Results

This analyses of the 4-H Study Wave 1 data were conducted to bring information to bear on the developmental systems view of the thriving process shown in Figure 18.1. In our initial analyses we focused on testing the empirical validity of the conception of positive development as instantiated

TABLE 18.4

Measurement Model of the Five Cs and PYD

Standardized ML Residual

Estimate Error

Confidence

1 Positive identity .91 .18

2 Self-worth .64 .50

Competence

3 Academic competence .51 .74

5 School engagement .72 .48

6 Social competence .46 .79

Character

7 Personal values .76 .42

8 Social conscience .79 .37

9 Values diversity .70 .51

10 Interpersonal values and skills .67 .54

Caring

11 Sympathy: Disadvantaged .72 .48

12 Sympathy: Loneliness .81 .30

13 Sympathy: Unfortunate .74 .46

14 Sympathy: Pain .80 .37

15 Sympathy: Rejection .76 .43 Connection

16 Family .60 .64

17 School .71 .40

18 Community .44 .81

19 Peers .43 .81 PYD

1 Confidence .77 .41

2 Competence .82 .33

3 Character .82 .32

5 Connection .91 .18

by the Five Cs, and proposed a measure of contribution for fifth graders that could be analyzed in relation to PYD. Future reports from the 4-H Study, which will capitalize on the longitudinal design of this research, will extend model testing to include the relationship between assets, contribution, and the growth of the Five Cs.

Descriptive analyses were conducted first to determine whether there was systematic variation in the measures described above with selected youth and parent background variables: sex, race/ethnicity, and social class. Accordingly, correlations were computed among the youth and parent background variables, such as youth sex, youth race/ethnicity (recoded as European American/nonEuropean American), household income, mother's education, and number of children in the household. Youth sex was not correlated with the other variables. Household income and mothers' education were significantly and positively correlated and both were negatively correlated with the number of children in the household. Being European American was positively correlated with household income and mothers' education and negatively correlated with number of children in the home.

Hierarchical multiple regression analyses were computed for each scale score as the dependent variable and sex, race/ethnicity, household income, mother's education, and number of children in the home as a set of independent variables. In addition, two-way interactions between race/ethnicity and sex, and race/ethnicity and household income were tested as a second set of independent variables. Race/ethnicity was assessed through the use of three dummy variables (European American, African American, Latino/a vs. the reference category of all other designations). These analyses are based on participants whose parents answered the PQ, with a maximum sample size of 1,117.

Because there were 48 measures included in these analyses, the significance level was adjusted to control for Type I error. Starting with a /»-value of .05, we adjusted our p-value to be .001, using the standard correction of p-value/N of analyses (.05/48). Using this corrected pp-value, none of the 2-way interactions were significant. Number of children as a variable also was dropped from the analyses because it never added a significant proportion of variance after other variables were included.

Youth sex and household income were significantly related to the measures in expected ways: Girls have higher scores for, and household income is positively related to, most of the measures. Once household income is controlled for there remained a few significant relationships with race/ethnicity. The race/ethnicity effects that were present show that Latino/a fifth graders reported greater parent involvement, value interpersonal relationships and skills, valued diversity in their relationships, and thought about the future more than fifth graders from other groups. Compared to other youth, African American fifth graders reported lower support from peers, and engaged in more delinquent behaviors but reported greater risk avoidance. European American fifth graders reported higher perceived social and physical competence, and greater self-worth than youth in the other race/ethnicity groups. Future work will explore the complex relationships among these variables within our model of positive youth development.

PYD AND THE FIVE CS

As we have noted, specification of the measurement model of the Five Cs proceeded through multiple steps. First, an extensive literature review was conducted to identify a set of measures that would serve as indicators for each of the Five Cs. Second, these measures were assessed in a pilot study involving 339 youth from five cities and towns in Massachusetts. Scales were assessed in relation to their ability to capture the essential definitions of the Five Cs developed for use in this study (see Table 18.1). Following the evaluation of the pilot results, the survey was revised to better represent the constructs. Third, and concurrent with Wave 1 data collection, the factor structure of internal and external developmental assets, as measured by the Search Institutes' PSL-AB measure (Benson et al., 1998), was reevaluated and restructured to reflect both empirical and substantive considerations (see Theokas et al., 2005). These modifications lead to revisions of the initial measurement model.

To accomplish these revisions, several of the authors independently categorized all scales included in the SQ as either an index of one of the Five Cs, an index of the sixth C of contribution, an index of internal assets, an index of external assets, an index of regulation, or as not relevant to any of the constructs (e.g., pubertal maturation, race, and sex were constructs placed by all authors/raters into this last category). When at least 80% of all raters categorized a measure as reflecting one of the constructs, this measure was considered as an operationalization of it.

Confirmatory factor analysis (CFA) was conducted to assess the degree to which the Five Cs/PYD model fit the data. Model-fitting analyses were conducted to assess the adequacy of the a priori model; subsequent analyses were used to assess model improvement following theoretically-sound modifications. LISREL 8.54 (Joreskog & Sorbom, 1996a), using Maximum Likelihood (ML) estimation on raw data within a PRELIS 2.0 (Joreskog & Sorbom, 1996b) file, was used for all CFA analyses.

The initial model contained 19 manifest indicators, five first-order latent factors, one for each of the Five Cs, and one second-order latent factor, representing the PYD construct. All hypothesized pathways were significant but the model had a relatively poor fit (x2 =1933, df = 147; RMSEA = 0.085; GFI = 0.89; CFI = 0.94; NNFI = 0.94). The Five Cs/PYD model was therefore retained and subjected to model-improving modifications.

