Child Effects on Positive Parenting Vary with Neighborhood Opportunity

Abstract

Prior theoretical and empirical research has highlighted links between positive parenting and the socioeconomic characteristics of the family’s neighborhood, but has yet to illuminate the etiologic origins of this association. One possibility is that the various predictors of parenting outlined by Belsky (1984; e.g., characteristics of the child, characteristics of the parent, contextual influences) may matter more in some neighborhood contexts than in others. To examine this possibility, we conducted genotype-environment interaction (GxE) analyses in a sample of 1,030 families of twins (average age 8 years; 51% male, 49% female; racial composition: 82% White, 10% Black, 1% Asian, 1% Indigenous, 6% multiracial) from the Twin Study of Behavioral and Emotional Development in Children in the Michigan State University Twin Registry. Neighborhood and parenting were assessed using multiple informants and assessment strategies (neighborhood and family informants, administrative data, videotaped parent-child interactions). Results pointed to strong evidence of etiologic moderation, such that child effects on positive mothering were prominent in neighborhoods with little opportunity and near zero in neighborhoods with ample opportunity. Such findings not only reframe the magnitude of child effects on the parenting they receive as context-dependent, but also indicate that mothers in impoverished neighborhoods may be more responsive to their children’s characteristics than mothers in neighborhoods with ample opportunity.

It is well known that specific parenting behaviors (i.e., positive parenting, parental control, or harshness with the child) often vary quite dramatically across families (Belsky, 1984), with key consequences for children’s mental health outcomes. Indeed, meta-analytic results support enduring associations between parenting and various forms of youth psychopathology (Pinquart, 2017; Yap & Jorm, 2015), while quasi-experimental (Burt et al., 2021; Klahr et al., 2011) and intervention (Tully & Hunt, 2015; Yap et al., 2016) studies have further suggested that these associations may be at least partially causal in origin.

Given the known sequalae of specific parenting behaviors, a growing body of research has sought to uncover why people parent the way that they do – that is, to illuminate the origins of positive parenting, parental control, and/or harshness. Extant theory (Belsky, 1984) has argued that parenting can be best understood as a complex and dynamic repertoire of behaviors embedded in an ecological network that includes the family context (e.g., the marital relationship, family financial stress), characteristics of the parent (e.g., personality), characteristics of the child (e.g., temperament), and the social context (e.g., ethnic, cultural, and community characteristics) (Kendler & Baker, 2007; Klahr & Burt, 2014; Kotchick & Forehand, 2002; Luster & Okagaki, 2005). For example, data has consistently suggested that, rather than being passive recipients of parenting, children’s behavior influences the parenting they receive (Anderson et al., 1986; Kendler & Baker, 2007; Klahr & Burt, 2014). Other work has found evidence of average differences in parenting practices both across and within nations (Chen et al., 1998; Porter et al., 2005). For example, mothers of preschool-aged children in China reported higher levels of protectiveness, encouragement of modesty, and use of shame along with lower levels of acceptance and democratic participation compared to mothers of preschool-aged children in the United States (Wu et al., 2002). Other research has indicated that, within the United States, higher SES parents tend to treat their children as autonomous equals whereas lower SES parents focused on obedience and conformity (Hoff et al., 2002).

Neighborhood characteristics have also emerged as a small but consistent correlate of parenting, especially for positive parenting behaviors. For instance, a prior systematic review (Cuellar et al., 2015) found that the majority of studies supported the presence of small but significant associations between protective parenting and neighborhood characteristics, while also noting that findings in the literature were somewhat mixed. Our lab (Burt et al., 2023) empirically expanded on the conclusions of Cuellar et al. (2015) by evaluating the specificity of these associations across elements of neighborhood and types of parenting behavior, respectively, in a sample of 1,030 child twin families (the same sample examined here). A series of specification curve regression analyses revealed that, even when controlling for family income, mothers residing in neighborhoods with limited socioeconomic opportunities (e.g., neighborhoods with high levels of poverty and blight) were less supportive of their children and were less likely to use explanation and praise during a difficult task. By contrast, measured forms of parental harshness, and most aspects of fathering, were unrelated or only minimally related to neighborhood context. Relatedly, a housing relocation randomized control trial (Moving to Opportunity) found that moving to a low poverty neighborhood reduced parental distress by 20% (Leventhal & Brooks-Gunn, 2003), an important finding in the context of the current study given that parental distress is a robust predictor of less supportive, harsher parenting (Conger et al., 1992). Such findings collectively document the clear existence of an association between neighborhood deprivation and positive parenting, over and above the level of family deprivation, although the theoretical mechanisms underlying this association remain unclear.

We suggest that the association between positive parenting and neighborhood characteristics can be understood as a specific manifestation of the way in which basic environmental cues shape behavior. We have previously proposed that, like all animals, humans detect and process cues in their surrounding environments related to the abundance of food, habitation, and other resources (Burt et al., 2023). These cues then influence the organism’s behavior, sympathetic and parasympathetic nervous activity, and ultimately, stress or recovery from stress (e.g., South et al., 2015). Studies of objective measures of the built environment (e.g., neighborhood audits, sensors, or geospatial data), for example, have shown that individuals in neighborhoods with more disorder have elevated cortisol, elevated heart rate during a laboratory stressor, higher perceived stress, and worse cognitive functioning (e.g., van Kamp et al., 2015). Put another way, we suspect that the absence of sufficient neighborhood resources and opportunities taps into biological systems related to the detection of security and other resources, and in doing so, shapes specific parenting behaviors.

How might this work in practice? One possibility relates to etiologic moderation (Purcell, 2002), whereby the etiologic origins of parenting behaviors differ across neighborhood contexts. Put another way, it is possible that the specific predictors of parenting outlined by Belsky (e.g., characteristics of the child, characteristics of the parent, contextual influences) may vary with neighborhood deprivation such that particular predictors matter more in some neighborhood contexts than in others. For example, cultural expectations regarding maternal support of her children may be more powerful in some socioeconomic contexts than in others. Alternately, child effects on maternal support could vary with neighborhood opportunity, such that the extent to which children elicit parental support varies systematically with level of neighborhood opportunity.

