Physical Discipline and Behavior Problems in African American, European American, and Hispanic...

Journal of Marriage and Family 64 (February 2002): 40–5340

VONNIE C. MCLOYD AND JULIA SMITH

University of Michigan

l

Physical Discipline and Behavior Problems in African

American, European American, and Hispanic Children:

Emotional Support as a Moderator

Using data collected over a 6-year period on asample of 1,039 European American children, 550African American children, and 401 Hispanicchildren from the children of the National Lon-gitudinal Survey of Youth, this study assessedwhether maternal emotional support of the childmoderates the relation between spanking and be-havior problems. Children were 4–5 years of agein the first of 4 waves of data used (1988, 1990,1992, 1994). At each wave, mothers reported theiruse of spanking and rated their children’s behav-ior problems. Maternal emotional support of thechild was based on interviewer observations con-ducted as part of the Home Observation for Mea-surement of the Environment. For each of the 3racial-ethnic groups, spanking predicted an in-crease in the level of problem behavior over time,controlling for income-needs ratio and maternalemotional support. Maternal emotional supportmoderated the link between spanking and problembehavior. Spanking was associated with an in-crease in behavior problems over time in the con-text of low levels of emotional support, but not inthe context of high levels of emotional support.This pattern held for all 3 racial-ethnic groups.

Center for Human Growth and Development, University ofMichigan, 300 North Ingalls, Ann Arbor, MI 48109([email protected]).

Key Words: behavior problems, emotional support, ethnic-ity, physical discipline.

Surveys indicate that the majority of Americanparents endorse corporal punishment as a child-rearing practice and use it to discipline their chil-dren (Day, Peterson, & McCracken, 1998; Gils-Sims, Straus, & Sugarman, 1995; Straus & Gelles,1986). Corporal punishment varies in intensityfrom spanking, typically defined as striking thechild on the buttocks or extremities with an openhand without inflicting physical injury, to physicalabuse, consisting of beatings and other forms ofextreme physical force that inflict bodily injury.Whereas spanking falls within the normativerange of socialization practices within the UnitedStates, physical abuse does not (Baumrind, 1997).Several studies have found that parental use ofphysical discipline is positively related to behav-ioral (e.g., aggression) and psychological (e.g.,dysphoria, low self-esteem) problems in childrenand adolescents, but these relations are markedlystronger in samples of clinically aggressive chil-dren (where frequency and intensity of physicaldiscipline tend to be higher than in nonclinicalsamples) and samples of children who have beenphysically abused (Dodge, Pettit, Bates, & Val-ente, 1995; Loeber & Stouthamer-Loeber, 1986;Strassberg, Dodge, Pettit, & Bates, 1994; Straus,Sugarman, & Gils-Sims, 1997).

Given its high prevalence in the United States,it is clear that physical discipline short of physicalabuse occurs within the context of a diversity ofparenting styles and behaviors. Consequently, ad-

41Physical Discipline

equate assessment of the effects of physical dis-cipline on children’s development may requiretaking into account the broader parenting contextwithin which physical discipline occurs. For ex-ample, studies indicate that physical discipline isunrelated to children’s socioemotional functioning(e.g., aggression, delinquency, self-esteem) oncedimensions of parenting associated with physicaldiscipline (e.g., parental rejection, low parental in-volvement) are taken into account (e.g., Larzelere,Klein, Schumm, & Alibrando, 1989; Rohner, Bo-urque, & Elordi, 1996; Simons, Johnson, & Con-ger, 1994). On the assumption that the affectivequality of parent-child relations is an importantmarker of parenting context, the present studyseeks to determine if the relation between physicaldiscipline and behavior problems is conditional onthe level of warmth and support mothers exhibittoward the child. We hypothesized that the rela-tion between physical punishment and behaviorproblems would be intensified in the context oflow levels of maternal support but attenuated inthe context of high levels of maternal support.

The notion that the effects of physical disci-pline depend on when and why parents use phys-ical discipline as well as on the affective contextwithin which physical discipline occurs has manyproponents, but direct empirical tests of these hy-pothesized moderation effects are surprisinglythin. Based on his synthesis of findings acrossstudies published in peer-reviewed journals, Lar-zelere (1996) concluded that nonabusive or ‘‘cus-tomary’’ physical discipline by parents tends to beassociated with positive or neutral outcomes inoffspring when the parenting context is marked byhigh levels of positive parental involvement, atendency to use physical discipline based on child-oriented rather than parent-oriented motives, con-sistent follow-through on disciplinary warnings,and absence of verbal putdowns and ridicule.However, Larzelere’s conclusion about markers ofthe parenting context that moderate the effects ofphysical punishment is not outcome specific andis based on a small number of primarily retro-spective studies with several methodological bi-ases and weaknesses.

More recently, Deater-Deckard and Dodge(1997) tested the affective quality of parent-childrelations as a moderator of the impact of physicalpunishment in their longitudinal sample of Afri-can American and European American families.They found that positive correlations between par-ents’ use of harsh physical discipline when chil-dren were 5 years old and children’s externalizing

behavior during kindergarten through sixth gradeas reported by teachers were considerably loweramong families characterized by high levels of pa-rental warmth and positive affect, compared tofamilies distinguished by low levels of parentalwarmth and positive affect. Although the moder-ation effect was not assessed in the conventionalform of a physical discipline by parental warmth/positive affect interaction effect, Deater-Deckardand Dodge’s analysis represents a strong test ofthe moderation hypothesis because it is based onprospective data and uses a reliable measure ofobserved parent-child warmth and multiple indi-cators of parental physical discipline. Extrapolat-ing from this finding, Deater-Deckard and Dodgespeculated that differences in the parenting con-text (e.g., prevalence and acceptability of physicalpunishment and parenting behaviors attendant tothis form of discipline) may account for evidencethat parents’ use of physical discipline is associ-ated with externalizing behavior problems amongEuropean American children but not among Af-rican American children. This race by physicalpunishment interaction was found when the mea-sure of externalizing behavior problems was basedon ratings from teachers and peers, but not whenit was based on maternal ratings (Deater-Deckard& Dodge, 1997; Deater-Deckard, Dodge, Bates,& Pettit, 1996).

