Child Development, July/August 2021, Volume 92, Number 4, Pages e457–e475

Capturing Environmental Dimensions of Adversity and Resources in the

Context of Poverty Across Infancy Through Early Adolescence: A Moderated
Nonlinear Factor Model
Meriah L. DeJoseph , and Robin D. Sifre C. Cybele Raver
University of Minnesota New York University

Clancy B. Blair Daniel Berry

New York University and New York University School of University of Minnesota
Medicine

Income, education, and cumulative-risk indices likely obscure meaningful heterogeneity in the mechanisms
through which poverty impacts child outcomes. This study draws from contemporary theory to specify multi-
ple dimensions of poverty-related adversity and resources, with the aim of better capturing these nuances.
Using data from the Family Life Project (N = 1,292), we leveraged moderated nonlinear factor analysis (Bauer,
2017) to establish group- and longitudinally invariant environmental measures from infancy to early adoles-
cence. Results indicated three latent factors—material deprivation, psychosocial threat, and sociocognitive
resources—were distinct from each other and from family income. Each was largely invariant across site,
racial group, and development and showed convergent and discriminant relations with age-twelve criterion
measures. Implications for ensuring socioculturally valid measurements of poverty are discussed.

Children growing up in poverty have a greater like- However, there is growing concern that these broad
lihood of experiencing a litany of adverse exposures indicators likely obscure meaningful heterogeneity
ranging from resource scarcity to domestic and in mechanisms by which poverty impacts child out-
neighborhood violence (Evans, 2004). While many comes (Duncan & Magnuson, 2003). This has
children growing up with material and psychoso- underscored the need for measures that capture
cial disadvantage flourish, others are placed at risk individual differences in experiences in the context
for early socioemotional and cognitive difficulties of poverty that account for the wide range in sever-
(Blair & Raver, 2012; Yoshikawa, Aber, & Beardslee, ity, timing, and type of exposure that may be driv-
2012). Identifying the mechanisms underlying these ing its impacts.
associations is central to strengthening community, An equally pressing concern is that these com-
school, and family supports for resilience and for mon indicators of poverty may not have the same
developing targeted intervention and policy efforts. meaning across time or across racial-ethnic groups,
Historically, income, maternal education, and leading to empirical imprecision and conclusions
other sociodemographic indicators have been that do not adequately characterize the experience
widely used as proxies of poverty-related adversity. of any sociocultural group (Borsboom, 2006). This
consideration is especially important given that
children and families of color are disproportionately
This research was supported as part of the National Institutes
of Health (NIH) grant R01HD081252 and P01 HD039667-01A1. represented among the poor due to pervasive struc-
We thank the many families and research assistants for making tural racial and social inequalities (Jiang & Koball,
this study possible. We also thank the members of the Bioecol- 2016; Raver & Blair, 2020). The intersections of race
ogy, Self-Regulation, and Learning Lab and the Cognitive Devel-
opment and Neuroimaging Lab for their thoughtful feedback on and class can be seen in the vast income and wealth
prior drafts. Meriah L. DeJoseph was supported by the NASEM racial disparities that exist: even when controlling
Ford Foundation Predoctoral Fellowship. Robin D. Sifre was sup- for education level, White household income is
ported by the National Science Foundation Graduate Research
Fellowship.
Correspondence concerning this article should be addressed to
Meriah L. DeJoseph, University of Minnesota, 51 East River © 2021 Society for Research in Child Development
Road, Minneapolis, MN 55455. Electronic mail may be sent to All rights reserved. 0009-3920/2021/9204-0034
dejos002@umn.edu. DOI: 10.1111/cdev.13504
e458 DeJoseph, Sifre, Raver, Blair, and Berry

twice the size and net worth is 13 times the size of Capturing the range of exposures associated with
African American households in the United States childhood poverty is a significant challenge, but
(Parker, Horowitz, & Mahl, 2016). Because tradi- clarifying certain domains of experience has pro-
tional poverty indicators like income are so inter- mise to yield significant empirical benefits. To
twined with these racial disparities, more direct illustrate the value of considering multiple poverty-
measures capturing resource scarcity and other relevant contextual factors as predictors of specific
proximal stressors may be more useful in its ability developmental adaptations, we draw from the
to represent the mechanistic pathway by which dimensional model of adversity and psychopathol-
low-income impacts child outcomes (Gershoff, ogy (DMAP; McLaughlin & Sheridan, 2016). DMAP
Aber, Raver, & Lennon, 2007; Parker et al., 2016). builds largely from adversity research over the last
Indeed, among more racially and ethnically diverse 20 years, which incorporates a wide range of scien-
samples, measured indicators of parents’ subjective tific methods such as observational human and
experience of hardship have been shown to make experimental animal neuroscientific models.
up to a fourfold increase in explained variance in Through those new methods, we have come to bet-
predicting family level outcomes compared to ter understand that deprivation (operationalized as
household income alone (Hurwich-Reiss, Wata- insufficient resources) and being exposed to condi-
mura, & Raver, 2019). tions that are threatening have both distinct biologi-
Reducing racial, social, and developmental bias cal and neurocognitive sequelae that allow for
in our measures is a major goal of developmental adaptation to a given set of environmental
science. Towards this end, rigorous methods testing demands. For example, studies of sensory depriva-
and adjusting for possible bias should be consid- tion (e.g., dark rearing; Wiesel & Hubel, 1965) and
ered alongside substantively relevant developmen- threat learning (e.g., repetitive foot shock; Raineki,
tal questions among diverse populations. In this Cort
es, Belnoue, & Sullivan, 2012) demonstrate
study, we draw from contemporary theoretical unique associations with cognitive neural develop-
models of adversity and leverage recent advances ment (Huttenlocher, de Courten, Garey, & Van der
in latent variable modeling to develop more Loos, 1982) and stress responsivity outcomes,
nuanced representations of the constraints and sup- respectively (Brunson, Eghbal-Ahmadi, Bender,
ports experienced by families struggling with eco- Chen, & Baram, 2001). Evidence from humans also
nomic adversity. suggests that deprivation and threat may exert dis-
tinct effects on neurodevelopment, which help to
provide malleable targets for intervention and pre-
Beyond Income: Capturing Environmental Heterogeneity
vention (e.g., McLaughlin et al., 2016; Rosen, Sheri-
in the Context of Poverty
dan, Sambrook, Meltzoff, & McLaughlin, 2018).
Two differing theoretical and empirical frame- Although many DMAP studies have used low-in-
works guide our approach to measuring dimen- come as an index of deprivation, studies leveraging
sions of poverty-related adversity and resources in large data sets are increasingly taking a more indi-
this study. First, by focusing on families’ resources, vidual difference approach that considers home,
we adopt an integrative, strengths-based approach school, and neighborhood contexts of poverty in
by examining the considerable heterogeneity in chil- their measurement (Ellwood-Lowe, Whitfield-Gab-
dren’s experiences that exist above and beyond rieli, & Bunge, 2020; Rosen, Amso, & McLaughlin,
their place on the income distribution (Frankenhuis, 2019). While there are many poverty-related envi-
Young, & Ellis, 2020; Hostinar & Miller, 2019). Such ronmental dimensions that could be examined, we
an approach highlights experiences in the context of focus this study on three representative constructs
poverty that are both more empirically precise and inspired by the DMAP—namely, material depriva-
less stigmatizing or marginalizing (Shafir, 2017; tion, sociocognitive resources, and psychosocial
Syed, Santos, Yoo, & Juang, 2018). Many families threat.
struggling with the uncertainty and strain of not
having enough money to meet material needs are
Measurement Invariance: Making Apples-to-Apples
nonetheless able to create environments that are
Comparisons
emotionally secure, affirming, and safe (Franken-
huis & Nettle, 2019). Clarifying these individual Critically, the validity and ultimate utility of
and family differences is central to understanding measures tapping environmental exposures hinge
both risk and resilience in children’s development on the extent to which these measures reflect com-
across the socioeconomic spectrum. mon substantive and quantitative scales across
 Capturing Enviornmental Dimensions e459