Child-based twin studies offer a compelling way to test these specific hypotheses due to the unique inferences that can be drawn. This is so because the unit of etiologic inference in these designs are the twin children, rather than their parents. Genetic and environmental influences on parenting in child-based twin designs are thus inferred from the genetic relatedness of the children (i.e., those who are being parented, rather than those who are doing the parenting). As a consequence, estimates of genetic influences within child-based twin designs serve as estimates of the child’s genetic makeup on the parenting they receive (Klahr & Burt, 2014). These kinds of indirect genetic effects on others’ behavior are typically conceptualized via evocative genotype-environment correlations (rGE) (Jaffee et al., 2012). Evocative rGE are defined as non-random or genetically-influenced responses from others in the individual’s environment, such that individuals evoke reactions from others that are consistent with their genetically-influenced characteristics. As one concrete example, it may be that children who behave aggressively evoke less nurturing reactions from their frustrated parents, reactions that are consistent with the child’s genetic predisposition towards aggression.

Child-based twin designs also allow us to disambiguate and quantify environmental influences on the parenting children receive. Non-shared environmental influences on parenting in child-based designs index the effects of non-genetic, child-specific characteristics on the parenting they receive, in addition to measurement error (Klahr & Burt, 2014). For example, one twin may suffer from an chronic illness while their co-twin does not and they receive differential parenting as a result (Caspi et al., 2004). By contrast, shared environmental influences within child-based designs are thought to capture those influences on parenting that are shared between siblings and act to increase similarity in the parenting that they receive regardless of the siblings’ degree of genetic relatedness (Klahr & Burt, 2014). These shared environmental influences are thought to estimate the effect of family-wide factors that would shape parenting similarly across all children in the family, including family socioeconomic status, broader cultural expectations, and/or characteristics of the parent (e.g., personality, cognitive ability). Shared environmental influences would also include any passive rGE (in which the parents’ own genetically-influenced behaviors underlie the parenting environment they provide for their children), as these effects are not thought to differ by zygosity. Put differently, because etiologic effects in child-based designs are necessarily calculated at the level of the child, broad cultural or family-wide influences on parenting in these designs would thus be included in estimates of the shared environment (Kendler & Baker, 2007; Klahr & Burt, 2014).

A child-based twin design is thus ideally positioned for answering our core question: does the etiology of positive parenting, including the specific contributions of child evocative rGE effects and broader cultural/parent effects, vary across neighborhood context? A standard univariate GxE (Purcell, 2002; genotype by environment interaction; defined as differential responsiveness to environmental risk as a function of genetic variability) in a child-based twin design would allow us to directly evaluate this possibility.

The current study sought to do just this, illuminating the extent to which the origins of supportive mother-child relationships, and specifically the respective roles of evocative child rGE and broader cultural influences, might vary with the level of opportunity in the broader neighborhood context. We conducted a series of univariate biometric GxE models with maternal support as the outcome and neighborhood opportunity as the moderator. We focused specifically on maternal support in these analyses for several reasons. First, and as noted above, prior work in these data strongly suggested that supportive mothering was the aspect of parenting most closely linked to the family’s neighborhood context (Burt et al., 2023). Second, prior meta-analytic work in child-based twin designs indicated that evocative rGE effects on mothers were generally stronger than those on fathers, accounting for 35–51% of the variance in mothering and only 15–34% in fathering (Klahr & Burt, 2014). Given all this, we thus focused specifically on supportive mothering in the current analyses. We hypothesized that both child effects (indexed here via evocative rGE or genetic influences) and broader cultural effects (indexed here via shared environmental influences) on maternal nurturance would vary with the extent of neighborhood opportunity, consistent with our hypothesis that the absence of sufficient neighborhood resources and opportunities taps into biological systems related to the detection of security and other resources, and in doing so, alters the origins of specific parenting behaviors. However, given the absence of prior work in this area, we did not have specific hypotheses as to the direction of these effects.

METHOD

Participants

The population-based Michigan State University Twin Registry (MSUTR) includes several independent twin projects. Participants in the current study were drawn from the Twin Study of Behavioral and Emotional Development in Children (TBED-C), a study within the MSUTR. The TBED-C includes a population-based arm (N=528 families), and an ‘under-resourced’ arm (N=502 families) for which inclusion criteria also specified that participating twin families lived in neighborhoods with neighborhood poverty levels at or above the Census mean at study onset (10.5%). All procedures were approved by the primary author’s institutional review board. Children provided informed assent. Parents provided informed consent for themselves and their children. This study was not pre-registered. Because of the language in the informed consent document, we cannot post these data publicly, but they can be obtained from the primary author upon reasonable request.

To recruit families, the Department of Vital Records in the Michigan Department of Health and Human Services (MDHHS) identified twins in our age-range via the MSUTR’s Michigan Twins Project, a population-based registry of twins in lower Michigan recruited via birth records. The Michigan Bureau of Integration, Information, and Planning Services database was used to locate family addresses no more than 120 miles from East Lansing, Michigan through parent drivers’ license information. Pre-made recruitment packets were then mailed to parents by the MDHHS. Parents who did not respond to the first mailing were sent additional mailings roughly one month apart until either a reply was received or up to four letters had been mailed.

This recruitment strategy for the TBED-C yielded an overall response rate of 57% for the under-resourced arm and 63% for the population-based arm. Other recruitment and sampling details can be found in prior publications (i.e., Burt & Klump, 2019). The two arms of the study were analyzed jointly for the current analyses, yielding 224 monozygotic (MZ) male twin pairs, 211 dizygotic (DZ) male pairs, 202 MZ female pairs, 206 DZ female pairs, and 187 DZ opposite-sex pairs. Collectively, the twins were 6–10 years old (mean age=8.02 years, although 30 pairs turned 11 by time of their participation) and were 48.7% female and 51.3% male. Across the full sample, participants endorsed racial identities in the following proportions: White, 82%; Black, 10%; Asian, 1%; Indigenous, 1%; multiracial, 6%. However, families in the under-resourced arm, but not the population-based arm, were more racially diverse than the local population (e.g., 14% Black and 77% White in the under-resourced arm versus 5% Black and 87% White in the population-based arm; 5% Black and 85% White in the local area Census).

PROCEDURES

TBED-C families completed their intake assessments between 2008 and 2015, although the vast majority of assessments (90%) were conducted between 2008 and 2013. Assessment teams consisted of two research assistants and at least one paid staff member and took 4–5 hours to complete (lunch was provided). Families completed questionnaires, interviews, and videotaped interactions, during the in-person assessment. Most assessments took place in our university laboratory. In the event that families were unable or unwilling to travel, however, assessments took place in participants’ homes (13%). Families with younger twins were more likely to complete home visits than on-site assessments visits (Cohen’s d=−.24, p=.01), as were families of color (d=−.21; p<.05). Similarly, families with fewer financial resources were more likely to participate in home visit versus on-site assessments (d=−.32; p=.004). However, families completing on-site assessments versus home visits did not differ in maternal education, zygosity of the twins, or sex of the twins (ds ranged from −.02 to .16, all ns). The on-campus parent-child interactions took place in offices that were arranged to resemble living rooms, with cameras inconspicuously installed in the ceiling. Home visit interactions took place in a family living space with a video camera placed on a tripod in the room. All families were compensated for their time and effort.

