Introduction

Emotion work is characterized as activities relevant to the enhancement of significant others’ emotional well-being and the giving of emotional support (e.g., offers of encouragement, listening closely to partner; Erickson 2005). Emotion work is an important construct in predicting divisions of household labor, childcare and relationship satisfaction, and contributions to family life (Erickson 2005; Holm et al. 2001; Pfeffer 2010; Stevens et al. 2006). Emotion work is distinguished from other kinds of work in that the former cannot be delegated to individuals outside of the partnership or group (Erickson 2005), and from emotional labor, or work occurring in the paid or public sections of the market economy (Hochschild 1989).

Using the gender relations perspective from feminist theorizing (Ferree 1990, 2010; Fox and Murry 2000; Risman 2004), we argue that emotion work is tied to relational power attributed to men and women through exaggerated gender differences and historically and socially constructed gender roles. Emotion work can underscore greater relational inequality between partners, including perceptions that women are held accountable for emotion work in ways that men are not (Daniels 1987), or where women consider the provision of emotion work to be work, whereas men consider emotion work to be part of the romantic relationship (Erickson 2005). Unequal emotion work is associated with dissatisfaction and dissolution of martial relationships when women put in the work to connect to their husbands and their husbands do not reciprocate (Duncombe and Marsden 1993). Despite these negative findings, emotion work can also allow for greater relational equality between partners, including positive feelings about the relationship when both members contribute to emotion work (Holm et al. 2001).

We examine how emotion work predicts relationship quality in a sample of women and men in romantic relationships in the U.S. We extend the existing knowledge about gender issues and relationship quality in several ways. First, we use the gender relations perspective of feminist theorizing to understand the association between emotion work and relationship quality for both partners in the romantic relationship. In this way, we focus our discussion on family rather than on women or men (Ferree 2010) while also advancing the feminist challenge of conventional views of “separateness and solidarity” (Ferree 1990, p. 870). Second, we use a daily diary approach to study both emotion work and relationship quality. Daily diary studies are important for untangling the lives of individuals in coupled relationships (Bolger et al. 2003). Third, we take a nuanced approach in understanding emotion work, separating emotion work into effects of trait (i.e., individuals’ average levels) and state (i.e., individuals’ daily fluctuations). Fourth, we examine emotion work in association with six unique types of relationship quality, including positive (love, commitment, satisfaction, closeness) and negative (ambivalence, conflict) relationship quality. This increased specificity permits a more nuanced understanding of the ways in which emotion work differentially influences relational quality.

All empirical studies we describe use U.S. samples unless otherwise noted. We recognize that conceptualizations of emotion work and relationships are not universal. We hope that readers from other countries may be able to use the information gained from this study as a starting point for examination of emotion work and relationship quality to further advance a discussion of gender issues evidenced in romantic relationships and within families in their own research and dialogue with one another.

Conceptual Framework

A feminist perspective is purposely inclusive of many meanings and practices, with core themes including power and agency, as manifested in family relationships (Allen et al. 2009). Feminism is central in understanding family relationships because families consist of the emotional and physical locations in which constructs such as power, conflict, and care come together to explain how people live their lives (Few-Demo et al. 2014).

Although feminist scholars have been challenged to study power in other contexts outside of the home (Collins 2000), the study of individuals within families on issues in the home remains a central one for feminist scholars (Hochschild 1989; Lachance-Grzela and Bouchard 2010). The understanding of which family members come into, maintain, or lose power, or the perceptions of power within families, is an important one for feminist scholars because the ability to understand such power dynamics predicts how both women and men act within interdependent relationships (Deutsch 2007; Few-Demo et al. 2014; Komter 1989 using data from The Netherlands; Sassler and Miller 2011).

These power and relational dynamics can be understood through the gender relations perspective of feminist theorizing (Ferree 1990). Gender is central (i.e., a structure, an institution) and is embedded in all social processes in everyday life (Risman 2004). That is, “the axis of feminist inquiry is gender, which consists of deeply ensconced social meanings and their derivative, power” (Fox and Murry 2000, p. 1160). Further, families are increasingly seen as places in which gender matters and can deeply affect women, men, and the realtionships between the two (Ferree 2010; Fox and Murry 2000).

To this end, the gender relations perspective focuses on the social construction of gendered behaviors, objects, and relationships (Ferree 2010). Gender is not a static norm or ideal, but rather “…a social relation characterized by power inequalities that hierarchically produce, organize, and evaluate masculinities and femininities through the contested but controlling practices of individuals, organizations and societies” (Ferree 2010, p. 424). In these ways, differences between women and men are seen as socially and culturally constructed, politically meaningful, understood within larger structures and levels that have their own practices and meanings, and deeply embedded in society (Ferree 2010; Fox and Murry 2000; Risman 2004).

A key way that gender is constructed is through societal and structural exaggeration of the differences between women and men (Ferree 1990). On average, there are many more similarities than differences between women and men; thus, by exaggerating differences between genders, social power is given to one gender over another in specific contexts (Ferree 1990; Fox and Murry 2000). For example, by creating a dichotomy that women are primarily nurturers within the home, whereas men are primarily providing financially for the family outside of the home (Ferree 1990), women are given more power in childrearing and men are given more power in careers and economic resources.

The power gained from constructing gender in this manner can be manifested in the form of emotion work between women and men. This pattern is likely due to exaggerated differences and expectations about the roles in relationships for women versus men, such as constructed expectations that because women are often more nurturing than men, all women are nurturing and all men are not (Davis and Greenstein 2009; Erickson 2005; Pfeffer 2010). The beliefs about exaggerated differences and expectations about gender roles in relationships are that women are more emotional than men and thus bear the majority of burden for maintaining the family, including doing most to all of the emotion work (Erickson 2005). Emotion work then becomes one of many ways gender is used to structure hierarchies both within and outside families, similar to child care and housework (Collins 2000). By conceptualizing emotion work as women’s work and subsequently devaluing it, gendered structures are reinforced through beliefs that women belong in the home, deferring to their husbands, while men should be doing paid work outside the home. This then “manufactures naturalized hierarchies” (Collins 2000, p. 158) by socializing individuals to believe that what occurs in families is a reflection of the natural order of society. Gendered beliefs that are reinforced in the family are then replicated in the public sphere: Men occupy more prestigious and valued positions and women serve under them in devalued, supportive positions; men are overrepresented in political systems, etc.

Although many relationships are more egalitarian as women have increasingly entered the workforce (Sayer et al. 2004), often women are still expected to perform, and do, the majority of family and emotion work in the home (Hochschild 1989; Lachance-Grzela and Bouchard 2010). In one study, despite how couples described their relationships in the language of equality (“we’re a partnership”, p. 86), none achieved actual equality (Knudson-Martin and Mahoney 1998). Instead, many of the women gave more and received less emotional nurturing than their partners, despite many of these couples being dual-earners (Knudson-Martin and Mahoney 1998). The gendered expectation that women are more nurturing than men is also demonstrated such that female partners of transgender men reported doing most of the emotion work in their romantic relationships (Pfeffer 2010, using data from the U.S and Canada).