groups and developmental time—broadly termed, maintain a common latent scale across groups (i.e.,
measurement invariance (MI; Meredith, 1993; Wida- apples to apples comparison), while also accounting
man, Ferrer, & Conger, 2010). When the underlying at least partially for group differences in: (a) the
meaning or scale of a measured construct differs extent to which an item is representative of higher
across groups or development—that is, when a versus lower levels of the latent construct (i.e., item
measure is noninvariant—it undermines one’s abil- difficulty), (b) the magnitude of the linear (continu-
ity to interpret group or developmental differences ous item) or nonlinear (categorical item) relation
because measurement differences are inherently between the latent construct and the item (i.e., item
confounded with substantive differences (Flake & discrimination), and (c) the stochastic properties of
Fried, 2019). In simple terms, it becomes unclear both the construct and the items (see Table S1).
whether one is comparing apples to apples or to However, in practical terms, the MGA approach
oranges. quickly becomes intractable as the number of
At the item level, noninvariance is often referred groups increases. MGA is also difficult to apply to
to as differential item functioning (DIF), and there continuous moderators—like age—without devising
are times when we might well expect DIF across often arbitrary age-groupings. Traditional MIMIC
groups or development. For instance, a scale item models are useful for such continuous moderators,
for sociocognitive resources might include and yet typically limit one’s invariance adjustment to
item like, “has a toy appropriate for grasping or item difficulty (see Bauer, 2017). Combined, these
mouthing.” Developmentally, one would expect limitations are particularly problematic for longitu-
this item to be endorsed far more often in a study dinal studies of diverse populations, in which mea-
of 6-month-olds, than during a longitudinal fol- surement noninvariance may exist simultaneously
low-up three years later, when such toys are no between groups (e.g., racial and gender category
longer as developmentally meaningful. However, groups) and developmental time (Curran et al.,
all other things being equal, if this item’s contribu- 2014).
tion to the construct is weighted equally over time Fortunately, recent psychometric innovations
—which is done implicitly in any summary score (e.g., Approximate MI; Alignment; Muth
en &
that fails to adjust the weight explicitly—one Asparouhov, 2014; see Davidov, Muthen, & Sch-
would erroneously conclude that sociocognitive midt, 2018 for review) have greatly improved the
resources decreased across early childhood. An detection of and adjustment for DIF. In this study,
analog extends to group differences. As a purely we leverage a novel methodological approach to
illustrative example, affluent families in the US DIF testing that is particularly well-suited for
likely have many of such conventional infant toys unbalanced longitudinal data collected from a
while families in poverty may have fewer conven- sociodemographically diverse population—moder-
tional toys, yet other nonconventional items that ated nonlinear factor analysis (MNLFA; Bauer,
fulfill the same developmental purpose. If this were 2017; Bauer & Hussong, 2009; Curran et al., 2014).
the case, then a typical summary score would sug- Conceptually, MNLFA combines the best aspects of
gest lower levels of sociocognitive resources for MGA and MIMIC into a single approach. Like
the low-income family, despite the fact that the MGA, MNLFA allows one to adjust for DIF across
true latent construct is identical across groups. As all parameters of typical psychometric interest (e.g.,
such, the cost of DIF can be high—spurious con- measurement intercepts, factor loadings, factor (co)-
clusions driven by measurement artifacts rather variances). However, like MIMIC models, MNLFA
than true differences in the outcome of interest allows one to readily extend these tests to multiple,
(Borsboom, 2006; Millsap, 1998). simultaneous moderators—either continuous or cat-
Traditional approaches to testing and adjusting egorical.
for DIF include Multiple Groups analysis (MGA;
Henseler, Ringle, & Sinkovics, 2009) and Multiple
The Present Study
Indicators, Multiple Causes (MIMIC) modeling
(J€
oreskog & Goldberger, 1975). Each approach has In this study, we leverage these recent method-
both strengths and weaknesses. For instance, MGA ological advances to address three key aims. Our
allows one to model heterogeneity in most of the first aim was to apply recent theoretical models of
parameters of typical psychometric interest to environmental adversity to a large sample of chil-
applied researchers (e.g., factor loadings; item inter- dren living in predominately low-income rural com-
cepts; factor (co)variances, item residual (co)vari- munities. We used three widely used measures: (a)
ances; factor means). In so doing, they allow one to The Economic Strain Questionnaire (ES; Conger,
e460 DeJoseph, Sifre, Raver, Blair, and Berry

Ge, Elder, Lorenz, & Simons, 1994) to measure dis- Carolina (NC) and three counties in central Penn-
tal sources of material deprivation; (b) the Home sylvania (PA) were selected to be indicative of the
Observation for Measurement of the Environment Black South and Appalachia, respectively. The FLP
(HOME; Bradley & Caldwell, 1984), to measure adopted a developmental epidemiological design in
more proximal sources of sociocognitive resources, which sampling procedures were employed to
and (c) the Conflict Tactics Scale (CTS; Straus & recruit a representative sample of 1,292 children
Gelles, 1990) to measure psychosocial threat in the whose families resided in one of the six counties at
form of intimate partner violence, both verbal and the time of the target child’s birth. Families were
physical. oversampled for low-income status in both states,
Our second aim was to test the extent to which and families identifying as African American were
the proposed measures are commensurable across oversampled in NC (African American families
key demographic covariates and developmental were not oversampled in PA because African
time. Due to the ways in which measures of pov- Americans made up < 5% of the population of the
erty are deeply confounded with structural racial target communities). On average, families manage
inequalities, we anticipated racial group member- with an income 190% of the federal poverty
ship to contribute to both measurement artifact and threshold. For a comprehensive description of the
latent mean differences in measures of poverty-re- sampling plan and recruitment procedures, see
lated adversity. We anticipated similar effects with Vernon-Feagans et al. (2013).
respect to age for the HOME given children’s Families were seen in 2- to 3-hr home visits
increasing independence away from caregivers when children were approximately 6 months, and
(Larson, Richards, Moneta, Holmbeck, & Duckett, 1, 2, 3, 5, 7, and 12 years of age. During the visits,
1996). We had no a priori hypotheses for gender the primary caregiver (in 99% of cases the mother),
category group membership, and thus considered completed questionnaires concerning family demo-
potential gender differences as exploratory. Our graphics, income, economic hardship, family con-
third aim was to investigate the utility of using flict, and child behavior. Trained home visitors
MNLFA above and beyond raw mean scores. To completed a series of measures related to the home
do this, we used multi-level growth curve analyses environment. At the 12-year visit, children com-
to examine differences between raw-score- and pleted computerized tasks measuring aspects of
MNLFA-derived estimates, with respect to develop- self-regulation. Overall, 49% children in this sample
mental change and group differences. We also were identified by their parents as female, 42%
examined the relationship between cognitive and were identified by their parents as African Ameri-
behavioral child outcomes and MNLFA-derived can, and 58% reside in NC. Demographics for miss-
estimates to investigate the predictive utility of ing and nonmissing cases were very similar at each
MNLFA scores. Based on the DMAP framework, time point; information on demographics, missing-
we hypothesized that: (a) higher levels of psychoso- ness, and retention rates over time can be found in
cial threat would be more positively correlated with Table S2).
greater behavioral problems but less so with cogni-
tive performance, while (b) lower sociocognitive
resources and greater material deprivation would Measures
show comparatively stronger relations with chil-
Material Deprivation
dren’s cognitive performance than with their behav-
ioral problems. Lastly, we expected that income Economic need and material sufficiency were
would be more strongly related to material depriva- reported by the primary caregiver across all time
tion and sociocognitive resources than to psychoso- points using the six-item Economic Strain Question-
cial threat. naire (Conger et al., 1994). Two items assess eco-
nomic need (the extent to which the family has
difficulty paying bills and runs out of money each
Method month) and four items assess economic sufficiency
(the extent to which the family feels it is able to
Participants
adequately meet its needs for housing, clothing,
The Family Life Project (FLP) was designed to food, and medical care). Scores were rated on a Lik-
study families in two of the four major geographi- ert-type scale (range 1–5). Higher scores indicated
cal regions of the United States with high poverty that families reported experiencing greater material
rates (Dill, 1999). Three counties in eastern North deprivation.
 Capturing Enviornmental Dimensions e461