MEASURES

Neighborhood Opportunity

Neighborhood characteristics were assessed using both administrative data and neighbor informant-reports. For the former, we made use of the 2010 Child Opportunity Index 2.0. The Child Opportunity Index (COI; Noelke et al., 2020) is a publicly available resource that combines 29 indicators of neighborhood opportunity (in this case, from the 2010 Census and the 2008–2012 American Community Survey data) across three domains: educational (e.g., quality of schools and social resources related to educational achievement), health/environmental (e.g., access to healthy food and greenspace, pollution and extreme heat), and socioeconomic (e.g., access to employment and neighborhood resources). Specific items in the COI are defined in https://www.diversitydatakids.org/sites/default/files/2020-02/ddk_coi2.0_technical_documentation_20200212.pdf (see their Table 2 for a summary reference). Internal consistency reliability for the COI was .85 across >72,000 census tracts nation-wide and .88 in the TBED-C. Nationally normed scores of each Census tract level were obtained using standard procedures (see https://data.diversitydatakids.org/dataset/coi20-child-opportunity-index-2-0-database). Specifically, all tracts nation-wide were ranked based on the COI overall score, and then converted to percentile groupings. The lowest 1% of tracts were assigned a score of 1, the next 1% were assigned a score of 2, and so on. These nationally normed percentiles groupings were applied to our data. Most families (57.9%) were the only participants in their Census tract. In the remainder, more than one participating family resided in a given tract, with 27.8%, 9.0%, 2.7%, 1.9%, and 0.6% for 2, 3, 4, 5, and 6 families sharing a given Census tract. Descriptive statistics for the COI are presented in Supplementary Table 1.

Table 2.

Fit Indices for the Etiologic Moderation of Positive Parenting by Neighborhood Opportunity Composite

Maternal Support variable	−2lnL	df	AIC	BIC	SABIC	DIC
Primary Analyses
Maternal support
Full ACE moderation	5429.77	1978	1473.77	−4110.93	−969.82	−2293.27
A moderation only	5429.85	1980	1469.85	−4117.80	−973.51	−2298.30
C moderation only	5433.46	1980	1473.46	−4115.99	−971.71	−2296.49
E moderation only	5436.64	1980	1476.64	−4114.40	−970.12	−2294.90
No moderation	5440.54	1981	1478.54	−4115.90	−970.03	−2295.48
Confirmatory analyses
Maternal Use of Praise and Explanation (coded from videotaped mother-child dyadic interactions)
Full ACE moderation	5001.17	1788	1425.17	−3595.53	−756.31	−1952.47
A moderation only	5003.39	1790	1423.39	−3601.24	−758.84	−1956.34
C moderation only	5006.72	1790	1426.72	−3599.58	−757.17	−1954.68
E moderation only	5003.76	1790	1423.76	−3601.06	−758.65	−1956.16
No moderation	5013.52	1791	1431.52	−3599.59	−755.59	−1953.77
Twin informant report of maternal support
Full ACE moderation	5484.68	1948	1588.68	−3971.10	−877.65	−2181.00
A moderation only	5485.25	1950	1585.25	−3977.70	−881.07	−2185.77
C moderation only	5488.02	1950	1588.02	−3976.31	−879.69	−2184.38
E moderation only	5486.28	1950	1586.28	−3977.18	−880.56	−2185.25
No moderation	5491.29	1951	1589.29	−3978.13	−879.92	−2185.28
Mother informant report of maternal support
Full ACE moderation	4651.63	1949	753.63	−4388.09	−1293.06	−2597.08
A moderation only	4655.03	1951	753.03	−4393.28	−1295.08	−2600.43
C moderation only	4652.84	1951	750.84	−4394.38	−1296.17	−2601.53
E moderation only	4654.23	1951	752.23	−4393.68	−1295.48	−2600.83
No moderation	4655.03	1952	751.03	−4396.73	−1296.93	−2602.96

Note. The best fitting model for a given set of analyses is highlighted in bond font and is indicated by the lowest AIC (Akaike’s Information Criterion), BIC (Bayesian Information Criterion), SABIC (sample size adjusted Bayesian Information Criterion), and DIC (Deviance Information Criterion) values for at least 3 of the 4 fit indices.

The protocol for the under-resourced arm of the TBED-C also included the recruitment and assessment of randomly chosen neighbors. Following the participation of a given family in the under-resourced arm of the study, we sent mailings to 10 randomly chosen addresses in that family’s Census tract, inviting one adult resident per household to complete a survey. When a particular address was no longer inhabited (i.e., the letter was returned as undeliverable), one attempt was made to find a replacement address. This approach resulted in a sample of 1,880 neighbors (63.2% women; 80.6% White, 11.6% Black, 7.8% other ethnic group memberships; average age of 52.6 with a range of 18–95 years). The response rate was 70%, of which 70% agreed to participate (for a final participation rate of 49%).

To determine average neighbor perceptions of neighborhood opportunity, we geocoded all neighbor and twin family addresses with a 99.9% success rate using an “.hmtl” code that uses Google Maps address data to assign coordinates. We then mapped the geocoded coordinates using ArcGIS v10.6 (ESRI, Redlands, CA). We verified the spatial accuracy of 20 random geocoded locations by comparing the tabular data to ensure that the assigned county and city names correspond with the Census tract found in the original dataset. Using the geocoded coordinates, we calculated averages within 5km of each twin family’s residential location using ArcMap software. Descriptive statistics for these various spatial covariates were then calculated using Stata v16 (College Station, TX). The mean number of neighbors living within 5 km (or 5000 m) of a given twin family was 13.09 (SD = 10.98), with a median of 10 and a range of 1 to 47. The mean distance to the nearest neighbor was 1437 m (SD = 1368 m), with a median of 855 m and a range of 0.25 m to 4992 m. Nearly all families across both sub-studies (N = 847) had at least one participating neighbor within 5 km.

Using these analytic procedures, we computed average neighbor perceptions of opportunity in their neighborhood using the Neighborhood Matters questionnaire (Henry et al., 2014). The Extent of Neighborhood Problems scale (α =.95) consists of 13 items assessing perceptions that graffiti, drugs, abandoned buildings, vandalism, etc., are a problem in their neighborhood (see supplemental materials for the item list). Items were rated on a 5 point scale. For the current study, this scale was reverse scored so that high levels indicated fewer neighborhood problems. The reverse scored 5km average indicator of neighborhood problems was correlated .51 with the COI. Descriptive statistics for the reverse-scored variable are presented in Supplementary Table 1.

