The Cross-Ethnic Equivalence of Parenting and Family Interaction Measures among Hispanic and Anglo-American Families George P. Knight, Jenn Yun Tein, Rita Shell, and Mark Roosa Arizona State University KNIGHT, GEORGE P.; TEIN, JENN YUN; SHELL, RITA; and ROOSA, MARK. The Cross-Ethnic Equiva-

lence of Parenting and Family Interaction Measures among Hispanic and Anglo-American Families. CHILD DEVELOPMENT, 1992, 63, 1392—1403. Recently there has been concern over the need for developmental research within ethnic minority populations and interest in family processes within, and variability across, ethnic groups. Unfortunately, most of the research using standard scales of family processes has sampled middle-class Anglo-Americans, and the potential absence of cross-ethnic measurement equivalence threatens the validity of the research using these scales with ethnic minority populations. This study reports confirmatory factor analyses and construct validity coefficients for several parenting and family interaction scales among Anglo-American and Hispanic 8-14-year-old children and mothers. The findings indicate that the Children's Report of Parental Behavior Inventory (except the hostile control subscale), the ParentAdolescent Communication Scale (open communication subscale only), and the Family Adaptability and Cohesion Evaluation Scales II appear to have sufficient cross-ethnic equivalence for English-speaking Hispanic samples. Further, the Family Routines Inventory and the problem communication subscale could benefit from additional scale development.

Recently there bas been serious consideration of how cross-cultural, and perhaps cross-ethnic, research can be informative to researchers interested in mainstream developmental issues (see Rogoff & Morelli, 1989). In addition, the changing characteristics of the U.S. population, particularly the growing Hispanic and black populations, have led to an understanding of the need for developmental research within other than white, middle-class samples (see Somerville, 1991). One developmental research area for which there has been considerable cross-cultural and cross-ethnic research is the investigation of socialization and family processes. There has also been a rapidly growing interest in socialization and family processes within Hispanic families (e.g., Keefe & Padilla, 1987; McGlintock, Bayard, & McGlintock, 1983; Vega, 1990). Even so, it is apparent that our understanding of socialization and family processes in Hispanic families, and the cross-etbnic differences

and similarities in these processes, is very limited (Staples & Miranda, 1980; Vega, 1990), perbaps because of difficulties associated with adapting or developing measures for use with Hispanic populations. That is, studies of socialization and family processes among Hispanics or comparative studies involving Hispanics may have been limited by the need for valid and cross-ethnically equivalent measures, Indeed, the research on the socialization and family processes in Hispanic families has produced somewhat inconsistent evidence regarding these processes. For example, Hispanic parents have often been empirically described as relatively warm, nurturant, egalitarian, and family oriented (e.g., Keefe, 1984; Mindel, 1980; Staples & Mirande, 1980; Vega, 1990). In contrast, some researchers have suggested that differential acculturation rates between parents and children lead to profound intergenera-

This study was funded by the National Institute for Mental Health grant 2-P50-MH39246-06 to support the Preventive Intervention Research Center at Arizona State University. The authors gratefully acknowledge the contributions made by Carolyn Berg, Marya Cota, Camille Garza, Katheryn Ocampo, Joanne Gersten, Nancy Groppenbacher, Lisa Jack, Lillian Lengua, Marcia Michaels, and Irwin Sandier. We also thank the two reviewers for providing very helpful comments. Requests for reprints may be directed to George P. Knight, Department of Psychology, Arizona State University, Tempe, AZ 85287. [Child Development, 1992,63,1392-1403. © 1992 by the Society for Research in Child Development, Inc. All rights reserved. 0009-3920/92/6306-0001$01.00]