McCord (1997) assessed parental warmth as amoderator of the effects of corporal punishmentin her sample of impoverished, urban boys fol-lowed over a period of four decades, but her out-comes were criminality and violence in adulthood,rather than child externalizing behavior. Corporalpunishment by fathers increased the likelihood ofcriminal behavior among sons, whereas corporalpunishment by mothers increased rates of vio-lence. Although maternal and paternal warmth re-duced the probability that sons would commit se-rious crimes, it had no effect on violence, nor didit moderate the effects of corporal punishment.Rarer still are studies that examine the effects ofthe context of corporal punishment on children’scognitive development. Using longitudinal datafrom the Infant Health and Development Program,a randomized clinical trial of low-birth-weight in-fants, Smith and Brooks-Gunn (1997) found thatpreschool girls who experienced high levels ofpunishment (i.e., hitting and scolding) in the con-text of low maternal warmth had significantlylower IQ scores than those who experienced lowlevels of punishment and high maternal warmth.However, these two groups did not differ from the

42 Journal of Marriage and Family

low punishment/low warmth group or the highpunishment/high warmth group.

Further evaluation of the moderation hypoth-esis is warranted in light of its prominence in thesocialization literature and the fact that direct testsof the hypothesis are sparse and have yielded con-flicting findings. Several features of the presentstudy permit a relatively stringent test of emotion-al support and warmth as moderators of the rela-tion between physical punishment and behaviorproblems: (a) we assess the relation between ma-ternal physical discipline and changes in child be-havior problems over time, (b) we directly testwhether maternal emotional support significantlyinteracts with physical discipline in its impact onchild behavior problems, (c) moderation effectsare examined within three different racial-ethnicgroups, and (d) moderation effects are estimatedafter introducing controls for important demo-graphic variables such as gender and income-to-needs ratios.

METHOD

Data and Sample

Data were a subset of the children of the NationalLongitudinal Survey of Youth (NLSY), an expan-sion of the fifth cohort of National LongitudinalSurveys of Labor Market Experience. Of the5,828 women who were originally sampled in1979 as part of this survey, 3,053 were identifiedas having had children (n 5 5,236) by the 1988round of surveys, which constitutes the base yearof this study. These children make up the poolfrom which the analytic sample of this study isdrawn. It should be noted that this sample is nota national representative sample of children.

Our analysis focuses on change in the BehaviorProblems Index (BPI), which was administered in1988, 1990, 1992, and 1994. This instrument canbe used with children as young as 4 years old, soour sample selection criterion for this study wasa child who was 4 in 1988. This decision gave usan analytic sample of the younger children of theNLSY sample, namely the 1,990 children whowere 2–3 years old in 1986 (and hence were 4–5years old in 1988 and 10–11 years old in 1994).The sample consists of 401 Hispanic, 550 AfricanAmerican, and 1,039 European American chil-dren. Roughly half of the children are girls (48%of the Hispanics, 52% of the African Americans,and 50% of the European Americans).

Measures

Behavior problems. The BPI was developed byPeterson and Zill (1986) using items drawn fromthe Child Behavior Checklist and other behaviorproblem checklists. Maternal reports on itemsabout the child’s behavior (e.g., ‘‘Child clings toomuch to adults’’) comprise the full scale. Al-though this scale has classically been divided intosubscales measuring internalizing and externaliz-ing problems (the subscales are correlated at r 5.70), the full scale has better reliability (.86) andmore interpretative clarity (Mott, 1998). Ratherthan argue for conceptual differences, we focuson the full scale to indicate level of behavior prob-lems in total.

In order to use this scale in a longitudinal anal-ysis, it has been modified so that the unit of anal-ysis corresponds to each child’s percentage on thetotal scale for each year. Growth curve modelingrequires comparable measures across each timepoint so that change from one time to the next hasmeaning. Thus, it was necessary to adjust thiscomposite measure so that the units had the samemeaning across time. To do this, we formed thefinal outcome measure in three steps: (a) The totalfor each child for each year was calculated. (b)This total was divided by the total possible foreach year’s scale and then multiplied by 100. Thiscomputation establishes a position or ‘‘level’’ ofbehavior problems for each child in each year,which can then be compared across years. (c) Wesubtracted the average level for each racial-ethnicgroup in 1988 (base year) from every score. Thisthird calculation translates every child’s score sothat it is ‘‘centered’’ at the group mean, givingeach group the same starting point for compari-son. This centering does not change the size ofthe metric, only the point of comparison.

Spanking. In 1988, interviewers noted whether ornot the mother hit the child during the course ofthe home observation. In subsequent years, thisobservation was not made. However, in all years,the survey items included an open-ended questionthat asked the mother to report how many timesshe spanked the focal child in the past week.These questions make up the measure of spankingused to predict children’s behavior problems. Thedata for 1988 generated a four-level ordinal scaleof spanking. At the bottom of the scale (coded 0)are mothers who reported not spanking their childin the given week of 1988. At the next level (cod-ed 1) are mothers who reported spanking their


child once. Neither of these groups had any moth-ers who were observed hitting their child duringthe interview. At the next level (coded 2) aremothers who reported spanking more than oncebut who were not observed hitting their child dur-ing the interview. At the top level (coded 3) aremothers who reported spanking more than onceand were observed hitting their child during theinterview. For each of the subsequent years (1990,1992, 1994), amount of spanking reported is as-sessed as a three-level ordinal scale (i.e., child notspanked, spanked once, or spanked more thanonce during week).

