An Evaluation of Positive Psychology Primals and Implications for Psychology

Abstract

This article describes two studies which test predictions derived from the Theory of Subjective Wellbeing Homeostasis in relation to self-report, correlational data. The theory proposes that all self-report data contain a common source of variance called Homeostatically Protected Mood (HPMood). It is therefore hypothesised that the magnitude of inter-correlations between self-report measures will diminish after their shared variance of HPMood has been removed. The first study analyses the extent to which the Primal World Beliefs (Primals) correlate with subjective wellbeing (SWB), before and after controlling for HPMood. The second study confirms the reliability of study 1 results. It also extends the analyses to examine the correlation between the brief 4-item Primals measure, and other related self-report measures, again before and after controlling for the HPMood. Analyses are based on data from approximately 1,000 participants collected at two timepoints. The hypotheses are supported, showing that both the Primals and other tested measures do not represent constructs that are independent from HPMood. This result throws doubt on the interpretation of correlations between Primals and the other self-report variables that were tested, bringing into question the validity of positive psychology overall.

1 Introduction

This report brings together two different approaches to understanding subjective wellbeing (SWB) within the discipline of psychological science. Both approaches were conceived around the turn of this century.

The earliest was Cummins (1995) who, after documenting the remarkable level of stability in population estimates of subjective wellbeing (SWB), suggested one possible cause “could be the existence of a psychological, homeostatic mechanism that maintains an average level of life satisfaction at around 75%SM [% of Scale Maximum, now referred to as ‘percentage points’ (pp)]” (pp.193–194).

A few years later, Seligman and Csikszentmihalyi (2000) introduced Positive Psychology to the discipline. “A science of positive subjective experience, positive individual traits, and positive institutions promises to improve quality of life … The exclusive focus on pathology that has dominated so much of our discipline results in a model of the human being lacking the positive features that make life worth living. Hope, wisdom, creativity, future mindedness, courage, spirituality, responsibility, and perseverance … The authors outline a framework for a science of positive psychology.”(abstract).

Both approaches to understanding SWB have yielded a substantial theoretical and empirical literature, much of which is overlapping and some of which is specific to each view. The purpose of this article is to examine and compare each approach, within the context of psychological science, most especially the extent to which positive psychological constructs are valid and independent from each other.

1.1 Background

Over the past 50 years, research in psychological science has revolutionized understanding of affect and subjective wellbeing (SWB). It began in earnest with two same-year publications (Andrews & Withey, 1976; Campbell et al., 1976) which set many of the measurement and conceptual parameters that have facilitated this rich seam of new psychological understanding. Outstanding achievements have included the creation of the affective circumplex (Russell, 1980), which provides a mathematically-derived basis for affect classification, while the many hundreds of articles by Ed Diener (1946 – 2021) create a formidable basis for building an understanding of SWB.

The most informative lines of research are the theoretical models which, due to their character, not only propose new ways of understanding the differential nature of affect, emotion, and mood, but also allow such proposals to be empirically tested. There are many relevant theories, most of which are centred around scale development for the negative forms of affect, operationalized as stress, anxiety and depression (see, for example, Lovibond & Lovibond, 1995). The attendant research is then concerned with testing whether such scales are reliable and valid measures of their intended constructs, often by involving only affectively negative variables. The exclusion of measured positive affect may represent an important limitation to understanding the full potential of these models.

This limitation is exemplified by one of the most influential of these theoretical proposals, as the Tripartite Model, developed by Clark and Watson (1991). This model proposes three factors as general distress, physiological hyperarousal (specific anxiety), and anhedonia (specific depression). A thoughtful review is offered by Laurent et al. (1999). They conclude that, while factor analytic work has provided support for the tripartite model, its ability to reliably differentiate anxiety and depressive disorders rests on the additional measurement of positive affect. That is, while anxiety is best characterized by elevated levels of physiological hyperarousal, depression is best characterized by low levels of positive affect.

A second theory-based view of affect is the Affect Infusion Model (Forgas, 1995). From the early 1980’s, Joseph Forgas has been researching the interaction of affect and cognition. By 1991 his research was starting to concentrate on affect and social cognition. Then, in 1995, he elaborated his Affect Infusion Model in a major journal, the Psychological Bulletin. This paper, which has received more than 5,000 citations, presents the Model as a ‘new integrative theory’ which is proposed as a ‘comprehensive explanation’ of the role of affect, both positive and negative, in social judgements.

Forgas’s, 1995 text mentions various constructs which are essential components of an affectively inclusive, homeostatic model, such as mood maintenance and the temporal infusion of affect into social judgements. However, this 1995 paper makes no mention of affect-related homeostasis. Contemporaneously, however, Cummins (1995) did make this connection, proposing the existence of a “psychological, homeostatic mechanism” as has already been noted.

Over the following period, 1996 to 2003, Cummins and colleagues published around 40 papers, many of which concerned some aspect of ‘SWB homeostasis theory’. In the following year, Forgas et al. (2004) published “An evolutionary model of managing moods: Evidence for the role of homeostatic cognitive strategies in affect regulation.” The authors fail to cite any reference to ‘Cummins’ or the developed literature on subjective wellbeing homeostasis at that time. Moreover, the central construct of ‘cognition’ in the Forgas model obviates the connection to ‘homeostasis’. Cognition is all about change, as in “The process of knowing, of acquiring knowledge and understanding through thought and through the senses” (Merriam-Webster, 2024). Therefore, since ‘homeostasis’ concerns stability around a ‘setpoint’ of whatever variable is being managed (see later), the idea of ‘homeostatic cognition’ in the absence of a setpoint is without merit.

Two other theoretical models concerning SWB management, which have survived into the current era, are those mentioned earlier as positive psychology and SWB homeostasis. Examining these two approaches, some quarter century after their proposal, provides an unusual opportunity for the development of scientific clarity. There are two reasons. First as has been mentioned, theories concerning affect in psychological science, tend to concentrate on the negative, as also noted by Seligman. In contrast, both Positive Psychology and Homeostasis Theory are primarily concerned with the management of positive affect. The second opportunity for developing theoretical clarity is that both theoretical approaches propose a central ‘core’ element that does not change. For positive psychology this ‘individual difference’ (genetic) component is the Primal World Beliefs (i.e. Primals), which are proposed to shape many personality and wellbeing variables (Clifton et al., 2019). For homeostasis theory, the core is a component of Subjective Wellbeing mentioned earlier, in the form of Homeostatically Protected Mood (HPMood). So it is logical that, since both Primals and HPMood are highly correlated with SWB, the first step in the examination of the two theoretical models will be to test the extent to which these two core constructs can be separated in their relationship with SWB. Each construct will now be examined in more detail.