Positive Parenting

Maternal support and involvement encompass a wide range of practices both within the home and within schools and communities. Here, our primary analyses focused on maternal investments in communication, closeness, and support in their relationship with their child as assessed via the parental involvement scale (12 items, detailed in the supplemental materials; e.g. “I praise my child when he/she does something well”; “My child talks about his/her concerns and experiences with me”) on the Parental Environment Questionnaire (PEQ; Elkins et al., 1997). Mothers individually rated their support and involvement with each of their participating twins, whereas twins individually rated their perceptions of support and involvement with their mother. Descriptive statistics for the mother and twin informant-reports are presented in Supplementary Table 1. Items were read to twins with reading levels under 5^th grade to assure comprehension. Each item was rated on a 4-point scale from definitely true to definitely false. The supportive parent-child relationship scale displayed good internal consistency reliability, with alphas between .68 and .87 across all individual informant-reports. Maternal and twin reports of mothering were each available for 98% twins. Consistent with prior work (Kobayashi et al., 2021), twin-reports were only modestly correlated with mother reports (r =.13, p < .01). To maintain consistency with prior work (Burt et al., 2005; Klahr et al., 2013), mother and twin perceptions were averaged to create a single composite measure of maternal support for our primary analyses. However, given the small association between them, we also analyzed each informant separately.

As a sensitivity analysis, we also evaluated coder ratings of videotaped parent-child interactions. Observer-ratings of parenting were obtained using 8-minute video-taped interactions of both mother-child dyads (mother-twin 1, mother-twin 2). Each mother-child dyad was asked to complete a mildly to moderately frustrating task (i.e., use an Etch-a Sketch to draw specific pictures, but parent and child may only use one dial each, thereby requiring cooperation; Deater-Deckard et al., 1997). All dyads completed the same task, although the specific pictures varied. The task was found to be a reliable and valid tool for assessing the parent-child relationship with school-age twins (Deater-Deckard et al., 1997).

Trained raters coded various aspects of parenting using the Twin Parent-Child Interaction System (PARCHISY; Deater-Deckard et al., 1997). Each observer received roughly 85 hours of training and was required to pass observation examinations before coding videotapes. Observers attended coder meetings for ongoing training and to prevent “drift.” Observer reliability was assessed by assigning 10% of all videos (N=200) to be rated by at least five observers, and then computing intraclass correlations across the ratings for each scale. To reduce rater bias, each mother-child dyad was coded by a research assistant who was blind to all participant data. Further, different coders rated each of the two mother-child dyads within a family, eliminating the possibility of shared method variance. Given our focus on protective aspects of parenting, the current study examined the Positive Control scale, which assesses the use of praise and explanation during the dyadic parent-child interaction. Inter-rater correlations for this scale were .88, and ratings were available for 1,856 twins. Descriptive statistics for the maternal explanations and praise variable are presented in Supplementary Table 1.

ANALYSES

Classical twin studies (see Neale & Cardon, 1992) leverage the difference in the proportion of genes shared between monozygotic or MZ twins (who share 100% of their genes) and dizygotic or DZ twins (who share an average of 50% of their segregating genes) to estimate additive genetic (in this case, evocative rGE; A), shared environmental (i.e., environmental factors that make twins similar to each other; C), and non-shared environmental (i.e., factors that make twins different from each other, including measurement error; E) contributions to a given phenotype. In the current study, we specifically fitted the ‘univariate GxE’ twin model (Purcell, 2002) to evaluate whether neighborhood opportunity moderated the etiology of maternal support (see Figure 1). Although prone to false positives when twin pairs are imperfectly correlated on the moderator (van der Sluis et al., 2012), Purcell’s univariate GxE model has been shown to be the most appropriate such model when the twins are perfectly concordant on the moderator (van der Sluis et al., 2012), as is the case here since our participants are children who reside in the same house as their co-twins. To circumvent the possibility that hidden rGE is masquerading as GxE in our analyses (in which genetic effects overlap across the moderator and the outcome), the moderator values were entered in a means model of each twin’s parenting data. Moderation was then modeled on the residual maternal support variance (i.e., that which does not overlap with neighborhood opportunity). In the first and least restrictive model, linear moderators (i.e., A₁, C₁, E₁) were added to the genetic and environmental paths (i.e., a, c, e; akin to intercepts) using the following equation: Unstandardized Variance_Total = (a + A₁(neighborhood opportunity))² + (c + C₁(neighborhood opportunity))² + (e + E₁(neighborhood opportunity))². We then fitted a series of more restrictive moderator models, constraining the moderators to be zero and evaluating reductions in model fit.

Figure 1.

Path diagram of the univariate GxE twin model.

Note. A, C, and E represent genetic, shared environmental, and non-shared environmental influences, respectively, on maternal support. For ease of presentation, the co-twin variables and paths are omitted here, though they are estimated in the models. The variance decomposition of maternal support is modeled as a function of neighborhood opportunity (the moderator, M). The moderator values were entered in a means model of maternal support. Linear moderation was then modeled on the residual maternal support variance (i.e., that which does not overlap with the moderator). The genetic and environmental paths function similar to intercepts in these models, in that they estimate the genetic and environmental variance in maternal support at the lowest level of neighborhood opportunity. The interaction terms (i.e., β_xM, β_YM, and β_ZM for a, c, and e paths, respectively) are added to the genetic and environmental paths, and are estimated separately for each component of variance.

To evaluate our statistical power to detect etiologic moderators of varying magnitude, Burt et al (2019) simulated data containing etiologic moderators ranging from small (0.2) to moderate (0.4) to large (0.6). Power analyses (reproduced from Burt et al., 2020 and plotted in Supplementary Figure 1) indicated that, in a sample of 1,001 twin pairs, we were sufficiently powered to detect small non-shared environmental moderators and moderate genetic moderators. However, we were underpowered to detect shared environmental moderators until they were large in magnitude (i.e., we had 80.5% power to detect C moderators of 0.6).

