The Value of Dimensional Models of Early Experience: Thinking Clearly About Concepts and Categories

Problem 1: Dimensional models do not advocate placing children into discrete categories of exposure

The first problem Smith and Pollak raise about dimensional models is that “subtypes of adverse experiences are fuzzy categories” (p. 71). After recasting dimensional models as specificity models, they contend that dimensional approaches advocate for placing children into “separate groups who have had [purportedly] different experiences” (p. 71). This framing fundamentally misrepresents the purpose of dimensional approaches, which seek to move beyond placement of children into single exposure categories as in specificity models and lumping all exposures together as in cumulative-risk. Instead, dimensional models seek to identify the shared mechanisms through which diverse early experiences influence different aspects of development (e.g., see Fig. 1 in McLaughlin & Sheridan, 2016). Smith and Pollak’s (2021) claim that threat and deprivation reflect “fuzzy categories” fundamentally misrepresents what dimensional models seek to accomplish (irrespective of whether they succeed or not).

Admittedly, numerous challenges exist in defining and operationalizing core underlying dimensions of early experience. This is particularly true for deprivation and unpredictability. Exposure to the dimension of threat/harshness has been operationalized as the number of different types of violence a child has encountered or the overall frequency of violence exposure. Deprivation has been studied primarily as a lack of learning opportunities and stimulation, but it may be useful to consider separate dimensions of cognitive, material, and emotional deprivation, each of which may have unique developmental consequences and mediating processes (Dennison et al., 2019; King et al., 2019). Measuring unpredictability presents numerous conceptual and methodological hurdles (Young et al., 2020), such as defining statistical properties of unpredictability in relation to social and non-social environmental factors. New measurement tools informed by dimensional models are sorely needed. Existing assessments were developed to assess the presence of discrete exposures (e.g., parental substance abuse, neglect) rather than dimensions of experience. Moreover, tools that measure children’s experience in their natural environments (e.g., devices that measure the child’s language environment [LENA] and caregiver–child proximity [TotTags]) will be particularly important for capturing variation in experiences of deprivation and unpredictability, which are difficult to assess using only self- or caregiver-reports (King et al., 2020).

Contrary to Smith and Pollak’s argument that dimensional models seek to place children into “separate categories” on the basis of different experiences, a central tenet of these models is that it is essential to measure and model multiple dimensions of experience simultaneously (Belsky et al., 2012; McLaughlin, 2020; McLaughlin & Sheridan, 2016). Nevertheless, they are correct in pointing out that in initial studies, threat, deprivation, harshness, and unpredictability have sometimes been measured using dichotomous indicators of relevant experiences (e.g., abuse reflecting the presence of threat, neglect reflecting the presence of deprivation). This practice reflects the difficulty of applying recently developed dimensional models to existing data collected using case-control designs aimed at identifying children with and without particular types of experiences. In such designs, children typically fall into two groups—those exposed to a relatively extreme form of adversity (e.g., abuse) and those who never encountered adversity, posing problems for modeling these experiences continuously. Shifting to a dimensional approach requires sampling strategies that capture not only children with the most severe experiences but also those in the mild-to-moderate range. As research on these topics has progressed, experiences of threat, deprivation, and unpredictability have been increasingly measured continuously, consistent with dimensional models (Goetschius et al., 2020; Hein et al., 2020; Lambert et al., 2017; Machlin et al., 2019; Miller et al., 2018, 2020).

In sum, dimensional models seek to identify and measure core features of environmental experience as dimensions that vary along a continuum of severity and occur to varying degrees in diverse forms of adversity (see Fig. S1 in the Supplemental Material available online). This affords the possibility of evaluating whether, how, and why these underlying aspects of experience shape developmental processes.

