There are two major distinctions made between subdimensions of SWB; the division between its cognitive and affective components, and the further division of affect into positive and negative components. An additional minor distinction involves happiness, which seems to contain elements of both cognition and affect.

The internal division in cognition was initially highlighted by Diener (1984), who differentiated between global life satisfaction, which measures cognitive evaluations of life as a whole, and domain satisfaction, which measures cognitive evaluations of specific areas of life such as work, relationships, and spirituality. Diener et al. (1999) recommended separating the two, though Schimmack (2008) has noted they are closely related, with correlations up to r = 0.70.

The affective component appears more complex. Positive and negative affect are often treated as a univariate construct (e.g., when a person feels good about their day it implies they do not feel bad about their day). However, evidence suggests they can also be treated as bivariate (e.g., a person can be frequently happy and also frequently sad, or experience very little of either emotion; see Schimmack, 2008). As Schimmack noted, the separability of positive and negative affect is influenced by factors like question content, temporal duration, and others, but even accounting for those, positive and negative affect remain distinct.

The separability of positive and negative affect has parallels elsewhere. For example, Rogge et al. (2017) demonstrated that positive and negative relationship evaluations can be treated as a bivariate scale with separable dimensions. And researchers are actively exploring whether well-being and its opposite, ill-being, should similarly be treated as separable (e.g., Iasiello and Van Agteren, 2020; Zhao and Tay, 2023).

Finally, research suggests that happiness contains elements of both cognition and affect. Happiness is often considered the quintessential positive emotion, but as Schimmack (2003) noted, it is absent from the PANAS scale developed by Watson et al. (1988), which is typically used to measure positive affect. To address the gap, Lyubomirsky and Lepper (1999) developed the Subjective Happiness Scale (SHS). However, the SHS taps global, cognitive appraisals; unlike emotion measures such as the PANAS, which ask people to report the frequency or intensity of specific emotions over a time interval, the SHS asks people to evaluate their general happiness, and also to situate their happiness in comparison with others.

Even when not measured using the SHS, happiness may be more cognitively relevant than other emotions; in an experience sampling study, Schimmack (2003) found that participants’ life satisfaction correlated higher with self-reports of happiness than all other positive emotions, suggesting that happiness overlaps more with cognitive variables like life satisfaction.

Additional research suggests that happiness is subject to cognitive modifiers; one such modifier is the fear of happiness (Joshanloo, 2013)—the belief that happiness can potentially lead to bad outcomes (see Yildirim, 2019, for a study of the downstream effects of fear of happiness on well-being). A second cognitive modifier is the externality of happiness, or the belief that happiness is uncontrollable and primarily caused by external variables (see Yildirim and Belen, 2019). Fear of happiness has been shown to predict both emotional and psychological aspects of well-being above and beyond personality variables, such as the activity of the Behavioral Inhibition System and Behavioral Activation System (Yildirim and Belen, 2018), suggesting that the cognitive component of happiness is consequential.

In practice, the separability of these different aspects of well-being—the division of affect into positive and negative components, as well as the division of SWB as a whole (and happiness in particular) into both cognitive and affective components—suggests that the dimensions comprising SWB may be subject to complicated internal dynamics which merit further inquiry.

Given the differences between the subdimensions comprising SWB it is reasonable to expect that the relationships between them may be complex even if they can be aggregated to form a single SWB score. In the present study we sought to model these relationships. Latent profile analysis (LPA), which can model complex patterns of interaction between variables parsimoniously by representing them as groups (called “profiles”), is ideal for this investigation (see Spurk et al., 2020). We therefore used measures of the five subdimensions of SWB as indicator variables for LPA, to identify subgroups within our sample, determine their magnitude, and explore their differences.

Latent profile analysis can return two types of results, referred to as level differences and shape differences between profiles. To offer a simplified case, imagine asking volunteers two questions – how happy their day was, and how social their day was. If a profile analysis of these volunteers returned three groups – one with high levels of both happiness and sociality, one with medium, and one with low, those would constitute level differences between profiles, where levels of indicators vary directly (or inversely) with each other across subgroups. Level differences indicate a simple, direct relationship between the two variables, and researchers may conclude that happiness and sociality are related in an uncomplicated way.

However, if the same analysis returned an extra group – one with very low levels of sociality and high levels of happiness, for example, breaking the norm – that profile would be said to have a shape difference, and would indicate a more complex relationship between the variables meriting further inquiry. Based on our review of the literature, we anticipate that LPA will reveal shape differences in profiles, in addition to normal level differences. We remain agnostic regarding which variables, specifically, will drive such shape differences.

