A r c h i v e d  I n f o r m a t i o n

Appendix C: Multivariate Analyses
in Chapters 5 and 6

This appendix discusses the multivariate regression analyses reported in Chapters 5 and 6 of this report: the analysis variables and method used; the rationale for the approach; and detailed analysis results tables. The text in this appendix primarily refers to the Chapter 6 analyses with participation rates as dependent variables; however, the same analytic approach was used for analyses reported in both chapters.

Analysis Method

Several major data analysis steps were involved in the multivariate analyses:

1. the creation of derived variables that "combined" related ESIS items into fewer variables representing higher-order constructs;
2. the systematic examination of statistical properties of all derived variables;
3. the identification of dependent and independent variables that represent key issues and topics associated with Even Start services, including preliminary correlational analyses of candidate variables to identify potential problems such as multicollinearity;
4. comparing results generated by several regression approaches, using a same set of variables, to select the most useful approach;
5. performing regression analyses on all dependent variables identified in steps 3 and 4; and
6. performing analyses of variance (ANOVAs) to further examine the independent variables that produced the strongest relationships with dependent variables in the regression analyses.

Derived variables were created by summing and/or averaging related ESIS items (e.g., averaging instructional hours offered across four levels of adult education and summing the types of organizations each project used as collaborating agencies). The derived variables achieved three important goals: (1) reducing to a manageable number the variables used in the multivariate analyses; (2) combining detailed ESIS data into variables that represent general concepts relevant to Even Start (e.g., home-based versus center-based services); and (3) increasing measurement reliability, where applicable.

The distributional characteristics of each derived variable were examined (e.g., frequencies, mean, standard deviation, minimum and maximum values, level of missing data). Variables with characteristics problematic for correlational analyses (e.g., restricted range) were further refined or eliminated from further analyses.

The selection of dependent and independent variables for multivariate analyses involved an iterative matching of available data (original ESIS variables and derived variables) with key concepts and topics pertinent to the evaluation. For example, from many data elements available concerning participant characteristics, a few variables expected to be relevant to program participation were selected (e.g., parent educational background and English proficiency). From data on project characteristics, we selected those that represented key program elements (e.g., service intensity, integration across service areas, flexibility of services) and those representing projects? organizational capacities (e.g., funding, staff, number of families served). The goal of this step was to identify all variables relevant to evaluation questions and, at the same time, to minimize redundancies among variables. The variable selection step also involved running rounds of regression analyses to identify and eliminate variables that consistently contributed minimally to multiple correlations.

Several multiple regression approaches were considered for analyzing the relationships between a wide variety of participant characteristics, project characteristics, and service delivery practices on one hand and families? participation patterns on the other. We performed exploratory analyses to select the approach that was appropriate for the type of data being analyzed and facilitated interpretation and reporting.

In the case of dichotomously coded dependent variables (e.g., items coded as yes or no), we tested three different regression methods: simple regression, probit model, and logit model. Given that all analyses produced similar results in terms of overall model fit and parameter estimates for individual independent variables, we reported the simple regression results for ease of interpretation (see Gruber and Madrian, 1997; Munnell et al., 1996).

The regression results reported in Chapters 5 and 6 are based on a stepwise regression model. The regression analyses were used to refine the selection of variables that most strongly "influence" a dependent variable. Once an independent variable (e.g., parent educational background) was thus identified, we grouped participants by different levels of that independent variable (e.g., 6th grade or less, 7th to 9th grades, etc.) and examined differences between the means of the dependent variable across these groups, generally by using the analysis of variance (ANOVA) method.

In the ANOVAs, we simply asked: how did families having different levels of the independent variable differ on a specific measure of participation? We did not partial out the potential influence of other variables entered in the regression analysis on the dependent variable for ease of interpretation. Instead, we examined interaction effects of two or more independent variables found to be related to the dependent variable in the regression analyses.