Further, the gender relations perspective takes into account historically constructed differences between women and men in the labor market; it is a myth that men have traditionally held the role of the provider (Ferree 1990). As a result of the construction of a sole provider model of relationships, there occurs differential power in heterosexual couples. Women may be more invested in relationships because they are often more financially dependent on the relationship (Baker and McNulty 2011), and consequently the bulk of emotion work rests on their shoulders. Thus, constructed gender differences and unequal divisions in the labor cycle may contribute to inequalities in emotion work (e.g., why emotion work is seen as work by women, but viewed as part of relational duties by men; Erickson 2005).

The sole provider model has the particular consequence of reinforcing structural inequalities between women and men because family work and the private sphere are considered of lower value than the public sphere of paid work, specifically because the spheres are gendered as feminine and masculine, respectively (Collins 2000; Osmond and Thorne 1993). Researchers have found that women in both same-sex and different-sex relationships do more emotion work than men to allow and encourage the sharing of personal thoughts, feelings, and emotions between relational partners (Umberson et al. 2015). When men do engage in emotion work, they describe it in masculinized terms, similar to how they would describe themselves as financial providers (Thomeer et al. 2015). In Thomeer et al.’s (2015) sample of heterosexual couples in which one partner had significant health problems, men explained doing emotion work because they were “being a rock” for their wives (p. 16). Still, the men in the sample did not report doing nearly as much emotion work as their wives, even when their wives were the ones with health problems. Further, men in this study overemphasized their wives’ femininity when they were caretakers, referring to their wives as nurses, but specifically ruled out being considered nurses to their wives. Such perceptions position emotion work as feminized and “highlight that a key part of the enactment of hegemonic masculinity [the culturally defined ideal of how men should behave] within marriage involved relationally contrasting it to emphasized femininity and wifehood” (p. 19). That is, in these marriages, roles for women and men were seen as exclusive and opposite, underscored by a strict binary of roles and attributes appropriate for each gender (Ferree 2010; Thomeer et al. 2015).

Emotion Work and Relationship Quality

Thus far, we have described emotion work in a more negative light (e.g., inequalities between romantic partners). But emotion work has positive contributions to relationships as well, such as possibilities of greater equality between partners. This pattern is especially evident when both members of the couple contribute to emotion work. As one example, in a study on couples seeking therapy, both men and women reported high relational satisfaction when emotion work was approximately equal between the two of them (Holm et al. 2001). In a related study, men’s provision of sensitive support (similar to emotion work) improved both their own and their wives’ marital outcomes (Jensen et al. 2013).

These studies provide evidence of the association between emotion work and relationship quality (i.e., relationship satisfaction and marital quality), as well as the inclusion of data from both members of the couple. Although other authors have studied emotion work, data are not necessarily from couples (e.g., men and women not in a relationship together; Erickson 2005), or data from only one partner are included (Erickson 1993; Wharton and Erickson 1995).

Here, we report data from both partners in the couple, permitting a more balanced examination of gender and emotion work dynamics between partners. In previous research on relationships and families, researchers often considered women to be the representatives of the family, to maintain the relationship, or to be the ones to provide emotion work for the couple (Erickson 2005; Few-Demo et al. 2014; Ogolsky and Bowers 2013). Yet, relationships and families do not only consist of women, or one member of the couple. Instead, families are characterized as a social group of two or more people who are interdependent, have long-term commitments to one another, and are constructed through social interactions and communication with one another (Baxter and Braithwaite 2006; Vangelisti 2004). Thus, in our examination of both members of the couple in the current study, we deconstruct the assumption that woman is synonymous with family (Ferree 2010; Few-Demo et al. 2014).

Emotion work should be beneficial to relationship quality when enacted by both members of the couple. This prediction is in line with earlier research in this area (e.g., Holm et al. 2001), as well as the capacity for both partners to engage in emotion work (e.g., letting one’s partner know that she or he was appreciated). Specific to emotion work in the current study, we include both actor effects, or how one’s behaviors impact one’s own outcomes, and partner effects, or how one’s partner’s behaviors impact one’s outcomes.

Beyond this inclusion of data from both members of the couple, we also add to the literature about emotion work and relationship quality in other ways. First, we examine positive relationship qualities other than relationship satisfaction (e.g., Holm et al. 2001), as well as other more negative relationship qualities that are understudied in the area of emotion work. Specifically, we examine emotion work in association with six unique types of relationship quality: commitment, satisfaction, closeness, love, as well as ambivalence and conflict. We specify these relationship quality variables as relationships involve both positive and negative experiences and interactions (Kelley et al. 1983; Totenhagen 2011; Totenhagen et al. 2012). Further, all of these study variables are common in studies of romantic relationships (e.g., Braiker and Kelley 1979; Kelley 1979; Reis et al. 2000). In previous work using variants of the sample we use in this paper, the authors found that all six of the relationship qualities that we examine here significantly varied within-person from day to day and are suitable for daily examination (Totenhagen 2011).

Second, we take a nuanced approach in conceptualizing emotion work. Here, we examine emotion work at the trait level (i.e., individuals’ average levels of emotion work across a week, or variation between individuals) as well as at the state level (i.e., individuals’ fluctuations around their individual mean level of emotion work, or variation within individuals; Bolger and Laurenceau 2013). Both types of variability are informative, as we can better understand the association between emotion work and relationship quality at between- and within-person levels. Using this design, we can gain knowledge of how certain levels of emotion work relate to overall levels of relationship quality (between-person differences; trait) and how levels of emotion work on a specific day relate to levels of relationship quality on that day (within-person differences; state). Said another way for state emotion work, on days in which individuals report emotion work that is higher or lower than their own average, we are able to examine whether this change impacts individuals’ relationship quality for that day.

Finally, we have alluded to some gender differences (e.g., Hochschild 1989; Lachance-Grzela and Bouchard 2010). Specific to the study of gender, emotion work, and relationship quality, few studies have been published (e.g., Holm et al. 2001; Pfeffer 2010). It may be that when both female and male partners report higher emotion work, relationship quality benefits. But it may also be that women and men have different expectations and reactions to emotion work (Daniels 1987; Erickson 2005; Thomeer et al. 2015). To this end, and as articulated below, we consider the influence of gender in associations between daily emotion work and each type of daily relationship quality.