Knight et al. tional conflict (Szapocznik, & Hernandez, 1988), and that the Hispanic family is punitive and authoritarian (Fromm & Maccoby, 1970; Minturn & Lambert, 1964; Rusemore & Kirmeyer, 1976). Further, Maslow and Diaz-Guerrero (1971) suggest that the authoritarian style of the Hispanic family may provide children with a greater sense of structure and security because of the clear norms these parents provide, thereby suggesting that specific behaviors may well be interpreted differently or have quite different functional meanings in different cultures. Vega (1990) has suggested that the inconsistencies in the findings may be due, in part, to the inadequacies of some of the measures used and/or nonequivalent measures across samples. Some of the research focusing on crosscultural or cross-ethnic differences in parenting and family interaction variables relies on interviews and/or standard scales (e.g., Lin & Fu, 1990; Okagaki & Sternberg, 1991) and is generally based on the assumption that these measures are cross-culturally or cross-ethnically equivalent. Once again, however, the issue of measurement equivalence is particularly important given the evidence of the variability in the culturally prescribed roles and practices associated with the family and in family structures (see, Keefe & Padilla, 1987; McClintock et al., 1983; and Vega, 1990). If the parenting and family interaction measures used in this cross-cultural and cross-ethnic research are not assessing the same underlying constructs across groups, then our understanding of the role of culture in socialization and the cultural/ethnic differences in parenting and family processes is limited. Similarly, recent research focusing on the prevention of mental health problems among children has often investigated parenting and family interaction variables and often hypothesized one of two causal roles regarding the impact of these variables on the mental health of children. Some researchers (e.g., Patterson & StouthamerLoeber, 1984; Robertson & Simons, 1989; Rutter, 1989; Wahler & Dumas, 1987; Wolchik, Ruehlman, Braver, & Sandier, 1989) have suggested that parenting and family experiences have a direct causal impact on mental health problems of children and mediate the influence of family life stressors (e.g., parental alcoholism, divorce, etc.) on the mental health of children. Some researchers (e.g., Sameroff, Seifer, Baldwin, & Baldwin, 1989; Wolchik et al., 1989) have

1393

also suggested that parenting and family variables moderate the impact of stressful life events on the mental health of children. These prevention researchers have necessarily sampled high-risk populations, which very often include substantial proportions of individuals from ethnic and racial minority backgrounds (e.g., Roosa, Gensheimer, Short, Ayers, & Shell, 1989). Given the nature of the existing psychometric evidence regarding the reliability and validity of the parenting and family interaction measures used in this research, the potential lack of measurement equivalence threatens the validity of the inferences generated. A number of different forms of measurement equivalence have been considered in the literature. Hui and Triandis (1985) have organized these different forms of equivalence into four basic types. Functional equivalence is evident when scores on a measure of a given construct have similar precursors, consequents, and correlates across ethnic groups. Equivalence of construct operationalization is evident when a construct is operationalized in the same procedure across ethnic groups, and this operationalization is equally meaningful across ethnic groups. Item equivalence is evident when each item has the same meaning across ethnic groups. And scalar equivalence is evident when a given score on the measure refers to the same degree, intensity, or magnitude of the construct across ethnic groups. Hui and Triandis (1985) have also provided a brief review of the strategies researchers have used to examine these types of measurement equivalence. The present study was designed to examine the item and functional cross-ethnic equivalence of several parenting and family interaction measures among Anglo-American and Hispanic samples by examining the internal structure congruence with crossgroup equivalence confirmatory factor analyses, and the comparison of the similarity of the interrelations among constructs across groups using structural equation modeling analyses. Item equivalence is not fully addressed in the present studies primarily because of the difficulty associated with obtaining Hispanic samples large enough for item response or response pattern analyses. The particular survey measures reported on here were used because the assessments were available from a larger project investigating family influences on children of alcoholics. How^ever, it is likely that the recent emphasis on obtaining normative data for

1394

Child Development

minority populations and the shift toward prevention strategies in child clinical research will result in an increasing reliance on these kinds of survey measures relative to observational measures.