Inspection of the modification indices suggested several changes to improve model fit. Considering that we found high intercorrelations among the indicators, modifications to allow correlated residual errors among the indicators were implemented. Specifically, residuals from three indicators, social competence, academic competence, and self-worth, were allowed to correlate on the assumption that scores on these scales, having all come from the Harter SPPC scale (Harter, 1998), shared method variance not accounted for by the model. The freeing of these residuals resulted in a significantly better model (x2 =1455, df = 144; p < .01; RMSEA = 0.073; GFI = 0.92; CFI = 0.96; NNFI = 0.95).

Next, residual errors were allowed to correlate between indicators within factors. Consistent with the definitions presented by Roth and Brooks-Gunn (2003a), within competence, residuals between grades and academic competence and between school engagement and social competence were allowed to correlate. Within character, residuals between personal values and social conscience and between values diversity and interpersonal values were allowed to correlate. Within caring, residuals between sympathy for disadvantaged and sympathy for unfortunate and between sympathy for rejected and sympathy for loneliness were allowed to correlate. Finally, within connection, residuals between connection to family and connection to community and between connection to school and connection to community were allowed to correlate. All together, these modifications significantly improved model fit (x2 =662, df = 136; p < .01; RMSEA = 0.048; GFI = 0.96; CFI = 0.98; NNFI = 0.98).

The modification indices suggested that for the revised model indicated that model fit could be further improved by correlating two more pairs of residuals: positive identity with academic competence and positive identity with social competence. Such relations may reflect the theoretically and empirically established relations between adolescent achievements in academic and social areas and their positive self-regard (e.g., Brown, 2004; Eccles, 2004; Harter, 1998). Following these modifications, model fit was again improved (x2 =552, df = 134; p < .01; RMSEA = 0.043; GFI = 0.97; CFI = 0.99; NNFI = 0.98). These modifications, however, were not sufficient to specify a model that optimally fit the data.

Inspection of the revised modification indices suggests that additional structure in the relationships among the first order factors has not been exhausted by either the correlated residuals among the respective indictors or their respective loading on the second-order factor of PYD. Specifically, two pairs of first-order factors, confidence/competence and character/caring, continued to share variance not accounted for by the model. Rather than specifying additional structure to the model, we retained the more parsimonious model described below, and allocated additional refinement of and evaluation of sample specific effects on the model to future waves of the 4-H study—waves that would allow us to take advantage of longitudinal data and retest control samples for purposes of cross-validating the model (cf. Browne & Cudeck, 1993; Cudeck & Browne, 1983).

The retained model is depicted in Figure 18.2. While the model Chi-Square was significant at 4-times the model degrees of freedom (x2 =552, df = 134, p < .01), it is sensitive to sample size. With large sample sizes, the x2 statistic becomes unreasonably powerful at detecting discrepancies between the model and the data and under realistic conditions perfect model fit is not to be expected (Bollen, 1989, pp. 266-269). Following prior recommendations (e.g., Raykov, Tomer, & Nesselroade, 1991; McDonald & Ho, 2002; Tomer & Pugesek, 2003), we evaluated a variety of fit indices. For this model, the Goodness-of-Fit index (GFI, Joreskog & Sorbom, 1996a), a measure of absolute fit, was 0.97, well above the 0.90 minimum criterion of close fit suggested by Hoyle and Panter (1995). The Comparative Fit Index (CFI, Bentler, 1990) was 0.99, suggesting that the specified model is 99% better than an independence model where all observed variables are assumed to be uncorrelated. Likewise, the Non-Normed Fit Index (NNFI, Bentler & Bonett, 1980), which takes into account model complexity and performs well with large sample sizes was 0.98, again indicating close fit. Finally, the Root Mean Square Error of Approximation (RMSEA, Steiger & Lind, 1980), which is a measure of fit per degree of freedom and is sensitive to model misspecifi cation (Hu & Bentler, 1995), was 0.043 with a 90% confidence interval of 0.039 to 0.047. A value of .05 or less indicates a close fit (Browne & Cudeck, 1993).

Standardized factor loadings for the 19 manifest variables ranged from .43 to .91, indicating that the Five Cs factors accounted for 18-83% of the indicators' variance. In turn, the second-order factor of PYD accounted for an average of 60% of the variance in the latent factors for the Five Cs. This explained variance (or common variance) ranged from 24% for Caring to 83% for Connection.

Latent factor scores for the Five Cs and PYD were calculated in LISREL 8.54 for use in remaining analyses (Joreskog, Sorbom, du Toit, & du Toit, 2001). These scores should be treated with caution since they are indeterminate, with individual-level rank ordering on a specified factor varying widely depending upon how the scores are calculated (Bollen, 1989). It should be noted, however, that correlations between the LISREL-computed factor scores and mean scores calculated from the standardized indicator variables (so called "coarse factor scores," e.g., Grice, 2001) were all high (> 0.93).

Figure 18.2 Retained factor model with standardized maximum likelihood estimates. Note: All estimates are significant at the 0.005 level.

In addition, hierarchical multiple regression analyses were computed using factor score as the dependent variable and sex, race/ethnicity, household income as predictor variables to provide comparable background information as for the indicator scale scores. Girls have higher scores than boys on caring, character, competence, connection, and PYD. European American and Latino/a youth have higher confidence scores than other youth. Youth from families with higher incomes have higher scores on all but the caring constructs. These scores, therefore, appear to be reliable and to have some predictive validity.

Confident Creatures

Confident Creatures

This is way better than drunken bravery. Literally Conquer Any Challenge Without Fear And Realize Your Full Potential By Tapping Into These Closely Guarded Secrets To Building Total Confidence. Overcome Your Fears And Achieve your Ideal Lifestyle With Ease Using These Keys To Building Ultimate Personal Confidence.

Get My Free Ebook


Post a comment