To test our hypotheses and research questions (RQs), we used MLM, or multilevel modeling. MLM is a form of regression in which multiple sources of interdependence are accounted for (e.g., data from both members of the couple; data from individuals measured each day across the week; Kenny et al. 2006). In contrast, a regression analysis assumes independence of sampling from the population, which would not be a correct assumption given the nested or interdependent data we have here. Further, to understand the perspectives of both members of the couple, we used Actor Partner Interdependence Models (APIM; Cook and Kenny 2005) to examine both actor and partner effects. Specifically, we examined how daily trait emotion work for individuals and partners and daily state emotion work for individuals and partners (the independent variables) predicted daily relationship quality (i.e., love, commitment, satisfaction, closeness, ambivalence, conflict; the dependent variables). We also included the impact of gender as a moderator between daily emotion work and daily relationship quality. That is, we created four interaction terms (i.e., Gender X Trait emotion work; Gender X State emotion work; Gender X Partner trait emotion work; Gender X Partner state emotion work) as further independent variables in predicting each type of daily relationship quality. In total, we analyzed six separate multilevel models; one for each daily relational quality variable.

Specific to these fixed effects, we pose these hypotheses and research questions (RQs):

  1. H1:

    Higher trait emotion work for both individuals and partners should be associated with higher love, commitment, satisfaction, and closeness, as well as lower ambivalence and conflict.

  2. RQ1:

    What is the influence of gender (i.e., women versus men) when examining associations between trait emotion work and each type of daily relationship quality?

  3. H2:

    On days in which individuals or their partners experience higher than average state emotion work, they should report higher love, commitment, satisfaction, and closeness, as well as lower ambivalence and conflict.

  4. RQ2:

    What is the influence of gender (i.e., women versus men) when examining associations between state emotion work and each type of daily relationship quality?

In addition to trait and state emotion work as associated with levels of relationship quality, we seek to examine associations between emotion work and the extent of volatility in relationship quality across a week. Volatility is a between-person factor, which means that some individuals may experience less variability (or more stability) in relationship quality, whereas other individuals experience more variability (or less stability) in relationship quality from day to day across a week. That is, relationships are often experienced in different ways, and it may be too simplistic to state that an individual is generally satisfied or not satisfied (Arriaga 2001; Berscheid and Lopes 1997). We seek to understand how trait (average) levels of emotion work predict differences in volatility of relationship quality between individuals.

The study of volatility in relationship quality is important because high fluctuations in relationship quality typically do not bode well for relationships (Karney and Bradbury 1995). For example, fluctuations in relationship quality are associated with eventual breakup in romantic couples (Arriaga 2001; Arriaga et al. 2006) as well as increasing psychological distress, decreasing life satisfaction, and depressive symptoms over time (Whitton and Whisman 2010; Whitton et al. 2014). To our knowledge, the associations between trait emotion work and volatility in relationship quality across a week have not been examined. We tested whether trait emotion work and partner trait emotion work predicted overall volatility in relationship quality. We also interacted these two variables with gender to test for potential gender differences. All of these tests were performed simultaneously in the models while testing the fixed effects mentioned previously in H1, RQ1, H2, and RQ2. Specific to the tests of volatility (which we further explain in the Results section), we pose the following RQs:

  1. RQ3:

    For individuals or partners who experience higher than average trait emotion work, what is the prediction of volatility in each type of relationship quality?

  2. RQ4:

    What is the influence of gender (i.e., women versus men) when examining associations between trait emotion work and volatility in each type of relationship quality?

In addition to the variables already specified, we also included several controls: Overall relationship quality, race/ethnicity, relationship length, and marital status. We chose these constructs as they are common controls in studies of emotion work and/or romantic relationships (Erickson 2005; Ruppel and Curran 2012; Totenhagen 2011; Totenhagen and Curran 2011; Totenhagen et al. 2012, 2013). For example, relationship length has been used to predict connections between individuals in romantic relationships, as well as if individuals should move forward with their relationship (Braiker and Kelley 1979). Further, researchers have documented the protective benefits of marriage compared to other relationship statuses such as cohabitation (e.g., Stanley et al. 2006). Relatedly, marriage is tied to education, such that college graduates are more likely to marry compared to individuals without a college degree (Cherlin 2010), as well as race/ethnicity (i.e., 81 % of non-Hispanic White women are predicted to marry by age 30 versus 52 % of non-Hispanic African American women; Cherlin 2010). Finally, parent status is important such that the presence of children can increase time for one or both adult partners specific to childcare and/or housework (Voydanoff 2007), while also sometimes adversely impacting romantic relationships between parents (Doss et al. 2009; Twenge et al. 2003). Originally we included age as a control variable, but given high correlations and multicollinearity with relationship length, we removed age from the models reported here.

Method

Procedure and Participants

Participants were 74 heterosexual couples (N = 148 individuals) in which both partners agreed to participate in a 7-day diary study and who had at least three matching days of data with their partners. Participants were recruited through two different departments (i.e., Family Studies and Human Development and Communication) at a large Southwestern university in the U.S. Students learned about the study from their instructor, a flyer about the study posted to the students’ class website, and/or a graduate student coming into the students’ classroom to tell them about the study. Students in these classes could earn extra credit by participating in the study themselves along with their romantic partners, or by passing a flyer on to other couples who participated and entered the students’ name for extra credit (e.g., friends, parents, roommates). Thus, not all participants were students (students comprised 61 % of the sample). Students who did not wish to participate or refer others to participate were offered a comparable alternative extra credit assignment expected to take approximately the same amount of time as participating in the study (e.g., finding a course content-relevant article to read and summarize). Because the surveys were online (described in more detail in this section), individuals did not have to live in the city or state in which the study took place. If individuals met the study criteria, they were eligible for participation. To qualify, both individuals in a couple were 18 years old or older, in a romantic relationship with their current partner for at least 6 weeks (e.g., married, dating, or cohabiting), and have their own e-mail address.

The sample was primarily Caucasian, followed by Hispanic (17.6 % for men and 14.9 % for women). Average relationship length was 7.18 years (SD = 9.86, Mdn = 2.50; range from 2 months to 36.6 years). Only three couples (7.69 % of the total sample) reported relationship lengths less than or equal to 6 months. Approximately 30 % of the couples were married. See Table 1 for additional details on men and women in the sample. We tested for gender differences in the demographic characteristics (i.e., MANOVAs, follow-up ANOVAs for each variable, and crosstab / chi-square tests). No significant gender differences (i.e., p < 0.05) emerged.

