections. Emotional costs in the negative asymmetry cluster were higher than in the positive asymmetry cluster, M diff = 1.04, SD = .36, p = .028, and the high-power symmetry cluster, M dif = 1.20, SD = .31, p = .001. Comparisons also showed that the reduction of emotional costs in the short term (T 1 to T 2 ) was marginally significant, M T1-T2 = .20, SD = .09, p = .07, and that the effect was stronger in the long term (T 2 to T 3 ), M T2-T3 = .40, SD = .11, p = .001.

Agreement Appraisals
In accordance with H3, in a repeated measures ANOVA, the asymmetry clusters significantly predicted all divorce process and outcome (agreement) evaluations by respondents on T 3 ; all Fs[3, 145] $ 3.36, all ps # .02, all, h 2 . .07, with the exception of the perceived (ex-)partner's contentment with the outcome of divorce, F(3, 145) = 1.82, p = .15, h 2 = .04. A first inspection shows that the most positive appraisals are given in the high-power symmetry cluster, and the most negative appraisals are given in the negative asymmetry cluster (see Table 7). Pairwise comparisons with Bonferroni corrections showed that those who perceived a power disadvantage on T 1 were less content with the process (p = .016, p = .010, and p , .001 for comparisons with the positive asymmetry, low-middle power symmetry, and high-power symmetry clusters, respectively), and with the (preliminary) agreements of their divorce (p = .031, p = .061, and p , .001 for comparisons with the positive asymmetry, low-middle power symmetry, and high power symmetry clusters respectively). They also scored lower than those in the high-power symmetry cluster scored on feeling that the agreements were sufficient and sufficiently detailed, p = .008, and would be sustainable in future, p = .051. 14 In addition, individuals in the negative asymmetry cluster were more likely than those in the positive asymmetry cluster, to feel that agreements were to the advantage of their ex-partner, p = .020, and less likely to feel that they were to their own advantage (p = .071, and p = .055, for comparisons with the positive asymmetry, and high-power symmetry clusters respectively). All other comparisons between the four clusters were not statistically significant, all p's . .178.
To test for perceived asymmetry in the agreement appraisals, we compared evaluations of own content with estimated content of the partner and evaluations of own advantage with estimated advantage of the partner. We found that individuals in the positive asymmetry cluster felt more content than they estimated their partners would feel, t(20) = 2.27, p = .034, d = .50. We found that those in the negative asymmetry cluster reported a disadvantage in the agreements compared with their partner, t(22) = À3.09, p = .005, d = À.64. There were no further statistically significant differences (all p's . .125).

Third-Party Involvement in Preparation of Agreement
Consistent with H4a, when respondents perceived a power disadvantage early in the divorce process, there was more involvement of lawyers in preparing the divorce agreement (47.8%), than among respondents in the other three clusters (13% to 29%) and much less negotiating without third parties (13%) than in the other three clusters (32-34%), v 2 (6, N = 149) = 13.33, p = .04, V = .21, see also Figure 6. Mediator guided negotiations were more popular in the symmetry clusters (51% for low-middle power and 52% for high power) than in the asymmetry clusters (38% and 39% for positive and negative asymmetry, respectively).

Third-Party Involvement in Bringing Divorce to Court
As expected, under H4b, for respondents who perceived a power disadvantage, an adversarial procedure with two lawyers was much more likely (57.9%) than in the other three clusters (12% to 27%), v 2 (3, N = 135) = 16.65, p , .001, V = .35 (see also Figure 7).

Discussion
Divorce is a major life event (Mancini et al., 2011;Sbarra et al., 2012), yet we know little about how the process of a divorce unfolds. This is especially true for the initial stages, which take place beyond the sight of courts, lawyers, and mediators. In this study, we examined the initial perceived power balance of divorcing parties and its development over time, as well as enduring effects of early perceptions of power disadvantages on emotional and financial costs. We also explored whether a web-based intervention (RW) that can reach people very early on in their divorce process, can affect the initial perceptions of (own) power. In line with our expectation, and regardless of whether it concerned an initial advantage or disadvantage, perceived asymmetry decreased over time. However, and also in line with expectations, we observed prolonged detrimental effects of early perceptions of power disadvantages in the medium and long term. In light of the decrease in perceived asymmetry, the pattern of negative effects Perceptions of Own Power on T 2 Note. Participant numbers in control group: n = 6, 22, 7, and 7, respectively; in intervention group: n = 23, 63, 57, and 24, respectively.