1.2 Primal World Beliefs

In their introduction to ‘Primals’, Clifton and colleagues (2019) provide a detailed description of the construct, the methodology used to select representative items, together with additional descriptive statistics (pp.86–87). In summary, this text introduces “a set of environment beliefs—primal world beliefs or primals—that concern the world’s overall character (e.g., the world is interesting, the world is dangerous). To create our measure, we systematically identified candidate primals … conducted exploratory factor analysis …and two confirmatory factor analyses …; examined sequence effects … and concurrent validity; and conducted test-retests … The resulting 99-item Primals Inventory (PI-99) measures 26 primals with three overarching beliefs—Safe, Enticing, and Alive (mean α 0.93)—that typically explain 55% of the common variance. These beliefs were normally distributed. They are stable …(19 month test–retest results averaged 0.88, 0.75, and 0.77, respectively); strongly correlated with many personality and wellbeing variables (e.g., Safe and optimism, r 0.61; Enticing and depression, r -0.52; Alive and meaning, r 0.54); and explained more variance in life satisfaction… than the BIG 5… In sum, the PI-99 showed strong psychometric characteristics, and the primals plausibly shape many personality and wellbeing variables” (abstract).

The list of 11 authors includes Martin E. P. Seligman and other esteemed academics. The methodology they used to produce their initial ‘Primals Inventory’, as summarized above, is well accepted within this area of psychology. Moreover, the conduct of their several exploratory sub-studies, has been generally well executed. Unfortunately, however, they have left a well-known Achille’s heal unguarded.

As observed by Meehl (1990), correlational data derived from self-reports, tend to naturally, and substantially, intercorrelate, across a wide range of self-evaluative topics. These “too high correlations with other tests purporting to measure different things” can potentially indicate lack of separation from a third [unknown] confounding variable (p. 84; Campbell & Fiske, 1959). Thus, it is worth taking a closer look into the standard methods of testing the construct validity in psychological science and its limitations.

1.3 Construct Validity in Psychological Science

There are several disciplinary features that distinguish science from philosophy. One of these is the definition of constructs in a manner that allows empirical testing of theoretical models, with gold standard evidence being that which leads to a valid statement of causation. Unfortunately, however, while psychological science claims many explanatory constructs, their empirical verification, especially in SWB research, tends to be weak. This is because the proposed models are largely, if not entirely, based on correlational statistics involving self-report data, which typically share some degree of variance (Cummins et al., 2018). This is particularly the case when the self-report data concern variables that are both proximal to the self, rather than distal, and also general/abstract rather than specific (Cummins et al., 2003). Thus, failing to account for this commonly shared variance in self-report data is likely to be the limiting factor in understanding the true relationships between measured constructs.

The conceptual fragility of such results was first noted by Thorndike (1920) and confirmed by many subsequent researchers. Of special relevance are two publications in the late 1980s. Webster and Starbuck (1988) observe “Everything correlates with everything else.” (p.113), while in a paper prepared in 1985, (Meehl, 1990: see footnote p.195) noted “…in social science everything correlates with everything to some extent, due to complex and obscure causal influences …” (p.207). Both sets of authors accompanied their observation with extensive analyses, with Webster and Starbuck concluding “the background relationships have roughly the same order of magnitude as the meaningful ones, partly because researchers' nonrandom behaviors construct meaningless background relationships. These meaningless relationships make induction untrustworthy.” (p.114). Meehl concludes “The net epistemic effect of these ten obfuscating influences [he lists nine others] is that the usual research literature review is well-nigh uninterpretable” (abstract). He famously named this obfuscating phenomenon the ‘‘Crud factor” (p.207).

These eminent researchers’ forthright observations have been empirically verified by numerous additional researchers, many using discriminant validity. Devised by Campbell and Fiske (1959), the test of discriminant validity determines the extent to which new concepts or measurements actually introduce unique variance, or whether they are simply variations on an established theme (i.e., confounding variable). The discriminant validity tests aim to verify the independence of measures from other unrelated constructs. It is typically tested using, what Maul refers to, as the ‘classic trinity of validation activities’. These are:

• (a). estimation of overall reliability or measurement precision, typically by Cronbach’s alpha,

• (b). some form of latent variable modelling, typically via exploratory factor analysis, and

• (c). estimations of associations between the measured variables and external variables, typically through a correlation matrix.

Maul discusses the question of why these classical statistics are insufficient to provide reliable validation. One reason, he suggests, is that the studies concerned are not testing theory. Because of this, virtually any results can be interpreted in favour of the instrument under investigation. But there is another explanation, which is that much of the overlapping variance between items, causing most of Maul’s (a), (b), and (c), comes from Meehl’s ‘crud factor’ (HPMood), as described above.

This concern is perhaps the strongest form of critique directed to results based on self-report data. That is, models formed by combining variables measured as positive psychological constructs, cannot be unequivocally interpreted. This is because the data forming such variables are inherently correlated with one another through the ‘crud factor’. Thus, the key verification of a construct as an independent predictor, requires the removal of this commonly shared variance. Only then can the correlated variables be demonstrated to be contributing their own unique variance to the relationship.

In conclusion, it is very easy to predict, to the point of inevitability, that data from a random collection of positive self-report variables will display a factorial structure, the factors will display internal consistency and reliability, and the variables will correlate with other positive self-report variables, satisfying the key validation criteria. However, to truly validate positive psychological constructs obtained through self-report data, it is paramount to control for the key theoretical confound in these relationships, namely the HPMood.

1.4 Primals as an Independent Construct

Given that Primals are strongly related to Subjective Wellbeing (SWB) (Clifton et al., 2019), the key test of difference between them is the extent to which this relationship survives after controlling for the common shared variance in self-report data, the ‘crud factor’. Unfortunately, however, the Primals’ authors have not conformed to the principle of using similar discriminants to control for such confounding. Instead, they demonstrate that Primals can be distinguished from variables with which they intuitively have little variance in common, such as objective measures (Cummins, 2000). For example, Clifton et al. (2019) show that the Primal ‘Alive’ correlates 0.03 with Body Mass Index, and that ‘funny’ correlates 0.01 with family income (p.90). In fact, there is no a priori reason to expect that either subjectively measured Primals variables would significantly correlate with such objective variables (see Andrews and Withey, 1976).

Their second demonstration involves the use of multiple regression, in which they regress the Big Five personality factors onto objective demographic variables in the first step, adding Primals in the second step. The authors report that the addition of Primals increased the variance accounted for. However, as in the demonstration above, objective demographic variables have a normally low connection to self-report variables (confirmed by Clifton et al., Table 3), and so are unsuited to be used as a test of discrimination. Adding self-report psychological variables in the second step is inevitably going to increase the overall variance accounted for.

In summary, these authors have failed to demonstrate that the variance contributed by the Primals measures can be distinguished from that contributed by other theoretically relevant positive psychological constructs. The following sections will advance this understanding, by demonstrating that the Primals’ variance cannot be distinguished from that of a more appropriate source, the HPMood, which dominates its relationship with subjective wellbeing.