times, 2 = 6 or more times. To capture overall expo-

Sociocognitive Resources
sure, irrespective of which partner was the perpe-
Home visitors reported on the presence of parent trator, we then took the highest of the two scores
responsiveness and opportunities for cognitive as the indicator for the measurement models. We
resources at each of the seven home visits using the removed seven items pertaining to physical vio-
HOME scale (Bradley & Caldwell, 1984). All items lence, given low base-rates (< 10%); the remaining
were scored such that a 1 on any one item indi- nine items were retained in the measurement mod-
cated the presence of a given resource and a zero els (Table 1).
indicated an absence. Three versions of the HOME
were used to ask developmentally appropriate
Moderator/Invariance Variables
questions over the seven time points; one item was
common across every time point, eight items were Child age was measured as chronological age, con-
unique to one time point, and all other items were tinuously scaled in years and centered at the mean age
administered on at least two time points. at the 6-month visit. Study site (NC = 1, PA = 0),
The HOME is intended to measure many aspects racial group membership (1 = Black, 0 = White), and
of a child’s home environment. For this study, we gender group membership (1 = Female, 0 = Male)
conducted a content analysis of the HOME items were also included as moderators.
and selected an initial set of 61 items that were
deemed substantively relevant to sociocognitive
Criterion Variables
resources—namely, items pertaining to develop-
mentally appropriate learning materials and parent- Executive function. Child performance on the
child interactions that help to scaffold children’s Hearts and Flowers Task (HF; Diamond, Barnett,
learning (Rosen et al., 2019). Specifically, we sur- Thomas, & Munro, 2007) measured at the age 12
veyed items corresponding to the responsivity, visit was used to examine convergent validity in
mom/child behaviors, and the learning and lan- the cognitive domain via executive function (EF).
guage materials/resources subscales established Participants are instructed to respond by pressing
elsewhere (Bradley, 1994; Bradley & Caldwell, the designated key on the same side of the stimulus
1984). We then retained items that demonstrated when the stimulus is a heart and on the opposite
reasonable variation (≥ 10% with either 0 or 1). This side when the stimulus is a flower. There are three
resulted in a final list of 29 items included in the conditions: (a) congruent, where only one rule
models (see Table 1). applies (press the key on the same side as the
heart), (b) incongruent, where participants must
remember another rule (press the key on the oppo-
Psychosocial Threat: Exposure to Intimate Partner
site side that the flower appears), and (c) mixed,
Violence
where congruent and incongruent trials are inter-
Primary caregivers reported on their own and mixed. The task included a total of 57 test trials
their partners’ use of verbal aggression and physi- across three blocks (12 hearts-only, 12 flowers-only,
cal aggression during the past 12 months (CTS— and 33 mixed). Stimulus display and response win-
Couple Form R; Straus & Gelles, 1990) across all dow duration were 750 ms, preceded by a 500 ms
time points. The CTS was collected if the primary fixation cross and 500 ms blank screen. Participants
caregiver reported that she was married or had a who were < 60% accurate on congruent only trials
partner at that time point (70%–82% of the sample; were removed (n = 38) from analyses. EF for this
see Table S1). Items assess the frequency with study was indexed by accuracy and reaction time
which the respondent and the partner used verbal on mixed trials.
acts that emotionally or psychologically hurt the Behavioral problems. Parent-reported behavioral
other party (e.g., “How often has he insulted or problems on the Strengths and Difficulties Ques-
swore at you?”) as well as the frequency with tionnaire (SDQ; Goodman, 2001) measured at the
which physical force was used as a means of age 12 visit were used to examine convergent valid-
resolving the conflict (e.g., “How often has he ity in the affective domain. Primary caregivers
pushed, grabbed, or shoved you?”). Items ranged reported the degree to which items were not true,
from 0 = never to 6 = more than 20 times in the somewhat true, and certainly true for their children.
past year. However, due to low base rates and While multiple subscales are created from this mea-
skew of the response distribution, item responses sure, we used the behavioral problems total score,
were collapsed and recoded as 0 = never, 1 = 1–5 which represents the mean response across all 25
e462 DeJoseph, Sifre, Raver, Blair, and Berry

Table 1
List of Items for Each Environmental Exposure and Time Points at Which Each Item Was Available

0.6 years 1.5 years 2 years 3 years 5 years 7 years 12 years

Material deprivation
How difficult is it for you to pay your family’s bills each X X X X X X X
month
Generally, at the end of each month, do you end up with. . . X X X X X X X
My family has enough money to afford the kind of home we X X X X X X X
need. Do you. . .
We have enough money to afford the kind of clothing we X X X X X X X
need. Do you. . .
We have enough money to afford the kind of food we need. X X X X X X X
Do you. . .
We have enough money to afford the kind of medical care we X X X X X X X
need. Do you. . .
Sociocognitive resources
Caregiver initiates verbal interchanges with visitor X X X X X
Caregiver spontaneously praises child, at least twice X X X X X X X
Caregiver kisses or caresses child, at least once X X X X X
At least 10 books are present X X X X X X
Muscle activity toys or equipment X X X
Caregiver provides toys for child to play with during visit X X X
Learning facilitators—mobile, table and chair, high chair, play X X X
pen
Complex eye-hand coordination toys X X X
Toys for literature and music X X X
Caregiver holds child close 10–15 min per day X X
CD player or tape recorder and 5 or more children’s CDs or X X
tapes are available to the child
10 or more books for adults are visible in the home X X
Caregiver regularly buys or receives 1 or more magazines X X
Fairly regular/predictable daily family schedule X X
Caregiver sometimes yields to child’s fears or rituals X X
Caregiver uses term of endearment for child when discussing X X
child
Child has access to musical instrument X X
Child has near access to 2 pieces playground equipment X X
Caregiver use complete sentences/long words to talk w/ X X
visitor
House has at least 2 pictures or art on walls X X
Family provides lessons/memberships for child’s talents X X
Child is encouraged to learn shapes X
Two or more toys which teach colors, size, and shape are X
available to the child
Three or more puzzles are available to the child X
Two or more games which help teach numbers are available X
to the child
Caregiver introduces visitor to child X
Child’s art work is displayed in some visible place in the X
home
Child is encouraged to learn to read a few words X
TV is used judiciously X
Psychosocial threat
Insulted or swore at him/her/you X X X X X X X
Sulked or refused to talk about an issue X X X X X X X
Stomped out of the room or house or yard X X X X X X X
 Capturing Enviornmental Dimensions e463

Table 1
Continued

0.6 years 1.5 years 2 years 3 years 5 years 7 years 12 years

Cried X X X X X X X
Did or said something to spite him/her/you X X X X X X X
Threatened to hit or throw something at him/her/you X X X X X X X
Threw or smashed or hit or kicked something X X X X X X X
Threw something at him/her/you X X X X X X X
Pushed, Grabbed, or shoved him/her/you X X X X X X X

items. This score ranges from zero to two, with Giordano, & Janssen, 2018). Specifically, the MNLFA
higher values reflecting more behavior problems. proceeded by, first, generating a calibration sample
Income. Income-to-needs ratios (INR) for each from the larger longitudinal sample by taking a
time point were calculated as the family’s reported single longitudinal observation from each case. This
total household income for a given year divided by rendered the calibration sample functionally cross-
the federal poverty threshold for that year, adjusted sectional and, in so doing, removed the dependencies
for the number of persons in the home. inherent to longitudinal data. This calibration sample
was used in all subsequent analyses, with the
exception of the final estimation model when the full
Analytic Plan longitudinal data set was used.
In a second step, the latent factors and variances
Study Aim 1: Apply Recent Theoretical Models of
were regressed on the covariates. Although these
Environmental Adversity for the Context of Poverty
structural relations do not speak directly to DIF,
The ES, HOME, and CTS questionnaires were per se, they need to be accounted for to accurately
chosen to represent environmental measures of assess DIF in the indicator loadings and intercepts.
material deprivation, sociocognitive resources, and As noted in the following section, these relations
psychosocial threat, respectively. In addition to also contribute important information to the factor
their substantive alignment with the DMAP dimen- scores extracted from these models. Because we
sions, these measures were also collected across all expected growth rates in our environmental con-
seven time points in the FLP sample, affording sub- structs to vary across children (i.e., random effects),
sequent testing for longitudinal MI. A series of as well as systematically across the levels of the
descriptive and graphical examinations of the data covariates (i.e., fixed effects), we examined these
were used for initial item screening. To test configu- structural relations using mixed models, as dis-
ral invariance we fitted separate common-factor, cussed under our third aim.
longitudinal CFA models for each of the constructs Actual DIF testing begins in the third step.
in Mplus (see Van de Schoot, Lugtig, & Hox, 2012). Specifically, because a model testing DIF simultane-
For each, we constrained all latent factor means ously across all loadings and indicator intercepts
and variances to zero and one, respectively, to iden- would be under-identified, MNLFA adopts a two-
tify the model, and all other parameters were freely step process: (a) DIF for the indicator loadings and
estimated over time. We then consulted the fit intercepts are first tested individually for each indi-
indices, parameter estimates, and modification cator, with the remaining indicators constrained to
indices for evidence of heterogeneous factor struc- invariance (i.e., no DIF). Then (b) the indicators
tures. deemed to be invariant (i.e., no DIF) in the first step
are constrained to invariance in the second step, in
order to identify a model in which DIF is tested
Study Aim 2: Model DIF Across Site, Gender and
simultaneously across the remaining indicators.
Racial Group Membership, and Developmental Time
Given multiple testing, this second step adopts Ben-
Based on the models generated in Aim 1, we con- jamini–Hochberg corrections to adjust for inflated
ducted MNLFA for each construct, using a slightly type I error rates. All models were fitted to the data
altered version of the automated MNLFA (aMNLFA; using a robust maximum likelihood estimator and a
Gottfredson et al., 2019) package available for the R logit (binary) or cumulative logit (ordinal) link
statistical analysis platform (Cole, Gottfredson, function.
e464 DeJoseph, Sifre, Raver, Blair, and Berry

