With over 7,000 citations at the end of 2021, Ryff and Keyes (1995) article is one of the most highly cited articles in the Journal of Personality and Social Psychology. A trend analysis shows that citations are still increasing with over 800 citations in the past two years.
Most of these citations are reference to the use of Ryff’s measure of psychological well-being that uncritically accept Ryff’s assertion that her PWB measure is a valid measure of psychological well-being. The abstract implies that the authors provided empirical support for Ryff’s theory of psychological well-being.
Contemporary psychologists contrast Ryff’s psychological well-being (PWB) with Diener’s (1984) subjective well-being (SWB). In an article with over 1,000 citations, Ryff and Keyes (2002) tried to examine how PWB and SWB are empirically related. This attempt resulted in a two-factor model that postulates that SWB and PWB are related, but distinct forms of well-being.
The general acceptance of this model shows that most psychologists lack proper training in the interpretation of structural equation models (Borsboom, 2006), although graphic representations of these models make SEM accessible to readers who are not familiar with matrix algebra. To interpret an SEM model, it is only necessary to know that boxes represent measured variables, ovals represent unmeasured constructs, directed straight arrows represent an assumption that one construct has a causal influence on another construct, and curved bidrectional arrows imply an unmeasured common cause.
Starting from the top, we see that the model implies that an unmeasured common cause produces a strong correlation between two unmeasured variables that are labelled Psychological Well-Being and Subjective Well-Being. These labels imply that the constructs PWB and SWB are represented by unmeasured variables. The direct causal arrows from these unmeasured variables to the measured variables imply that PWB and SWB can be measured because the measured variables reflect the unmeasured variables to some extent. This is called a reflective measurement model (Borsboom et al., 2003). For example, autonomy is a measure of PWB because .38^2 = 14% of the variance in autonomy scores reflect PWB. Of course, this makes autonomy a poor indicator of PWB because the remaining 86% of the variance do not reflect the influence of PWB. This variance in autonomy is caused by other unmeasured influences and is called unique variance, residual variance, or disturbance. It is often omitted from SEM figures because it is assumed that this variance is simply irrelevant measurement error. I added it here because Ryff and users of her measure clearly do not think that 86% of the variance in the autonomy scale is just measurement error. In fact, the scale scores of autonomy are often used as if they are a 100% valid measure of autonomy. The proper interpretation of the model is therefore that autonomy is measured with high validity, but that variation in autonomy is only a poor indicator of psychological well-being.
Examination of the factor loadings (i.e., the numbers next to the arrows from PWB to the six indicators) shows that personal relationships has the highest validity as a measure of PWB, but even for personal relationships, the amount of PWB variance is only .66^2 = 44%.
In a manuscript (doc), we challenged this widely accepted model of PWB. We argued that the reflective model does not fit Ryff’s own theory of PWB. In a nutshell, Ryff’s theory of PWB is one of many list-theories of well-being (Sumner, 1996). These theories list a number of desirable attributes that the creator of a list considers to be necessary and sufficient for high well-being. As Sumner (1996) points out, the key problem of list theories is that they vary considerably from author to author and lack a deeper theory that explains why some desirable attributes are included or excluded from a list. Moreover, if such a deeper theory existed, it would provide a much stronger definition of well-being than the enumeration of specific items.
This theory of well-being implies a different measurement model in which arrows point from the measured variables to the construct of PWB. In psychometrics, these models are called formative measurement models. There is nothing unobserved about formative constructs. They are merely a combination of the measured constructs. The simplest way to integrate information about the components of PWB is to average them. If assumptions about importance are added, the construct could be a weighted average. This model is shown in Figure 2.
The key problem for this model is that it makes no predictions about the pattern of correlations among the measured variables. For example, Ryff’s theory does not postulate whether an increase in autonomy produces an increase in personal growth or a decrease in personal relations. At best, the distinction between PWB and SWB might imply that changes in PWB components are independent of changes in SWB components, but this assumption is highly questionable. For example, some studies suggest that positive relationships improve subjective well-being (Schimmack & Lucas, 2010). As a result, the theory provides no explanation for the presence of a single factor that produces positive correlations among the six PWB scales. Even today, nobody has provided a theory for this empirical finding, let alone tested it.
Jason and I did exactly that. Starting with the observation that PWB is measured with self-reports, we postulated that ratings of PWB items are influenced by the fact that some individuals provide inflated ratings on desirable attributes. This response style is called halo bias and was discovered over 100 years ago (Thorndike, 1920). It is present in ratings of personality and subjective well-being (Anusic et al., 2009; Kim et al., 2012; Schimmack & Kim, 2020). However, personality psychologists routinely ignore this fact and pretend that self-ratings are nearly perfectly valid measures of the intended construct.