Problem 2: Co-occurrence of adversities does not mean that it is impossible to examine differential effects

A second claim of Smith and Pollak is that the co-occurrence of adversity poses a fundamental problem for dimensional models. However, dimensional models are predicated on the understanding that adversities co-occur (Belsky et al., 2012; McLaughlin, 2020; McLaughlin et al., 2014). Without addressing this co-occurrence by assessing multiple dimensions of experience, it would be easy to misattribute variance associated with a particular adversity to another co-occurring one. For this reason, dimensional models stipulate measuring multiple dimensions of adversity simultaneously and only then examining their associations (distinctly and jointly) with developmental outcomes. Any study examining a single form of adversity alone with no consideration of other aspects of adversity would not be considered to be taking a dimensional approach and would be instead applying a specificity approach.

One fundamental—and widely appreciated—concern about co-occurrence discussed by Smith and Pollak is that it can introduce problems of multicollinearity in statistical analysis. They present a few examples of adversity studies documenting co-occurrence rates in the moderate range. More comprehensive approaches to examining adversity co-occurrence—including prevalence rates in population-based studies and meta-analyses—detect associations in the small-to-moderate range (Matsumoto et al., 2020; McLaughlin et al., 2012). There is debate about the specific thresholds used to determine multicollinearity, but the cutoff most often used for correlations among predictors that is likely to result in problematic variance inflation is strikingly large (.80) and substantially larger than the observed associations among even the most strongly co-occurring adversities. Moreover, a growing number of statistical approaches beyond multiple regression have been implemented to account for adversity co-occurrence when evaluating the associations of multiple adversity types with developmental outcomes. These include latent-class analysis (Ballard et al., 2015), network models (Goetschius et al., 2020; Sheridan et al., 2020), and bifactor models that characterize the unique and shared variance in early experiences of adversity (Briggs-Gowan et al., 2019).

In sum, the overlap between multiple adversities is not sufficiently high that it creates problems for disentangling whether particular aspects of experience have differential associations with developmental outcomes, and there are numerous strategies for handling this co-occurrence. The fact that adversities co-occur is not a reasonable justification for assuming that all adversities are created equal.

Problem 3: Consistent differences in the downstream consequences of different dimensions of early experience have been observed

The third critique advanced by Smith and Pollak is that “It is not clear from extant data that there are consistent and replicable effects associated with different types of early childhood adversities” (p. 74). We contend that their effort to substantiate this claim empirically is lacking, although we acknowledge that much of the relevant research is relatively recent. Indeed, a number of recent empirical studies designed specifically to evaluate propositions of dimensional models yield evidence consistent with these ideas. This includes work supporting the predicted distinctions between threat and deprivation in their associations with a range of developmental outcomes, such as amygdala reactivity to threat, aversive learning, cognitive control, and even pubertal timing (Goetschius et al., 2020; Hein et al., 2020; Lambert et al., 2017; Machlin et al., 2019; Miller et al., 2018, 2020; Peckins et al., 2020; Rosen et al., 2019; Sheridan et al., 2017, 2020; Sumner et al., 2019; Sun et al., 2020; Wolf & Suntheimer, 2019). Perhaps the strongest evidence comes from systematic reviews and meta-analyses that document clearly divergent associations of threat and deprivation with neural structure and function (McLaughlin et al., 2019) and measures of biological aging, including pubertal timing and cellular aging (Colich et al., 2020).

Accumulating evidence also supports predicted distinctions between dimensions of harshness and unpredictability. For example, unique associations of environmental harshness and unpredictability with numerous life-history traits have been observed, including mating and relationship outcomes, parenting, risk-taking, effortful control, and temporal discounting (Belsky et al., 2012; Griskevicius et al., 2013; Simpson et al., 2012; Sturge-Apple et al., 2017; Szepsenwol et al., 2015, 2017, 2019). Moreover, multiple studies indicate that young adults growing up in more unpredictable environments display enhanced abilities for flexibly switching between tasks or mental sets and for tracking novel environmental information, particularly when in a mindset of stress/uncertainty (Mittal et al., 2015; Young et al., 2018). These results underscore the theoretical claim that developmental exposure to harsh and unpredictable environments not only induces trade-offs with costs to mental and physical health but also enhances skills for solving problems that are ecologically relevant in such environments (Ellis et al., 2017, 2020). The fact that no such effects emerged among individuals who grew up in harsher environments underscores, again, specificity in developmental consequences of different dimensions of adversity.