Lastly, the parameter estimates from this final modification indices for each construct suggested
calibration-sample model are applied as model con- homogeneity in the respective factor structures over
straints in a model fitted to data from the entire time. Specifically, although statistically significant
sample. Individual- and time-specific factor scores v2 values indicated imperfect fit—common with
(expected a posteriori estimates) are then extracted large samples—fit indices were within common
from this final model. As such, these factor scores thresholds (ES: v2 = 1,858.17, df = 665, p < .001,
represent an individual’s estimated DIF-adjusted comparative fit index [CFI] = .94, root mean square
level of the construct, given his/her age, gender error of approximation [RMSEA] = .038; HOME:
group membership, racial group membership, and v2 = 4,396.77, df = 2,254, p < .001, CFI = .89,
research site. For a detailed description of these RMSEA = .028; CTS: v2 = 3,677.43, df = 1,687,
steps in the aMNLFA statistical package, see p < .001, CFI = .96, RMSEA = .031), and the modifi-
Gottfredson et al. (2019). cation indices did not suggest regions of strain sug-
gestive of a heterogeneous factor structure over
time. Rank-order stability in the constructs ranged
Study Aim 3: Compare Empirical Differences Between
from .10 to .79 (ES: .10-.22; HOME: .30–.55; CTS:
Raw-Score- and MNLFA-Based Estimates With Respect
.36–.79).
to Developmental Change, Group Differences, and
Predictive Utility
To evaluate the advantages of the MNLFA Study Aim 2: Model DIF Across Site, Gender and
scores, mean proportion scores of ES, HOME, and Racial Group Membership, and Developmental Time
CTS were calculated at each time point and graphi-
Material Deprivation
cally compared against factor scores. We also fitted
a series of mixed linear models (Lmer package in R DIF testing via MNLFA indicated that the com-
3.3.1; Kuznetsova, Brockhoff, & Christensen, 2017), mon-factor model for parent-reported material
to test the extent to which the random and system- deprivation (indexed via ES questionnaire) was
atic differences in the growth of each of these con- invariant across gender and racial group member-
structs differed across the factor- versus raw-score ship, site, and age (Figure S1). Accounting for false-
scales. Specifically, we first specified a series of discovery rate, the factor variances, factor loadings,
growth models to establish the optimal growth and indicator intercepts were statistically indistin-
function and (co)variance structure for the given guishable across levels of the moderators.
construct (see Singer & Willett, 2002). We subse-
quently tested the extent to which growth in each
Sociocognitive Resources
construct varied as a function site and race. Model
comparisons were conducted using likelihood ratio The MNLFA results indicated that the common
tests and second-order Akaike information criteria items across time in the HOME data were largely
(AIC). invariant, with the exception of three items tapping
To test convergent validity and predictive utility, caregiver responsivity (“Caregiver kisses or caresses
partial correlations (controlling for covariates child at least once”) and learning materials (“Presence
retained in final MNLFA models) between the raw of learning facilitators—table and chair, high chair, play
and MNLFA scores and three theoretically relevant pen” and “CD player or tape recorder and 5 or more
constructs of interest were examined: (a) EF children’s CDs or tapes are available to the child”).
indexed by accuracy and reaction time on mixed Specifically, caregiver responsivity (B = .651,
HF blocks, (b) behavioral problems indexed by the p < .001) and learning facilitators (B = 1.135,
total problems subscale on the SDQ, and (c) pov- p < .001) tended to be endorsed less as children
erty as indexed by INR. aged, whereas materials requiring a CD player
tended to be endorsed more as children aged
(B = .159, p < .001). All longitudinal DIF effects
Results were on the intercepts only (i.e., partial scalar non-
invariance). All other parameters were invariant
Study Aim 1: Apply Recent Theoretical Models of
across age, gender group membership, racial group
Environmental Adversity for the Context of Poverty
membership, and site (Figure S2). Accounting for
Preliminary longitudinal CFAs with each con- false-discovery rate, the factor variances, factor
struct suggested longitudinal configural invariance, loadings, and indicator intercepts were statistically
such that model fit, the model parameters, and the indistinguishable across levels of the moderators.
 Capturing Enviornmental Dimensions e465

instance, by the range of MNLFA scores available

Psychosocial Threat
for every individual who had the lowest score on
The MNLFA for the CTS data indicated no evi- the raw-variable composite (Figure 1). Despite these
dence of DIF for the items tapping exposure to inti- differences in variation, the two scales manifested
mate partner violence. All other parameters were in very similar substantive conclusions with respect
invariant across age, gender, and racial group mem- to random and systematic differences in growth.
bership, and site (Figure S3). For example, irrespective of scale type, families
tended to show statistically significant and substan-
tively similar quadratic declines in material depriva-
Study Aim 3: Comparison of MNLFA-Derived Factor
tion over time. For both the factor scores and the
Scores and Raw-Variable Composites
raw-variable composites, the magnitudes of the
Although our MNLFA models tested and instantaneous linear slopes were strongest at
adjusted for fixed effects for age, site, and racial 6 months (Bmnlfa = .057, p < .001; Braw = .037,
group on the latent factors, the MNLFA approach p < .001) and leveled off at similar rates across
precluded our abilities to test random variation in childhood (Bmnlfa = .003, p < .001; Braw = .002,
growth or systematic differences in growth as a p < .001). Indeed, the estimated points of inflection
function of the covariates (i.e., cross-level interac- for these curves were quite similar across the mea-
tions; see Figure S4 for raw individual growth tra- sures (MNLFA = 10 years; raw = 9.75 years). Scal-
jectories). To address these questions and ing on their respective unconditional within-person
subsequently compare these estimates across the variances, these inflection points represent an
MNLFA factor scores and raw-variable composites, approximate .39 and .34, standard deviation
we fitted a series of nested mixed linear models to decrease in material deprivation from families’ aver-
both the MNLFA factor scores and the raw-variable age 6-month levels. Similarly, regardless of scale
composites—first, establishing plausible growth type, there was evidence that African American fami-
functions and covariance structures and then testing lies tended to show higher levels of material depriva-
individual differences in these growth rates as a tion (Bmnlfa = .419, p < .001; Braw = .235, p < .001).
function of site, and racial group (see Singer & Wil- These relations were statistically constant over time
lett, 2002). Because gender category did not show for both the MNLFA and raw-variable composite
up as a meaningful moderator in the MNLFA or scores, and corresponded to standardized effect sizes
subsequent analyses, we do not discuss it further. of d = .64 and d = .48, respectively. No site differ-
Model comparisons were based on likelihood ences were evident in material deprivation.
ratio tests and second-order AIC comparisons, There were also some noteworthy differences
where statistically nonsignificant fixed and random across the MNLFA factor score and raw-variable
effects were constrained to zero for parsimony composites. For example, despite the fact that there
(Table 2). As the MNLFA factor scores and raw- was little indication of DIF in the loadings or the
variable composites are on different scales, the indicator intercepts for our measure of psychosocial
parameter estimates cannot be compared directly. threat, simply treating threat as a reflective latent fac-
We, thus, provide descriptive comparisons of the tor led to different substantive conclusions compared
model-implied growth functions and standardized with those derived from the raw-variable composite.
effect sizes, in lieu of inferential comparisons. All Specifically, with the raw-variable composite, we
standardized estimates are scaled on the square found that, on average, families tended to show sta-
root of the unconditional variance for the given tistically significant quadratic decay in psychosocial
measure (i.e., factor score vs. raw composite), based threat across childhood (Braw_lin = .053, p < .001;
on the most relevant level of analysis (i.e., between- Braw_quad = .003, p < .001) and comparatively higher
person inferences are scaled on between-person levels of reported threat in PA (Braw = .134,
variation and within-person inferences are scaled p < .001) and among African American families
on within-person variation). (Braw = .092, p < .001; d = .44) at the 6-month wave
As expected, the correlations between the (Figure 2). Neither research site nor racial group
MNLFA factor scores and the raw variable compos- membership was predictive of changes in psychoso-
ites were very strong (material deprivation: r = .96; cial threat.
sociocognitive resources: r = .74; psychosocial Identical tests with the MNLFA factor scores
threat: r = .97). However, the MNLFA scores indicated similar racial group disparities at
demonstrated much larger individual variability 6 months (Bmnlfa = .304, p < .001; d = .47), as well
than did the raw scores. This is exemplified, for as a similar overall quadratic functional form as
 e466
Table 2
Baseline and Final Growth Models for MNLFA (Left) and Raw Mean Scores (Right) for Each Measure Making Up the Material Deprivation, Sociocognitive Resources, and Psychosocial Threat
Dimensions

MNLFA scores Raw mean scores

Baseline models Final models Baseline models Final models

ES HOME CTS ES HOME CTS ES HOME CTS ES HOME CTS

Fixed effects
(Intercept) 0.043 0.073 0.042 0.044 0.082 0.013 2.192 0.770 0.636 2.169 0.860 0.791
(.028) (.021)*** (.034) (.028) (.021)*** (.038) (.015)*** (.004)*** (.010)*** (.002)*** (.006)*** (.016)***
Age 0.057 0.476 0.138 0.057 0.476 0.118 0.037 0.023 0.053
(.007)*** (.007)*** (.007)*** (.007)*** (.007)*** (.010)*** (.005)*** (.002)*** (.003)***
Site 0.301 0.198 0.134
(North (.052)*** (.063)** (.002)**
DeJoseph, Sifre, Raver, Blair, and Berry