We used ratings of personality items in Ryff’s data (the MIDUS study, wave 2), to identify the halo factor in personality ratings. We then examined how much of the variance in PWB scales reflects the actual construct (e.g., variance in the autonomy factor is reflected in the autonomy scale), and how much of the variance reflects halo. The results are shocking for most of the PWB scales.
Halo Variance in PWB Scales
To conclude, JPSP has published two highly cited articles that fitted a reflective measurement model to PWB indicators. In the desk-rejected manuscript, Jason Payne and I presented a new model that is grounded in theories of well-being and that treats PWB dimensions like autonomy and positive relations as possible components of a good life. Our model also clarified the confusion about Diener’s (1984) model of subjective well-being.
Ryff et al.’s (2002) two-factor model of well-being was influenced by Ryan and Deci’s (2001) distinction between two broad traditions in well-being research. “one dealing with happiness (hedonic well-being), and one dealing with human potential (eudaimonic well-being; Ryan &
Deci, 2001; see also Waterman, 1993)” (Ryff et al., 2002, p. 1007). We argued that this dichotomy overlooks another important distinction between well-being theories, namely the distinction between subjective and objective theories of well-being (Sumner, 1996). The key difference between objective and subjective theories of well-being is that objective theories aim to specify universal aspects of a good life that are based on philosophical analyses of the good life. In contrast, subjective theories reject the notion that universal criteria of a good life exist and leave it to individuals to create their own evaluation standards of a good life (Cantril., 1965). Unfortunately, Diener’s tripartite model of SWB is difficult to classify because it combines objective and subjective indicators. Whereas life-evaluations like life-satisfaction judgments are clearly subjective indicators, the amount of positive affect and negative affect implies a hedonistic conception of well-being. Diener never resolved this contradiction (Busseri & Sadava, 2011), but his writing made it clear that Diener stressed subjectivity as an essential component of well-being.
It is therefore incorrect to characterize Diener’s concept of SWB as a hedonic or hedonistic conception of well-being. The key contribution of Diener was to introduce psychologists to subjective conceptions of well-being and to publish the most widely used subjective measure of well-being, namely the Satisfaction with Life Scale. In my opinion, the inclusion of PA and NA in the tripartite model was a mistake because it does not allow individuals to choose what they want to do with their lives. Even Diener himself published articles that suggested positive affect and negative affect are not essential for all people (Suh, Diener, Oishi, & Triandis, 1998). At the very least, it remains an empirical question how important positive affect and negative affect are for subjective life evaluations and whether other aspects of a good life are even more important. At least, this question can be empirically tested by examining how much eudaimonic and hedonic measures of well-being contribute to variation in subjective measures of well-being. This question leads to a model in which life-satisfaction judgments are a criterion variable and the other variables are predictor variables.
The most surprising finding was that environmental mastery was a strong unique predictor and a much stronger predictor than positive affect or negative affect (direct effect, b = .66).
In our model, we also allowed for the possibility that PWB attributes influence subjective well-being by increasing positive affect or decreasing negative affect. The total effect is a very strong relationship, b = .78, with more than 50% of the variance in life-satisfaction being explained by a single PWB dimension, namely environmental mastery.
Other noteworthy findings were that none of the other PWB attribute made a positive (direct or indirect) contribution to life-satisfaction judgments. Autonomy even was a negative predictor. The effects of positive affect and negative affect were statistically significant, but small. This suggests that PA and NA are meaningful indicators of subjective well-being because the reflect a good life, but provide no evidence for hedonic theories of well-being that suggest positive affect increases well-being no matter how it is elicited.
These results are dramatically different from the published model in JPSP. In that model an unmeasured construct, SWB, causes variation in Environmental Mastery. In our model, environmental mastery is a strong cause of the only subjective indicator of well-being, namely life-satisfaction judgments. Whereas the published model implies that feeling good makes people have environmental mastery, our model suggests that having control over one’s life increases well-being. Call us crazy, but we think the latter model makes more sense.
However, it is near impossible to publish research that is critical of published work, especially if citations show that psychologists like a construct and are willing to overlook fundamental problems with constructs and measures. We received a desk rejection from JPSP, which just shows that journals do not feel responsible for articles in their journals that require corrections. JPSP even declined to correct an article that used deceptive practices to get significant results (Bem). A new open science journal of personality with a high rejection rate also rejected our article because it is just more interesting to publish new bullshit than to correct old bullshit. In short, psychology doesn’t have a system of self-correction, which is really the hallmark of science. Few researchers can afford to invest massive amounts of time and energy on correcting others mistakes, which does not advance their careers. As a result, psychology is filled with false claims, invalid measures, and nobody can do anything about it. In short, psychology is not a science, until it develops a culture of competition for quality.