to evaluate each item for cultural relevance (i.e., the degree to which the behaviors and attitudes reflected in the items were applicable in the Hispanic culture). In the second sitting, each of the panel members received a list of items from the CRPBI, PAC, FACES II, and FRI in a random order. Panel memPreliminary Item Evaluation bers were required to identify the underlyPrior to the study, there ^vas a three- ing construct assessed by each item in a mulphase examination of the parenting and fam- tiple choice format. In the third sitting, each ily interaction measures by small panels panel member was given a list of the items (ranging from four to seven persons) of fe- for each construct and asked to evaluate the males with some training in measurement underlying construct being measured by who were also members of an Hispanic cul- each set of items, first in a free-report format ture. The four inventories of parenting and and then in a multiple-choice format. Panel family variables evaluated by these expert members were then told the theoretically panels were.- (1) the Children's Report of Pa- correct variable label for each set of items rental Behavior Inventory (CRPBI), (2) the and asked, "Do you think that this set of Parent-Adolescent Communication Scale items adequately reflects this construct? (PAC), (3) the Family Adaptability and Co- Please explain your response below." hesion Evaluation Scales II (FACES II), and The panel members identified three re(4) the Family Routines Inventory (FRI). These inventories were used in this investi- jection items, three cohesion items, three gation because they are likely to be most adaptability items, and seven family rouuseful in the large-scale field assessments tines items as potentially irrelevant or as that will include ethnic minorities. A more questionably relevant for the Mexican Amerthorough report of the preliminary item ex- ican culture. The explanations provided by panel members of the lack of relevance for amination is available upon request. each item fell into one of two categories: (1) The CRPBI (Schaefer, 1965) was de- the item itself, or some wording or phrasing signed to assess: acceptance (8 items), rejec- in the item, either has an ethnically specific tion (10 items), inconsistent discipline (8 meaning or has unclear meaning to members items), control (8 items), and hostile control of the Mexican American culture; and (2) the (8 items). The PAC (Barnes & Olson, 1982) item was worded poorly, or vague terms was designed to assess the qualities of open were used, such that it is unlikely that sub(10 items) and problem (10 items) parent- jects would understand the meaning of the child communication. Although the PAC item regardless of their ethnicity. was designed to be used with adolescents, the items and verbal level apply to younger There were 19 items for which a majorchildren (e.g.. If I were in trouble I would ity of the members of the panel did not accutell my parent; or My parent is a good lis- rately predict the correct subscale. These tener). The FACES II (Olson, Portner, & were one acceptance item, one rejection Bell, 1982) was designed to assess family co- item, two open communication items, three hesion (i.e., the emotional bonding that fam- problem communication items, one coheily members have toward one another: 16 sion item, seven adaptability items, and four items) and adaptability (i.e., the ability of family routines items. Particularly, the open the family to change in response to situa- communication, problem communication, tional and development stress: 14 items). family adaptability, and family routines The FRI (Boyce, Jensen, James, & Peacock, scales have a substantial percentage of items 1983; Jensen, James, Boyce, & Hartnett, (20%, 30%, 50%, and 15%, respectively) with 1983) was designed to assess the extent to limited face validity. which the family maintains daily rituals (26 Although one panel member did not items) and the importance placed on maincorrectly identify the acceptance construct taining daily rituals (26 items). and one member did not correctly identify The preliminary item examination was the control construct in free report, all panel accomplished in three sittings. In the first members were satisfied with all of the sitting, each panel member received a se- CRPBI constructs in the multiple-choice quential list of the PAC, FACES II, CRPBI, and checked-agreement responses. All panel and FRI items (in this order, but without any members provided accurate free reports, scale or subscale identification) and asked correct forced-choice responses, and agree-

Knight et al. ment responses for the open communication subscale. However, there was considerable dissatisfaction with the problem communication subscale: Three panel members did not provide reasonably accurate free reports, two did not identify the theoretically correct forced choice, and two disagreed with the theoretically prescribed construct label. Panel members generally provided accurate free reports, correct fixed choices, and agreed with the theoretically prescribed labels for the adaptability and cohesion scales; the only exception was one free report that could not easily be construed as equivalent to adaptability. Although the panel members generally suggested that these subscales were face valid, several members did note some discomfort with assuming that the items regarding flexibility truly represented adaptability among Mexican American families. Finally, for the FRI, panel members uniformly provided accurate free reports and correct fixed choices, and agreed with the theoretically prescribed construct labels. The general conclusions to be drawn from the preliminary item examination is that there appears to be a small subset of items from the rejection, adaptability, cohesion, and FRI subscales that are likely of limited item equivalence (as defined by Hui & Triandis, 1985) because the behaviors or attitudes represented in these items are of limited applicability or generalizability to the Hispanic family. There were also items on nearly every subscale, but particularly the open communication, problem communication, family adaptability, and family routines subscales, that are likely of limited face validity. Further, the concerns expressed by several panel members regarding entire subscales likely represent limited functional equivalence (again as defined by Hui & Triandis, 1985) for the problem communication and adaptability scales. In general, however, one must interpret these few problem items and panel concerns within the context of the general judgments of equivalence for most items and scales.

The Study The study was designed to assess the item equivalence and functional equivalence of the latent structure and subscale intercorrelations among Hispanic and AngloAmerican samples. If these 11 parenting and family interaction subscales are measuring the same psychological constructs in Hispanic and Anglo populations, items for each subscale should load similarly across

1395

samples, and subscale scores should have similar construct validity coefficients represented by similar intercorrelations among subscales across samples.