Mx (Neale et al., 2003) was used to fit the GxE models to the data using Full-Information Maximum-Likelihood techniques. When fitting models to raw data, variances, covariances, and means are first freely estimated to get a baseline index of fit (minus twice the log-likelihood; −2lnL). Model fit was evaluated using four information theoretic indices that balance overall fit with model parsimony: the Akaike’s Information Criterion (AIC; Akaike, 1987), the Bayesian Information Criteria (BIC; Raftery, 1995), the sample-size adjusted Bayesian Information Criterion (SABIC; Sclove, 1987), and the Deviance Information Criterion (DIC; Spiegelhalter et al., 2002). The lowest or most negative AIC, BIC, SABIC, and DIC among a series of nested models is considered best. Because fit indices do not always agree (e.g., they place different values on parsimony, among other things), we reasoned that the best fitting model should yield lower or more negative values for at least 3 of the 4 fit indices. When two models both provided a reasonable fit to the data (i.e., each fit best by 2 of 4 fit indices), their respective −2lnL were compared to determine the better fitting model (−2lnL differences are chi-square distributed). A statistically significant chi-square value would indicate that the less restrictive of the two models provided a better to the data.

Because non-normal distributions in the outcome variable can lead to spurious findings, we evaluated our positive parenting variables for evidence of skew. Although the informant-report composite of maternal support was determined to be normally distributed (skew −0.70), the praise and explanation variable was not (skew was −1.15). The latter was thus transformed using log₁₀ prior to analysis (skew following transformation was −0.12). Twin sex, age, and race/ethnicity were regressed out of the positive parenting variables, in keeping with prior recommendations (McGue & Bouchard, 1984). Although the interpretation of standardized or proportional ACE estimates may be useful in some cases, it is generally recommended that unstandardized or absolute ACE estimates be presented (Purcell, 2002). We thus standardized all positive parenting variables to have a mean of zero and a standard deviation of one to facilitate interpretation of the unstandardized values.

To ensure that our findings reflected neighborhood opportunity per se, rather than family income, we regressed family income out of both the COI and reverse-scored neighborhood problems variables (their r’s were .35 and .31, respectively). We also regressed population density from our neighborhood opportunity variables, as urban areas tended to have more opportunities than rural areas (their r’s were −.27 and −.42, respectively). We then saved the residuals for analysis, flooring each moderator at 0 and dividing by its maximum to provide a continuous measure of neighborhood opportunity that ranged from 0 to 1. Given the correlation between the two indices of neighborhood opportunity, we also created a composite of the two measures of opportunity by averaging them together. If one was missing, the other was score was used in place of the average. This composite index of neighborhood opportunity was also floored to have a range of 0 to 1. Our primary analyses focused on the neighborhood opportunity composite, with confirmatory analyses focusing on the individual COI and reverse-scored neighbor informant report variables.

In addition to standard plots of unstandardized heritability estimates at each level of the moderator, we also plotted the results for maternal support by location across the state of Michigan, employing recent modeling innovations (Shero et al., in press). Specifically, we applied a twin-based spatial weighting procedure (Davis et al., 2012; Reed et al., 2022) to compute the prior referenced best-fitting models at a variety of target locations throughout Michigan. This approach works by first establishing a variety of target locations throughout the state at which our models were to be tested, followed by computing distances between each twin pair’s geographic location (jittered to protect privacy) and the target locations. The inverse of these distances were then applied as weights, such that twin pairs that live in closer proximity to a given target location received more weight for that target location’s estimate. With those weights applied, separate MZ and DZ weighted covariance matrices were then computed for each target location, which were then applied to the models tested earlier. The results of this analysis supplied us with location-specific effect estimates, which were then mapped to their jittered geographic locations.

RESULTS

DESCRIPTIVES

As already demonstrated in Burt et al. (2023), neighborhood opportunity evidenced small but significant associations with maternal support (rs were .07 for neighbor informant-reports, .12 for the full COI, and .11 for the opportunity composite; all p<.05). Although these associations are quite small, it is important to note that the magnitude of the phenotypic correlation between a moderator and an outcome has no bearing on the presence or magnitude of etiologic moderation (Purcell, 2002), as moderation is modeled only on that variance that does not overlap across the outcome and the moderator.

Intraclass correlations are presented in Table 1, separately by zygosity and level of neighborhood opportunity (dichotomized at the median). In neighborhoods with less opportunity, MZ intraclass correlation was larger than DZ intraclass correlation, pointing to the possible presence of evocative rGE on maternal support in those contexts. In neighborhoods with more opportunity, however, the MZ correlation was equivalent in magnitude to the DZ correlation, suggesting an absence of children’s evocative rGE on maternal support in these contexts.

Table 1.

Sample sizes and intraclass correlations for maternal support by neighborhood opportunity

Number of twin families			Less Opportunity (below the median)		More Opportunity (above the median)
Total	MZ	DZ	MZ	DZ	MZ	DZ
1,013	418	595	.51*	.36*	.42	.41

Note. For these preliminary analyses, neighborhood opportunity was dichotomized at the median.

^*indicates that the monozygotic (MZ) correlation is significantly larger than the dizygotic (DZ) correlation at p<.05.

PRIMARY GXE ANALYSES

Formal tests of moderation were conducted next. Fit indices are reported in Table 2, and model fit results are presented in Table 3. As seen there, there was clear evidence that neighborhood opportunity moderated the magnitude of children’s genetic influences on the maternal support they received. The A only moderation model emerged as the best fitting model according to all 4 fit indices. The estimated A moderator was large in magnitude (i.e., −.66 in the full ACE moderation model and −.68 in the A-only moderation model). When the parameter estimates for maternal support were plotted (see Figure 2), children’s genetic influences were observed to contribute far more to the level of maternal support they experienced when they resided in neighborhoods with fewer opportunities. However, these influences dissipated as neighborhood opportunities increased, such that children’s genetic influences on the maternal support they received were negligible in neighborhoods with ample opportunity. By contrast, the C and E moderators were very small in magnitude and could be dropped from the best-fitting models. That said, the shared and non-shared environmental paths (akin to intercepts) were estimated to be moderately-to-large in magnitude and significantly larger than zero, collectively suggesting that environmental influences were important to the etiology of maternal support regardless of the amount of opportunity in the neighborhood.

Table 3.