1.5 Defining Subjective Wellbeing

Cummins (2022) describes the derivation of a novel definition for Subjective Wellbeing (SWB) created with assistance from the first publicly available version of ChatGPT (Generative Pre-trained Transformer) (openai.com/blog/chatgpt/). This definition is:

“Subjective wellbeing (SWB) is a broad term, referring to a person's overall sense of wellbeing, happiness, and life satisfaction. It is a subjective and personal concept, based on each person's own perceptions and feelings about their life. SWB includes emotional, cognitive, and affective components. The cognitive component is nested within emotion, and is therefore unavailable for independent measurement, while the affective component is present as both a short-term state and long-term mood. SWB is influenced more strongly by subjective than objective factors. The strongest three influences are an emotionally intimate relationship, a sense of purpose in life, and having financial adequacy. The level of SWB is normally under homeostatic control.”

There are many parts to this definition. However, a test of its adequacy is that the parts, both severally and collectively, are consistent with the empirically demonstrated character of SWB. That is, the terms and constructs involved in this definition can be used to explain the auto-correlation of self-report variables. The sequence of ideas to achieve this explanation is as follows:

• 1. The key variables describing the structure and functioning of Subjective Wellbeing (SWB) are affects, which come in two forms as emotion and mood. Affect is the term used to describe both emotion and mood in combination, within a general context (Buchanan, 2007).

• 1.1. Emotions are complex, acute, affective-cognitive responses to percepts (Russell, 2003) and thoughts (Wyer et al., 1999). Emotions may be high or low intensity (Russell, 2003), involving cognitive and somatic reactions to specific environmental or cognitive antecedents (Cohen et al., 2008; Scherer, 2000).

• 1.2. Moods are a primitive form of affect, genetic in origin, object-free, and chronic (Ekkekakis, 2013; Oatley & Johnson-Laird, 1987; Russell & Feldman Barrett, 1999). They are stable (Chaplin et al., 1988), low intensity affects (Cohen et al., 2008; Russell, 2003), forming an unchanging, affective background for each individual (Buchanan, 2007), and often not having a specific known antecedent of origin (Bower & Forgas, 2000; Cohen et al., 2008; Ekkekakis, 2013). Moods have less well-defined behavioural response tendencies than Emotions (Larsen, 2000). One empirically demonstrated mood that fulfills all the above criteria for the description of a mood is Homeostatically Protected Mood (Capic et al., 2018; Cummins, 2010, 2017; Cummins et al., 2014a).

• 2. Homeostatically Protected Mood (HPMood) is a simple, genetically determined mood, with a level that does not change for each person. This stability is consistent with HPMood being the phenotype of an individual difference, representing a set-point for each person. Such setpoints have been demonstrated and confirmed (Capic et al., 2018; Cummins et al., 2014a).

• 2.1. HPMood comprises three affects (content, happy, alert) (Cummins, 2010), all closely related to one another within the positive-deactivated quadrant of the affective circumplex (Hartley-Clark, 2014). The purpose of HPMood is to provide a weak, positively activated, affective background to consciousness, such that HPMood is present in most thoughts that concern self-evaluations. Due to the influence of HPMood on conscious evaluations of the self, people generally feel positive about themselves.

• 3. Subjective Wellbeing (SWB) is remarkably stable because it has HPMood at its core, and the level of HPMood does not change. Variation in the level of SWB is caused by its emotion content (Davern et al., 2007).Thus, a response to a SWB item, such as ‘How satisfied are you with your life as a whole’, is an amalgam of HPMood and emotion (Capic et al., 2018).

• 3.1. Homeostatic control represents the extent of variation in the measured level of SWB, while under the influence of Subjective Wellbeing Homeostasis. Effective homeostatic control strives to maintain the level of SWB close to setpoint for each person (Cummins, 2017). Because of this homeostatic control, the chronic level of SWB normally varies by only a few percentage points for each person (Anglim et al., 2015).

The combination of these component parts allows the generation of a theoretical explanation for the normally high stability of measured levels of SWB (see: Khor et al., 2023).

1.6 Subjective Wellbeing Homeostasis

This second approach to the management of affect is described by the Theory of Subjective Wellbeing Homeostasis, which has received much coverage in the pages of the Australian Centre on Quality of life Bulletin over the past five years (Cummins, 2022, 2023), and elsewhere (Cummins, 2017, 2021) . The basis of this theoretical approach to understanding SWB lies in the science of affect, and the pioneering work of Russell (2003: cited > 5,000 times), who coined the term core affect, to describe a neurophysiological state experienced as a basic feeling (mood). Russell describes core affect using the homeostatic analogy of felt body temperature. It is always there, it can be accessed when attention is drawn to it, extremes are most obvious, and it exists without words to describe it.

This idea was subsequently operationalized (Davern et al., 2007) and theoretically contextualized (Cummins, 2017) into a simple mood under homeostatic control. Now referred to as Homeostatically Protected Mood (HPMood), this central mood comprises an amalgam of three affects (content, happy, alert), and has been demonstrated to be remarkable stable under normal conditions, chronically varying by only a few percentage points for each person (Anglim et al., 2015). It is proposed that such stability is consistent with HPMood being the phenotype of an individual difference, representing a genetically-determined, stable set-point level for each person (Capic et al., 2018; Cummins et al., 2014a), as also mentioned above.

The potential implications of SWB homeostasis and HPMood, for understanding psychological measurement involving self-report data, are considerable. The overall implication is that a ubiquitous aspect of biological functioning, homeostasis, has been additionally demonstrated within psychology. If this claim is accepted, the basic properties of biological homeostasis can be anticipated to have parallels within psychology. One of these is the explanation of why SWB data display the ubiquitous content of stable HPMood. This is the variable with a level that is set, and defended by, genetically endowed, homeostatic processes. Then, because the level of HPMood is set at a different level for each person, this predictable source of shared variance influences the statistical relationships between self-report variables. This process is now further elaborated.

1.7 Homeostatically Protected Mood contaminates all relationships involving self-report data

The logical sequence in support of this statement is as follows:

• 1.1. All self-report evaluations are made within a single moment of consciousness. Each of these conscious moments comprises an amalgam of variable emotion (cognition + acute affect) and a constant, idiosyncratic level of HPMood for each person. Consider, then, what happens to the shared variance between variables [the basis for correlations] when survey data from a group of people are combined. Each response datum will share two types of variance with other data as:

• Common variance coming from the same individual person in the form of HPMood. This variance will be present in all self-evaluative responses produced by the same individual, causing those responses to share variance (correlate) with one another. Note that these correlations are caused by HPMood quite irrespective of the emotional (cognitive, acute affect) content of the data. This effect of HPMood is proposed to represent Meehl’s Crud Factor (Meehl, 1990), as has been stated.

• The second form of shared variance between self-report variables is not HPMood. For example, responses to ‘happy’ and ‘satisfied’ will share variance that reflects their elicitation of a common cognitive/affective reaction to each word. The level of this shared variance that is ‘unique’ to each variable pair, will likely differ between individuals and over time. However, this is normally a fragile source of shared variance to demonstrate and will be normally dominated by the shared variance arising from HPMood, as demonstrated by several studies (Lai & Cummins, 2013).