When LPA is conducted it is considered best practice to arrange the resulting profiles according to their scores on a relevant outcome variable (Spurk et al., 2020). This serves two purposes; the first purpose of such an outcome is to establish the discriminant validity of the profiles by demonstrating that they each predict different levels of the outcome variable. The second purpose is to situate the profile solution inside of the larger nomological network of constructs related to the indicators.

In our case, our indicator variables were all subdimensions of SWB. Therefore it was important to pick an external criterion variable that is theoretically important relative to SWB. As such, in the present study, we selected the Couples Satisfaction Index (CSI; Funk and Rogge, 2007) as an outcome measure to use for validating the profile solution.

The CSI is a measure of relationship quality focused on a person’s evaluation of their partner and of the relationship overall. It makes an ideal criterion variable in the present study for two reasons. First, romantic relationship satisfaction is widely considered to be one of the most important predictors of SWB (see Proulx et al., 2007, for a meta-analytic review of the relationship between the two). Second, research conducted in the past ten years suggests that romantic relationship satisfaction may also be a complex construct. While it is traditionally treated as univariate, anchored on one end by positive evaluations and on the other by negative, research by Rogge et al. (2017) suggests that relationship satisfaction may also contain nuances that emerge when positive and negative relationship qualities are treated as bivariate dimensions.

While the original CSI developed by Funk and Rogge (2007) is a univariate measure, the later research conducted by Rogge et al. (2017) suggests that the construct of relationship satisfaction as a whole may be fruitful ground for the exploration of complex relationships between the positive and negative aspects of well-being. By taking such a broad and consequential relationship—one traditionally considered to share a linear relationship with SWB—and using the indicators to explore whether the relationship is more complex and marked by profile arrangements with shape differences, LPA may contribute new knowledge to the understanding of the relationship between SWB and couple satisfaction.

We conducted a secondary analysis of a preexisting dataset (see Kaufman et al., 2022b). A nationally representative sample of U.S. residents was recruited (via Dynata), with informed consent, to participate in an online survey. The initial sample consisted of 3,699 people who were further filtered using five attention-check questions to ensure that participants were engaged with the study. Participants were only included if they passed all five attention checks. This reduced the sample to 2,000. This was further trimmed to 1,487 participants who indicated that they were in a relationship; single participants were excluded in this study because the outcome variable used to organize profiles was couple satisfaction. The participants were near-evenly split between males (49.7%) and females (50.3%). Regarding race and ethnicity, the sample consisted of 68.3% White/Caucasian participants, 17.2% Hispanic, 10.6% Black, 4.6% Asian, and 1.6% who classified themselves as “Other.” In terms of education and socioeconomic status, 62% of participants had at least a college degree and 59.3% reported an income of $75,000 or higher. Participant demographics differed slightly from the original sample which included people who were married or in relationships; participants in the present study tended to be more wealthy and more educated. The demographics for the full sample of singles in the present study can be seen in Table 1. The demographics for the original sample can be seen in Kaufman et al. (2022b, Study 2).

All variables were converted to Z-scores. We used LPA to identify profiles based on our indicators using Mplus (version 8.1). We ran a series of sequential models ranging from one to six profiles. We determined the best model solution by evaluating information criteria and likelihood ratio tests. Once we found an optimal solution, individuals were assigned profiles based on the greatest probability of group membership. Finally, we used the manual three-step auxiliary BCH approach to test for differences in couple satisfaction using a Wald chi-square test (Asparouhov and Muthén, 2014).

The present findings have theoretical and practical implications for researchers and practitioners who work with well-being. At the theoretical level our study has the potential to inform current debates about how to measure and conceptualize SWB. The debate over whether the positive and negative aspects of SWB should be treated as separate from each other has been ongoing and has occurred at all levels, from the level of positive and negative affect as subdimensions of SWB to broader questions of whether well-being and its converse, ill-being, should themselves be treated as separable (Schimmack, 2008; Iasiello and Van Agteren, 2020; Zhao and Tay, 2023).

The results of our analysis suggest that the univariate and bivariate perspectives are both useful. Our earlier research of SWB (see Kaufman et al., 2022a) suggests that positive and negative aspects of well-being can be combined into an essentially unidimensional (or univariate) measure. Doing so is often desirable due to the versatility and comprehensibility of such an approach. However, the present study reveals key LPA shape differences by examining positive constructs (e.g., positive affect, life satisfaction) and negative affect separately (i.e., as bivariate dimensions). The debate over the structure of SWB will likely continue, but the present study suggests that the univariate and the bivariate approaches may each have their benefits, depending on context and research goals. While most researchers will likely be content with using a univariate approach, there may be cases where the additional insight provided by a bivariate approach is valuable—for example, if researchers’ aim is to maximize the explanation of variance in an outcome, or if they have a compelling theoretical reason to believe that the Ambivalent or Indifferent profiles relate in a unique way to a construct.