Table 1 Participant demographic characteristics
Full size table

We collected all data via a secure Internet-based system. We directed participants to a website in which they created a unique couple ID that linked the two partners and indicated the individual’s gender. Individuals were instructed to complete surveys separately from their partners. The first time they logged onto the system, participants completed informed consent, demographic information, and an initial survey. Individuals were then asked to log onto the website at approximately the same time each day for seven consecutive days to complete daily surveys, at which time they were instructed to think about survey items they had experienced in the past 24 hours. We chose 7 days because of the desire to increase the chances that participants would remain in the study rather than experience fatigue (e.g., Bolger et al. 2003) and drop out, and because it is common to collect data over the course of a week in daily diary studies (e.g., Neff and Karney 2005; Ruppel and Curran 2012; Surra et al. 2009; Totenhagen 2011; Totenhagen and Curran 2011; Totenhagen et al. 2012; Totenhagen et al. 2013; Witt and Wood 2010; Young et al. 2012). One advantage of using online methods is that every submission is time and date stamped (Ogolsky et al. 2009). We used this information to remove any entries that were deemed invalid due to duplicate entries (i.e., submitting data more than once in a single day). The time and date stamps also permitted us to identify daily entries in which partners did not match (i.e., both did not submit an entry on a particular day). Only couples who completed and matched on at least 3 days were retained in the final sample (Totenhagen et al. 2012, 2013).

Although some participants completed more than 7 days, we restricted the sample to the first 7 days of data for each individual. Individuals completed an average of 5.12 days of data (84 % completed at least 4 days) for a total of 758 person-days of data. We used Proc Mixed in SAS, which can handle missing data on the outcome variable using restricted maximum likelihood; however, we dropped from the analysis any days of data that contained any missing predictor data (i.e., daily emotion work and control variables), leaving 732 person-days of data. Depending on which relationship quality construct was examined (i.e., love, commitment, satisfaction, closeness, ambivalence, or conflict), there was a small amount of missing data on each of the outcome variables (ranging from 1 day missing on some to 14 days on one of them) that was handled by restricted maximum likelihood estimation in SAS Proc Mixed.

Measures

Emotion Work

Using the measure by Erickson (2005), each day we asked participants to rate how often in the past 24 hours they engaged in each of eight items with their partner on a 1 (never) to 5 (very often) point scale. Sample items include “Initiated talking things over” and “Recognized the importance of my partner’s feelings even if my partner did not share the feelings with me.”

We averaged items to produce an overall emotion work score each day for each participant (alphas = 0.88 for both men and women). We grand-mean centered participants’ scores, as this generally reduces multicollinearity between predictors, allows the predicted intercept in our models to represent the average value of the outcome for the average man or woman in the sample, and allows effects of this variable to be interpreted as the meaning of effects when individuals are above or below average on the predictor (Aiken and West 1991). Then we created the trait (between-person) and state (within-person) versions of emotion work. We created the trait portion by computing each participant’s mean across the 7 days (for women, M = 0.00, SD = 0.66; for men, M = 0.00, SD = 0.66). We created the state portion by subtracting each participant’s mean from each day’s score, which conceptually produces deviations from one’s usual level of emotion work (for women, M = 0.00, SD = 0.57; for men, M = 0.00, SD = 0.54). Higher scores on trait emotion work represent greater average emotion work across a week, and positive scores on state emotion work represent greater emotion work on a particular day than one’s own average level of emotion work across a week.

Relationship Quality

Each day we asked participants to rate six relational qualities via single items on a 1 (not very much or just a little) to 7 (very much or a lot) point scale, with a score of 4 indicating neutral (Totenhagen et al. 2012, 2013). We retained the relationship quality variables as distinct in line with our previous research in this area in which we demonstrated unique predictions to certain constructs of relationship quality (e.g., commitment) over others (Totenhagen et al. 2012, 2013).

Participants rated how they felt today with respect to each quality, including love, closeness, satisfaction, commitment, ambivalence, and conflict (e.g., “Today, how satisfied were you with your relationship with your partner?”; “Today, how ambivalent, or uncertain did you feel about the future of your relationship with your partner?”). Higher scores indicate more daily positive relationship quality (i.e., greater love, closeness, satisfaction, commitment) as well as more daily negative relationship quality (i.e., greater ambivalence and conflict). See Table 2 for means and standard deviations by gender.

Table 2 Correlations and descriptive information
Full size table

Control Variables

Controls included overall relationship quality, race/ethnicity (Caucasian = 1), education (attended at least some college = 1), relationship length in months, parent status (have children = 1), and marital status (married = 1). We measured overall relationship quality at baseline as the average of the same six relational quality items (e.g., for satisfaction, the item was worded as “Overall, how satisfied are you with your relationship with your partner?”), with negative items reverse scored (female alpha = 0.84, male = 0.85). We mean-centered relationship quality and relationship length before including them in analyses.

Results

Descriptive Statistics and Correlations

Descriptive statistics and between-person correlations are in Table 2 (categorical demographic control variables are in Table 1). We tested for gender differences in the study and control variables (i.e., MANOVAs, follow-up ANOVAs, and crosstab / chi-square tests for categorical variables). No significant gender differences (i.e., p < 0.05) emerged. Specific to correlations, and as expected, all correlations of the study variables were significant between women and men (with these correlations indicated on the diagonal).

Multilevel Models: Plan of Analysis

As noted at the end of the Introduction, we have distinguishable dyadic data across 7 days, with individuals (women and men) nested within couples and crossed with time. Given this nesting (or interdependence) of data we used MLM using Proc Mixed in SAS 9.3.

Our data have a three-level structure: from lowest to highest level, day (Level 1), individual (Level 2), and dyad (Level 3). Level 3 has no variability in a standard 3-level dyadic data MLM model, as there are only two possible members in each dyad. Thus, we created dummy coded variables for women and men (Bolger and Laurenceau 2013), which we entered on the random line in the model. This specification of the terms, women and men, appropriately accounts for the structure and interdependencies in the data and error structure, and allows for interindividual differences (random effects) in women’s and men’s intercepts on relationship quality. In sum, these specifications make the model a two-level model with day (Level 1) and women and men specified within couples (Level 2). Also, we allowed for random variability in the slope between state emotion work and daily relationship quality, as substantively not all individuals should react identically to the same level of emotion work.

To increase parsimony in the models, we utilized a one-intercept model and entered an indicator of gender (1 = male, 0 = female; similar to Campbell et al. 2005). We chose this method, rather than use of the two-intercept model with dummy coded variables for women and men in the fixed effects (as done in Bolger and Laurenceau 2013), as the latter would have given us two separate intercepts (one for women and one for men) as well as separate estimates for all of the study variables for women and men. That is, the single intercept model provides a more parsimonious option for testing the significance of gender differences as opposed to reporting effects separately for women and men.