• 1.2 The above analysis predicts that, very commonly, what is measured as a ‘correlation’ between variables is not a valid indicator of unique variance overlap. For example, the correlation between self-esteem and extraversion does not measure the strength of some unique form of variance held in common by both variables. Rather, the correlation mainly reflects the shared variance of HPMood. This conclusion is consistent with many empirical demonstrations. For example, Ruyg (2016) studied the correlations between the ‘homeostatic buffers’ of self-esteem, optimism, and perceived control with two measures of SWB. Using raw scores, the buffers together accounted for around 13% of variance in SWB. However, this dropped to a negligible 1% after controlling for HPMood.

While other sources of shared variance between variables have been suggested, such as social desirability and acquiescence, none have been demonstrated to exert a comparable influence as that of HPMood. Social desirability has been discounted in the context of SWB measurement by adults, by Diener (1984) who concludes that none of the self-report scales he reviews “ show high social desirability effects’ (p.551). In relation to acquiescence, the manual for the Personal Wellbeing Index (International Wellbeing Group, 2024) recommends that “Prior to conducting statistical analyses, it is essential that raw data are checked for ‘response sets’ (3.1.1 Data cleaning)”.

In conclusion, studies in psychological science based on self-report data are easy to interpret. The measured variables will predictably intercorrelate with one another at impressive levels, allowing all manner of insightful explanations as to their meaning for theory development. The only fly in this ointment is HPMood which, unfortunately, renders such explanations either invalid or greatly exaggerated. Meehl’s (1990) ‘Crud Factor’ rules.

1.8 Implication of the Homeostatically Protected Mood confounding effects

It makes for a very sobering reflection that much of the psychological literature based on self-report data and correlational statistics cannot be simply interpreted due to the hidden confounding influence of HPMood. Yet this crucial information has been very slow to be accepted within general psychological science. Evidence of this uncomfortable fact is readily attained. The originators of this understanding were senior academics, highly regarded within their discipline. Yet the citation record for the papers giving strength to this alarming insight reflect not an academic approach response, but avoidance. The Webster and Starbuck (1988) publication has been cited about 130 times and Meehl (1990) about 800. In other words, while their papers have not been ignored, their combined effect on the published literature has been relatively small. Why is this so?

Perhaps the message is too destructive to be acceptable. Even in contemporary times, the great majority of the empirical literature in psychological science is based on self-report data, often accompanied by sophisticated statistical analyses based on correlations. So, it is clearly not in the self-interest of these authors to even mention the ‘inevitable auto-correlation’ as an ‘inconvenient truth’ (Gore, 2006) that might reduce their results to irrelevance. Besides, despite their accompanying analyses, neither Webster & Starbuck nor Meehl offer a theoretically useful reason for the auto-correlation phenomenon. As one consequence, even empirical researchers who are curious about the phenomenon have had nothing tangible to work with. But this excuse is no longer valid. As far as is known to the current authors, no valid challenge has been made to the demonstrations of setpoints for HPMood (Capic et al., 2018; Cummins et al., 2014a, b), the ubiquitous presence of HPMood in self-report data (Cummins et al., 2018), or the management of self-reported feelings concerning the self by homeostasis (Cummins & Nistico, 2002).

While these conclusions apply most strongly to measured subjective wellbeing which, by its very nature, can only be based on self-report data, the influence of HPMood extends to all self-report data to some degree. So, it is surely time for psychological science to grasp this nettle, to confront the challenge, work to further understand the auto-correlation problem, and to regard the removal of HPMood from correlational data as an imperative prior to the interpretation of results.

1.9 The Current Study

The preceding text has demonstrated the substantial danger to validity that is inherent in self-report data. That is, the variables measured by such data have an inherent tendency to intercorrelate. It is proposed that this tendency is caused by their content of Homeostatically Protected Mood. In sympathy with this description, the following hypothesis is to be tested: That the positive correlation between Primals and Subjective Wellbeing will diminish after the shared variance of Homeostatically Protected Mood has been removed.

2 Method

2.1 Participants

Cross-sectional data were provided from 2001 participants who completed the 5th longitudinal survey (Long-5) of the Australian Centre on Quality of Life (ACQOL) at Deakin University in September 2022. This is not a representative sample of the population. While each respondent was initially recruited by a random population invitation, into one of the preceding waves of the Australian Unity Wellbeing Index (AUWI), the respondents who volunteered to continue into Long-5 are a minority, representing a third of the original sample initially recruited (for recruitment methodology to the AUWI phone survey see Frykberg et al., 2023). Each respondent was invited to be recontacted for follow-ups in subsequent ACQOL online annual surveys. Participants who agreed to be recontacted were emailed an online survey at each subsequent wave of longitudinal data collection. Participants were advised that the participation in the longitudinal survey is voluntary and that their agreement was indicated by returning a completed questionnaire. As a result of this recruitment process, the attrition from the previous wave was 34.8% and the sample is older (M = 62.3, SD = 15.3) compared to the average age of the adult Australian population (M = 48.1, SD = 18.7) (Australian Bureau of Statistics, 2021).

2.2 Measures

2.2.1 Homeostatically Protected Mood

Homeostatically Protected Mood (HPMood) is a measure of each person’s genetically determined mood setpoint, represented by an amalgam of the three affects from the circumplex model of affect which account for the most variance in SWB (Blore et al., 2011; Davern et al., 2007; Tomyn & Cummins, 2011). These affects are: ‘content’, ‘happy’ and ‘alert’. Survey participants were asked to indicate how they generally feel each day in relation to these affects by rating each one from zero ‘Not at all’ to 10 ‘Extremely’. The statements were worded as: “I generally feel…[content/happy/alert]”. Cronbach's alpha for the HPMood measure was 0.85, indicating high internal consistency.

2.2.2 Subjective Wellbeing

Subjective wellbeing was measured using two measures. First is the widely used single-item Global Life Satisfaction (GLS), which asks participants to rate their response to a question “How satisfied are you with your life as a whole?” from zero “No satisfaction at all” to 10 “completely satisfied”.

The second measure employed was the Personal Wellbeing Index (PWI), computed as a composite score derived from satisfaction ratings across seven life domains, which together account for most of the variance in GLS (International Wellbeing Group, 2024).These domains encompass satisfaction with Standard of Living, Health, Achieving in Life, Relationships, Safety, Community Connectedness and Future Security. Participants were asked to rate their satisfaction levels on a scale from zero ‘no satisfaction at all’, to 10 ‘completely satisfied’. The questions were worded as: ‘How satisfied are you with…[your standard of living; your health; what you are currently achieving in life; your personal relationships; how safe you feel; your future security;]’. Cronbach's alpha for the PWI was 0.86, indicating high internal consistency.