Method Sample.—The original sample consisted of 303 8-14-year-old children (M = 10.5, SD = 1.08, with only one 8-year-old and one 14-year-old participating), and either one or both of their parents. Parental participants included 146 fathers (or other male head of household) and 291 mothers (or other female head of household) from English-speaking households. Targeted recruitment areas encompassed 10 schools in three center city school districts. The school districts were centrally located in a large southwestern U.S. metropolitan area. Ethnic composition of the current sample was 60% Anglo, 20% Hispanic, 13% black, 4% Native American, and 2% other. The present report is based on a subset of 231 children and mothers from the original sample. The 231 children and mothers consisted of 70 Hispanic children and mothers (nearly all of whom were Mexican American) and 161 Anglo-American children and mothers. The distribution of children's ages was very comparable across ethnic groups. The Anglo-American mothers reported a mean of 13.3 (SD = 2.02) years of education for themselves, 14.1 (SD = 2.18) years of education for their spouse, and a mean family income in the $20,000 to $25,000 range (with 56% of the families at or below this range). The Hispanic mothers reported a mean of 11.1 (SD = 2.58) years of education for themselves, 11.6 (SD = 3.46) years of education for their spouse, and a mean family income in the $10,000 to $15,000 range (with 60% of the families at or below this range). Among the Hispanic mothers: 14.3% were first-generation Americans (i.e., were born in Mexico and immigrated to the United States), 28.6% were second generation, and 50.0% were third generation or beyond. Information regarding the migration for 7.1% of the mothers was not available; however, all of the participating children were at least second generation. Further, 97% of the mothers were descended from families in Mexico (i.e., were Mexican American). Procedures.—This study was part of a larger project to study families of children who did and did not participate in a schoolbased prevention program for children of al-

1396

Child Development

coholics (Michaels, Roosa, & Gensheimer, 1991). Thus, families were recruited from the communities around those schools that offered the prevention program. Families whose children had at least shown an interest in joining the prevention program (see Gensheimer, Roosa, & Ayers, 1990), who were presumed to be at bigb risk for alcobol abuse, were oversampled using several methods. Once recruited, in-home sessions were scheduled and conducted by trained professional interviewers. One interviewer was assigned to the target child and one to each participating parent. Parent(s) and the target child completed the measures in separate rooms and away from other family members. Upon arrival at the home, interviewers described the study and obtained informed consent from the parent(s) and assent from the target child. All measures were administered using laptop computers, with the survey questions displayed on the screen at the same time the intervie\ver was reading them aloud to the participant. The parent(s) and children responded on a separate keypad so that their responses were not observable to tbe interviewer. Each session took approximately 2 hours, which included scheduled breaks to minimize fatigue. Tbe four inventories of parenting and family interaction examined in the preliminary item examination (i.e., the GRPBI, PAG, FAGES II, and FRI) were completed by eacb participant as a part of the total survey (the FAGES II was completed by parents only). Results The purpose of this study was to assess the cross-ethnic equivalence of latent structures and intercorrelations among constructs. The analysis strategy involves a three-step plan: the first two steps were designed to provide evidence regarding item equivalence, and the final step was designed to provide evidence regarding functional equivalence. The relatively large number of items representing the 11 intercorrelated subscales of tbe four parenting and family interaction measures probibits an assessment of the full latent structure for the entire set of measures because such a large model would require an extremely large sample size to achieve the necessary ratio of cases to paths. Therefore, the first step in the analysis strategy was to include the items for each subscale in separate maximumlikelibood confirmatory factor analyses for each group (i.e., Anglo mothers, Hispanic

mothers, Anglo children, and Hispanic children). The second step in the analysis strategy was to include the item covariances for each subscale in separate cross-group equivalence, maximum-likelibood confirmatory latent structure analyses. Finally, the similarity of correlations among subscale scores was examined across ethnic groups, and the scale covariances ^vere included in a cross-group equivalence, maximum-likelihood structural model. Table 1 presents a summary of the results of the initial confirmatory factor analyses. A more thorough report of the results upon which this summary is based is available upon request. Gonfirmatory factor analysis using structural modeling procedures is designed to test the viability of theoretically prescribed factors. Tbis table presents a summary of the results of severalfitindexes: the chi-square for model fit, the BentlerBonett non-normed fit index (BBNN), and the comparative fit index (GFI). A good modelfitis characterized by a nonsignificant chi-square; however, the chi-square statistic is heavily influenced by sample size and is often significant even in good-fitting models. Good model fit is also characterized by fit indexes (BBNN and GFI) above .90 (.80 and above is considered adequate but marginal fit). Theoretically, afitindex of 1.00 is considered a better fit than is possible witb any otber model (although for mathematical reasons afitindex will occasionally be sligbtly greater than 1.00). The BBNN and GFI fit indexes are not strongly influenced by sample size (Bentler, 1990; Marsh, Balla, & McDonald, 1988) and are the most informative indicators of model fit in the present moderate sample size investigation. Given these characteristics, the BBNN and GFI indexes were used to characterize the adequacy of model fit in Table 1. The model fit indexes show at least a marginal fit for the GRPBI scales, except for the rejection subscale completed by the Anglo-American and Hispanic mothers, and the hostile control subscale completed by the Hispanic children. For the PAG, tlie measurement model fit at least adequately for the open communication subscale completed by the Anglo-American and Hispanic children, and for the problem communication subscale completed by each group except the Hispanic children. For the FAGES II, the measurement model had an acceptable fit for tbe cohesion subscale completed by the Anglo-American mother and the adaptability subscale completed by the His-