To test all hypotheses and RQs, we analyzed six separate MLM models; one for each daily relational quality variable (i.e., love, commitment, satisfaction, closeness, ambivalence, conflict). Specific to fixed effects (i.e., H1, RQ1, H2, RQ2), we entered controls, trait and state emotion work (both actor and partner), gender, and interactions with gender into the models. Specific to volatility (i.e., RQ3, RQ4), we utilized Hoffman’s (2007) methods to examine both actor and partner effects of trait level emotion work on the heterogeneity (or overall daily volatility) in daily relationship qualities. To predict residual variability, we entered actor and partner trait emotion work, as well as gender and interactions with gender, as predictors on the repeated line in Proc Mixed. The residual variability was exponentiated to allow for linear prediction of the variance component by trait levels of emotion work. Upon evaluation of the models, we trimmed all nonsignificant interactions with gender. Finally, we calculated degrees of freedom in all models using the Satterthwaite approximation, which takes into account the amount of interdependence for any outcome variable and adjusts both between- and within-person effects accordingly. This also explains why the degrees of freedom for each fixed effect may vary widely. Below we describe the general equations for our MLM models:

Level 1:

$$ \begin{array}{l}Rel. Qualit{y}_{ti} = {\beta}_{0i}+{\beta}_{1i} Da{y}_{ti}+{\beta}_{2i} State\ Emotion\ Wor{k}_{ti}\hfill \\ {}\kern3.36em +{\beta}_{3i} Partner\ State\ Emotion\ Wor{k}_{ti}+{e}_{ti}\hfill \end{array} $$

Level 2:

$$ \begin{array}{l}{\beta}_{0i} = {\gamma}_{00} + {\gamma}_{01} Gende{r}_i + {\gamma}_{02} Trait\ Emotion\ Wor{k}_i+{\gamma}_{03} Partner\ Trait\ Emotion\ Wor{k}_i\hfill \\ {}+{\gamma}_{04} Trait\ Emotion\ Work* Gende{r}_i+{\gamma}_{05} Partner\ Trait\ Emotion\ Work* Gende{r}_i\kern3.46em +{\gamma}_{06} Baseline\ Rel. Qualit{y}_i+{\gamma}_{07} Caucasia{n}_i+{\gamma}_{08} Colleg{e}_i+{\gamma}_{09}Rel. Lengt{h}_i+{\gamma}_{010} Parent\ Statu{s}_i+{\gamma}_{011} MaritalStatu{s}_i+ males*{\mu}_{0i}\hfill \\ {}+ females*{\mu}_{0i}\hfill \\ {}{\beta}_{1i}={\gamma}_{10}\hfill \\ {}{\beta}_{2i}={\gamma}_{20}+{\gamma}_{21} Gende{r}_i+{\mu}_{2i}\hfill \\ {}{\beta}_{3i}={\gamma}_{30}+{\gamma}_{31} Gende{r}_i\hfill \end{array} $$

Heterogeneous Variance:

$$ \begin{array}{l}\hfill {\sigma}_{ei}^2 = {\alpha}_0 \exp \Big({\alpha}_1 Gende{r}_i + {\alpha}_2 Trait\ Emotion\ Wor{k}_i + {\alpha}_3 Partner\ Trait\ Emotion\ Wor{k}_i\hfill \\ {}\hfill \kern5em +{\alpha}_4 Trait\ Emotion\ Work* Gende{r}_i + {\alpha}_5 Partner\ Trait\ Emotion\ Work* Gende{r}_i\Big)\hfill \end{array} $$

At Level 1, we have the equation describing the within-person relationship of daily relationship quality (\( Rel.{Quality}_{ti} \)) to the daily predictors, state emotion work (\( {\beta}_{2i} \)) and partner state emotion work (\( {\beta}_{3i} \)). The predicted value for relationship quality for each individual “i” on a given occasion “t” is a function of the individual’s average relationship quality on day 1 (intercept, \( {\beta}_{0i} \)), the linear slope of day (\( {\beta}_{1i} \)), deviations around the individual’s average emotion work (\( {\beta}_{2i} \)), deviations around the individual’s partner’s average emotion work (\( {\beta}_{3i} \)), and residual variation in relationship quality (\( {e}_{ti} \)).

At level 2, we entered between-person predictors, trait emotion work, and between-person random effects. The average relationship quality score (\( {\beta}_{0i} \)) is a function of the overall sample average relationship quality score (\( {\gamma}_{00} \)), gender (\( {\gamma}_{01} \)), trait emotion work (\( {\gamma}_{02} \)), partner trait emotion work (\( {\gamma}_{03} \)), the interactions of trait emotion work with gender (\( {\gamma}_{04} \) and \( {\gamma}_{05} \)), controls (\( {\gamma}_{06} \) to \( {\gamma}_{011} \)), and random variation around the sample average for males and for females (\( males*{\mu}_{0i} \) and \( fe males*{\mu}_{0i} \)). The linear slope in relationship quality over days (\( {\beta}_{1i} \)) is the average sample linear slope across days (\( {\gamma}_{10} \)). The effect of state emotion work (\( {\beta}_{2i} \)) is a function of the sample average state emotion work effect (\( {\gamma}_{20} \)), gender (\( {\gamma}_{21} \)), and random variation (\( {\mu}_{2i} \)). The effect of partner state emotion work (\( {\beta}_{3i} \)) is a function of the sample average partner state emotion work (\( {\gamma}_{30} \)) and gender (\( {\gamma}_{31} \)). We did not include random effects at Level 2 for the linear slope (\( {\beta}_{1i} \)) and partner state emotion work (\( {\beta}_{3i} \)) as the model could not converge with these included and therefore had to be simplified.

Finally, the overall residual variability across days (within-person variability not accounted for by any of the predictors) for each individual “i” (\( {\sigma}_{ei}^2 \)), what we call volatility, was modeled by level-2 variables, including gender (\( {\alpha}_1 \)), trait emotion work (\( {\alpha}_2 \)), partner trait emotion work (\( {\alpha}_3 \)), and interactions with gender (\( {\alpha}_4 \) and \( {\alpha}_5 \)).

In Table 3, we report unstandardized estimates from the results of our MLM models. As the categorical control variables (e.g., Caucasian, married) and effects of gender on the intercept in relationship quality were generally nonsignificant, the intercept values can be interpreted such that the average individual on an average day in this sample experienced fairly high positive relationship quality (\( {\gamma}_{00} \) range from 5.38 to 6.24, dfs range from 66 to 106, ps < 0.001) and low negative relationship quality (\( {\gamma}_{00} \) range from 2.47 to 2.61, dfs range from 122 to 127, ps < 0.001). Men and women differed only on feelings of commitment, with men reporting lower overall levels of commitment than women (\( {\gamma}_{00} \) = −0.25, df = 57, p < 0.01).

Table 3 Multilevel models of daily relationship quality predicted by daily emotion work and gender
Full size table

For the controls, and as expected, higher baseline relationship quality was significantly associated with higher love, commitment, satisfaction, and closeness, and less ambivalence and conflict. Having attended at least some college was significantly associated with lower ambivalence and conflict. Longer relationship length was significantly associated with lower relationship satisfaction. Finally, parental status (i.e., having children) was significantly associated with greater relationship satisfaction.