2.2.3 Primal World Beliefs

Primal World Beliefs (“Primals”), are defined as beliefs about the basic character of the world (Clifton et al., 2019). The Primals scale used in this study comprises 4 items from the 18 item Primals Inventory (PI-18) (Clifton & Yaden, 2021), agreed by members of the Positive Psychology Center (see acknowledgements) as forming the essence of the Primals construct. These include: ‘I tend to see the world as pretty safe’; ‘No matter where we are or what the topic might be, the world is fascinating’. ‘Everything happens for a reason and on purpose.’ and ‘Most things in the world are good.’ Respondents were asked to rate their level of agreement with each item, from zero ‘Do not agree at all’ to 10 ‘Completely agree’. Cronbach's alpha for the 4-item Primals’ measure was 0.66, indicating moderate internal consistency. Because of this, the Primals items in Study 1 will be examined individually.

2.3 Statistical Analyses

Descriptive statistics were calculated in Stata (Version 15.1), and the size of the change between correlations of unadjusted and adjusted pairs of variables was calculated in Excel following the procedure described in Zou (2007) (see Zou’s Excel calculator). Data cleaning and standardization of all continuous measures to percentage point scale (i.e., 0–100) were performed as per the Personal Wellbeing Index (PWI) Manual (International Wellbeing Group, 2024). Data cleaning involved identifying and removing response sets where respondents consistently provided either a minimum (0) or maximum (10) score across all seven domains of the PWI. Such uniform responses may result from acute emotional states, acquiescence, lack of comprehension or intentional response patterns. Consequently, four participants were excluded from the analysis.

Descriptive statistics were calculated for all variables, and the means for Global Life Satisfaction (GLS) and the Personal Wellbeing Index (PWI) were compared to the Australian normative data (Khor et al., 2023). The influence of HPMood on each pair of self-report variables was estimated from the significant change in the size of the correlations between six variables (i.e., 4 Primals items, GLS and PWI) from before (i.e., unadjusted) to after removing the shared variance with HPMood (i.e., adjusted).

To remove shared variance with HPMood from each of the variables, separate linear regression analyses were conducted. Each of the six variables were regressed onto the composite HPMood measure. Residual values from each regression model were saved as separate variables in the dataset, representing the variance in each variable not shared with HPMood.

The difference between the unadjusted and adjusted correlation coefficients was estimated following the procedure described by Zou (2007), where correlations between the unadjusted and adjusted pairs of variables were treated as dependent and non-overlapping. The correlations were treated as dependent because they involve the same group of participants, and non-overlapping because they are compared between two different pairs of variables (i.e., unadjusted pair and adjusted pair), without an overlapping third variable (eg. Var1 vs Var3 and Var1 vs Var3). Thus, this estimation followed a two-step process. First, difference between unadjusted and adjusted correlations were calculated as r₁—r₂. Second, the significance of the differences between correlations of unadjusted and adjusted pairs of variables was estimated based on non-overlapping confidence intervals (CI’s). The formula for estimating the CIs of the difference between correlations includes the following three elements:

• 1. Correlation coefficients between two pairs of variables (i.e., one unadjusted and other adjusted) that are being compared (i.e., r₁ and r₂).

• 2. Lower and upper confidence intervals of each of the two correlations (i.e., lower = l; upper = u). These confidence intervals are first calculated for Fisher’s z transformed correlation coefficient (r’) as shown in Formula 1 and then converted to confidence intervals for Pearson’s correlation coefficient (r), as shown in Formula 2. In Formula 1, ^z_lower and ^z_upper are the lower and upper confidence intervals for r’; n is the sample size; ^z_critical is the critical value in a z distribution beyond which the null hypothesis is rejected. In Formula 2, the e^value are the inverted confidence intervals for z scores.

  $$\begin{array}{c}zlower = {r}{\prime}-{\sqrt{1/(n-3)}}_{({z}_{critical})}\\ zupper = {r}{\prime}+{\sqrt{1/(n-3)}}_{(zcritical)}\end{array}$$

  (1)
  $$\begin{array}{l}l = ({e}^{2 \times z}lower -1) / ({e}^{2 \times z}lower +1)\\ u= ({\text{e}}^{2 \times z}upper -1) / ({\text{e}}^{2 \times z}upper+1)\end{array}$$

  (2)

• 3. Correlation between a pair of correlations (i.e., r₁ and r₂), accounting for correlations between all other possible pairs of unadjusted and adjusted variables in the correlation matrix. Thus, in a correlation matrix of 4 × 4 variables (Table 1) four additional correlations are accounted for (i.e., r3-r6).

  Table 1 All correlations between a pair of non-adjusted and adjusted variables

The correlation between pairs of correlations are calculated as per Formula 3:

$$\begin{array}{c}corr({r}_{1},{r}_{6})=cov({r}_{1},{r}_{6})/\sqrt{\text{var}({r}_{1})\text{var}({r}_{6})}=\\ \left[.5{r}_{1}{r}_{6}({r}_{2}+{r}_{3}+{r}_{4}+{r}_{5})+{r}_{2}{r}_{5}+{r}_{3}{r}_{4}-({r}_{1}{r}_{2}{r}_{3}+{r}_{1}{r}_{4}{r}_{5}+{r}_{2}{r}_{4}{r}_{6}+{r}_{3}{r}_{5}{r}_{6})\right]/\left[(1-{r}_{1}{}^{2})\times (1-{r}_{6}{}^{2})\right]\end{array}$$

(3)

The above three elements are included in the Formula 4 below, to estimate the CIs of the difference between the correlations of unadjusted and adjusted pairs of variables. In this formula, l₁ & u₁ are lower and upper CIs for r₁ (i.e., correlation between unadjusted pair of variables) and l₆ & u₆ are lower and upper CIs for r₂ (i.e., adjusted pair of variables).

$$\begin{array}{c}\text{L }= {r}_{1}-{r}_{6}-\sqrt{{({r}_{1}-{l}_{1})}^{2}+{({u}_{2}-{r}_{6})}^{2}-2corr({r}_{1}{r}_{6})({r}_{1}-{l}_{1})({u}_{2}-{r}_{6})}\\ \text{U }= {r}_{1}-{r}_{6}+\sqrt{{({u}_{1}-{r}_{1})}^{2}+{({r}_{6}-{l}_{6})}^{2}-2corr({r}_{1}{r}_{6})({u}_{1}-{r}_{1})({r}_{6}-{l}_{6})}\end{array}$$

(4)

To address the risk of a Type 1 error in the correlations between variables, the Bonferroni method was employed, which involved dividing the alpha level by the total number of correlations conducted. The number of tests is calculated as shown in Formula 5.

$$Adjusted alpha level = [number of variables * (1 - number of variables)] / 2$$

(5)

Thus, significance of change in correlations was estimated based on CI’s reflecting the Bonferroni adjusted alpha levels.

The magnitude of the difference was also presented as Cohen’s q measure of effect size, which is recommended for estimation of difference between two correlation coefficients (Cohen, 2013). The Cohen’s q is a difference between Fisher’s z transformed Pearson’s correlations coefficients. Cohen proposes the following categories for the interpretation: < 0.1: no effect; 0.1 to 0.3: small effect; 0.3 to 0.5: intermediate effect; > 0.5: large effect.