The 1988 measure of spanking is used in ourexamination of spanking-related differences inproblem behavior during the base year. However,to capture the impact of spanking on change inbehavior problems over time, we use hierarchicallinear modeling (HLM) techniques, with spankingas the outcome, to create two measures of the‘‘spanking environment’’ of the child. One mea-sure, the Baysian estimate of the overall intercept,measures the average amount the child wasspanked per week, experienced over 6 years. Forexample, one child might have been spanked anaverage of once a week, another four times aweek, another not at all. The second measure, theBaysian estimate of the slope, estimates thechange in spanking experienced as the child aged.For example, if a child was spanked more at 4years of age than at 10 years of age, this measurewould be negative, whereas if a child was spankedmore as he or she got older, this measure wouldbe positive. If the amount of spanking stayed thesame over 6 years, this measure would have avalue of 0. These two measures are used as pre-dictors in examining the change in behavior prob-lems over the 6 years studied. This use of HLMto create child-level measures reduces the auto-correlation and error misestimation associatedwith simple averages. A full description of themethod can be found in Bryk and Raudenbush(1992).

Emotional support. The measure of maternal emo-tional support is based on five items from theHome Observation for Measurement of the En-vironment (HOME; Caldwell & Bradley, 1980).The HOME observations were conducted duringinterviewer visits in each year of the study. Theitems concern the amount of warmth and emo-tional support displayed toward the child by themother over the course of the observation as as-sessed by the interviewer (e.g., mother caressed,

kissed, or hugged child during visit; conveyedpositive feeling about child; answered child’squestions or requests verbally; encouraged childto contribute to the conversation). The scale wasinternally standardized by NLSY to be compara-ble across time as the child aged.

As was done with amount of spanking, the1988 emotional support scale was used as a cov-ariate for the baseline behavior problem level. Inaddition, we constructed (using HLM analysiswith emotional support as the outcome) an esti-mate for each child of the overall average emo-tional support over the 6 years, and the change insupport experienced by that child over the 6 yearsof the study. For example, one child pulled fromthe sample had a comparatively high overall levelof emotional support over the 6 years (AverageSupport 5 1.62 SD) and very little change overthat time (Change 5 0.02 SD). Another child hadan average level of support across the 6 years (Av-erage Support 5 0.05 SD), but it dropped sub-stantially over that time (Change 5 21.2 SD).

Controls. Because of gender differences in behav-ior problems, gender is taken into account in allanalyses. Similarly, although the sample cohortwas initially selected within a limited age span,differences could still emerge related to age inmonths. Thus age (in months) was also controlledin all initial analyses. However, unlike gender, agewas not a significant predictor in any analysis andwas subsequently dropped from the final analyses.

Controls also were introduced for family in-come. We derived an income-to-needs ratio by di-viding the family’s total income by the officialpoverty threshold for that year. (The poverty levelfor each child comes from NLSY and is drawnfrom the federal level declared for that year ad-justed for family size.) This ratio estimates familyincome in a metric that takes poverty as its criticalreferent. For example, a ratio of 1 indicates thatthe family’s income is exactly at the poverty levelfor that year, whereas a ratio of 2 indicates thatthe family has income twice that of the povertythreshold. As was done with amount of spanking,the 1988 income-needs ratio was used as a cov-ariate for the baseline behavior problem level.Similarly, to capture the family’s economic well-being over the 6-year period of the study, HLMwas used to provide an estimate for both the av-erage level of the income-to-needs ratio over thisperiod and the change in income-to-needs ratioexperienced by families during this time.


Analytic Method

We analyzed these data in stages, building fromsimple bivariate comparisons to full multivariateanalysis of change. We use one-way analysis ofvariance (ANOVA) to examine racial-ethnic dif-ferences in the BPI (item level and composite),environmental conditions, and controls. In theseanalyses, we have adjusted the significance levelbecause of multiple tests, so that only those dif-ferences found at a significance level of p , .001are reported. In addition, we examine the unad-justed correlations among all measures used in theanalyses to provide some sense of how these fac-tors may be operating in tandem. It should be keptin mind, however, that these latter relations are notnecessarily sustained in the multivariate context(precisely because of some of the multicollinearityinvolved between predictors).

The analyses of behavior problems are all donewith growth curve analysis using HLM. This ana-lytic method estimates a change function for eachchild by nesting individual measures within person.It then uses parameters of that change function—the ‘‘intercept’’ and ‘‘slope’’ of each child’s line—as a person-level outcome on which characteristicssuch as gender, spanking, family income, and emo-tional support can be explored. This type of two-level model is analytically appropriate for lookingat nested data structures, such as this case ofchange in behavior over time nested within person.A more complete discussion of this analytic meth-od can be found in Bryk and Raudenbush (1992).We provide only an overview here.

HLM examines variance within and betweenpersons, similar to the way that repeated measuresANOVA does. However, it does this by runningwhat amounts to a small regression equation foreach person, of the form Y 5 b0 1 b1(Time) 1E. When the outcome of this regression is behavior,b0 estimates each individual child’s behavior levelat Time 0 (1988), and b1 measures the change inthat child’s behavior over time (from 1988 to1994). These two parameters are conditionally re-lated, in that each is adjusted for the other.

These two adjusted estimates about a child’sbehavior are then lifted to the next level of theanalysis to be the outcomes on which other childcharacteristics are predicted. For example, theHLM equation looking at individual differencesin initial behavior would take the form:

b0 5 g00 1 g01(Female)

1 g02(Spanking in 1988) 1 etc.

Similarly, the equation looking at change in be-havior would take the form:

b1 5 g10 1 g11(Female)

1 g12(Average Level of Spanking)

1 etc.

Each parameter (termed gamma to distinguishfrom the betas in Level 1) provides an estimate ofthe effect of that variable. For example, g01 pro-vides an estimate for the difference in initial be-havior problems between boys and girls, whereasg11 estimates the gender difference in change inbehavior over time. Using this method, we canexamine individual differences related to both ini-tial behavior (b0) and change in behavior overtime (b1) without the problems of autocorrelationamong common measures, which confounds mostrepeated measures analyses (for further discussionof this issue, see Collins & Horn, 1991).