Figure 4
Perceptions of Own Power on T 1 Note. Participant numbers in control group: n = 7, 27, 9, and 9, respectively; in intervention group: n = 43, 99, 89, and 38, respectively. was remarkably persistent. First, individuals who reported a power disadvantage early on in their divorce reported higher levels of negative emotions including stress, anxiety, anger, and humiliation. We note the parallel with equity theory which tells us that an experienced lack of equity (balance in the ratio of contributions and benefits of each party) in a romantic relationship is associated with distress such as anger, sadness, and resentment toward the partner (Hatfield et al., 2008). Second, early perceptions of power disadvantages predicted lower assessments of quality and sustainability of the agreements and perceived disadvantages in the agreement. This is noteworthy, not only because it may predict further litigation (Koel et al., 1994) but also because we know that negative divorce experiences can translate into negative coparenting relationships (Visser et al., 2017). Remarkably, the group who saw themselves at a disadvantage in power early in the divorce as well as at a disadvantage in the agreements later in the divorce, reported no difference in contentment about the agreement between them and their (ex-)partner. This was mirrored by individuals who saw themselves at a power advantage early on. They reported that they were more content about the divorce agreement than their (ex-)partner was but saw no difference in advantages in the divorce agreement. These findings suggest that divorcing partners anchor their assessment of what a good outcome would be on what they can expect to gain or lose (Welsh, 2004). Effect sizes of the decline in asymmetry over time as well as the enduring effects of asymmetry on emotions and evaluations of the agreement ranged from medium to large effects. Finally, perceived power disadvantages early in the divorce process predicted more involvement of lawyers, less direct negotiations with the other party, and more adversarial procedures in court, all of which are associated with higher financial costs during the divorce and a higher potential for further litigation in postdivorce families (Quek Anderson et al., 2022). Effect sizes for third party involvement were small to medium. With respect to the RW intervention, effect sizes were less clear owing to a lack of power and multicollinearity. Our sample results suggest that only individuals who experience a power disadvantage, and who use the RW intervention, are empowered (report higher perceived power) during the early weeks of their Emotional Costs per Cluster Over Time Note. Participant numbers on T 1 , T 2 , T 3 respectively: Positive asymmetry: n = 41, 30, 21, Low-middle power symmetry: n = 126, 85, 66, High power symmetry: n = 98, 64, 48, Negative asymmetry: n = 47, 31, 23. Means and standard deviations can be found in Table  S2 in the online supplemental materials. divorce. In other words, those who face a power disadvantage might be most susceptible to an early web-based intervention like RW Shnabel & Nadler, 2008).

