2050 words

Introduction

“Variance explained” (VE) is a statistical concept which is used to quantify the proportion of variance in a trait that can be accounted for or attributed to one or more independent variables in a statistical model. VE is represented by “R squared”, which ranges from 0 to 100 percent. An r2 of 0 percent means that none of the variance in the dependent variable is explained by the independent variable whereas an r2 of 100 percent means that all of the variance is explained. But VE doesn’t imply causation, it merely quantifies the degree of association or predictability between two variables.

So in the world of genetics, heritability and GWAS, the VE concept has been employed as a fundamental measure to quantify the extent to which a specific trait’s variability can be attributed to genetic factors. One may think that it’s intuitive to think that G and E factors can be separated and their relative influences can be seen and disentangled for human traits. But beneath its apparent simplicity lies a philosophically contentious issue, most importantly, due to the claim/assumption that G and E factors can be separated into percentages.

But I think the concept of VE in psychology/psychometrics and GWAS is mistaken, because (1) it implies a causal relationship that may not exist; (2) implies reductionism; (3) upholds the nature-nurture dichotomy; (4) doesn’t account for interaction and epigenetics; and (5) doesn’t account for context-dependency. In this article, I will argue that the concept of VE is confused, since it assumes too much while explaining too little. Overall, I will explain the issues using a conceptual analysis and then give a few arguments on why I think the phrase is confused.

Arguments against the phrase “variance explained”

While VE doesn’t necessarily imply causation, in psychology/psychometrics and GWAS literature, it seems to be used as somewhat of a causal phrase. The phrase also reduces the trait in question to a single percentage, which is of course not accurate—so basically it attempts at reducing T to a number, a percentage.

But more importantly, the notion of VE is subject to philosophical critique in virtue of the implications of what the phrase inherently means, particularly when it comes to the separation of genetic and environmental factors. The idea of VE most often perpetuates the nature-nurture dichotomy, assuming that G and E can be neatly separated into percentages of causes of a trait. Thus this simplistic division between G and E oversimplifies the intricate interplay between genes, environment and all levels of the developmental system and the irreducible interaction between all developmental resources that lead to the reliable ontogeny of traits (Noble, 2012).

Moreover, VE can be reductionist in nature, since it implies that a certain percentage of a trait’s variance can be attributable to genetics, disregarding the dynamic and complex interactions between genes and other resources in the developmental system. Therefore, this reductionism fails to capture the holistic and emergent nature of human development and behavior. So just like the concept of heritability, the reductionism inherent in the concept of VE focuses on isolating the contributions of G and E, rather than treating them as interacting factors that are not reducible.

Furthermore, we know that epigenetics demonstrates that environmental factors can influence gene expression which then blurs the line between G and E. Therefore, G and E are not separable entities but are intertwined and influence each other in unique ways.

It also may inadvertently carry implicit value judgements about which traits or outcomes are deemed desirable or significant. In a lot circles, a high heritability is seen as evidence for the belief that a trait is strongly influenced by genes—however wrong that may be (Moore and Shenk, 2016). Further, it could also stigmatize environmental influences if a trait is perceived as primarily genetic. This, then, could contribute to a bias that then downplays the importance of environmental factors which would then overlook their importance and potential impact in individual development and behavior.

This concept, moreover, doesn’t provide clarity on questions like identity and causality. Even if a high percentage of variance is attributed to genetics, it doesn’t necessarily reveal the causal mechanisms or genetic factors responsible, which then leads to philosophical indeterminancy regarding the nature of causation. Human traits are highly complex and the attempt to quantify them and break then apart into heat percentages or variances explained by G and E vastly oversimplifies the complexity of these traits. This oversimplification then further contributes to philosophical indeterminancy about the nature and true origins (which would be the irreducible interactions between all developmental resources) of these traits.

The act of quantifying variance also inherently involves power dynamics, where certain variables are deemed more significant or influential than others. This, then, introduces a potential bias that may reflect existing societal norms or power structures. “Variance explained” may inadvertently perpetuate and reinforce these power dynamics by quantifying and emphasizing certain factors over others. (Like eg the results of Hill et al, 2019 and Barth, Papageorge, and Thom, 2020 and see Joseph’s critique of these claims). Basically, these differences between people in income and other socially-important traits are due to genetic differences between them. (Even though there is no molecular genetic evidence for the claim made in The Bell Curve that we are becoming more genetically stratified; Conley and Domingue, 2016.)

The concept of VE also implies a kind of predictive precision that may not align with the uncertainty of human behavior. The illusion of certainty created by high r2 values can lead to misplaced confidence in predictions. In reality, the complexity of human traits often defies prediction and overreliance on VE may create a false sense of certainty.