Smith and Pollak additionally contend that evidence for any developmental effects shared across adversity experiences somehow invalidates dimensional models. One principle central to dimensional models is that different dimensions of experience will influence children in ways that are at least partially distinct (see McLaughlin, 2020; McLaughlin et al., 2014). Thus, dimensional frameworks do not make the same claims as specificity models (i.e., that different exposures are associated with effects that are completely different). Treating them as if they do is equivalent to debating a straw man.

To make their case, Smith and Pollak highlight studies that show similar associations of threat and deprivation with stress response system functioning—specifically, the autonomic nervous system and hypothalamic-pituitary-adrenal (HPA) axis, as well as hippocampal structure. These arguments are puzzling, as advocates of dimensional models have long made clear that alterations in these stress-response systems are at least one common pathway influenced by many forms of adversity (McLaughlin, 2020; McLaughlin & Sheridan, 2016; McLaughlin et al., 2014). Such alterations include both recurring hyperarousal and hypoarousal of stress response systems, which regulate behavior in distinctive way (Del Giudice et al., 2011). Because of well-established deleterious effects of glucocorticoids on hippocampal neurons, any form of adversity that recurrently up-regulates the HPA axis is likely to affect hippocampal structure and function. Indeed, this is borne out in a recent systematic review of the literature (McLaughlin et al., 2019). To emphasize, dimensional models do not stipulate that threat and deprivation exert unique effects on stress-response systems—or on all developmental mechanisms. Instead, they argue that some developmental pathways, but not others, are uniquely influenced by particular dimensions of early experience.

Problem 4: Stress-response systems are not the only biological mechanism through which the environment influences development

Smith and Pollak’s final critique of dimensional models is that different types of adverse early environments are not biologically meaningful. In making this claim, they argue that (a) categories of exposure such as abuse and neglect are unlikely to map onto biology, (b) stress response systems are not responsive to particular types of experiences, and (c) children’s interpretation of events may be more important in shaping neurobiology than the actual experiences.

First, dimensional models focus not on categories of exposure but rather on core features of the environment that occur to varying degrees in a range of different exposures and experiences. These models assume that these features of the environment will be associated with some pathways that are shared (i.e., disruptions in stress response systems) and some that are unique to particular types of experiences. For example, the idea that the brain responds in specific and unique ways to the presence of threat is uncontroversial. The ability to identify threats in the environment and mobilize defensive responses to them is essential for survival. Decades of animal and human neuroscience research supports the presence of neural circuits, conserved across species, that respond to environmental threats and orchestrate defensive responses (LeDoux, 2003, 2012; Phelps & LeDoux, 2005). A systematic review demonstrates that early-life experiences of threat are consistently associated with changes in the structure and function of these networks (e.g., reduced amygdala volume, elevated amygdala responses to threat cues; McLaughlin et al., 2019). These findings are consistent with substantial evidence that children who have encountered threatening early environments exhibit heightened perceptual sensitivity to anger, increased accuracy in identifying angry (but not other) facial expressions, and attentional biases to threat cues—all results that have not been observed in children who experience deprivation (Pollak et al., 2005, 2009; Pollak & Sinha, 2002; Pollak & Tolley-Schell, 2003).

Central to dimensional models is the claim that the magnitude of these effects should scale with the intensity and duration of exposure to threat, and existing evidence is consistent with that idea (Ganzel et al., 2013; Machlin et al., 2019; Marusak et al., 2015; McLaughlin et al., 2015). It is noteworthy that systematic reviews provide no evidence that an absence of cognitive and linguistic stimulation, for example, influences these same threat-related neural systems (McLaughlin et al., 2019).