Limitations
This study has some notable limitations. First, the analysis of the effect of the intervention is based on a quasi-experimental pre-post design, because it was impossible to select participants randomly or to assign participants randomly to the intervention and control group. Although we included covariates to account for differences in demographic and divorce related variables as well as perceptions of power on the first measurement, considering our study design and the differences in attrition rates between the intervention and control groups, we cannot exclude selection effects.
Second, our analyses of the development of perceptions of power balance over time and correlations to negative emotions, agreement appraisals, and third-party involvement are of a descriptive kind. Whether there is a causal link between these factors will have to be examined in future research. We note that gender (Kaufman & Pulerwitz, 2019;Raley & Sweeney, 2020), children (Ogolsky et al., 2019), and different types of power sources (Pratto et al., 2011) are likely important factors in these processes.
Third, we used clusters based on perceived balance or imbalance of power to draw a broad picture, while avoiding the pitfalls of using difference scores (Johns, 1981). Cluster analysis lends itself to creating clarity and is specifically suited to this study, as the correlations  Third-Party Involvement in Bringing Divorce to Court per Cluster Note. Positive asymmetry: n = 19, Low-middle power symmetry: n = 57, High power symmetry: n = 40, Negative asymmetry: n = 19. between the clusters and the negative effects are relatively large and clear and are easily made visible using the clusters. At the same time, we do not suggest that people who go through divorce can be divided into these four clear-cut categories (Marquand et al., 2016). Future studies should aim for higher participant numbers, allowing for analyses that are more sophisticated.
Fourth, the population of this study was limited to divorcing partners who did not report violence in the relationship. Yet, a distinction between structural and incidental violence, or in intensity of violence, mutuality of violence, and whether children were victimized or not, was not made but could be potentially relevant to more precisely determine whether and to which population the current results can be generalized (Davidson & Beck, 2017;Watson & Ancis, 2013). A study that would include individuals who experience partner violence would require a study design with access to dedicated help.
Fifth, the measures we used in this study were taken from various fields of psychology as well as from practice and were mostly short to avoid overburdening respondents who were going through divorce. In future studies, a more fine-grained analysis, particularly of the emotions during the divorce process would be valuable to better understand the type and extent of emotional costs of an early perception of power disadvantage. Finally, and in relation to the measurement of perceived power, these measurements were limited to the divorce process and viewed from the perspective of one of the divorcees. Future research should strive to create a more complete picture by recruiting couples rather than individuals, and include other methods next to self-report questionnaires, for example measuring actual power bases as observed by independent observers (Tan et al., 2019). Research could also look more closely into the temporal context of power dynamics in the divorce process. Note that we did not expect nor observe a decline in absolute power-dependence. An important reason for this is that our research endeavors focus exclusively on the divorce process itself. During this process, feelings of dependence are inherently present, as the parties continuously need each other to attain their goals and ultimately reach an agreement. Of course, when a relationship ends, there is a process of detachment, which in most cases would reduce mutual dependence. However, research has shown that shifts in power divisions are likely to start earlier, when relationship quality declines (Fine & Sacher, 1997;Hatfield et al., 2008;Rusbult et al., 2011;Sprecher et al., 1998). In addition, the literature on postdivorce families shows that in most families, there is a level of enduring dependence on the other parent and power dynamics between coparents continue to evolve (Harman et al., 2021;Ogolsky et al., 2019). It is likely that the decrease in mutual dependency between separating romantic partners could best be observed when we take into account both the final stages of a relationship and the postdivorce dynamic when the power division crystallizes into a more long-term state of weaker dependence.

Implications for Policy
This study included a first exploration of the impact of the Rechtwijzer (RW) intervention with a focus on asymmetry of power and empowerment. This framework offers a template for how to think about effects of these programs for different groups, with a focus on potentially vulnerable groups. With the launch of the RW website, there were concerns of the legal aid field that those with a power disadvantage would not benefit from online assistance (Smith, 2015). These concerns are only scarcely reflected in our research findings. Although individuals who used the RW intervention perceived a slightly higher level of own power than those in the control group, those who experienced a power disadvantage seem to benefit most from using the intervention. This exploration of the effect of RW suggests that web-based interventions that reach individuals early on could potentially offer a form of empowering help that is beneficial particularly to vulnerable/disadvantaged parties (Van Dijk et al., 2016).
In terms of implications for the field of divorce support (legal as well as social/psychological), this study shows the importance of early process perceptions and interventions. We highlight two elements. First, our findings show that perceptions of power, which can be easily measured by way of self-report, are a promising and strong early indicator for a costly divorce process, which could be very useful to professional third parties. The perception of a power disadvantage early in the process increases the risk of emotional and economic costs, which warrants alertness to these perceptions by third parties.
Second, this study dealt with early process dynamics. The RW intervention mostly targets the early stage in a divorce process where individuals orient themselves toward the complex procedure. Importantly, RW was able to reach individuals in the early stages of their divorce, long before they had received any guidance from professional support parties such as lawyers or mediators. Policymakers can make use of these relatively minimal early interventions that can be made available online to help individuals start out on a constructive conflict path. This does not mean that these types of interventions will make third-party support obsolete for all their users, although it might do for some. Rather, these interventions offer an opportunity to provide help before third parties come into the process.
A final observation with respect to field implications was that across groups, participants in our study indicated that filling out the questionnaire helped them to see their divorce process more clearly. This unanticipated effect of our study suggests that doing research can function as an intervention in itself. In particular, sufficient reflection on the psychological aspects of the divorce process could offer important benefits for the parties involved (Larson & Sbarra, 2015).

Conclusion
This study is unique in that it focused on power perceptions during the divorce process rather than during relationship decline or in postdivorce families. It empirically demonstrated the dynamic nature of the divorce process, the important role and consequences of early power perceptions, and the potential of web-based interventions that target individuals in the very first stages of a divorce process to empower those who are most vulnerable.