3 Results

Table 2 presents the descriptive statistics for all variables used in the analyses. It is notable that both measures of Subjective Wellbeing (SWB) approximated the bottom of their normal range as Global Life Satisfaction (GLS:-0.20 percentage points: pp) and the Personal Wellbeing Index (PWI: + 0.57 pp). The most divergent domains lying below their normal range were Health (-2.98 pp) and Achieving (-2.69 pp). Those highest above their normal range are Safety (+ 6.54 pp) and Standard (+ 1.52 pp). These variations were likely due to the older age of the sample. All were within 1 response unit (10 pp) of their normal range and therefore unlikely to affect the correlational results to follow.

Table 2 Descriptive Statistics (2022 Survey)

Tables 3 and 4 present analyses which test the hypothesis that bivariate correlations involving Primals and SWB variables (GLS and PWI), will decrease after the HPMood shared variance has been removed as a covariate.

Table 3 Pearson's correlations between HPMood, Primals and GLS (unadjusted and adjusted)

Table 4 Pearson's correlations between HPMood, Primals and PWI (unadjusted and adjusted)

Column 2 in Table 3 demonstrates raw-score correlations with HPMood. This revealed substantial (0.8) correlations with GLS and PWI, and with Primals between 0.22 and 0.38.

Column 3 in Table 3 (GLS-unadjusted) and Column 2 in Table 4 (PWI-unadjusted) show similar correlations. All unadjusted correlations were significant at a Bonferroni adjusted alpha level of 0.001 (i.e. 0.0002).

Column 4 in Table 3 in (GLS-adjusted) and Column 3 in Table 4 (PWI-adjusted) show the correlations after HPMood variance has been removed from each variable. The GLS vs PWI correlations remained significant and moderate (0.40). However, the correlations between Primals and GLS have become non-significant, while the Primals vs PWI correlations diminished substantially, by 61%-76%.

Column 5 in Table 3 and Column 4 in Table 4 show differences between unadjusted-adjusted correlations and the significance of this change. All correlations have been significantly reduced following the removal of HPMood variance (p < 0.001).

Column 6 in Table 3 and Column 5 in Table 4 show the percentage change between unadjusted and adjusted correlations. These showed a 45% reduction for both GLS and PWI, and a minimum reduction of 61% for the Primals.

Column 7 in in Table 3 and Column 6 in Table 4 show magnitude of the difference between correlations of unadjusted and adjusted pairs of variables using Cohen’s q statistic. The reduction in correlations after removal of shared HPMood variance was large between PWI and GLS, intermediate between the Primals items and GLS, with the lowest difference between GLS and the Primals item of purpose. Reductions in correlations between all Primals items and the PWI were small but significant.

In summary, the significant unadjusted correlations between Primals and GLS became non-significant following the adjustment, when the variance contributed by HPMood was removed. Likewise, the correlations between the PWI and Primals were substantially reduced in the presence of HPMood. Thus, the hypothesis that the positive correlation between Primals and Subjective Wellbeing will diminish after the shared variance of Homeostatically Protected Mood is removed, has been supported.

4 Discussion

The objective of this study was to test the extent to which the correlational relationships between Primals, Global Life Satisfaction (GLS), and the Personal Wellbeing Index (PWI) remain after the shared variance of Homeostatically Protected Mood (HPMood) has been removed. The results reported in Tables 3 and 4 support this prediction.

The very high correlation between GLS and PWI of 0.80, accounting for 64% of the shared variance, is confirmation of raw-score, convergent validity, for both forms of scale. This close connection was intended by the formative construction of the PWI (Diamantopoulos & Winklhofer, 2001) as the first level deconstruction of GLS. More surprising is the residual r = 0.44 (19.36% of the variance accounted for, after HPMood has been removed. The source of this residual shared variance is uncertain but worthy of further investigation.

The results shown in Table 2, also confirm the significant raw-score correlation between each of the 4 Primal World Belifs (Primals) and HPMood (range 0.22 to 0.38), GLS (range 0.19 to 0.30), and between the Primals and the PWI (range 0.23 to 0.37). However, after the shared variance of HPMood has been removed, the correlations between the Primals and GLS become non-significant (0.01 to 0.03), while those between the Primals and PWI were significantly reduced (0.08 to 0.14).

Taken together, these results confirm the hypothesis that ‘the positive correlation between Primals and Subjective Wellbeing will diminish after the shared variance of Homeostatically Protected Mood has been removed. This is evidenced by non-significant, adjusted correlations between Primals and GLS, after the removal of HPMood variance, and the much-reduced Primals vs. GLS correlations. There are, however, some aspects of this study that give pause to simple acceptance of this result.

First, the sample was one of convenience, recruited from previous surveys, and not representative of the general population. Thus, the psychological characteristics influencing these respondent’s decisions to continue responding to successive surveys are unknown. Second, the respondents were older (LONG5: M = 62.3, SD = 15.3; LONG6: M = 61.3, SD = 16.2) than the normative range (M = 50.7, SD = 17.4; Khor et al., 2023). Table 2 shows that both Satisfaction with Health and Achieving were lower in this sample than is normal, consistent with a high proportion of elderly people who have retired from work. On the other hand, Standard and Safety domains were both above their normal ranges, consistent with the availability of superannuation and pensions, coupled with high levels of social security from Government agencies. However, there is no known reason to expect these sample characteristics to account for the correlational results that were being reported.

In summary, these results confirm the hypothesis that correlations between the Primals and the other measured, self-report variables’ can mainly be attributed to the confounding variable of HPMood. However, two kinds of confirmatory data are required to ascertain that HPMood is the confounding factor as hypothesised. One is to allow replication of the current results involving Primals and SWB measures (i.e. GLS and PWI). The second is to broaden the variables under investigation beyond Primals, to determine the extent to which other self-report variables also display auto-correlation due to the pairs sharing a large proportion of HPMood variance.

5 Study Two: Replication and Extension

Study 2 follows on directly from Study 1. It uses the same methodology and tests similar hypotheses, but this time employing an expanded set of dependent variables. These variables in Study 2 have been selected based on the three criteria. First, that the variable had shown strong correlations with measures of other wellbeing and psychological constructs (Clifton & Yaden, 2021). Second, that the data were self-reported, facilitating a response containing Homeostatically Protected Mood (HPMood). Third, that the measures varied in their content and presentation.

The rationale for this study lies within the composition of Subjective Wellbeing (SWB) as defined by the Theory of Subjective Wellbeing Homeostasis and detailed in the ‘Defining Subjective Wellbeing’ section of the Study 1 introduction. Briefly, SWB is an amalgam of HPMood and emotion (Capic et al., 2018), both present at each conscious moment, at a varying degree. The less emotion is present in consciousness at the time of responding, the greater will be the relative (default) content of HPMood. It is assumed that this logic will apply more broadly to the measurement of a range of self-report variables. Moreover, the level of their shared variance with HPMood is expected to be similar across all measures, regardless of the number of items, temporal frame of reference, response type and psychological domain used in each measure. One caveat, however, is that measures which are more abstract and personal are expected to share more variance with HPMood (Cummins et al., 2018), and therefore exhibit greater reductions following the removal of HPMood variance.