Second, Smith and Pollak present a remarkably narrow view of biological mechanisms through which environmental experiences influence development. They argue that “stress-response systems are not sensitive to specific types of experience,” and we have already made clear that dimensional models do not presume otherwise, although patterns of hyper- versus hypoarousal do have coherent developmental antecedents (Del Giudice et al., 2011). Further, one might assume, on the basis of their arguments, that stress response systems are the only route by which adverse environmental experience influences the brain. This is deeply inconsistent with existing evidence on experience-driven plasticity mechanisms that influence neurodevelopment independent of the stress response system.

Experience-driven plasticity involves several well-established biological mechanisms through which environmental experiences exert relatively specific influences on learning and neurodevelopment, including experience-expectant and experience-dependent learning. These concepts are reviewed in depth elsewhere (Gabard-Durnam & McLaughlin, 2020; McLaughlin & Gabard-Durnam, in press; Nelson & Gabard-Durnam, 2020). Experience-driven plasticity mechanisms produce substantial changes in behavior and neural circuits through myelination and synaptic pruning that eliminates inefficient and unnecessary connections in response to particular types of environmental inputs, some occurring during specific windows of heightened neuroplasticity known as sensitive periods (Fu & Zuo, 2011; Takesian & Hensch, 2013).

Exposure to the specific environmental experiences is required to initiate the plasticity underlying experience-expectant and experience-dependent learning, and the timing, quality, and intensity of those experiences determines the amount of learning and plasticity that occur (Kolb & Gibb, 2014; Werker & Hensch, 2015). This has been amply demonstrated in animal models (e.g., an absence of light input to the retina shifts organization of primary visual cortex; Hubel & Wiesel, 1970) and human studies (e.g., language exposure shapes later phonemic perception; Kuhl et al., 2003). Beyond decades of work in animal models, evidence from human cognitive neuroscience demonstrates that specific types of environmental experiences have specific effects on neural circuits. For example, children’s linguistic experiences (measured observationally in their natural environments) are related to the structure and function of circuits specialized for language processing (e.g., activity in Broca’s area during a language processing task, white-matter integrity of the arcuate fasciculus; Romeo, Leonard, et al., 2018; Romeo, Segaran, et al., 2018). This simple example demonstrates clearly the idea of experience-dependent plasticity in the brain that is specific to neural circuits that process particular types of information. This is difficult to reconcile with Smith and Pollak’s argument that “the nature or type of adverse experiences is not directly tied to a specific neurobiological response or outcome” (p. 76). Although such specificity is unimpeachable in the domains of basic sensory and motor processing and language, much remains to be understood in the domain of higher-order cognition (Rosen et al., 2019; Sheridan & McLaughlin, 2016). Identifying which inputs are of primary importance for shaping association cortex—at which developmental periods—is a task that requires time and careful scientific inquiry. The importance of a robust and well-characterized scientific theory is that empirical studies can evaluate and refine initial predictions. Dismissing the idea that experience-driven learning and plasticity is a mechanism through which adversity influences neural development and focusing entirely on the stress response system will not advance these important scientific goals.

Finally, Smith and Pollak argue that the way children interpret their experiences may be more important than the objective experience in shaping neurobiology. This idea is interesting and worthy of empirical investigation, although existing data that speak directly to this issue are sparse and suggest that correlations between appraisals and physiology are small in magnitude (Denson et al., 2009; Mauss et al., 2005). In our view, the way children construe their experiences may be important in shaping some developmental mechanisms, though certainly not all. The experience-driven plasticity mechanisms thought to underlie many developmental consequences of deprivation have little to do with children’s interpretation of their experiences. An absence of expected inputs from the environment, such as a lack of exposure to linguistic input early in life, does not need to be interpreted by children as stressful to have lasting effects on neural architecture, learning, and cognitive abilities any more than an absence of visual input has to be appreciated by children to affect vision. In sum, extensive evidence supports the notion that specific types of environmental experiences are associated with changes in particular brain circuits and that the neurodevelopmental consequences of early-life adversity are not restricted to stress response systems.