A common way to address the causal directiondifficulty is to predict an outcome by a measurepreceding that outcome, for example, predictingchild behavior problems in 1990 by spanking in1988. We use a variant of this process, but theproblem is more complicated because both behav-ior problems and spanking are changing overtime. We address this problem using a lagged-change estimation process, in which the changefunctions for both behavior (described previously)and spanking are estimated over time, with thespanking change assessed at the time point priorto the behavior problem estimates. This strategyfor examining cross-estimated change functions isdescribed more completely in Duncan and Rau-denbush (1999).

We conducted our analysis in two stages, de-signed first to identify the impact of spankingwithout controls, and second to examine thechange in impact after controlling for environ-mental conditions (i.e., family income and mater-nal emotional support). In both stages, we exam-ined results separately for Hispanic, AfricanAmerican, and European American subsamples.

RESULTS

Racial and Ethnic Differences

Table 1 shows comparisons for the critical predic-tor variables (spanking, poverty, and maternalwarmth) for each year broken down by racial-eth-nic group, with differences between groups tested


TABLE 1. RACIAL-ETHNIC DIFFERENCES IN SPANKING, INCOME-NEEDS RATIO AND MATERNAL EMOTIONAL SUPPORT

OVER THE 6 YEARS STUDIED

Variable HispanicAfrican

AmericanEuropeanAmerican

1988 levelsAmount spanked (M)

Percentage not spankedPercentage spanked once last week, not observed hitPercentage spanked more than once last week, not observed hitPercentage spanked more than once last week, observed hit

.69a

48a

2419a

10

.91b

36b

2627b

11

.65a

49a

2814a

10Family income-needs ratio (M)Maternal emotional support scale (M)

1.74a

87.5a

1.48b

76.6b

2.18c

90.3a

1988–1994 average levelsAmount of spanking per week (M)Family income-needs ratio (M)Maternal emotional support scale (M)

.58a

1.96a

103.6a

.97b

1.49b

93.1b

.50a

2.37c

107.7a

1988–1994 change in levelsChange in spanking (M)

Percentage whose amount of spanking declinedPercentage whose amount of spanking increasedPercentage whose amount did not change

2.12a

24a

24a

52a

2.25b

41b

18a

41b

2.09a

17c

44b

39b

Change in family income-needs ratio (M)Percentage who stayed in povertyPercentage who dropped into povertyPercentage who rose out of povertyPercentage never in poverty

.17a

23a

9a

13a

55a

2.05b

35b

11a

12a

42b

.16a

10c

5b

7b

78c

Change in maternal emotional support scale (M)Percentage whose support decreasedPercentage whose support increasedPercentage whose support stayed constant

14.1234730

15.5224830

13.6185131

Note: Values with unique subscripts are significantly different, p , .001.

by one-way ANOVA (for continuous variables)and contingency table analysis (for categoricalvariables). In 1988, fewer African American chil-dren were ‘‘never hit’’ and more were spankedmore than once a week than European Americanor Hispanic children. Interestingly, whereas Eu-ropean American mothers increased their amountof spanking from 1988 to 1994 (44% reported anincrease), more African American children expe-rienced a decline in spanking over the 6 years(41%), and only 18% experienced an increase.Hispanic children experienced change as well, butthe same portion experienced a drop in spankingas experienced an increase (24%).

In addition, in 1988, African Americans had alower family income-needs ratio and lower levelsof emotional support than did Hispanics or Eu-ropean Americans. Hispanics had a lower averageincome-needs ratio than European Americans, butHispanics were similar to European Americans inlevels of emotional support. Similar patterns arepresent for the average levels estimated from1988–1994. In general, environmental advantageappears more concentrated among European

American families, even though the original sam-pling of the NLSY targeted disadvantaged womenof each racial-ethnic group.

Bivariate Relationships AmongCovariates and Outcomes

Table 2 shows a correlation matrix that relateseach measure examined in the HLM analyses.These comparisons allow one to consider a varietyof relations between covariates used in these anal-yses. For example, the relation between spankinglevel and emotional support in 1988 was moder-ately negative (r 5 2.23), whereas the relationbetween income-needs ratio and emotional sup-port in 1988 was moderately positive (r 5 .29).In general, relations among variables supportthose reported in other studies concerning covar-iates of spanking and behavior.

Net Impact of Spanking on Change inBehavior Problems

Table 3 shows the results of the first stage of HLManalyses, examining the impact of spanking on


TA

BL

E2.

CO

RR

EL

AT

ION

SA

MO

NG

VA

RIA

BL

ES

WIT

HIN

POO

LE

DSA

MP

LE

Chi

ld’s

Age

Am

ount

ofSp

anki

ng19

88

Cha

nge

inSp

anki

ng19

88–1

994

Inco

me-

Nee

dsR

atio

1988

Cha

nge

inIn

com

e-N

eeds

1988

–199

4E

mot

iona

lSu

ppor

t19

88

Cha

nge

inE

mot

iona

lSu

ppor

t19

88–1

994

Beh

avio

rPr

oble

ms

1988

Cha

nge

inB

ehav

ior

Prob

lem

s19

88–1

994

Fem

ale

Chi

ld’s

age

Am

ount

ofsp

anki

ng19

88C

hang

ein

span

king

1988

–199

4In

com

e-ne

eds

ratio

1988

2.0

31.

002

.06*

*2

.05*

*1.

00

.004

.01

2.2

3***

1.00

2.0

003

2.0

32

.10*

**2

.01

1.00

.002

2.0

22

.09*

**2

.01

2.1

5***

.07*

*.2

4***

2.2

3***

.01

.29*

**

2.0

12

.13*

**.0

9***

2.0

6*2

.05

2.1

1***

.03

.46*

**2

.005

2.1

4***

.02

2.0

42

.06*

.09*

*2

.02

Cha

nge

inin

com

e-ne

eds

ratio

1988

–199

41.

00.0

2.0

7**

2.0

7**

.01

Em

otio

nal

supp

ort

1988

1.00

2.5

0***

2.1

8***

2.0

2C

hang

ein

emot

iona

lsu

ppor

t19

88–1

994

1.00

.01

2.0

9***

Beh

avio

rpr

oble

ms

1988

1.00

2.3

3***

*p

,.0

5.**

p,

.01.