In summary, the aim of Study 2 was to replicate the results from Study 1 in a different data set and test the following hypotheses in a range of self-reported measures commonly used in psychology:

• (a) As in study 1, ‘the positive correlation between Primals and Subjective Wellbeing (SWB) will diminish after the variance each variable shares with Homeostatically Protected Mood (HPMood) has been removed.

• (b) This diminution of shared variance will also be evidenced by the other bi-variate correlations with SWB.

6 Method

6.1 Participants

Cross-sectional data were provided from 1003 participants who completed the 6th longitudinal survey (Long-6) of the Australian Centre on Quality of Life at Deakin University. The Long-6 survey was conducted in September/October 2023 following the same recruitment strategy as for Study 1. The attrition in this survey from the previous survey was 23.4%. The mean age of the sample was again older (M = 61.3, SD = 16.2) than the average age of the adult Australian population (M = 48.1, SD = 18.7) (Australian Bureau of Statistics, 2021).

6.2 Measures

In Study 2, the data from fifteen self-reported instruments were examined. Of those, four had been previously scrutinized in Study 1, namely: Homeostatically Protected Mood (HPMood), Personal Wellbeing Index (PWI), Global Life Satisfaction (GLS), and Primal World Beliefs (Primals). In addition, 11 new measures were included. While all measures reported self-reported data, they differed across five different aspects:

• 1. Number of items comprising each measure: ranged from single-item to multiple-item measures (3–7).

• 2. Level of abstraction: some items were more abstract (‘satisfaction with life as a whole’; ‘most things are good’) and others more specific (“new ideas distract me’).

• 3. Temporal frame of reference: some items referred to the present (‘how satisfied are you with your life’), some to the future (‘I am always optimistic’), some to life in general (‘How you generally feel’).

• 4. Response types: (a) degree of satisfaction, (b) extent of feeling, (c) extent of agreement with statements.

• 5. Psychology domain: some items referred to a stable personality trait (‘I see myself as outgoing’), some to action (‘I ask others for help’), and some to specific psychopathology (‘How depressed do you feel’).

Together these measures were selected to demonstrate the effect of HPMood on a broad range of self-report data. Table 5 summarises all the measures used in Study 2, number of items in each, response mode and their rated target (eg., life, life domains, affect, personality, etc.).

Table 5 Measurement scales summary: Number of items, response option and the rated target

6.2.1 National Wellbeing

National Wellbeing was measured using two variables. One was the single question ‘How satisfied are you with your life in Australia?’ (Global National Wellbeing – GNW). This is a national adaptation of GLS. The other was the National Wellbeing Index (NWI), modelled on the PWI,

comprising satisfaction with six domains: the economy, the environment, social conditions, government, business, and national security (Frykberg et al., 2023). The responses were rated from zero (‘No satisfaction at all’) to 10 (‘Completely satisfied’). Cronbach's alpha for the NWI was 0.82, indicating high internal consistency.

6.2.2 Extraversion

Extraversion was measured using 3 items, adapted from the Ten Item Personality Inventory (TIPI; Gosling et al., 2003). Following the advice of Herzberg and Brähler (2006), the items were simplified to become single positive statements (eg., ‘I see myself as extraverted’). The responses were rated from zero (‘Disagree’) to 10 (‘Agree’). Cronbach's alpha (0.82) indicated high internal consistency.

6.2.3 Optimism

Optimism was measured using three positively-worded items from the Life Orientation Test-Revised (LOT-R) scale (Scheier et al., 1994). For example, ‘In uncertain times I usually expect the best’. The responses were rated from zero (‘Disagree’) to 10 (‘Agree’). Cronbach's alpha (0.89) indicated high internal consistency.

6.2.4 Self-esteem

Self-esteem was measured using Rosenberg Self-Esteem Scale (RSES) (Rosenberg, 1965) using five positively worded items. For example: ‘I feel that I have a number of good qualities.’ The responses were rated from zero (‘Disagree’) to 10 (‘Agree’). Cronbach's alpha (0.91) indicated high internal consistency.

6.2.5 Grit

Grit was described as having two factors: Consistency and Perseverance (Duckworth et al., 2007: p.1087). Items used in this study included the three from each factor that are most general and have the highest face validity. The Grit-consistency items were: ‘New ideas and new projects sometimes distract me from previous ones’, ‘I become interested in new pursuits every few months’, and ‘I have difficulty maintaining my focus on projects that take more than a few months to complete’. The Grit-perseverance items were: ‘I finish whatever I begin’, ‘I am a hard worker’, ‘I am diligent’. The responses were rated from zero (‘Disagree’) to 10 (‘Agree’). Cronbach's alpha for Grit-consistency and Grit-perseverance measures were 0.69 and 0.76, indicating adequate internal consistency, respectively.

6.2.6 Perceived Control

Perceived Control was measured along two dimensions representing primary and secondary control (Holloway, 2003), with positively worded items representing each dimension. Participants rated their agreement with statements relating to how they cope when something bad happens to them. For example, the item ‘I ask others for help or advice’ reflects primary control, and the item ‘I remember that the situation will improve if you are patient’ reflects secondary control. The responses were rated from zero (‘Disagree’) to 10 (‘Agree’). Cronbach's alpha for primary and secondary control measures were 0.69 and 0.76, indicating adequate and good internal consistency, respectively.

6.2.7 Mental Health

Mental health was measured using a single question: ‘How do you rate your mental health?’. This measure was adapted from the 40th Australian Unity Wellbeing Index cross-sectional survey (Frykberg et al., 2023). The responses were rated from zero (‘Very poor’) to 10 (‘Very good’).

6.2.8 Loneliness

Loneliness was measured using a single question: ‘How lonely do you generally feel?’. This measure was adapted from the 40th Australian Unity Wellbeing Index cross-sectional survey (Frykberg et al., 2023). The responses were rated from zero (‘I never feel lonely’) to 10 (‘I feel lonely all the time’).

6.3 Statistical Analyses

As in Study 1, descriptive statistics in Study 2 were calculated using Stata (Version 15.1), and the size of the difference between correlations of unadjusted and adjusted pairs of variables was calculated in Excel, following the same procedure specified in Zou (2007) (see Zou’s Excel calculator). Data cleaning and standardization of all continuous measures to percentage point scale (i.e., 0–100) were performed as per the Personal Wellbeing Index (PWI) Manual and as described in Study 1 Statistical Analysis section. Consequently, eight participants who consistently rated all seven PWI items as either zero or 10 were excluded from the analysis.

The descriptive statistics were calculated for all 15 variables, and the means for Global Life Satisfaction (GLS) and the Personal Wellbeing Index (PWI) were compared to the Australian normative data (Khor et al., 2023).

7 Results

A requirement for reliability comparisons between Study 1 and Study 2 is that their common variables have statistically similar means and standard deviations. These statistics are presented in Table 6.