We also have what I call the “veil of objectivity” argument. This argument challenges the notion that VE provides an entirely objective view. Behind the numerical representation lies a series of subjective decisions, like the selection of variables to the interpretation of results. From the initial selection of variables to be studied to the interpretation of their results, researchers exercise subjective judgments which then could introduce biases and assumptions. So if “variance explained” is presumed to offer an entirely objective view of human traits, then the numerical representation represents an objective measure of variance attribution. If, behind this numerical representation, subjective decisions are involved in variable selection and results interpretation, then the presumed objectivity implied by VE becomes a veil masking underlying subjectivity. So if subjective decisions are integral to the process of VE, then the presumed objectivity of the numerical representation serves as a veil concealing the subjective aspects of the research process. So if the veil of objectivity conceals subjective decisions, then there exists a potential for biases and assumptions which then would influence the quantitative analysis. Thus, if biases and assumptions are inherent in the quantitative analysis due to the veil of objectivity, then the objectivity attributed to VE is compromised, and a more critical examination of subjective elements becomes imperative. This argument of course is for “IQ” studies, heritability studies of socially-important human traits and the like, along with GWASs. In interpreting associations, GWASs and h2 studies also fall prey to the veil of objectivity argument, since as seen above, many people would like the hereditarian claim to be true. So when it comes to GWAS and heritability studies, VE refers to the propagation of phenotypic variance attributed to genetic variance.

So the VE concept assumes a clear separation between genetic and environmental factors which is often reductionist and unwarranted. It doesn’t account for the dynamic nature and influence of these influences, nor—of course—the influence of unmeasured factors. The concepts oversimplification can lead to misunderstandings and has ethical implications, especially when dealing with complex human traits and behaviors. Thus, the VE concept is conceptually flawed and should be used cautiously, if at all, in the fields in which it is applied. It does not adequately represent the complex reality of genetic and environmental influences on human traits. So the VE concept is conceptually limited.

If the concept of VE accurately separates genetic and environmental influences, then it should provide a comprehensive and nuanced representation of factors that contribute to a trait. But the concept does not adequately consider the dynamic interactions, correlations, contextual dependencies, and unmeasured variables. So if the concept does not and cannot address these complexities, then it cannot accurately separate genetic and environmental influences. So if a concept can’t accurately separate genetic and environmental influences, then it lacks coherence in the context of genetic and behavioral studies. Thus the concept of VE lacks coherence in the context of genetic and behavioral studies, as it does not and cannot adequately separate genetic and environmental influences.

Conclusion

In exploring the concept of VE and it’s application in genetic studies, heritability research and GWAS, a series of nuanced critiques have been uncovered that challenge its conceptual coherence. The phrase quantifies the proportion of variance in a trait that is attributed to certain variables, typically genetic and environmental ones. The reductionist nature of VE is apparent since it attempts to distill interplay between G and E into percentages (like h2 studies). But this oversimplification neglects the complexity and dynamic nature of these influences which then perpetuates the nature-nurture dichotomy which fails to capture the intricate interactions between all developmental resources in the system. The concepts inclination to overlook G-E interactions, epigenetic influences, and context-dependents variablity further speaks to its limitations. Lastly, normative assumptions intertwined with the concept thenninteouvde ethical considerations as implicit judgments may stigmatize certain traits or downplay the role and importance of environmental factors. Philosophical indeterminancy, therefore, arises from the inability of the concept of VE to offer clarity on identity, causality, and the complex nature of human traits.

So by considering the reductionist nature, the perpetuation of the false dichotomy between nature and nurture, the oversight of G-E interactions, and the introduction of normative assumptions, I have demonstrated through multiple cases that the phrase “variance explained” falls short in providing a nuanced and coherent understanding of the complexities involved in the study of human traits.

In all reality, the issue of this concept is refuted by the fact that the interaction between all developmental resources shows that the separation of the influences/factors is an impossible project, along with the fact that we know that there is no privileged level of causation. Claims of “variance explained”, heritability, and GWAS all push forth the false notion that the relative contributions of genes and environment can be be quantified into the causes of a trait in question. However, we know now that this is false since this is conceptually confused, since the organism and environment are interdependent. So the inseparability of nature and nurture, genes and environment, means that the The ability for GWAS and heritability studies to meet their intended goals will necessarily fall short, especially due to the missing heritability problem. The phrase “variance explained by” implies a direct causal link between independent and dependent variables. A priori reasoning suggests that the intracacies of human traits are probabilistic and context-dependent and it implicated a vast web of bidirectional influences with feedback loops and dynamic interactions. So if the a priori argument advocates for a contextual, nuanced and probabilistic view of human traits, then it challenges the conceptual foundations of VE.

At the molecular level, the nurture/nature debate currently revolves around reactive genomes and the environments, internal and external to the body, to which they ceaselessly respond. Body boundaries are permeable, and our genome and microbiome are constantly made and remade over our lifetimes. Certain of these changes can be transmitted from one generation to the next and may, at times, persist into succeeding generations. But these findings will not terminate the nurture/nature debate – ongoing research keeps arguments fueled and forces shifts in orientations to shift. Without doubt, molecular pathways will come to light that better account for the circumstances under which specific genes are expressed or inhibited, and data based on correlations will be replaced gradually by causal findings. Slowly, “links” between nurture and nature will collapse, leaving an indivisible entity. But such research, almost exclusively, will miniaturize the environment for the sake of accuracy – an unavoidable process if findings are to be scientifically replicable and reliable. Even so, increasing recognition of the frequency of stochastic, unpredictable events ensures that we can never achieve certainty. (Locke and Pallson, 2016)