Carolina)
Race 0.419 .970 0.283 0.419 0.992 0.304 0.235 0.206 0.092
(Black) (.040)*** (.028)*** (.052)*** (.040)*** (.027)*** (.064)*** (.030)*** (.009)*** (.024)***
Interactions
Age 9 Age 0.003 0.027 0.007 0.003 0.027 0.006 0.002 0.002 0.003
(.000)*** (.000)*** (.001)*** (.000)*** (.001)*** (.001)*** (.000)*** (.000)*** (.000)***
Age 9 Site 0.055
(.018)**
Age 9 Race 0.018 0.040
(.019) (.004)***
Age2 9 Site 0.003
(.001)*
Age2 9 Race 0.003 0.003
(.001)* (.000)***
Random effects
(Intercept) 0.389 0.148 0.403 0.424 0.158 0.501 0.237 0.015 0.094 0.260 0.013 0.127
Age 0.003 3.28 e-04 0.003 0.002 1.85 e-05 0.001
Age 9 Age 5.57 e-06 4.99 e-09
Residual 0.493 0.461 0.397 0.449 0.457 0.346 0.260 0.034 0.095 0.232 0.032 0.077
AIC 18,395.6 16,809.6 13,524.8 18,290.7 16,789.5 13,347.6 13,671.5 2,597.6 4,942.7 13,393.4 3,164.7 4,261.5
Log likelihood 9,191.8 8,398.8 6,755.4 9,137.4 8,383.7 6,660.8 6,832.7 1,301.8 2,468.4 6,688.7 1,595.3 2,121.7

Note. baseline models for the moderated nonlinear factor analysis (MNLFA) scores are conditional on the variables included in the final MNLFA scoring models, as mean impacts
are embedded in the estimation of the final MNLFA scores. ES = Economic Strain Questionnaire; HOME = Home Observation for Measurement of the Environment; CTS = Conflict
Tactics Scale; AIC = Akaike information criteria.
*p < .05. **p < .01. ***p < .001.
 Capturing Enviornmental Dimensions e467

Figure 1. Graphs illustrating enhanced individual variability from MNLFA scores. Top panel: Scatterplot of MNLFA scores against cor-
responding raw mean scores colored by age (rounded for better plotting). Bottom panel: The complete distribution of MNLFA scores
associated with the lowest mean score on each of the measures (ES: mean = 1; HOME: mean = 0; CTS: mean = 0). MNLFA = moder-
ated nonlinear factor analysis; ES = Economic Strain Questionnaire; HOME = Home Observation for Measurement of the Environment;
CTS = Conflict Tactics Scale.

those established with the raw-variable composite. p < .001; d = .31) and largely limited to early
However, the results using the MNLFA factor infancy (Figure 2).
scores showed systematic differences in families’ Differences in families’ trajectories of sociocogni-
trajectories of psychosocial threat, as a function tive resources (via HOME scale) also differed
race. Specifically, the MNLFA-derived data indi- between the MNLFA factor scores and the raw-
cated that, although African American families variable composites. Based on the raw-variable
showed higher levels of psychosocial threat at composite, on average, African American families
6 months, they also showed more rapid quadratic tended to have fewer sociocognitive resources than
declines. By the age-12 wave, these growth differ- White families at the 6-month assessment
ences led to a substantial decrease in the racial (Braw = .206 , p < .001; d = 1.68) and to largely
group disparity observed at the 6-month wave (i.e., maintain these lower levels through the age-12
dage6mo = .47 to dage12 = .18). assessment (Figure 2). In addition to starting with
The MNLFA-derived data also showed differ- more sociocognitive resources, on average, the raw-
ences with respect to site. Recall that the results variable data suggested that White families tended
from the raw-variable composite indicated that PA to show stronger declines (Braw_linear = .040,
families tended have higher levels of psychosocial p < .001; Braw_quad = .003, p < .001) in their
threat that were stable over time (d = .44). The sociocognitive resources from 6-months to approxi-
results from the MNLFA-derived data, in contrast, mately age 6 (inflection = 5.75 years). As a function
indicated that 6-month psychosocial threat was of these varying trajectories, the 6-month disparity
more pronounced for NC families (Bmnlfa = .198; between racial groups was comparatively smaller
e468 DeJoseph, Sifre, Raver, Blair, and Berry

Figure 2. Group trajectory estimates from mixed effects models for MNLFA scores (left panel) and raw mean scores (right panel).
Colored lines indicate the predicted change curves for African American (blue) and White (red) children, superimposed on the raw data
points. MNLFA = moderated nonlinear factor analysis; ES = Economic Strain Questionnaire; HOME = Home Observation for Measure-
ment of the Environment; CTS = Conflict Tactics Scale.

when it reached its smallest point at approximately In contrast, the results using MNLFA-derived data
the age of 6.5 (d = .59)—though, still nontrivial in showed some noteworthy differences. Similar to the
absolute terms. raw-variable composite results, there was evidence
 Capturing Enviornmental Dimensions e469

Table 3
Partial Correlations Between Material Deprivation, Sociocognitive Resources, Psychosocial Threat, and Theoretically Relevant Outcomes (Account-
ing for Covariates Retained in Respective Final Moderated Nonlinear Factor Analysis [MNLFA] Models)

MNLFA scores Raw means

1 2 3 1 2 3

1. Material deprivation
2. Socicognitive resources .13*** .11***
3. Psychosocial threat .19*** .10*** .23*** .05***
4. H&F mixed block: % accuracy .08*** .13*** .02 .08*** .14*** .02
5. H&F mixed block: latency .01 .01 .02 .01 .01 .03
6. Behavioral problems .23*** .15*** .17*** .24*** .15*** .17***
7. Income-to-needs ratio .33*** .22*** .05*** .33*** .23*** .06***

***p < .001.

of a large disparity between racial groups in material deprivation, sociocognitive resources, and
sociocognitive resources at 6 months (Bmnlfa = .992, psychosocial threat). As expected, the factor scores
p < .001; d = 1.81 Unlike the raw-variable compos- were only modestly correlated with each other
ite results, though, the statistically nonsignificant (r = .13 to .19), suggesting that the material depri-
interactions with linear and quadratic age indicated vation, sociocognitive resources, and psychosocial
that the 6-month disparity in sociocognitive threat constructs are largely distinct. As expected,
resources was rather stable over time. material deprivation and sociocognitive resources
were moderately associated with family INR
(r = .33 and .22, respectively), whereas psychoso-
Relations With Criterion Measures
cial threat showed only a weak relation with INR
To provide a sense of construct validity for the (r = .05). We also examined potential overlap in
newly formed constructs, we examined partial cor- factor score distributions using typical INR cutoffs.
relations between the respective mean factor score Among families with an INR at or below 1, 38% fell
values (over time) and a number of criterion vari- above the total sample mean for sociocognitive
ables, adjusting for racial group and age (and site resources and 55% fell below the total sample mean
for the CTS; Table 3). First, we examined associa- for psychosocial threat exposure (Figure 3). These
tions between our environmental factors (i.e., descriptive findings highlight the large proportion

Figure 3. Density plots for MNLFA-derived scores. Red distributions represent families that fall at or below an income-to-needs ratio of
1 (i.e., 100% of the federal poverty threshold) and blue represent those above 1. Overlap between low- and higher-income groups high-
lights what it lost when grouping families by income cutoffs. MNLFA = moderated nonlinear factor analysis; ES = Economic Strain
Questionnaire; HOME = Home Observation for Measurement of the Environment; CTS = Conflict Tactics Scale.
e470 DeJoseph, Sifre, Raver, Blair, and Berry