Unstandardized parameter estimates for the moderation models of positive parenting by neighborhood opportunity

	PATHS			LINEAR MODERATORS
	a	c	e	A1	C1	E1
Full ACE moderation model
Maternal support	0.85* (0.23, 1.25)	0.55* (0.12, 0.92)	0.70* (0.54, 0.88)	−0.66 (−1.35, 0.21)	−0.05 (−0.65, 0.52)	0.02 (−0.24, 0.26)
Maternal use of praise	0.58 (−1.02, 1.02)	0.48 (−0.85, 0.85)	0.95* (0.76, 1.14)	−0.33 (−0.95, 0.95)	−0.18 (−0.76, 0.58)	−0.20 (−0.47, 0.08)
Twin informant-report	0.65 (−1.04, 1.04)	0.32 (−0.77, 0.77)	0.90* (0.72, 1.09)	−0.35 (−1.06, 1.06)	−0.01 (−0.81, 0.81)	−0.11 (−0.39, 0.17)
Mother informant-report	0.45 (−1.20, 1.20)	0.94* (0.73, 1.47)	0.37* (0.26, 0.58)	−0.13 (−2.18, 2.18)	−0.20 (−1.04, 0.15)	0.09 (−0.23, 0.33)
Best-fitting moderation model
Maternal support	0.86* (0.59, 1.12)	0.52* (0.35, 0.63)	0.71* (0.67, 0.76)	−0.68* (−1.28, −0.26)	--	--
Maternal use of praise	0.84* (0.53, 1.09)	0.30 (−0.47, 0.47)	0.82* (0.77, 0.87)	−0.62* (−1.07, −0.25)	--	--
Twin informant-report	0.75* (0.43, 0.99)	0.29 (−0.48, 0.48)	0.83* (0.78, 0.88)	−0.48* (−1.05, −0.10)	--	--
Mother informant-report	0.38* (0.26, 0.47)	0.82* (0.76, 0.88)	0.43* (0.40, 0.46)	--	--	--

Note. A, C, and E (upper and lower case) respectively represent genetic, shared, and non-shared environmental influences on positive parenting. Because the lowest level of the neighborhood opportunity was coded as 0, genetic and environmental contributions to positive parenting at this level can be obtained by squaring the path estimates (i.e., a, c, and e). At higher levels, linear moderators (i.e., A₁, C₁, E₁) were added to the paths as follows: Unstandardized Variance_Total = (a + A₁(neighborhood opportunity))² + (c + C₁(neighborhood opportunity))² + (e + E₁(neighborhood opportunity))². An asterisk indicate that the parameter estimate was significant at p<.05.

Figure 2.

Changes in the etiology of maternal support with levels of neighborhood opportunity for the best-fitting model.

We also plotted the genetic moderation results by location across the state of Michigan (see Figure 3). Genetic moderator values tended to be relatively small in the (often wealthy) suburbs of Detroit, but larger in the city of Detroit itself. We also observed very large moderator values in the middle of the state, in and around the Lansing area, and large moderators on the Western side of the state. The Western part of the state is generally rural, although we note that the city of Grand Rapids and its suburbs are also included. These values correspond to the level of neighborhood opportunity in various locations across the state of Michigan.

Figure 3.

Distribution and histogram of the A moderator estimates for maternal support by neighborhood opportunity in locations across Michigan (jittered to protect privacy).

CONFIRMATORY ANALYSES

We conducted three sets of confirmatory analyses. First, we examined whether these results persisted to a separate and presumably more objective measure of protective mothering – maternal use of explanation and praise during an 8-minute recording of each mother-child dyad. Fit indices are reported in Table 2, and model fit results are presented in Table 3 and plotted in Supplementary Figure 2. This was an important within study replication, both in terms of gauging the specificity of the above findings to a particular aspect of protective mothering but also because this particular variable also demonstrated consistent associations with neighborhood deprivation in Burt et al. (2023). As above, the A only moderation model emerged as the best fitting model according to all 4 fit indices. The estimated A moderator was moderate-to-large in magnitude (i.e., −.33 in the full ACE moderation model and −.62 in the A-only moderation model). When the parameter estimates for maternal explanation and praise were plotted (see Supplementary Figure 2), children’s genetic influences were again observed to contribute far more to the origins of maternal use of explanation and praise in neighborhoods with fewer opportunities.

Second, we evaluated whether and how neighborhood opportunity moderated the etiology of twin and mother informant-reports of maternal support, respectively. As seen in Tables 2 and 3, there was clear evidence of negatively signed genetic moderation when evaluating twin informant reports. The A only moderation model again emerged as the best fitting model, and the estimated A moderator was moderate-to-large in magnitude (i.e., −.35 in the full ACE moderation model and −.48 in the A-only moderation model). Children’s genetic influences on twin reports of the maternal support they received thus appeared to be accentuated in neighborhoods with little opportunity, and decreased as those opportunities increased. By contrast, maternal informant reports of her support yielded no evidence of moderation. The best-fitting model by 3 of the 4 fit indices was the no moderation model, indicating no evidence of significant etiologic moderation of maternal perceptions of support by neighborhood opportunity.

As a final check on our results, we sought to constructively replicate our primary findings for maternal support across each of the two individual indices of neighborhood opportunity. Results are presented in Tables 4 and 5. As seen there, the pattern of results were quite similar to those reported above. For neighbor informant reports of opportunity, the best-fitting model across all four fit indices was the A-only moderation model. The genetic moderator was very large (−1.00 and lower) and statistically significant in both the A moderation and full ACE moderation models. For the COI, the best-fitting model according to 3 of the 4 fit indices was the A-only moderation model. And although the A moderator in the best-fitting model was consistent with those reported for the opportunity composite and the neighbor informant reports of opportunity, it was equivalent in magnitude to the C moderator in the full ACE moderation model. As a check on these somewhat unexpected findings, we repeated these analyses using the maternal use of explanation and praise variable. Results are presented in Tables 4 and 5. For both the COI and the neighbor informant reports of opportunity moderators, the A only moderation model emerged as the best fitting model, and the estimated A moderator was moderate-to-large in magnitude in both the full ACE moderation and the A-only moderation models (i.e., ranging from −.52 to −.90). We thus conclude that our primary findings were also largely robust the specific operationalization of neighborhood opportunity.

Table 4.