  1. H1:

    Trait emotion work (individuals’ average levels) and relationship quality (Fixed effects)

    As hypothesized, individuals with higher average levels of emotion work (trait level) across a week reported higher scores on average on love, commitment, satisfaction, and closeness (\( {\gamma}_{02} \) values range from 0.49 to 0.59, dfs range from 114 to 136, ps < 0.001), and lower scores on ambivalence (\( {\gamma}_{02} \) = −0.27, df = 126, p < 0.05). This association did not emerge for conflict (\( {\gamma}_{02} \) = −0.14, df = 131, ns). Unexpectedly, we found no significant partner effects for trait emotion work (\( {\gamma}_{03} \)) in predicting the relational qualities.

  2. RQ1:

    What is the influence of gender (i.e., women versus men) when examining associations between trait emotion work and each type of daily relationship quality? (Fixed effects)

    We found no significant gender interactions (see \( {\gamma}_{04} \) and \( {\gamma}_{05} \)).

  3. H2:

    State emotion work (individuals’ daily fluctuations) and relationship quality (Fixed effects)

    As hypothesized, on days when individuals reported higher scores on emotion work than their average level, they also reported greater love, commitment, satisfaction, and closeness (\( {\gamma}_{20} \) values range from 0.25 to 0.57, dfs range from 43 to 60, ps < 0.01). Contrary to the hypothesis, daily fluctuations in emotion work were not significantly associated with daily fluctuations in negative relationship qualities (ambivalence and conflict) (\( {\gamma}_{20} \) values range from 0.01 to −0.23, dfs range from 43 to 58, ns).

    Partner effects (i.e., effects of fluctuations in one’s partner’s reports of emotion work) emerged for one’s own feelings of satisfaction and closeness (both \( {\gamma}_{30} \) values = 0.16, dfs = 566 and 550 respectively, ps < 0.05). Individuals experienced greater satisfaction and closeness on days when their partner reported higher emotion work than their typical amount.

  4. RQ2:

    What is the influence of gender (i.e., women versus men) when examining associations between state emotion work and each type of daily relationship quality? (Fixed effects)

    We found only one significant gender interaction for partner state emotion work on love (\( {\gamma}_{31} \) = 0.27, df = 488, p < 0.01). Unlike women (\( {\gamma}_{30} \) = 0.08, df = 296, ns), men reported greater feelings of love on days when their partner reported higher emotion work than their average level (\( {\gamma}_{30}+{\gamma}_{31} \) = 0.08 + 0.27 = 0.35, df = 224, p < 0.01).

  5. RQ3:

    Does trait level emotion work predict overall volatility in relationship quality? (Volatility)

    As the extent of volatility across a week in relationship quality is a between-person factor, we examined whether trait emotion work predicted differences in the amount of volatility in relationship quality. We summarize the pattern of findings overall, while noting that gender differences emerged. That is, findings were qualified by higher level interactions with gender, and we refer the reader to the next section (i.e., RQ4) for specific estimates for women and men.

    Overall, individuals with higher trait emotion work reported lower volatility in love, commitment, satisfaction, and closeness (see the findings for RQ4 for specific estimates for women and men, which often differed). Trait emotion work largely did not predict volatility in ambivalence or conflict (although see one potential difference by gender in RQ4). Often, individuals whose partners reported higher trait emotion work reported similar patterns of lower volatility in love, satisfaction, and closeness, but not commitment (although some gender differences emerged and estimates differed by gender; see RQ4).

  6. RQ4:

    What is the influence of gender (i.e., women versus men) when examining associations between trait emotion work and volatility in each type of relationship quality? (Volatility)

    Several gender differences emerged, both for actor and partner effects. For actor effects, men experienced greater daily volatility in feelings of commitment than women (\( {\alpha}_1 \) = 0.50, z = 3.92, p < 0.001). Further, the impact of trait emotion work on volatility differed for women and men. Women and men who reported higher average emotion work across a week experienced lower overall daily volatility in love, commitment, satisfaction, and closeness (\( {\alpha}_2 \) estimates range from −0.28 to −1.18, z values range from −2.32 to −6.40, ps < 0.05), and this effect was stronger for men than women on volatility in love (\( {\alpha}_4 \) = −0.95, z = −3.97, p < 0.001) and commitment (\( {\alpha}_4 \) = −1.08, z = −3.93, p < 0.001). We also saw gender differences for ambivalence, such that men who reported higher average emotion work experienced greater daily volatility in ambivalence (\( {\alpha}_2+{\alpha}_4 \) = −0.10 + 0.51 = 0.41, z = 2.70, p < 0.01), whereas women experienced no significant effect (\( {\alpha}_2 \) = −0.10, z = −0.72, ns).

    For partner effects, we found significant interactions between partner emotion work and gender (\( {\alpha}_5 \) estimates range from 0.46 to 1.21, z values range from 1.99 to 5.34, ps < 0.05). For women, having a partner who reported higher average emotion work predicted women’s experiences of lower volatility in love, satisfaction, and closeness (\( {\alpha}_3 \) estimates range from −0.37 to −0.72, z values range from −2.18 to −4.30, ps < 0.05; not significant for commitment although in the same direction, \( {\alpha}_3 \) = −0.25, z = −1.46, ns). For men, having a partner who reported higher average emotion work predicted men’s experiences of greater volatility in love (\( {\alpha}_3 \) + \( {\alpha}_5 \) = −0.51 + 1.21 = 0.70, z = 4.12, p < 0.001) and commitment (\( {\alpha}_3 \) + \( {\alpha}_5 \) = −0.25 + 0.84 = 0.59, z = 2.99, p < 0.01). Unlike for women, partner trait emotion work was not significantly associated with volatility in men’s feelings of satisfaction (\( {\alpha}_3 \) + \( {\alpha}_5 \) = −0.37 + 0.46 = 0.09, z = 0.58, ns) or closeness (\( {\alpha}_3 \) + \( {\alpha}_5 \) = −0.72 + 0.96 = 0.24, z = 1.59, ns).

Discussion

We used the gender relations perspective from feminist theorizing to examine how daily emotion work predicts daily relationship quality for both partners. We chose emotion work because this construct extends the discussion away from economic or relative resources (Lachance-Grzela and Bouchard 2010) to relational dynamics within the family (Deutsch 2007; Ferree 1990), while also focusing on activities specific to the enhancement of others’ emotional well-being and to the provision of emotional support (Erickson 2005).

Specifically, we examined how individuals’ average levels (trait) and daily fluctuations (state) of emotion work predict various types of relationship quality. We also examined how trait emotion work predicts volatility in relationship quality (or overall daily variability across a week). We discuss when patterns of gender differences between emotion work and relationship quality were minimal (e.g., fixed effects) compared to when patterns of gender differences were more robust (e.g., volatility). Below we discuss three patterns of results.