Knight et al,

1397

TABLE 1 SUMMARY OF RESULTS OF THE INITIAL CONFIRMATORY FACTOR ANALYSIS FOR EACH OF THE ELEVEN PARENTING AND FAMILY INTERACTION SUBSCALES

SCALE AND CONSTRUCT

Anglo American Mothers

CRPBI: Acceptance Adequate Rejection Poor Inconsistent discipline Good Control Good Hostile control Adequate PAC: Open communication Poor Problem communication ... Adequate FAGES II: Cohesion Adequate Adaptability Adequate FRI: Extent Very Poor Importance Poor

Hispanic Mothers

Anglo American Children

Hispanic Children

Cood Poor Adequate Adequate Adequate

Adequate Cood Good Good Adequate

Adequate Cood Adequate Good Very Poor

Poor Good

Good Adequate

Good Very Poor

Poor Good

N.A. N.A.

N.A. N.A.

Very Poor Very Poor

Adequate Adequate

Poor Adequate

NOTE.—BBNN and CFI indexes above .90 = good, between .80 and .89 = adequate, between .60 and .79 = poor, below .60 = very poor.

panic mothers and, to a lesser extent, AngloAmerican mothers. For the FRI, the measurement model tended to produce poor fit indexes, with a marginal fit occurring only for the extent subscale completed by AngloAmerican children and the importance subscale completed by the Anglo-American and Hispanic children. Table 2 presents the results of the assessments of fit across groups (i.e., across Anglo-American and Hispanic mothers, across Anglo-American and Hispanic children, and across all four groups). Since the adult version of the PAG assesses the parent's perception of their child's behavior, while the child version assesses the child's perception of the parent's bebavior, it is not appropriate to test the equivalence of the latent structure across mother and child groups. In addition, given the generally poor fit for the FRI in the initial analyses, the FRI was not included in any of these analyses. The results indicate a reasonably comparable measurement model fit across AngloAmerican and Hispanic mothers for the acceptance, control, hostile control, and problem communication subscales, and a near marginal fit for tbe inconsistent discipline subscale. However, tbere was not a reasonably comparable fit across the AngloAmerican and Hispanic mothers for the rejection, open communication, cohesion, and adaptability subscales. For the comparison

of fit across Anglo-American and Hispanic children, there was an adequate fit across groups for all subscales except for tbe hostile control and problem communication subscales (neither of which had an adequate fit within the sample of Hispanic children). In contrast to the reasonably comparable fit across groups noted above, it appears that only for the acceptance subscale was there a comparable fit across all four groups. Table 3 and Table 4 present the intercorrelations among the scores obtained on the parenting and family interaction subscales for the Anglo-American and Hispanic mothers and children, respectively. Those correlations that are significantly different across ethnic group are noted with one or two asterisks in these tables, depending on tbe level of significance of the difference assessed through the r-to-z transformation procedure. As can be seen in these tables, there are a few cases in which the correlations differ significantly across ethnic group. However, in some of these cases the significantly different correlations are not interpretively different. For example, although the correlations between control and hostile control are significantly different for the AngloAmerican and Hispanic mothers, botb are positive and each is significantly different from zero. It is also important to note that among the mothers only four out of 55 correlation pairs were significantly different (p
CO i n

bO5 i-H 1—

CO t 00 r CO CO

O5 CO O5 C>q " 0 0 ^ t ^ O5 C

< 00 cq CO oq CO 00 o> 00

^

S

G5

52-

oS

nee

g

The cross-ethnic equivalence of parenting and family interaction measures among Hispanic and Anglo-American families.

Recently there has been concern over the need for developmental research within ethnic minority populations and interest in family processes within, a...
1MB Sizes 0 Downloads 0 Views