Fit indices for the etiologic moderation of positive parenting by specific indices of neighborhood opportunity

Positive Parenting variable	−2lnL	df	AIC	BIC	SABIC	DIC
Maternal Support
Moderation by neighbor informant-reports of neighborhood problems (reverse-scored)
Full ACE moderation	4430.75	1622	1186.75	−3221.91	−646.49	−1731.39
A moderation only	4432.70	1624	1184.70	−3227.64	−649.04	−1735.28
C moderation only	4435.19	1624	1187.19	−3226.39	−647.80	−1734.03
E moderation only	4442.24	1624	1194.24	−3222.86	−644.27	−1730.51
No moderation	4443.80	1625	1193.80	−3225.44	−645.26	−1732.16
Moderation by COI score
Full ACE moderation	5304.91	1928	1448.91	−3976.26	−914.61	−2204.55
A moderation only	5305.60	1930	1445.60	−3982.80	−917.96	−2209.24
C moderation only	5306.69	1930	1446.69	−3982.25	−917.42	−2208.70
E moderation only	5307.71	1930	1447.71	−3981.74	−916.91	−2208.19
No moderation	5311.41	1931	1449.41	−3983.33	−916.91	−2208.86
Maternal Use of Praise and Explanation
Moderation by neighbor informant-reports of neighborhood problems (reverse-scored)
Full ACE moderation	4127.42	1454	1219.42	−2746.17	−437.67	−1410.03
A moderation only	4130.15	1456	1218.15	−2751.42	−439.75	−1413.45
C moderation only	4134.66	1456	1222.66	−2749.17	−437.49	−1411.19
E moderation only	4131.67	1456	1219.67	−2750.66	−438.98	−1412.69
No moderation	4144.73	1457	1230.73	−2747.44	−434.18	−1408.55
Moderation by COI score
Full ACE moderation	4871.03	1744	1383.03	−3489.36	−720.05	−1886.74
A moderation only	4872.07	1746	1380.07	−3495.64	−723.15	−1891.17
C moderation only	4872.74	1746	1380.74	−3495.30	−722.82	−1890.84
E moderation only	4873.96	1746	1381.96	−3494.69	−722.21	−1890.23
No moderation	4878.04	1747	1384.04	−3496.05	−721.97	−1890.66

Note. The best fitting model for a given set of analyses is highlighted in bond font and is indicated by the lowest AIC (Akaike’s Information Criterion), BIC (Bayesian Information Criterion), SABIC (sample size adjusted Bayesian Information Criterion), and DIC (Deviance Information Criterion) values for at least 3 of the 4 fit indices.

Table 5.

Unstandardized parameter estimates for the moderation of positive parenting by specific indices of neighborhood opportunity

	PATHS			LINEAR MODERATORS
	a	c	e	A1	C1	E1
Full ACE moderation model
Moderation by neighbor informant-reports of neighborhood problems (reverse-scored)
Maternal support	1.08* (0.39, 1.52)	0.61* (0.07, 1.04)	0.59* (0.41, 0.81)	−1.14* (−1.91, −0.12)	−0.11 (−0.74, 0.61)	0.20 (−0.10, 0.49)
Maternal use of praise	0.77 (−1.29, 1.29)	0.55 (−1.08, 1.08)	1.02* (0.77, 1.29)	−0.52 (−1.27, 1.27)	−0.30 (−1.09, 1.09)	−0.30 (−0.69, 0.08)
Moderation by COI score
Maternal support	0.56 (−0.02, 0.92)	0.60* (0.24, 0.87)	0.76* (0.63, 0.89)	−0.18 (−0.96, 0.62)	−0.18 (−0.75, 0.39)	−0.08 (−0.31, 0.15)
Maternal use of praise	0.81 (−1.06, 1.06)	0.06 (−0.87, 0.87)	0.83* (0.70, 1.01)	−0.80 (−1.53, 1.53)	0.48 (−1.58, 1.58)	−0.01 (−0.32, 0.23)
A-only moderation model
Moderation by neighbor informant-reports of neighborhood problems (reverse-scored)
Maternal support	0.98* (0.61, 1.37)	0.55* (0.40, 0.65)	0.73* (0.68, 0.77)	−1.00* (−1.82, −0.39)	--	--
Maternal use of praise	1.09* (0.75, 1.40)	0.25 (−0.47, 0.47)	0.82* (0.76, 0.88)	−0.90* (−1.45, −0.43)	--	--
Moderation by COI score
Maternal support	0.69* (0.43, 0.89)	0.51* (0.32, 0.63)	0.71* (0.67, 0.76)	−0.44* (−0.97, −0.08)	--	--
Maternal use of praise	0.65* (0.35, 0.95)	0.37 (−0.51, 0.51)	0.83* (0.77, 0.89)	−0.58* (−1.59, −0.09)	--	--

Note. A, C, and E (upper and lower case) respectively represent genetic, shared, and non-shared environmental parameters on positive parenting. Because the lowest level of the neighborhood opportunity was coded as 0, the genetic and environmental contributions to positive parenting at this level can be obtained by squaring the path estimates (i.e., a, c, and e). At higher levels, linear moderators (i.e., A₁, C₁, E₁) were added to the paths using the following equation: Unstandardized Variance_Total = (a + A₁(neighborhood opportunity))² + (c + C₁(neighborhood opportunity))² + (e + E₁(neighborhood opportunity))². Bold font and an asterisk indicate that that parameter estimate was significant at p<.05.

DISCUSSION

The goal of the present study was to evaluate whether the origins of positive parenting varied with the level of neighborhood opportunity. Results pointed to robust evidence that the contributions of children’s genetic influences on the parenting they received varied dramatically across neighborhood context. Maternal support was primarily a function of evocative rGE processes in disadvantaged neighborhoods with little opportunity, but was primarily environmental in origin in neighborhoods with high levels of opportunity. These findings persisted across coder ratings of maternal use of explanation and praise during short interactions, various operationalizations of neighborhood opportunity (including neighbor informant reports, a Census tract composite variable, and a composite of the two), and twin informant-reports of maternal support, although they did not persist to maternal informant-reports of her support.

Such findings significantly extend our understanding of the origins of positive mothering behavior, and the links between mothering and neighborhood context in particular. Burt et al. (2023) and Cuellar et al. (2015) both concluded that neighborhood context was negatively associated with positive parenting behaviors, but were not able to evaluate whether the origins of mothering might vary also across contexts. Separately, a prior meta-analysis of child-based twin and adoption studies of parenting behavior (Klahr & Burt, 2014) demonstrated that genetic influences at the level of the child make significant and moderate contributions (35–51%) to the mothering they receive, but did not examine whether these estimates varied across environmental context. The current study advances both of these prior conclusions, indicating that child effects/evocative rGE were a primary driver of positive parenting in disadvantaged neighborhoods with little opportunity, but not in neighborhoods with high levels of opportunity, where environmental and/or cultural effects exerted the largest effects.

The current study benefited from a number of strengths that increase confidence in our results, including the use of a strong sampling frame (birth records) enriched for exposure to neighborhood disadvantage and our extensive and state-of the-science measurements of neighborhood and parenting. Despite these strengths, several limitations are noted. First, the cross-sectional and non-experimental nature of our study design, combined with the absence of an independent replication sample, limits the strength of our causal inferences. Parents select and shape their social surroundings and are also shaped by them, an interactive process that belies strong assumptions regarding directionality when evaluating neighborhood effects. Future work should seek to replicate our findings in longitudinal data, and to extend our understanding of causality in the association between parenting and neighborhood opportunity via experimental studies.