Exhibit C.1 presents the variables included in the regression analyses reported in Chapters 5 and 6 and their distributional characteristics. Not all variables listed were in every regression analysis; some variables were used as dependent variables in Chapter 5 analyses and as independent variables in Chapter 6 analyses. Exhibit C.2 presents bivariate, simple correlations among variables used in regression analyses at the adult participant level. In analyses at the family and child participant levels, the magnitude of correlations among variables was similar (generally very low) to those shown in Exhibit C.2.

Exhibit C.1: Variables Used in Chapters 5 and 6 Multiple Regression Analyses

Variable	Variable Name	Minimum-Maximum Values	Mean	SD
Participation Measures
Number of instructional home visits in which family participated	HOMEVIST	0-159	7.15	9.51
Hours/month of adult education participation	AE_PRTHR	0-1988	96.22	158.41
Hours/month of parenting education participation	PE_PRTHR	0-982	27.66	41.98
Child (did, did not) participate in ECE for 10-12 months	ECE10_12	Dichotomous 1, 0	0.22	0.42
Family (did, did not) participate in all core services	ALL_CORE	Dichotomous 1, 0	0.93	0.26
Family was continuing at year-end	RETAINED	Dichotomous 1, 0	0.56	0.50
Family completed goals and left the program	COMPLETD	Dichotomous 1, 0	0.15	0.36
Family Characteristics
Age of parent	ADULTAGE	14-91	28.49	8.57
New vs. continuing family	NEW97	Dichotomous 1, 0	0.60	0.49
Highest grade reached by parent prior to enrollment in Even Start	EDUCA-TION	0-16	9.55	3.04
Parent with limited English proficiency	LEP	Dichotomous 1, 0	0.33	0.47
Family with 4 or more Need Indices	NEEDY	Dichotomous 1, 0	0.43	0.19
Number of support services family received during the year	SUPPORT	0-9	2.89	2.06
Single-parent vs. non-single-parent family	ONEPAR	Dichotomous 1, 0	0.36	0.48
Project Characteristics
Rural vs. non-rural community/service area	RURAL	Dichotomous 1, 0	0.50	0.50
Total project funds in 1996-97 (in thousand dollars)	TOT_FUNDS	23-943	285.72	153.13
Project age	PROJAGE	1-8	4.59	1.99
Number of families served in 1996-97	PROJSIZE	0-400	55.81	44.67
Extent of interagency collaboration	COLABSUM	0-9	4.65	2.74
Barriers experienced by project in program implementation	BARRIERS	2-85	30.36	12.51
Staff Resources and Qualifications
Number of Even Start paid staff	NO_STAFF	0-61	10.26	7.26
Proportion of instructors with college or higher education	INS_HIED	0-1	0.75	0.35
Proportion of instructors with five or more years of experience	INS_HIEX	0-1	0.50	0.38
Days/year of inservice training per staff	AVGDAYS	0.5-11	6.93	2.77
Service Intensity and Delivery Practices
Adult education hours offered per month	AE_HRMO	0-160	30.62	23.75
Parenting education hours offered per month	PE_HRMO	0-128	18.75	17.13
Early childhood education hours offered per month	ECE_HRMO	0-160	42.72	33.41
Ratio of home-based instruction hours offered to total hours offered	HB_RATIO	0-1	0.21	0.24
Individually-tailored vs. standardized instruction	INDIV	1-5	2.18	0.70
Group activities vs. learners working alone	GROUP	1-5	2.80	0.62
Learner- vs. instructor-selected instruction	LEARNER	1-5	3.26	0.72
Extent of functional literacy incorporated into adult education curriculum	FUNCLIT	1-3	2.23	0.43
Extent of parenting education activities (variety and proportion of families affected)	COMP_PE	11-60	55.79	5.15
Extent of integration of services across core service areas	INT_ALL	2-4	2.78	0.47
Transitional services offered to children	TRANSERV	0-12	5.61	3.33
Flexibility of service delivery schedule	FLEXSERV	0-3	2.17	0.81

-###-

[Appendix B: Exhibit B-11]

[Exhibit C-2]