***

p,

.001

.

change in behavior over time after taking onlygender into account. These analyses were con-ducted independently on the Hispanic, AfricanAmerican, and European American subsamples,as well as on the pooled sample. As described inthe Method section, the numbers in this table arethe gamma estimates of the impact of these child-level predictors on the initial level of behaviorproblems (b0), shown in the first half of the table,and the change in behavior over time (b1), shownin the second half of the table. Because a valueof 0 in our converted construction of the BPI cor-responds to the group mean, the negative interceptfor this table indicates that the comparison group(male, not hit) is slightly below the overall groupmean for 1988.

The first part of Table 3 shows that the amountthat a child was hit or spanked in 1988 is unre-lated to the level of behavior problems reportedin 1988, after taking into account (a) the overalllevel of behavior problems over time and (b) gen-der (g 5 2.08 for Hispanics, .02 for AfricanAmericans, and .01 for European Americans).This result differs substantially from the correla-tion shown in Table 2 (r 5 .46) and points outthe critical need to examine relations that changeover time in a way that appropriately apportionsthe within- and between-persons variance. This re-sult should be interpreted as a lack of relationshipbetween spanking and initial level of problems,relative to the overall level per person. In otherwords, when one looks at the 1988 level of be-havior problems in the context of an individualperson’s starting point, the amount that person wasspanked at the time is unrelated to comparativelyhigh or low levels. There are some comparativelylow-problem-level children who are getting hitfairly often and some high-problem-level childrenwho are not getting hit. In this baseline compari-son, only gender makes a significant difference,with parents reporting lower levels of problem be-havior for girls compared to boys.

However, being spanked clearly increases thelevel of problem behavior over time. The secondpart of Table 3 shows that all groups had an in-crease in behavior problems over the 6-year pe-riod (g 5 1.35 for Hispanics, 1.27 for AfricanAmericans, and 1.54 for European Americans).These gains did not differ by gender. However,children who experienced more spanking duringthis time period had a larger increase in behaviorproblems (g 5 1.97 more for Hispanics, 1.55 forAfrican Americans, and 1.97 for European Amer-icans). This effect was estimated independently


TABLE 3. HIERARCHICAL LINEAR MODEL RESULTS ON THE NET IMPACT OF BEING HIT ON CHANGE IN BEHAVIOR

PROBLEMS FOR EACH RACIAL-ETHNIC GROUP

HispanicAfrican


PooledSample

Average behavior problem level, 1988FemaleAmount hit in 1988

2.6523.17a

2.08

2.3822.83a

.02

2.7822.37a

.01

2.5922.71b

.01Average change in behavior problems, 1988–1994

FemaleAverage level of spanking, 1988–1994Change in level of spanking, 1988–1994

1.35a

.321.97a

1.14a

1.27a

.601.55a

.98a

1.54a

2.091.97a

.98a

1.40b

.261.80b

1.01b

aSignificant effect within racial-ethnic group, p , .001. b Significant effect within pooled sample, p , .001.

FIGURE 1. IMPACT OF AMOUNT AND CHANGE IN SPANKING ON CHANGE IN BEHAVIOR PROBLEMS FROM 1988 TO 1994FOR POOLED SAMPLE OF CHILDREN

from the impact of changing the amount of spank-ing, which was also a factor. Children who ex-perienced an increase in spanking also had greaterincreases in behavior problems (g 5 1.14 for His-panics, 0.98 for African Americans, and 0.98 forEuropean Americans). Because these effects areadditive, the most ‘‘at-risk’’ child (according tothis analysis) would be one who had both highlevels of spanking and experienced an increase inspanking from 1988 to 1994. Using the pooledsample estimates, the average change in behaviorproblems for such a child would be 1.40 (the basechange) plus 1.80 (the effect for higher level ofspanking) plus 1.01 (the effect for the spankingincrease), or an increase of 4.21 per year. Giventhat the overall standard deviation on this index is8.71, this increase represents a change of abouthalf a standard deviation every year.

Figure 1 shows a graphic representation ofthese results, drawn from the results in Table 3.We highlight the effects by focusing on two sub-samples from the full analysis. The first panel inFigure 1 represents the estimated impact of theamount of spanking on change in behavior prob-lems, focusing on the subsample of children forwhom spanking remained constant over the 6-yearperiod (about 42% of the overall sample). Thispanel shows that, whereas all parents report someincrease in level of behavior problems over thisperiod, those children who experience higher lev-els of spanking on a weekly basis have a muchgreater increase over time.

The second panel in Figure 1 shows the esti-mated impact of change in the amount of spankingexperienced by children over 6 years, this timeusing the subsample of children who were


TABLE 4. HIERARCHICAL LINEAR MODEL RESULTS ON THE PREDICTORS OF GROWTH DIFFERENCES IN BEHAVIOR

PROBLEMS OVER TIME FOR EACH RACIAL-ETHNIC GROUP

HispanicAfrican


PooledSample

Average behavior problem level, 1988FemaleAmount hit in 1988Income-needs ratio in 1988c

Level of emotional support in 1988d

.0521.99a

.262.82a

22.61a

2.7521.65a

.282.14a

24.71a

.0221.43a

.202.71a

24.69a

2.0521.63b

.252.67b

24.02b

Average change in behavior problems, 1988–1994FemaleAverage level of spanking, 1988–1994Change in level of spanking, 1988–1994Income-needs ratio, 1988–1994a

Change in income-needs ratio, 1988–1994Level of emotional support, 1988–1994d

Change in emotional support, 1988–1994

2.25a

.08

.85a

.62a

2.092.75a

21.25a

2.65a

2.03a

.48a

.72a

.54a

.042.62a

21.19a

2.69a

2.19a

2.12.90a

.57a

2.062.65a

2.262.66a

2.11b

.16

.82b

.57b

2.032.66b

2.83b

2.67b

Interaction between emotional supportand level of spanking, 1988–1994 2.62a 2.68a 2.35a 2.54b

aSignificant effect within racial-ethnic group, p , .001. b Significant effect within pooled sample, p , .001. c This measureis in original metric, where 1 5 income exactly at poverty level but has been centered to the sample mean. Thus, 0 indicatesthe average income-needs ratio for that sample, and 1 unit above 0 corresponds to an increase in income above the samplemean of the amount of designated poverty level. d This measure was z-scored to the sample mean and standard deviation.