Second, maternal and child perceptions of maternal support demonstrated only modest associations with one another (r = .13). We strongly suspect that the small size of this correlation (and perhaps the lack of GxE replication for maternal informant-reports) stems in part from the clear variance range restriction observed for maternal informant-reports, which is well known to suppress correlations. As shown in Supplementary Table 1, observed twin reports of maternal support ranged from 16 to 48 (out of a possible range of 12 to 48), with a mean of 39.8 and a standard deviation (SD) of 5.78. By contrast, maternal reports ranged from 30 to 48, with a mean of 43.2 and a SD of 3.06. Put another way, the maternal SD was only 52.9% that of the twins. This stands in stark contrast to the similar SD’s observed for the other PEQ scale (parent-child conflict), which saw SDs of 5.92 for mother informant-reports and 6.03 for twin informant-reports. We thus suspect that maternal perceptions of the support they provide may be affected by social desirability bias. That said, there does appear to be some meaningful signal there as well. Partial correlation analyses indicated that both mother and twin informant-reports incremented the other in the prediction of the COI and reverse-scored neighbor informant reports of problems (partial correlations ranged from .05 to .10, all p<.05), and in the prediction of Teacher Report Form (TRF) teacher reports of DSM-oriented child conduct problems (partial correlations for each informant, controlling for the other informant, were estimated at −.09, both p<.01). We thus created a composite and examined each informant individually.

Next, our results should be considered specific to middle childhood, and may not extend to other developmental stages, including adolescence or toddlerhood. This is potentially important since parenting is known to manifest quite differently across child development. As one example, maternal warmth has been shown to decrease as the child transitions to adolescence, whereas parent-child conflict increases (Paikoff & Brooks-Gunn, 1991; Steinberg, 1987). Another limitation centers on the fact that the children in our study were twins. Although twins are representative of singleton populations on most behavioral and relational traits (e.g., Christensen & McGue, 2020), parenting two children exactly the same age may well impose additional hurdles for mothers that influence the support they provide. It remains unclear, however, whether or how this may have influenced our findings.

Relatedly, because our sample primarily identified as white non-Hispanic, we controlled for parental race and ethnicity in our analyses rather than conducting analyses separately for those with specific marginalized identities. Although necessary to assure sufficient statistical power, this approach is less than ideal for (at least) three reasons. First, parenting behaviors and their outcomes vary to some extent by race and ethnicity. For example, parents that identify as Native American may be more likely than parents from other racial and ethnic groups in the United States to emphasize interdependence over independence (MacPhee et al., 1996), while parents who identify as Black have been shown to engage in higher levels of parental control (Richman & Mandara, 2013), with more positive consequences for their children, than parents who identify as white (Dunbar et al., 2017). Second, none of the parenting measures examined here have been explicitly validated for use with participants from marginalized populations. It is thus unknown whether these parenting measures function as intended in racialized populations. Third, neighborhood opportunity is not randomly distributed across neighborhoods, and may be a direct legacy of historical structural racism (e.g., redlining, blockbusting, placement of freeway construction) (Nardone et al., 2020). The modern-day legacies of these racist practices from the past are many and include less neighborhood greenspace (Nardone et al., 2021) and higher vacancy (Sadler & Lafreniere, 2016), among others. Given all this, the current findings may not generalize to parenting in families with racialized identities. Future research should explicitly evaluate this important issue.

IMPLICATIONS

Our results indicated that positive parenting was primarily evocative rGE in origin in neighborhoods with little opportunity, an effect that persisted over and above family income and urbanicity. Such findings suggest an especially important role for child effects on maternal support in under-resourced contexts. In neighborhoods with high levels of opportunity, by contrast, maternal support was primarily environmental in origin, indicating a far more limited role for child effects and a more central role for broader environmental effects in advantaged contexts. Such findings have several implications. First, they indicate that, rather than speculating as to the determinants of parenting behavior in general, theory regarding the origins of parenting behavior should instead conceptualize the origins of parenting behavior as context-bound, or at the very least, consider how the various elements of their general theory might be differentially important across different neighborhood contexts.

Building on this possibility, we found that mothers in impoverished neighborhood contexts were more responsive to their children’s genetically influenced characteristics than were mothers in wealthier areas with more opportunity. Such findings push against common stereotypes of mothers in impoverished areas, and indicate that, although they may be slightly less supportive on average (Burt et al., 2023; Cuellar et al., 2015), their support is far more likely to be elicited or evoked by their child’s genetically influenced characteristics. This may reflect targeted efforts on the part of those mothers to bolster their children in the face of limited opportunities available in their neighborhood context. Alternately, it may reflect the fact that mothers under stress may need to conserve their emotional resources, and thus are more likely provide supportive parenting only when elicited by the child. By contrast, mothers in neighborhoods with ample resources appear to provide positive parenting regardless of the child’s behavior. Such findings could indicate that sociocultural pressures in favor of high maternal nurturance and support are particularly strong in advantaged neighborhoods with lots of opportunity, or that it is simply easier to be unconditionally supportive and nurturing with your children when resources are plentiful. In either case, mothers in advantaged contexts appear to engage in slightly higher levels of positive parenting but they do so in a less child-driven way (i.e., regardless of whether or not their child is eliciting that type of parenting). Future work should seek both to identify the specific child characteristics that evoke positive parenting in under-resourced neighborhoods and to examine the consequences of these differentially responsive maternal styles for children’s outcomes.

Finally, the finding that evocative rGE effects vary with the broader neighborhood context has profound implications for our understanding of genotype-environment interplay. Prior theory regarding GxE and rGE has by and large considered rGE and GxE as two separate and independent forms of interplay, in keeping with early equations (Plomin et al., 1977). Subsequent work (Purcell, 2002) further suggested that rGE can masquerade as GxE, and that rGE and GxE can jointly underlie the links between a given environmental experience and an outcome. We know of no theoretical or empirical study, however, to evaluate whether rGE are themselves moderated by the broader environmental context. The current study found compelling evidence of just this sort of process, and in doing so, significantly advances our understanding of genotype-environment correlations as dynamic and context-dependent. Future studies of rGE should be mindful of the fact that their results may well be context specific.

Public Significance Statement:

Our findings confirm the long-running notion that children are active participants in the parenting they receive. However, we find that child effects on positive parenting are also context-dependent, such that child effects on positive parenting are especially pronounced in under-resourced neighborhood contexts.