It is not difficult to imagine, for example, that those in the Indifferent profile may be over-represented in a population of patients exhibiting flat affect, while those in the Ambivalent profile may be over-represented in a sample of patients who have difficulties controlling their emotions (for example, those who are prone to both anger and excitement). While these are only examples, and need to be tested, the point is that there are compelling situations where researchers may prize the extra explanatory power offered by approaches that treat the positive and negative components of SWB as separable. And, finally, we note that this conceptualization of well-being aligns well with the theoretical ideal of happiness advanced by Diener and Biswas-Diener (2008), where the goal of individuals should not be to maximize positive emotion and eliminate negative emotion endlessly, but accept that the two will exist alongside each other, sometimes independent, sometimes balanced.

At the practical level, our present findings offer implications for practitioners who deal with well-being in clinical, educational, and organizational settings. First, our findings suggest that in real-world settings it may not be advisable to assume that individuals who present with high negative affect are necessarily lacking in the positive aspects of well-being. The presence of our Ambivalent profile suggests that a portion of the population experiences high levels of both positivity and negativity—and even more noteworthy, the Ambivalent profile also ranked second highest on our external criterion variable of Couple Satisfaction, suggesting that in at least one important area of life (relationships), some people can experience high levels of negativity and still rank highly on important aspects of well-being relative to their peers. Further research is needed to determine how far this principle can be generalized, but we note here that our findings echo previous researchers’ conclusions (e.g., Bateson et al., 2011) that reflexively seeking to reduce negativity without inquiring about its broader context is a sub-optimal clinical approach.

Our research has limitations that we would encourage researchers to account for when interpreting our results and when designing future studies. First, our sample is representative of American coupled adults. Further research is needed to establish that our Indifferent and Ambivalent profiles have cross-cultural equivalents, estimate their prevalence, and determine how they relate to external criteria across cultures. We therefore caution against generalizing, but also suggest cross-cultural research as a potentially fruitful avenue for future exploration.

Second, LPA is methodologically distinct from more common, variable-centric forms of analysis. LPA is useful for identifying complex relationships between multiple variables by rendering them as groups (Spurk et al., 2020), but the Indifferent and Ambivalent profiles found here can also be explored using other approaches. In particular, SWB research may benefit if researchers explore bivariate scales that model positive and negative aspects of well-being as orthogonal dimensions, similar to the methods used by Rogge et al. (2017) to model relationship satisfaction. If successful, it would provide conceptual replication of the Ambivalent/Indifferent distinction found here, and may lead to the development of simpler ways to measure it.

A final limitation is our choice of dimensions. We accepted for the purposes of this analysis that the five subdimensions established in our previous research were sufficient as indicators. However, some researchers may prefer the original three dimensions recommended by Diener (1984), while others may wish to explore additional dimensions. Future research should use LPA to determine whether the profile solution found here replicates with different indicators. If so, it would further support the Ambivalent/Indifferent distinction.

While the sample used in the present study is both large and nationally representative, the findings here still represent data from only a single study. A natural follow-up step to this research is to verify that this profile solution replicates when different measures of the dimensions of SWB are used (such as the PANAS or SPANE, for positive and negative affect), as well as replicating and providing further validation for the profile solution using other outcomes. In particular, we think it would be fruitful to examine our profile solution in relation to other forms of satisfaction such as job satisfaction, as well as consequential life outcomes such as depression, anxiety, perceived health, and flourishing. As a final note, it may be beneficial to expand the range of outcomes tested beyond measures of satisfaction and well-being to a wider range of constructs to determine if there are situations where the Ambivalent and Indifferent profiles are especially relevant. For example, in the present analysis the Ambivalent and Indifferent profiles predicted moderate-to-high levels of couple satisfaction, but it is possible that profiles like these, where levels of positivity and negativity are either muted or exaggerated, may produce markedly different results in how a person is perceived by their relationship partner. This may be a fruitful avenue for further investigation.

Our demographic table also hints at future research possibilities. The demographic information suggests that men are over-represented in the Indifferent profile while women are over-represented in the Ambivalent profile. While much research has been done on the relationship between SWB and demographic categories such as race, gender, and socioeconomic status, we think that there is also potential for research studying how demographic variables relate to our Ambivalent and Indifferent categories.