Emotion Work Matters

First, emotion work predicted most kinds of relationship quality in this diverse set of couples (individuals in dating, cohabiting, and married relationships). Typically, emotion work has been examined with samples of married couples (e.g., Erickson 2005). We demonstrate that emotion work, characterized by activities that enhance others’ emotional well-being, is a construct of importance for individuals in multiple kinds of relationships.

Gender Differences as Minimal

Second, when examining fixed effects of emotion work specific to trait (i.e., average levels) or state (i.e., daily fluctuations) and levels of relationship quality, gender differences between emotion work and relationship quality were minimal. Instead, across gender, we found that both trait and state emotion work predicted higher average scores on, and positive daily increases in, the four types of positive relationship quality: love, commitment, satisfaction, and closeness. The only exception in which women and men significantly differed was for partner state emotion work on love: Unlike women, men reported greater feelings of love on days when their partner reported higher emotion work than their average level.

When focused on trait or state emotion work and relationship quality overall, our findings support the conclusions by feminist scholars that gender differences are often exaggerated and there are more similarities than differences between men and women (Ferree 1990; Fox and Murry 2000). Further, these findings offer support for how the context being studied between women and men (or the “gendered relational context,” Umberson et al. 2015, p. 543) may be more influential than the gender of the participants in understanding a construct like emotion work (Umberson et al. 2015). Replication of our study with same-gender couples and couples with individuals of less studied gender identities (e.g., transgender individuals) may further reveal whether it is gender or the gendered relational context at play when it comes to associations between emotion work and relationship quality.

Because of the robust pattern of results for individuals (actor effects), we were surprised that the patterns for partner effects (effects of average level and fluctuations in one’s partner’s reports of emotion work) were more limited. We found only that on days when their partner reports higher emotion work than their typical amount, individuals experience greater satisfaction and closeness. In other studies of romantic relationships, authors have noted that partner effects are sometimes less frequent (or nonexistent) compared to actor effects (e.g., Gable et al. 2003; Kelley 1979; Overall et al. 2012, using New Zealand data; Totenhagen et al. 2013).

From an attributional standpoint, individuals often overestimate how much they do in the relationship, while underestimating how much their partners do (Kelley 1979), perhaps explaining the fewer significant partner effects versus actor effects. Individuals may perceive that they are doing high amounts of emotion work, but those perceptions may not be shared by the partner. The extent to which perceptions of emotion work are shared and how this level of similarity predicts relationship quality is an important one for future research. Recall the study by Holm et al. (2001) in which both women and men were satisfied with their relationship when levels of emotion work were equal or balanced between them. Using a daily diary approach, future researchers could examine how different patterns of daily emotion work (e.g., how the amount of similarity changes from day to day between partners; daily reciprocity between partners versus across the span of a week; how degree of similarity between partners interacts with the level of daily emotion work) predict daily relationship quality.

Gender Differences as Robust and Patterned

Third, because researchers have identified fluctuations or volatility in relationship quality to bode poorly for relationships (e.g., Arriaga 2001; Karney and Bradbury 1995; Whitton et al. 2014), we extended this literature by focusing on patterns of volatility in relationship quality as predicted by trait emotion work and gender. For actor effects, both women and men who reported higher average emotion work across a week experienced lower overall daily volatility in love, commitment, satisfaction, and closeness, although this effect was stronger for men than women on volatility in love and commitment.

Specific to the latter part of this finding, these men seem to be experiencing a “status bonus” (Cottingham et al. 2014, p. 4), such that when men report higher average emotion work across a week, they are not only “shielded from the negative effects of covering emotion” (Cottingham et al. 2014, p. 1), but they benefit in terms of lower volatility in love and commitment compared to women. That is, men’s privileged status often shields them from having to enact emotion work as often as women (Cottingham et al. 2014; Hochschild 1983).

For partner effects, the patterns of findings were quite different between women and men. For women, having a partner who reported higher average emotion work predicted lower volatility in love, satisfaction, and closeness. In contrast, we found the opposite direction of effects for men. That is, for men, having a partner who reported higher average emotion work predicted greater volatility in love and commitment. Although we posed research questions rather than hypotheses, it seems puzzling to find greater volatility in love and commitment for men who have partners who report higher average emotion work. We say this as emotion work refers to activities specific to the enhancement of others’ emotional well-being and with the provision of emotional support (Erickson 2005). Further, given that emotion work is considered as work, or effort expended, it needs to be “managed, focused, and directed so as to have the intended effect on the care recipient” (Erickson 2005, p. 349).

It may be that these men feel overbenefitted when their female partners report higher average emotion work. Overbenefitting is a form of inequality that can occur when individuals receive greater benefits relative to contributions in comparison to their partner, and it has been linked with higher feelings of guilt (Guerrero et al. 2008; Walster et al. 1978) and possible feelings of inefficacy, dependence, and indebtedness (Gleason et al. 2003). Connected to gender and power, it may be that the male partner perceives more emotion work performed on his behalf than he desires or needs, which connects to his feelings of lack of relational control or the female partner having more relational power. Relatedly, it may be that men perceive higher trait levels of emotion work by their female partners as demanding and a threat to their power, although the female partners may not intend their emotion work to be perceived in this way.

Further, though conceptualizations of masculinity as rational and nonemotional are shifting to recognize changes in men’s roles in family life (e.g., househusbands, fathers as caretakers; Smith 1998), men who feel relationally overbenefitted and dependent may feel that they are not performing appropriately under hegemonic masculinity, the dominant form of masculinity in a given context. These men may be attempting to construct a more conventional masculinity and recreate gender stratification in their relationship more closely matched to that of dominant society (Connell and Messerschmidt 2005; Hanlon 2012).

Yet by following these rules of hegemonic masculinity, men may be experiencing disadvantages in at least two ways: first, in terms of lower quality romantic relationships (Wade and Donis 2007) and second, in terms of the benefits offered by care work (here, emotion work), including feelings of being emotionally fulfilled (Hanlon 2012). However, men are unlikely to change unless these benefits offset losses of privilege that occur as relationships become more equal (Hunt and Hunt 1987). This set of patterns elucidates Risman’s (2004) discussion of the importance of work that pinpoints the direction and strength of relationships between genders on certain dimensions. That is, these patterns from the current study could help identify the sites in which change between genders could occur, in which “habitualized” gender routines could be rejected, and how such “rebellion” could change institutions themselves (Risman 2004, p. 434).

