Structural Model. The bundle designability landscape showed a single maximum “hotspot” at (ΔZ, ΔΦ) = (21.5 Å, −33.8°) (Fig. 1C) corresponding to bundle placement in which each pair of helices was offset in both translation and rotation to accommo- date the gentle left-handed supertwist of the bundles (SI Ap- pendix, Fig.
Structural Model. After confirming the fit of indicator measures on latent constructs and the overall fit of the data to the measurement model, I move on to fit the structural model (the path analytic model), which hypothesizes causal relationships between latent and observed variables. Figure 3-3 depicts the structural model used for the analyses presented in Chapters Five and Six. As mentioned above, I use a maximum likelihood estimation routine to generate estimates for each path specified in the figure. The model includes paths for both direct and indirect impacts on learning in early childhood. This specification assesses how poverty works in creating the cognitive skill gap in early childhood, what other factors contribute to the cognitive skill gap, independently of poverty, and how these many factors mediate the relationship between poverty and cognitive skill growth. Part of SEM’s appeal stems from the ability to compare models to one another assessing which model most accurately fits the data. In order to gain the most parsimonious and meaningful model, I test different theoretical pathways in order to find the model that best fits the data and theory. In both steps of structural equation modeling, researchers take account of how well the hypothesized model fits the data. Social scientists have developed many fit statistics for structural equation modeling over the past twenty years (Garson 2008). While XXXX 17.0 presents approximately 25 measures of goodness of fit, my analyses follow the recommendations of Jaccard and Wan (1996), who suggest reporting at least one measure of fit from each of three families of tests. In most cases below, I report the χ2 statistic, the Root Mean Square Error of Approximation (RMSEA), and one of the baseline measures of fit (e.g., Comparative Fit Index [CFI]). When comparing hypothesized models to one another, I also report one of the information theory measures (Bayesian Information Criterion [BIC]). Typically, values for the RMSEA below .05 and values above .9 for the fit indices indicate an acceptable model fit, although these guidelines are flexible, particularly for models with many variables (see Garson 2008). While I sometimes present the χ2 statistic, it is biased against models that have a large sample size and will frequently return a significant result even though the model may fit the data (Hu and Xxxxxxx 1999). In Chapter Six, I run the base model developed in Chapter Five for each specific racial group. The measuremen...
Structural Model. Table 3-1. Descriptive Characteristics of Variables in Chapters Four Through Six.5 Variable Name Variable Description Value Minimum Maximum Mean SD Cognitive Skills dvmental1 Cognitive Skills at 9 months 32.04 131.17 74.99 9.96 dvmental2 Cognitive Skills at 24 months 92.35 174.14 125.53 10.98 City 0.35 Suburb 0.35 Town 0.15 Rural 0.15 0 (No) 1 (Yes) 0.68 0 (No) 1 (Yes) 0.80 locale1 Collapsed locale code from census data (extrapolated from locale3) nomove12 Family did not change zipcodes between 9 and 24 homeng Language spoken at home is English numsib1 Number of siblings living in the same household as hhsize1 Number of people living in the same residence as the 0 6 (6 or More) 0 9 (9 or More) 1.10 1.14 4.46 1.44 mstatus Marital status of parents in household Married 0.66 Separated 0.03 Divorced 0.03 Widowed 0.00 Never Married