spanked once a week or less in 1988 (about 25%of the overall sample). Note that the middle linein both panels is the same group. The comparisonin the second panel shows that, in addition to theincrease in behavior problems experienced by allchildren at this level, there is added an impact onbehavior problems related to change in amount ofspanking. Compared to children for whom spank-ing remained constant over the 6-year period, chil-dren who experienced a decrease in amount ofspanking over time have a smaller increase in be-havior problems, whereas those who experiencedan increase in amount of spanking have a greaterincrease in behavior problems.

Three critical conclusions can be drawn fromthese results. First, children who experience morespanking on average have a greater increase inbehavior problems over time. Second, when theamount of spanking experienced decreases as thechild ages, the impact on the growth of behaviorproblems is blunted, whereas an increasingamount of spanking appears to accelerate thegrowth of behavior problems. Finally, when look-ing at the net impact of spanking on the devel-opment of behavior problems, there are no differ-ences related to race-ethnicity. The impact is thesame across all groups.

Combined Effects of Spanking and HomeEnvironment on Behavior Problems

Table 4 shows all of the main effects explored inthe HLM analysis separately for each racial-ethnic

group—namely, those of being spanked, the in-come-needs ratio for the relevant years, and thelevel of emotional support in each relevant year.Each column is a separate and independent sam-ple, and the main-effects HLM model is run iden-tically on each racial-ethnic group, as well as onthe pooled sample. Hence, it is appropriate tocompare effects within groups as well as to ex-amine between-group differences in both the basiccurve and the factors influencing change for eachgroup. We discuss each set of findings separately,although all are analyzed simultaneously. We ex-plored the possibility that each of the home en-vironment measures might interact with one ormore of the spanking measures. However, onlyone interaction—that between the average level ofemotional support and the average amount ofspanking from 1988 to 1994—was significant.This is the only interaction left in the final model.It is discussed in the section concerning the im-pact of emotional supportiveness.

Adjusted impact of spanking on change in behav-ior problems. Because the behavior compositewas centered to each group’s 1988 mean, the in-tercept in this main-effect model can be interpret-ed as the average deviation from the group meanin 1988 behavior problems observed for malechildren who reportedly were not hit in 1988 andhave an average 1988 income-needs ratio and lev-el of emotional support. Of the three groups, Af-rican American children have the lowest adjusted


level of behavior problems, followed by Hispanicsand then European Americans, although the dif-ference between the latter two groups is not sig-nificant. In addition, although the overall changein behavior problems is an average increase foreach racial-ethnic group, the change is slightlylower for African American children. In otherwords, after adjusting for spanking, income-needsratio, and level of emotional support, AfricanAmerican children have a lower average level ofbehavior problems and a smaller increase overtime.

The effect of spanking is fairly consistentacross racial-ethnic groups. Once again, we ob-serve no difference in behavior problems in 1988comparing children who were and were not hit orspanked. In addition, the patterns observed con-cerning the impact of amount and change inspanking over the 6-year period are the same asobserved previously, though slightly attenuated bythe other factors considered in the model. Notably,this relation continues to show no differences re-lated to racial-ethnic background. In short, evenafter one takes into account relative poverty andemotional support, spanking has a similar impacton the development of behavior problems overtime, regardless of a child’s racial-ethnic back-ground.

Income-needs ratio. As shown in Table 4, for bothEuropean American and Hispanic children thereis a strong relation between the income level ofthe family and the level of reported behavior prob-lems of the child during the base year, with moth-ers of children living in more affluent families re-porting lower levels of child behavior problems.There is an association between income level andbehavior problems for African American children,as well, but it is not nearly as large. Interestingly,once this initial impact is in place, the averagelevel of a family’s income-needs ratio over theperiod from 1988 to 1994 does not make a dif-ference in the change in behavior problems overthat time. This result suggests that the impact onbehavior is largely constant—children who arelower in problem behavior in 1988 maintain thatlevel compared to children who are higher. Thisobservation holds only for those families whoseincome remained essentially constant.

On the other hand, children who experienceda drop in income-needs ratio in their familyshowed a corresponding rise in level of problembehavior, and vice versa. To illustrate this relation,we focus on children who were not spanked

throughout the 6-year period, and compare thechange in behavior problems for those who startedat their sample’s average income-need ratio anddropped to those who started at the average androse. These patterns are shown for each racial-ethnic group in Figure 2. In general, change inincome level is related to a change in behaviorproblems, after taking spanking and emotionalsupport into account. The overall impact is slight-ly smaller for African American children than foreither Hispanic or European American children.

Maternal emotional support. There are several im-portant observations to be made about the resultsconcerning the impact of emotional support on be-havior problems. When considering the cross-sec-tional results from 1988 (base year), there is anegative relation between support and behaviorproblems. As shown in Table 4, children who re-ceive more emotional support from their mothershave fewer reported problems. This relation isstronger for African American and EuropeanAmerican children (g 5 24.71 and 24.69, re-spectively) than it is for Hispanic children (g 522.61). In addition, for two of the groups, thereis a benefit gained by emotional supportiveness onthe development of behavior problems over time.Unlike what was observed with family income,supportiveness has additional impact on thechange in African American and Hispanic chil-dren’s behavior, contributing to a smaller increaseover time (g 5 21.19 and 21.25, respectively).This relation is not observed for European Amer-ican children, however (g 5 2.26). Similarly,when emotional support increases, it has a buff-ering effect on the development of problems,whereas a drop in support has an additional im-pact by increasing the level of behavior problemsover these years. This result is consistent acrossthe three racial-ethnic groups (g 5 2.65, 2.69,and 2.66 for Hispanic, African American, andEuropean American children, respectively) andsuggests that adding emotional support to the fam-ily context is always beneficial, no matter the ageof the child or the timing involved.