of families providing sociocognitively supportive managing at or below 100% of the threshold (i.e.,
and safe environments despite financial hardship— income-to-needs ≤ 1). While many DMAP studies
nuance that would have been obscured if making categorize such income poverty as deprivation, we
group comparisons by income. sought to create a more nuanced measurement
Next, we examined the (adjusted) associations model that captures the full range of variation in a
between these latent factor scores and the age-12 family’s level of both distal and proximal
child outcomes. Across the board, these relations resources. A series of CFA models resulted in an
were rather modest. As expected, psychosocial threat uncorrelated three-factor model representing three
was comparatively more strongly correlated with distinct—albeit moderately related—dimensions of
children’s behavior problems (r = .17, p < .001) than material deprivation (indexed as a perceived lack
either measure of EF (rcorrect = .02, rlatency = .02; of financial resources) and sociocognitive resources
Fisher’s z = 3.85, p < .001, and z = 4.87, p < .001, (indexed as amount of social and cognitive
respectively). However, contrary to our hypotheses, resources in the home), as well as a dimension of
the remaining environmental measures were rather psychosocial threat (indexed as exposure to inti-
diffusely correlated across the cognitive and social mate partner violence).
outcomes. Sociocognitive resources were modestly
associated with fewer behavior problems (r = .15,
Findings for Aim 2: Models Examining DIF Across
p < .001) and better EF accuracy (r = .13, p < .001).
Site, Gender and Racial Group Membership, and
Material deprivation was somewhat more strongly
Developmental Time
correlated with worse behavior (r = .23, p < .001)
problems and lower EF accuracy (r = .08, p < .001). In line with our second aim, we examined the
Correlations between the raw mean scores and crite- extent to which our derived constructs were com-
rion variables were largely similar to MNLFA scores mensurate across child demographic covariates
(Table 3). using Bauer and Curran’s MNLFA approach
(Bauer, 2017; Gottfredson et al., 2019). By and large,
we found that group and developmental differences
in the meaning of the scales (i.e., noninvariance)
Discussion
were minimal. Some scales were more invariant
This study sought to test for group and longitudi- than others, however. Our measures of material
nal MI among environmental dimensions of adver- deprivation and psychosocial threat showed no evi-
sity and resources in a population-based sample of dence of DIF, irrespective of age, gender group
children and families, the majority of whom membership, racial group membership, or site. This
reported living on low incomes relative to the U.S. was a notable finding, given that rigorous analytical
national poverty line. This study answers calls to procedures suggested that the items on these two
employ less biased and more rigorous statistical measures mean the same thing across key demo-
methods to measure children’s environmental expo- graphic covariates. Sociocognitive resources, as
sures in the context of poverty across developmen- measured using partially overlapping versions of
tal time, beyond traditional indicators such as the HOME (Bradley & Caldwell, 1984) over time,
income or cumulative risk. To this end, we applied however, showed some longitudinal DIF. Part of
and expanded current models of adversity by gen- this was unavoidable, given discontinuities in the
erating three distinct latent dimensions of environ- items included in versions of the HOME intended
mental exposures across seven time points that for different age bands. However, there was also
spanned child age 6 months–12 years. some evidence of DIF for common items—specifi-
cally three items tapping caregiver responsivity
and learning materials, suggesting that certain
Findings for Aim 1
aspects of sociocognitive resources in the home
Our first aim was to test the extent to which are more or less likely to be endorsed as children
theoretically informed constructs outlined in the get older. This is in line with developmental work
DMAP framework (McLaughlin & Sheridan, 2016) showing increasing independence and time spent
would be supported empirically in sample of chil- outside of the home as children transition into
dren and families growing up in low-income, rural adolescence (Larson et al., 1996). Crucially,
contexts. On average, children in our sample were MNLFA scores adjust for DIF and therefore gener-
in households managing with an income of 190% ate more empirically accurate estimates in subse-
of the federal poverty threshold, with 30% quent modeling.
 Capturing Enviornmental Dimensions e471

that many families in our sample are providing a

Findings for Aim 3: Disparities Between Racial Groups
safe and cognitively stimulating home environment
At the latent construct level, average racial group despite economic hardship, and this nuance in indi-
(as determined by families who self-identified as vidual differences is a nontrivial improvement from
members of a particular race category) differences comparisons made by income alone. Taken
were found across all three constructs. On average, together, our findings provide support for growing
African American children in our sample experi- consensus in the field to move towards adaptation-
enced more material deprivation, fewer sociocogni- based models that account for the entire spectrum
tive resources, and more psychosocial threat than of variation in experiences that children in poverty
White children. We speculate that these differences are exposed to (Frankenhuis & Nettle, 2019; Hosti-
presumably reflect larger systemic inequalities and nar & Miller, 2019). Doing so affords a greater
longstanding racial income disparities stemming understanding of the ways families are able to
from historical forms of racial segregation. Even in maintain high levels of resilience in the face of eco-
this predominantly low-income sample, on average, nomic adversity.
the income of White families was nearly twice that
of African American families (mean income-to-
Findings for Aim 3: Criterion Validity
needs: 2.39 and 1.26, respectively). Such macro-level
forms of social stratification have been argued to Although individual differences in children’s
influence group-level variation in families’ values experiences and resources in the context of poverty
and behaviors, resulting in a uniquely adaptive cul- were clear, how such exposures are related to speci-
ture that better meets the specific contextual fic adaptive outcomes across multiple domains
demands faced by a given group (Coll et al., 1996). remains somewhat unclear. As hypothesized, psy-
In addition to the challenges of parenting in pov- chosocial threat—as indexed by inter-partner vio-
erty (Magnuson & Duncan, 2002), families of color lence—was modestly associated with heightened
additionally experience a host of stressors related to levels of behavior problems, yet was unrelated to
discrimination and bias (e.g., police brutality, lower children’s cognitive outcomes. Thus, there was
quality healthcare) that undoubtedly exert cascad- potentially some evidence of developmental speci-
ing effects on resource scarcity and caregivers’ well- ficity, even if the effect size was modest in absolute
being (Raver & Blair, 2020). While direct examina- terms. While the cognitive measure we used tapped
tion of these individual- and structural-level forms executive functioning skills specifically, the behav-
of racial and class inequalities is beyond the scope ioral problems scale had some limitations. First,
of this article, our findings underscore the need for there was shared rater bias in the psychosocial
applying an intersectional framework in future threat measure and the child behavior problems
work (Syed & Ajayi, 2018). scale. Second, behavior problems tapping both
internalizing and externalizing may have been too
broad, and a more emotionally salient task (e.g.,
Findings for Aim 3: Heterogeneity in Exposure to
fear conditioning) at this age may be necessary to
Adversities
better disentangle the effects of these dimensions on
A significant contribution of our derived mea- domain-specific outcomes (McLaughlin et al., 2016;
surement model is best revealed in descriptive find- Raver, Blair, Garrett-Peters, & Family Life Project
ings highlighting the ways different children whose Investigators, 2015). Contrary to our hypotheses
families reported very similar incomes nonetheless that material deprivation and sociocognitive
experienced a wide array of differing types of stres- resources would be more strongly associated with
sors and resources. Overall, many children in our cognitive than with affective outcomes, the modest
sample experienced high levels of material depriva- to moderate relations were rather largely diffuse
tion across time, yet a relatively smaller proportion across these outcomes. This may be because mate-
experienced low levels of sociocognitive resources rial and financial resources likely increase exposure
over time and even fewer were exposed to intimate to conditions that pose a threat to the child (e.g.,
partner violence (Figure 3). The three derived fac- being forced to live in higher crime neighborhoods
tors were also minimally to moderately related to in order to afford rent), which may explain why the
each other; in other words, these forms of poverty- material deprivation factor was comparatively more
related adversity do not always occur together. strongly related to our affective outcome. Nonethe-
Income-to-needs was only moderately correlated less, the degree to which children had access to
with our latent constructs, underscoring the ways sociocognitive resources in the home was more
e472 DeJoseph, Sifre, Raver, Blair, and Berry

strongly related to EF compared to the other two different conclusions. Specifically, growth patterns
latent factors. However, these relations were small for MNLFA scores for sociocognitive resources
in magnitude, perhaps because such resources mat- were nearly opposite of the raw mean scores. We
ter more at earlier stages of cognitive development, also found differences with respect to racial dispari-
which we were unable to examine in this study. ties between the two scoring approaches. Raw
Despite this, the specificity of these relations aligns mean scores for psychosocial threat suggested race
with prior literature suggesting that compared to differences were constant over time, whereas the
threat, access to developmentally appropriate learn- MNLFA scores suggested these differences become
ing materials and a variety of scaffolded experi- smaller over time. The opposite pattern was true
ences with caregivers shapes individual differences for sociocognitive resources. For studies with policy
in cognitive outcomes (e.g., Rosen, Amso, & or clinical implications, careful measurement work
McLaughlin). ensures our conclusions are driven by substantive
differences on a given construct rather than by
measurement artifacts. This is essential to avoid
Value Added: Comparison of MNLFA-Derived Factor
wasting tax-payer dollars, individuals’ time, and in
Scores and Raw-Variable Composites
extreme circumstances, the risk of contributing to
Why use MNLFA scores in research examining racial oppression. While we found minimal evi-
the developmental effects of poverty-related envi- dence for DIF in this study, this may not always be
ronmental exposures, as opposed to raw-item the case in studies representing more diverse popu-
sums? With regard to methodological benefits, lations. Accurate measurement of our constructs
MNLFA scores are continuously and normally dis- also have important theoretical implications. Since
tributed, and demonstrate a large increase in indi- MNLFA scores permit apples-to-apples compar-
vidual variability due to the ways in which this isons, developmental researchers can more accu-
approach incorporates unique information on indi- rately and rigorously examine the role of timing
vidual differences (i.e., scores adjust for which and sensitive periods, while simultaneously
items were endorsed at each time point and by accounting for heterogeneity in dose and severity of
whom). Moreover, in contrast to raw-item propor- environmental exposures. Such analyses using the
tion scores, variation in the type and severity of current MNLFA-derived scores are currently under-
items are accounted for using item-level weighting way from our research team.
—which may be one of the reasons we found large
differences in developmental trajectories for the
Strengths, Limitations, and Conclusions
HOME measure, for example. Such increased varia-
tion also increases statistical power and thus This study has notable strengths including the
increases the ability to accurately detect a true use of a large, diverse sample of children in low-in-
effect. Lastly, the flexibility of this approach is come contexts, tests of longitudinal MI across seven
undeniable. This is illustrated by its ability to incor- time points, and the use of multiple measures that
porate multiple longitudinal measures with a sub- better illustrate the multidimensional nature of pov-
stantial amount of nonoverlap which can be erty-related adversity. However, there were a num-
leveraged for large-scale multi-site developmental ber of limitations. First, the measures included in
data—enhancing generalizability and external valid- this analysis represent only three poverty-relevant
ity (Curran, Cole, Giordano, et al., 2018). In addi- contextual factors. While this was a limitation of
tion, MNLFA scores can be used in subsequent available data across the seven time points, addi-
secondary models and have been shown to produce tional factors such as racial discrimination, social
less biased estimates compared to CFA scores (Cur- support, and neighborhood safety should be exam-
ran, Cole, Bauer, Cole, Bauer, Rothenberg, & Hus- ined in future studies. Second, because the CTS
song, 2018). measure was only collected among caregivers with
On a substantive level, these scores empirically partners who rarely (if ever) reported experiencing
disentangle race and age, minimizing bias and thus more severe forms of intimate partner violence,
permitting more developmentally and sociocultur- findings related to psychosocial threat exposure are
ally appropriate conclusions about the effects of not generalizable to single-parent households or
poverty-related exposures on child adaptive out- households that may be experiencing severe domes-
comes. This is reflected in our observed differences tic violence. Third, the HOME measure provides a
in growth trajectories between the scoring very rough and inadequate proxy for the many
approaches, which resulted in substantively vibrant ways that families engage their children in
 Capturing Enviornmental Dimensions e473

