TABLE 2.

Illustrative Hybrid Design/Model-Based Analysis Using the High School and Beyond (HSB) Data Set

	1. Original, Model-Based Analysisa	2. Hybrid Analysis, Intermediate Step	3. Hybrid Analysis, Final Step
Fixed Effects	Accounts for Clustering; Partially Accounts for Stratification	Model 1 Plus Weights to Account for Disproportionate Selection	Model 2 Plus Covariates to Fully Account for Stratification
INTERCEPT	7.27 (.06)**	7.59 (.11)**	7.76 (.12)**
CSES	2.09 (.07)**	2.16 (.11)**	2.16 (.11)**
MEANSES	4.43 (.14)**	4.37 (.25)**	3.60 (.28)**
SECTOR	−0.06 (.24)	−0.01 (.36)	0.15 (.36)
CSES × MEANSES	0.62 (.17)**	0.39 (.28)	0.39 (.28)
CSES × SECTOR	−1.50 (.19)**	−1.63 (.27)**	−1.63 (.28)**
HISPANIC			−0.96 (.32)**
BLACK			−1.83 (.29)**
HISPANIC × SECTOR			0.25 (.59)
BLACK × SECTOR			−1.42 (.74)
Variance Components
τ 00	1.86 (.15)**	2.05 (.27)**	1.73 (.24)**
τ 01	0.31 (.13)*	0.27 (.16)	0.18 (.16)
τ 11	0.29 (.10)**	0.53 (.26)*	0.55 (.26)*
σ 2	21.89 (.25)**	21.09 (.24)**	21.08 (.34)**

^aThe model-based analysis results in Column 1 differ somewhat from those of Raudenbush and Bryk (2002, chap. 4) and Singer (1998) for two reasons. First, our variables were taken directly from HSB's 1982 public-use datafile for the sophomore cohort (see online Appendix and www.icpsr.umich.edu). Raudenbush and Bryk (2002) constructed and used factor score composites of 1980 and 1982 datafile variables for sophomore and senior cohorts (Lee & Bryk, 1989). Second, we used all public and Catholic schools and they used a random subset. MATHACH = math achievement; MEANSES = school average socioeconomic status; SECTOR = Catholic or public school; CSES = school mean centered child socioeconomic status; BLACK = high % Black enrollment; HISPANIC = high % Hispanic enrollment.

^*p < .05.

^**p < .01.