Finally, and most critical to this investigation,we found a significant interaction effect betweenemotional support and level of spanking over the6-year period for each racial-ethnic group (Table4). Consistent with prediction, emotional supportmoderated the impact of spanking. Figure 3 showsfor each racial-ethnic group a comparison betweenchildren who were not spanked and those whowere spanked more than once a week in low-sup-


FIGURE 2. EFFECT OF CHANGE IN FAMILY INCOME ON CHANGE IN CHILDREN’S LEVEL

OF BEHAVIOR PROBLEMS OVER TIME

portive (1 or more SDs below average) and high-supportive (1 or more SDs above average) homeenvironments. Among children with low levels ofemotional support, both groups have a high levelof problems, and those getting spanked have agreater increase in behavior problems. Althoughthe pattern is the same for all three groups, therelation between spanking and increase in behav-ior problems is strongest for Hispanic children andweakest for European American children. Con-versely, among children with high levels of emo-tional support, there is almost no relation betweenspanking and change in behavior problems overtime.

In general, these results suggest strongly that,although spanking can have a negative impact onchildren’s socioemotional functioning over time,this effect is moderated by the emotional contextin which such spanking occurs. When spankingoccurs in a context of strong overall emotionalsupport for the child, it does not appear to con-tribute to a significant increase in behavior prob-lems. However, without this support in place, be-havior problems tend to increase in response toincreases in spanking.

DISCUSSION

In keeping with a host of studies reporting nega-tive associations between various forms of coer-cive control by parents and socioemotional ad-justment in children (McCord, 1997; Rohner,Kean, & Cournoyer, 1991; Strassberg et al., 1994;

Straus et al., 1997), the present study found thatspanking predicted increases in problem behaviorover the 6-year period after controlling for gender,income-need ratio, and maternal emotional sup-port. Although African American children ascompared to European American and Hispanicchildren were more likely to be spanked and werespanked more frequently—a finding that concurswith earlier studies (Day et al., 1998; Deater-Deckard et al., 1996)—we found no evidence thatthe relation between spanking and behavior prob-lems is related to race or ethnicity. In contrast tothe present findings, some research indicates thatspanking is predictive of externalizing problemsamong European American children, but not Af-rican American children (Deater-Deckard et al.,1996; Gunnoe & Mariner, 1997). Research isneeded to determine if this pattern of conflictingfindings is due to differences among studies in thesource of information about the child’s behaviorproblems (e.g., mothers, teachers, peers), in themeasure of spanking, or in the extent to whichanalyses take account of the impact of children’sexternalizing behavior on mothers’ tendency tospank (McLeod, Kruttschnitt, & Dornfeld, 1994).

The present findings indicate that the effects ofspanking on problem behavior depend partly onwhether spanking is administered in the contextof high emotional support of the child. Our con-fidence in the reliability of this moderation effectis buttressed by the fact that it was replicated inseparate analyses based on three different racial-


FIGURE 3. INTERACTION BETWEEN AMOUNT OF SPANKING AND HIGH AND LOW MATERNAL SUPPORT ON CHANGE IN

BEHAVIOR PROBLEMS OVER TIME

ethnic groups. Emotional support may moderatethe impact of spanking by influencing the child’sinterpretation of physical discipline. The childmay be less likely to view spanking as harsh, un-

just, and indicative of parental rejection when re-lations with the parent are generally warm andsupportive. Previous research indicates that theimpact of physical punishment on children’s psy-


chological adjustment is partly mediated throughperceived parental rejection (Rohner et al., 1996;Rohner et al., 1991). It is also reasonable to as-sume that hostile attributional biases about the in-tentions of others—one of the psychological fac-tors underlying aggressive behavior (Weiss,Dodge, Bates, & Pettit, 1992)—may also be lesslikely to develop when physical punishment ismeted out in the context of positive parent-childrelations. Notwithstanding the moderating effectof maternal warmth and emotional support, thefindings of the present study do not negate someof the reasons cited by scholars for discouragingparental reliance on physical punishment. For ex-ample, as Simons et al. (1994) point out, whenparents use physical punishment, there is alwaysthe possibility, however miniscule, that the en-counter will escalate to the point that excessiveforce is used and the child is physically injured.Furthermore, physical punishment typicallyevokes anger and emotional distress in the child,which, over time, may diminish positive feelings.

Several limitations of the present study need tobe acknowledged. First, our measure of childproblem behavior is based on maternal report, andas such, it may not be a good indicator of problembehavior in the school setting or other settings be-yond the home. Second, the fact that mothers wereinformants about both spanking and child problembehavior may have inflated the relation betweenthese variables. Third, our measure of spankingdoes not incorporate information about intensity,timing, or contingencies of spanking. It is impos-sible to know whether ‘‘spanking’’ as adminis-tered by some parents constituted excessive phys-ical force that falls outside the normative range ofsocialization practices. It may be that low mater-nal support covaries with more extreme forms ofphysical discipline not captured in our measure.In addition, our measure does not reflect the extentto which spanking episodes are or are not preced-ed by the use of reason to gain the child’s com-pliance (e.g., reasoning), a factor that may mod-erate the impact of reasoning and physicaldiscipline alike (Larzelere, Sather, Schneider, Lar-son, & Pike, 1998). Finally, our measure of spank-ing is limited by the possibility that some parentswho spanked their children regularly did not havethe occasion to do so during the previous weekand by the possibility that parents who typicallydid not use physical discipline spanked their chil-dren during the previous week (Day et al., 1998).It is likely that the first bias is balanced by the

second bias, but we cannot determine this withany certitude.

Longitudinal research that relies on multipleinformants about child behavior problems anduses a variety of methodologies to capture reliabledifferences in the manner in which parents ad-minister physical discipline would be especiallyvaluable. We did not systematically explore dif-ferences in the antecedents of internalizing versusexternalizing behavior, because these two domainsof behavior were highly correlated. Nonetheless,this is potentially an important line of inquiry thatmay provide insight into some intriguing ques-tions. As one example, the relative contribution ofspanking to internalizing versus externalizing be-havior may depend on the child’s temperamentand degree of identification with the discipliningparent.