cognitively stimulating and emotionally supportive Blair, C., & Raver, C. C. (2012). Child development in the
practices. Furthermore, we were limited to home context of adversity: Experiential canalization of brain
visitor- and parent-reported measures as child-re- and behavior. American Psychologist, 67, 309. https://
ported measures were not collected until much doi.org/10.1037/a0027493
Borsboom, D. (2006). When does measurement invariance
later. Using mixed methods approaches to capture
matter? Medical Care, 44, S176–S181. https://doi.org/
children’s subjective lived experiences as they
10.1097/01.mlr.0000245143.08679.cc
mature into adolescence is another important future Bradley, R. H. (1994). The Home Inventory: Review and
direction for understanding how both external and reflections. San Diego, CA: Academic Press.
internal resources in the context of poverty shape Bradley, R. H., & Caldwell, B. M. (1984). The HOME
developmental adaptations. Finally, to the extent Inventory and family demographics. Developmental Psy-
that sample characteristics including the nonurban chology, 20, 315. https://doi.org/10.1037//0012-1649.
and predominantly low-income regions in which 20.2.315
our sample resides, the generalizability of our find- Brunson, K. L., Eghbal-Ahmadi, M., Bender, R., Chen, Y.,
ings is an important question that can best be & Baram, T. Z. (2001). Long-term, progressive hip-
addressed through further analyses of data sets pocampal cell loss and dysfunction induced by early-
life administration of corticotropin-releasing hormone
with similar types of variables to those analyzed
reproduce the effects of early-life stress. Proceedings of
here.
the National Academy of Sciences of the United States of
As researchers seek to examine the ways in America, 98, 8856–8861. https://doi.org/10.1073/pnas.
which the type and timing of poverty influence 151224898
neurodevelopmental mechanisms among diverse Cole, V. T., Gottfredson, N. C., Giordano, M., & Janssen,
groups of children, care must be taken to ensure T. (2018). Automated fitting of moderated nonlinear factor
chosen measures are reliable and socioculturally analysis (MNLFA) through the Mplus program. R package
valid. Our field is facing long overdue calls to version 1.0.
action to address the grave harms and barriers that Coll, C. G., Crnic, K., Lamberty, G., Wasik, B. H., Jenkins,
families and children face in coping with the toxic R., Garcia, H. V., & McAdoo, H. P. (1996). An integra-
convergence of racism and social and economic tive model for the study of developmental competen-
cies in minority children. Child Development, 67, 1891–
inequality. Without sound methods, we cannot be
1914. https://doi.org/10.2307/1131600
confident in interpreting our substantive results,
Conger, R. D., Ge, X., Elder, G. H. Jr, Lorenz, F. O., &
and novel methods do not advance the field unless Simons, R. L. (1994). Economic stress, coercive family
appropriately applied to substantively meaningful process, and developmental problems of adolescents.
questions. Here we highlight an innovative mea- Child Development, 65, 541–561. https://doi.org/10.
surement approach that revealed three distinct 1111/j.1467-8624.1994.tb00768.x
environmental exposures in the context of poverty Curran, P. J., Cole, V. T., Bauer, D. J., Rothenberg, W. A.,
—providing tangible targets for the creation of indi- & Hussong, A. M. (2018). Recovering predictor–crite-
vidualized interventions as our nation works to rion relations using covariate-informed factor score esti-
ameliorate income poverty and structural racial, mates. Structural Equation Modeling: A Multidisciplinary
social, and economic inequalities. In sum, careful Journal, 25, 860–875. https://doi.org/10.1080/10705511.
2018.1473773
consideration of the statistical validity of poverty-
Curran, P. J., Cole, V., Giordano, M., Georgeson, A. R.,
related constructs will result in more trustworthy
Hussong, A. M., & Bauer, D. J. (2018). Advancing the
conclusions and therefore more informative policies study of adolescent substance use through the use of
and programs aimed to enable children to thrive. integrative data analysis. Evaluation & the Health Profes-
sions, 41, 216–245. https://doi.org/10.1177/
0163278717747947
References Curran, P. J., McGinley, J. S., Bauer, D. J., Hussong, A.
M., Burns, A., Chassin, L., . . . Zucker, R. (2014). A
Bauer, D. J. (2017). A more general model for testing mea- moderated nonlinear factor model for the development
surement invariance and differential item functioning. of commensurate measures in integrative data analysis.
Psychological Methods, 22, 507. https://doi.org/10.1037/ Multivariate Behavioral Research, 49, 214–231. https://
met0000077 doi.org/10.1080/00273171.2014.889594
Bauer, D. J., & Hussong, A. M. (2009). Psychometric Davidov, E., Muthen, B., & Schmidt, P. (2018). Measure-
approaches for developing commensurate measures ment invariance in cross-national studies: challenging
across independent studies: Traditional and new mod- traditional approaches and evaluating new ones. Socio-
els. Psychological Methods, 14, 101. https://doi.org/10. logical Methods & Research, 47, 631–636. https://doi.
1037/a0017642 org/10.1177/0049124118789708
e474 DeJoseph, Sifre, Raver, Blair, and Berry