Taken together, these findings specific to emotion work and volatility in relationship quality underscore the importance of the gender relations perspective, or how gender is a social relation in which the focus should be on the social construction of gendered behaviors and relationships (Ferree 2010). That is, if we had only focused on average levels or daily fluctuations (fixed effects, like our findings that across gender, both trait and state emotion work predicted higher average scores on, and positive daily increases in, the four types of positive relationship quality) we would have concluded that gender differences were more or less absent from the discussion of daily emotion and daily relationship quality. Yet when we consider how emotion work predicts volatility over the course of the week, gender differences are robust and patterned (e.g., our finding that having a partner who reported higher average emotion work predicted lower volatility in love, satisfaction, and closeness for women versus greater volatility in love and commitment for men). By examining periods of time across days, we can see the ways in which gendered behaviors influence romantic relationship outcomes and demonstrate the way that gender is constructed and based on many levels and institutions (Ferree 2010).

Thus far, our patterns have been specific to positive relationship quality. We found only a few significant results for emotion work and ambivalence: Individuals with higher trait emotion work experienced lower scores on ambivalence, and men who reported higher average emotion work experienced higher daily volatility in ambivalence, whereas women experienced no significant effect here. We found no significant patterns for emotion work and conflict.

Given the limited research on emotion work and relationship quality (e.g., Holm et al. 2001; Pfeffer 2010), we do not know why patterns are more robust for positive rather than negative relationship quality. We have found in other research, however, that uplifting daily events had same-day impacts only on positive relationship quality, and not negative relationship quality (Totenhagen et al. 2012). Further, in research regarding social exchanges, researchers have found a similar pattern in that both positive and negative social exchanges impact positive mood, whereas negative mood is impacted only by negative social exchanges (e.g., Rook 2001). Our findings suggest the need to study both positive and negative relationship quality.

Limitations, Future Directions, and Conclusion

We note some limitations of our study. All participants were heterosexual; how such results apply to samples of lesbian, gay, bisexual, transgender, and queer individuals is beyond the scope of our study. In our review of emotion work, we did find one study (Pfeffer 2010) in which the author used qualitative methods to examine how female partners of transgender men experienced housework and emotion work. However, the couples in the study by Pfeffer (2010) can still be considered other-gender/mixed-gender couples based on gender identity even if they did not identify as heterosexual. Because we have argued that families reproduce societal inequities within themselves (Few-Demo et al. 2014), an analysis of emotion work and gender within heterosexual couples provided the most clarity; however, examining if and how same-gender couples also reproduce these inequities would advance discussions of gender and power.

We also acknowledge that the complexities of intersectionality were underrepresented in our study, leaving unanswered questions about many levels of exclusion and marginalization versus privilege (e.g., Risman 2004; Shields 2008) within romantic relationships. Intersectionality is defined as the “relationships among multiple dimensions and modalities of social relations and subject formations” (McCall 2005, p. 1771). In thinking about intersectionality and feminism, and specific to emotion work and related topics such as housework and childcare, many published studies are specific to married individuals (see Ferree 2010), leaving unknown the experiences of individuals in dating relationships or unmarried cohabitors. Much work is still needed to “bring the standpoint of marginalized persons and groups into the research design” and to “embrace different forms of family” (Ferree 2010, p. 427 and 430). To this latter point, we included responses from individuals who were married, dating, and cohabiting. For cohabitors in particular, issues of marginalization may occur (e.g., on average, people with less education are more likely to cohabit; Cherlin 2010) and gender differences in cohabitation are documented (e.g., female cohabitors “waiting to be asked” in relationship transitions; Sassler and Miller 2011, p. 19).

In regards to areas for future research, relationship status as a moderating variable of emotion work and relationship quality for couples would be important to study. For example, would associations between emotion work and relationship quality differ between daters versus cohabiting and married individuals given that the former do not live together but the latter do? Or might it be that such associations matter less depending on living arrangements and more on relational commitment to one another (daters and marrieds vs. cohabitors; see Stanley et al. 2006)? As another area of future research, more information about the extent to which perceptions of emotion work are shared by both partners within the couple is needed.

While participants were diverse in relationship lengths and types, they lacked diversity in other constructs (e.g., 70 % were Caucasian; 88 % reported at least some college or more). Returning to intersectionality, in the study by Pfeffer (2010), transmen lacked cisgender (i.e., someone whose gender aligns with the sex they were assigned at birth) privilege for identifying as transgender; however, they found privilege in their relationships through identifying as men. This example illustrates the importance for future researchers to investigate how both relational partners reinforce or subvert societal inequalities and power dynamics in their own relationships. Further, like the vast majority of the literature we reviewed specific to emotion work, our sample was also specific to individuals in the U.S. To this end, we do not assume that the results of our study represent universal truths that are independent of culture and cultural differences. Thus, we acknowledge the point by Shields (2008) of “not enough information” (p. 305), and we will continue to learn from intersectional feminists and incorporate this focus into our future research.

We chose 7 days of diary data to have a span of weekdays and a weekend, and also not to overly burden our participants, in line with other researchers who do daily diary work (e.g., Bolger et al. 2003). It may be that our study week was unique to the participants in some ways of which we were not aware. It may also be that a longer period of time may have been needed to assess the relationship between emotion work and negative relationship quality. Related to this point, although our data allow us to answer questions about daily variability, we could not examine how daily levels or volatility in emotion work influence long-term changes in relationship quality. Finally, following the example of others (e.g., Laurenceau et al. 2005), and our previous research (e.g., Totenhagen et al. 2012, 2013), we chose to use single-item measures in efforts to minimize participant fatigue. Nevertheless, single-item measures have limitations, including the inability to report reliability and the possibility for reduced variability (although we found significant variability across days in all of our relational quality variables; Totenhagen 2011).

We also note several strengths. From the gender relations perspective from feminist theorizing, we focused on emotion work as a construct of power, rather than more traditional concepts of economic or relative resources (Erickson 2005; Lachance-Grzela and Bouchard 2010). We examined emotion work and relationship quality using data from both members of the couple. Although others have studied emotion work, those data are not always from couples or from both members of the couple (Erickson 1993; Wharton and Erickson 1995). Further, we examined emotion work and relationship quality as daily constructs, while furthering our understanding of relationship quality as average levels, daily fluctuations, and volatility. When we examined emotion work for partners in predicting volatility in relationship quality we uncovered unique gender differences, which have important implications in understanding both greater equality and inequality in romantic relationships.

In sum, we make several contributions to the literature as to how gender and emotion work are integral constructs in understanding relational dynamics between partners. In general, we suggest that partners engage in emotion work as it is associated with more positive relationship quality on a daily basis and across the span of a week. We also suggest that romantic partners pay attention to gender stratification in their relationship and interactions with one another, especially in understanding the association between emotion work and volatility across the course of the week in multiple types of relationship quality. For example, specific to volatility, when women reported higher trait emotion work, their male partners reported greater volatility in love and commitment. Both partners are encouraged to maintain open lines of communication regarding emotion work, and the intentions of their emotion work, as part of efforts to balance inequities and work toward greater equality in the relationship.