NOTE

Support for this research was provided by the Depart-ment of Psychology and Center for Human Growth andDevelopment at the University of Michigan. The au-thors express sincere appreciation to Sheba Shakir andAutumn Kelly for their assistance in the preparation ofthis manuscript.

REFERENCES

Baumrind, D. (1997). Necessary distinctions. Psycho-logical Inquiry, 8, 176–229.

Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchicallinear models: Applications and data analysis meth-ods. Newbury Park, CA: Sage.

Caldwell, B. M., & Bradley, R. H. (1984). Home Ob-servation for Measurement of the Environment. LittleRock, Arkansas: University of Arkansas.

Collins, L. M., & Horn, J. L. (Eds.). (1991). Best meth-ods for the analysis of change: Recent advances, un-answered questions, future directions. Washington,DC: American Psychological Association.

Day, R. D., Peterson, G. W., & McCracken, C. (1998).Predicting spanking of younger and older children bymothers and fathers. Journal of Marriage and theFamily, 60, 79–94.

Deater-Deckard, K., & Dodge, K. A. (1997). External-izing behavior problems and discipline revisited:Nonlinear effects and variation by culture, context,and gender. Psychological Inquiry, 8, 161–175.

Deater-Deckard, K., Dodge, K. A., Bates, J., & Pettit,G. (1996). Physical discipline among African Amer-ican and European American mothers: Links to chil-dren’s externalizing behaviors. Developmental Psy-chology, 32, 1065–1072.

Dodge, K., Pettit, G., Bates, J., & Valente, E. (1995).Social information processing patterns partially me-diate the effect of early physical abuse on later con-duct problems. Journal of Abnormal Psychology,104, 632–643.

Duncan, G. J., & Raudenbush, S. W. (1999). Assessing


the effects of context in studies of child and youthdevelopment. Educational Psychologist, 34, 29–41.

Gils-Sims, J., Straus, M., & Sugarman, D. (1995).Child, maternal, and family characteristics associatedwith spanking. Family Relations, 44, 170–176.

Gunnoe, M. L., & Mariner, C. L. (1997). Toward a de-velopmental-contextual model of the effects of paren-tal spanking on children’s aggression. Archives of Pe-diatrics and Adolescent Medicine, 151, 768–775.

Larzelere, R. E. (1996). A review of the outcomes ofparental use of nonabusive or customary physicalpunishment. Pediatrics, 98, 824–831.

Larzelere, R. E., Klein, M., Schumm, W. A., & Ali-brando, S. (1989). Relations of spanking and otherparenting characteristics to self-esteem and perceivedfairness of parental discipline. Psychological Reports,64, 1140–1142.

Larzelere, R., Sather, P., Schneider, W., Larson, D., &Pike, P. (1998). Punishment enhances reasoning’s ef-fectiveness as a disciplinary response to toddlers.Journal of Marriage and the Family, 60, 388–403.

Loeber, R., & Stouthamer-Loeber, M. (1986). Familyfactors as correlates and predictors of juvenile con-duct problems and delinquency. In M. Tonry & N.Morris (Eds.), Crime and justice (Vol. 7, pp. 219–339). Chicago: University of Chicago Press.

McCord, J. (1997). On discipline. Psychological Inqui-ry, 8, 215–217.

McLeod, J., Kruttschnitt, C., & Dornfeld, M. (1994).Does parenting explain the effects of structural con-ditions on children’s antisocial behavior? A compar-ison of Blacks and Whites. Social Forces, 73, 575–604.

Mott, F. (1998). The NLSY children 1992: Descriptionand evaluation. Columbus: Center for Human Re-source Research, Ohio State University.

Peterson, J., & Zill, N. (1986). Marital disruptions, par-ent-child relationships and behavior problems in chil-

dren. Journal of Marriage and the Family, 48, 295–307.

Rohner, R. P., Bourque, S. L., & Elordi, C. A. (1996).Children’s perceptions of corporal punishment, care-taker acceptance, and psychological adjustment in apoor, biracial Southern community. Journal of Mar-riage and the Family, 58, 842–852.

Rohner, R. P., Kean, K. J., & Cournoyer, D. E. (1991).Effects of corporal punishment, perceived caretakerwarmth, and cultural beliefs on the psychological ad-justment of children in St. Kitts, West Indies. Journalof Marriage and the Family, 53, 681–693.

Simons, R., Johnson, C., & Conger, R. (1994). Harshcorporal punishment versus quality of parental in-volvement as an explanation of adolescent malad-justment. Journal of Marriage and the Family, 56,591–607.

Smith, J., & Brooks-Gunn, J. (1997). Correlates andconsequences of harsh discipline for young children.Archives of Pediatrics and Adolescent Medicine, 151,777–786.

Strassberg, Z., Dodge, K. A., Pettit, G. S., & Bates, J.E. (1994). Spanking in the home and children’s sub-sequent aggression toward kindergarten peers. Devel-opment and Psychopathology, 6, 445–461.

Straus, M., & Gelles, R. (1986). Societal change andchange in family violence from 1975 to 1985 as re-vealed by two national surveys. Journal of Marriageand the Family, 48, 465–479.

Straus, M., Sugarman, D., & Gils-Sims, J. (1997).Spanking by parents and subsequent antisocial be-havior of children. Archives of Pediatrics and Ado-lescent Medicine, 151, 761–767.

Weiss, B., Dodge, K. A., Bates, G. E., & Pettit, G. S.(1992). Some consequences of early harsh discipline:Child aggression and a maladaptive social informa-tion processing style. Child Development, 63, 1321–1335.

Physical Discipline and Behavior Problems in African American, European American, and Hispanic...

Documents

Transcript of Physical Discipline and Behavior Problems in African American, European American, and Hispanic...