Diamond, A., Barnett, W. S., Thomas, J., & Munro, S. to better predict caregiver well-being. Journal of Child
(2007). Preschool program improves cognitive control. and Family Studies, 28, 753–764. https://doi.org/10.
Science, 318, 1387–1388. https://doi.org/10.1126/scie 1007/s10826-018-01304-5
nce.1151148 Huttenlocher, P. R., de Courten, C., Garey, L. J., & Van
Dill, B. T. (1999). Poverty in the rural US: Implications for der Loos, H. (1982). Synaptogenesis in human visual
children, families, and communities. Literature review pre- cortex—Evidence for synapse elimination during nor-
pared for the Annie E. Casey Foundation. mal development. Neuroscience Letters, 33, 247–252.
Duncan, G. J., & Magnuson, K. A. (2003). Off with https://doi.org/10.1016/0304-3940(82)90379-2
Hollingshead: Socioeconomic resources, parenting, and Jiang, Y., & Koball, H. (2018). Basic facts about low-income
child development. In M. H. Bornstein & R. H. Bradley children: Children under 18 years, 2016. New York, NY:
(Eds.), Socioeconomic status, parenting, and child develop- National Center for Children in Poverty, Columbia
ment (pp. 83–106). New York, NY: Psychology Press. University Mailman School of Public Health.
Ellwood-Lowe, M. E., Whitfield-Gabrieli, S., & Bunge, S. J€
oreskog, K. G., & Goldberger, A. S. (1975). Estimation of
A. (2020). What is an adaptive pattern of brain activity a model with multiple indicators and multiple causes
for a child? It Depends on their Environment. bioRxiv. of a single latent variable. Journal of the American Statis-
https://doi.org/10.1101/2020.05.29.124297 tical Association, 70, 631–639. https://doi.org/10.2307/
Evans, G. W. (2004). The environment of childhood pov- 2285946
erty. American Psychologist, 59, 77. https://doi.org/10. Kuznetsova, A., Brockhoff, P. B., & Christensen, R. H. B.
1037/0003-066x.59.2.77 (2017). lmerTest Package: Tests in linear mixed effects
Flake, J. K., & Fried, E. I. (2019). Measurement schmeasure- models. Journal of Statistical Software, 82(13), 1–26.
ment: Questionable measurement practices and how to avoid https://doi.org/10.18637/jss.v082.i13
them. https://doi.org/10.31234/osf.io/hs7wm Larson, R. W., Richards, M. H., Moneta, G., Holmbeck,
Frankenhuis, W. E., & Nettle, D. (2019). The strengths of G., & Duckett, E. (1996). Changes in adolescents’ daily
people in poverty. Current Directions in Psychological interactions with their families from ages 10 to 18:
Science, 29(1), 16–21. https://doi.org/10.1177/ Disengagement and transformation. Developmental Psy-
0963721419881154 chology, 32, 744. https://doi.org/10.1037//0012-1649.
Frankenhuis, W. E., Young, E. S., & Ellis, B. J. (2020). The 32.4.744
hidden talents approach: Theoretical and methodologi- Magnuson, K. A., & Duncan, G. J. (2002). Parents in pov-
cal challenges. Trends in Cognitive Sciences, 24, 569–581. erty. https://doi.org/10.4324/9780429398995-9
https://doi.org/10.1016/j.tics.2020.03.007 McLaughlin, K. A., & Sheridan, M. A. (2016). Beyond
Gershoff, E. T., Aber, J. L., Raver, C. C., & Lennon, M. C. cumulative risk: A dimensional approach to childhood
(2007). Income is not enough: Incorporating material adversity. Current Directions in Psychological Science, 25,
hardship into models of income associations with par- 239–245. https://doi.org/10.1177/0963721416655883
enting and child development. Child Development, 78(1), McLaughlin, K. A., Sheridan, M. A., Gold, A. L., Duys,
70–95. https://doi.org/10.1111/j.1467-8624.2007.00986.x A., Lambert, H. K., Peverill, M., . . . Pine, D. S. (2016).
Goodman, R. (2001). Psychometric properties of the Maltreatment exposure, brain structure, and fear condi-
strengths and difficulties questionnaire. Journal of the tioning in children and adolescents. Neuropsychopharma-
American Academy of Child & Adolescent Psychiatry, 40, cology, 41, 1956. https://doi.org/10.1038/npp.2015.365
1337–1345. https://doi.org/10.1097/00004583-20011100 Meredith, W. (1993). Measurement invariance, factor anal-
0-00015 ysis and factorial invariance. Psychometrika, 58, 525–543.
Gottfredson, N. C., Cole, V. T., Giordano, M. L., Bauer, https://doi.org/10.1007/bf02294825
D. J., Hussong, A. M., & Ennett, S. T. (2019). Simplify- Millsap, R. E. (1998). Group differences in regression
ing the implementation of modern scale scoring meth- intercepts: Implications for factorial invariance. Multi-
ods with an automated R package: Automated variate Behavioral Research, 33, 403–424. https://doi.org/
moderated nonlinear factor analysis (aMNLFA). Addic- 10.1207/s15327906mbr3303_5
tive Behaviors, 94, 65–73. https://doi.org/10.1016/ Muth
en, B., & Asparouhov, T. (2014). IRT studies of
j.addbeh.2018.10.031 many groups: The alignment method. Frontiers in Psy-
Henseler, J., Ringle, C. M., & Sinkovics, R. R. (2009). The chology, 5, 978. https://doi.org/10.3389/fpsyg.2014.
use of partial least squares path modeling in interna- 00978
tional marketing. In New challenges to international mar- Parker, K., Horowitz, J., & Mahl, B. (2016). On views of
keting (pp. 277–319). Emerald Group. https://doi.org/ race and inequality, Blacks and Whites are worlds apart.
10.1108/s1474-7979(2009)0000020014 Washington, DC: Pew Research Center. http://www.
Hostinar, C. E., & Miller, G. E. (2019). Protective factors for pewsocialtrends.org/2016/06/27/on-views-of-race-and-
youth confronting economic hardship: Current challenges inequality-blacks-and-whites-are-worlds-apart/
and future avenues in resilience research. American Psy- Raineki, C., Cort
es, M. R., Belnoue, L., & Sullivan, R. M.
chologist, 74, 641. https://doi.org/10.1037/amp0000520 (2012). Effects of early-life abuse differ across develop-
Hurwich-Reiss, E., Watamura, S. E., & Raver, C. C. ment: Infant social behavior deficits are followed by
(2019). Beyond income: Expanding our empirical toolkit adolescent depressive-like behaviors mediated by the
 Capturing Enviornmental Dimensions e475

amygdala. Journal of Neuroscience, 32, 7758–7765. Journal of Developmental Psychology, 9, 486–492. https://
https://doi.org/10.1523/JNEUROSCI.5843-11.2012 doi.org/10.1080/17405629.2012.686740
Raver, C. C., & Blair, C. (2020). Developmental science Vernon-Feagans, L., Cox, M., Willoughby, M., Burchinal,
aimed at reducing inequality: Maximizing the social M., Garrett-Peters, P., Mills-Koonce, R., . . . Bauer, P. J.
impact of research on executive function in context. (2013). The Family Life Project: An epidemiological and
Infant and Child Development, 29, e2175. https://doi. developmental study of young children living in poor
org/10.1002/icd.2175 rural communities. Monographs of the Society for Research
Raver, C. C., Blair, C., Garrett-Peters, P.; Family Life Pro- in Child Development, i–150, 43773265. https://doi.org/
ject Key Investigators. (2015). Poverty, household 10.1111/mono.12048
chaos, and interparental aggression predict children’s Widaman, K. F., Ferrer, E., & Conger, R. D. (2010). Facto-
ability to recognize and modulate negative emotions. rial invariance within longitudinal structural equa-
Development and Psychopathology, 27, 695–708. https:// tion models: Measuring the same construct across time.
doi.org/10.1017/s0954579414000935 Child Development Perspectives, 4(1), 10–18. https://doi.
Rosen, M. L., Amso, D., & McLaughlin, K. A. (2019). The org/10.1111/j.1750-8606.2009.00110.x
role of the visual association cortex in scaffolding pre- Wiesel, T. N., & Hubel, D. H. (1965). Extent of recovery
frontal cortex development: A novel mechanism linking from the effects of visual deprivation in kittens. Journal
socioeconomic status and executive function. Develop- of Neurophysiology, 28, 1060–1072. https://doi.org/10.
mental Cognitive Neuroscience, 39, 100699. https://doi. 1152/jn.1965.28.6.1060
org/10.1016/j.dcn.2019.100699 Yoshikawa, H., Aber, J. L., & Beardslee, W. R. (2012). The
Rosen, M. L., Sheridan, M. A., Sambrook, K. A., Meltzoff, effects of poverty on the mental, emotional, and behav-
A. N., & McLaughlin, K. A. (2018). Socioeconomic dis- ioral health of children and youth: Implications for pre-
parities in academic achievement: A multi-modal inves- vention. American Psychologist, 67, 272. https://doi.org/
tigation of neural mechanisms in children and 10.1037/a0028015
adolescents. NeuroImage, 173, 298–310. https://doi.org/
10.1016/j.neuroimage.2018.02.043
Shafir, E. (2017). Decisions in poverty contexts. Current Supporting Information
Opinion in Psychology, 18, 131–136. https://doi.org/10.
Additional supporting information may be found in
1016/j.copsyc.2017.08.026
Singer, J. D., & Willett, J. B. (2003). Applied longitudinal
the online version of this article at the publisher’s
data analysis: Modeling change and event occurrence. website:
Oxford University Press.
Straus, M. A., & Gelles, R. J. (1990). New scoring methods Figure S1. MNLFA Final Model Output for the
for violence and new norms for the Conflict Tactics Material Deprivation Latent Factor (Eta)
Scales. Physical Violence in American Families: Risk Fac- Figure S2. Final MNLFA Model for the Sociocog-
tors and Adaptations to Violence in, 8, 341–367. https:// nitive Resources Latent Factor (Eta)
doi.org/10.4324/9781315126401-4 Figure S3. Final MNLFA Model for the Psy-
Syed, M., & Ajayi, A. A. (2018). Promises and pitfalls in chosocial Threat Latent Factor (Eta)
the integration of intersectionality with development Figure S4. Random Subset of n = 75 Illustrating
science. New Directions for Child and Adolescent Develop-
Raw, Individual Growth Trajectories for MNLFA
ment, 2018, 109–117. https://doi.org/10.1002/cad.20250
Syed, M., Santos, C., Yoo, H. C., & Juang, L. P. (2018).
Scores (Left Panel) and Raw Means (Right Panel)
Invisibility of racial/ethnic minorities in developmental Table S1. Cross-Walk Between Invariance/
science: Implications for research and institutional prac- Equivalence Testing Terminology in the Context of
tices. American Psychologist, 73, 812. https://doi.org/10. CFA With Continuous Observed Items and CFA/2-
31234/osf.io/hg9fm Parameter IRT With Binary Items, Respectively
Van de Schoot, R., Lugtig, P., & Hox, J. (2012). A check- Table S2. Sociodemographic Descriptives, Attri-
list for testing measurement invariance. European tion, and Missingness Over Time
This document is a scanned copy of a printed document. No warranty is given about the
accuracy of the copy. Users should refer to the original published version of the material.