Table 6 Descriptive statistics (2023 Survey)

The Table 6 shows that the two samples were generally comparable to one another on all variables with the exception of optimism, which was 3.6 pp lower in 2023. While the declines were consistent across all self-reported measures in 2023, compared with 2022 levels, these were not significant for some key variables. These included Homeostatically Protected Mood (HPMood), Global Life Satisfaction (GLS), Personal Wellbeing Index (PWI), Primals, Secondary Control and Mental Health. These variables, with the exception of Mental Health, remained steady to within 1 percentage point (pp) across the two surveys.

In summary, these changes reflect a substantial worsening of economic life in Australia in the period between the two sets of data collection. Recovery from Covid-19 lockdowns, consequential business failings, and high inflation adversely affected the lives of many people. Interestingly, however, the two measures of SWB (i.e., GLS and PWI) remained homeostatically protected and their levels remained steady. In summary, for the most part these different levels between studies were small, including the measure of Primals, which did not differ in magnitude between the studies.

7.1 Testing the Hypotheses

• (a) As for study 1, ‘the positive correlation between Primals and Subjective Wellbeing (SWB) will diminish after the variance each variable shares with Homeostatically Protected Mood (HPMood) has been removed.

• (b) This diminution of shared variance will also be evidenced by the other bi-variate correlations with SWB.

The results for correlations with GLS are presented in Table 7 and for PWI in Table 8. From these tables, it is evident that all measures conformed with the expected reduction in correlations following the removal of HPMood variance. The clear outlier was the ‘Grit-consistency of interest’, which unlike all other psychological measures, including the Grit-perseverance, did not correlate significantly with either SWB or HPMood. One reason for the lack of relationship is that these items were negatively worded, and reverse-scored. Recent research shows that such items are processed differently from positively worded questions (Zuanazzi et al., 2024). These authors find that respondents go through a two-stage process of converting an initial, automatic positive response, to a negative: “negation does not invert the representation of adjectives (i.e., “not bad” represented as “good”); furthermore, decoding accuracy for negated adjectives was found to be significantly lower than that for affirmative adjectives. Overall, these results suggest that “negation mitigates rather than inverts the neural representations of adjectives” (abstract). It seems possible that the content of HPMood was lost in this process.

Table 7 Pearson’s correlations between 15 variables with HPMood and GLS (unadjusted and adjusted)

Table 8 Pearson’s correlations between 15 variables with HPMood and PWI (unadjusted and adjusted)

Column 2 in Table 7 shows that HPMood correlated significantly and in the appropriate direction, with all other variables. The extremes were 0.41 with Primals and 0.81 with the PWI.

Column 3 in Table 7 and Column 2 in Table 8 shows that unadjusted GLS and PWI correlated significantly and in the appropriate direction, with all other variables. For GLS, the extremes were 0.35 with Extraversion (0.36 with Primals) and 0.79 with PWI. For correlations with PWI, the extremes were 0.41 with Primals and 0.81 with HPMood.

Column 4 in Table 7 and Column 3 in Table 8 show that, with the exception of the 0.40 correlation between PWI and GLS, all other adjusted correlations with GLS and PWI, respectively, were small (GLS: 0.00–0.17; PWI: 0.06–0.32). Bonferroni correction has been applied to the interpretation of these residual bi-variate correlations, which sets 0.008 as the minimum level of statistical significance. Using this criterion, 8/13 and of the adjusted correlations with GLS have remained significant, while for PWI all remained significant.

Column 5 in Table 7 and Column 4 in Table 8, show that all of the adjusted correlations were significantly lower (GLS: p < 0.001; PWI: p < 0.0002) than their unadjusted form.

Column 6 in Table 7 and Column 5 in Table 8, show that the correlations with all unadjusted and adjusted variables with GLS reduced by 77%-100%, with the smallest reduction between PWI and GLS, which reduced by 49%. Reductions in correlations between unadjusted and adjusted variables with PWI were somewhat smaller ranging from 49%-78%.

Column 7 in Table 7 and Column 6 in Table 8, show that the magnitude of these reductions ranged from intermediate (eg., Primals vs GLS: 0.30; Primals vs PWI: 0.29; Primals vs Loneliness: 0.29) to large (eg., PWI vs GLS: 0.63; Self-esteem and GLS: 0.62).

Conclusion: The Study 2 hypotheses were fully supported for GLS and partially for PWI. While all correlations with GLS and PWI significantly reduced following the removal of HPMood variance, most correlations with GLS became non-significant, while correlations with PWI all remained significant.

Overall, the results from the Study 2 confirmed those of the Study 1, that using raw scores, the shared variance of HPMood inflated the correlations between Primals and SWB. These results also demonstrated that this same influence of HPMood applied to correlations involving other self-report variables, except for the Grit-consistency measure, which has been discussed.

8 Discussion and Conclusion

The primary purpose of this research was to test the hypothesis that ‘the positive correlation between Primal World Beliefs and Subjective Wellbeing will diminish after the variance each variable shares with Homeostatically Protected Mood has been removed’. The secondary purpose was to test the extension of the hypothesis to include a mixed collection of other self-report variables. Both the primary and the secondary hypotheses were supported.

The potential relevance of these findings extends deeply into psychological science, where self-report data are commonly used as the basis for complex multivariate analyses. The results from such analyses may now be questioned in terms of their validity. Whereas authors have generally assumed that their correlations represent unique shared variance between the variables concerned, there is now an alternative explanation to be considered. That is that many, if not all, of the correlating self-report variables are linking their variances for a quite different reason. Their statistical linkage is through their shared content of Homeostatically Protected Mood (HPMood).

The mechanism whereby such correlations can occur has been covered in the introduction. Briefly, the process involves (a) the personal stability of HPMood level for each respondent; (b) The presence of HPMood within each measured self-report by each respondent; (c) This intra-person consistency across measured variables causes grouped survey data to demonstrate inter-variable correlation.

While the named causal agent for this correlational process is original, as HPMood, the more general idea of some underlying and unknown cause for such data auto-correlations has been suggested by many eminent authors. In brief, Thorndike (1920) was the first to note such auto-correlation, referring to it as ‘A constant error in psychological rating: A Halo effect’ (p.28), while some 50 years later Meehl (1990) referred to it as the ‘Crud factor’ (p.207). Most recently, almost a century after Thorndike, the active agent causing such auto-correlations was identified as Homeostatically Protected Mood (Capic et al., 2018; Cummins, 2010, 2017; Cummins et al., 2014a) in the more general context of Subjective Wellbeing Homeostasis. This latter research not only provided a new name for the phenomenon, but also an explanation for its occurrence, as explained above.

In conclusion, the results from both studies confirm the early observations giving rise to Thorndike’s Halo Effect and Meehl’s Crud factor. The implication for psychological research using self-report data is that any statement of correlation between variables must include a statement of the auto-correlation variance caused by HPMood. To paraphrase these authors in contemporary terms, ‘In psychological science, self-report variables tend to significantly intercorrelate due to the influence of Homeostatically Protected Mood.’ Thus, a corollary statement of consequence might be ‘The interpretation of empirical analyses, involving raw, self-report correlational data, are likely to be invalid’.