|
Modeling Reciprocal Relationships Among Program Outcomes
|
| Presenter(s):
|
| Omolola A Adedokun, Purdue University, oadedok@purdue.edu
|
| Timothy J Owens, Purdue University, towens@purdue.edu
|
| Abstract:
There is no doubt that evaluation quality is enhanced by the application of advanced statistical methods to appropriate program designs. Although programs targeting psychological processes often produce outcomes that may reciprocally affect each other, a notable limitation in the estimation of variable-oriented program models is that reciprocal relationships among outcomes are hardly explored. Using data from waves I and II of the National Longitudinal Study of Adolescent Health, this study employs the method of structural equation modeling to illustrate the estimation of reciprocal relationships among variables. Specifically, we estimated the reciprocal effects of self-esteem and academic performance. The model included instrumental variables that are expected to exercise direct effects on one of a pair of reciprocally affected variables but not the other. The models were first tested for the full sample, and a “stacked” model was then estimated to compare the estimates of the reciprocal paths between boys and girls.
|
|
Interpreting Differences in Covariance Structures Among Clinical and Demographic Subgroups for a Model Describing Perceived Barriers to Seeking Help for Abused Elder Women
|
| Presenter(s):
|
| Frederick Newman, Florida International Unversity, newmanf@fiu.edu
|
| Laura Seff, Florida International Unversity, lseff@bellsouth.net
|
| Richard Bealaurier, Florida International Unversity, rbeaulau@fiu.edu
|
| Abstract:
The study was designed to contrast the perceived barriers to help-seeking for female victims and non-victims of domestic abuse age 50+ who were not in the service system. 445 women self-administered a 78 item survey in small groups. We employed structural equation modeling to develop and confirm our model. We then tested for differences in coefficient weightings and in the co-variance structures as a function of victimhood and the demographic characteristics. Six factors were identified as contributing to the overall perceived barrier score, accounting for 84% of the variance with excellent fit statistics (?2/df=1.527, CFI=.989, RMSEA=.034). The six factor coefficients predicting overall perceived barrier scores were not significantly different by level of subgroups. However, there were significant differences in the covariances among the six factors among the various subgroups. The discussion will focus on interpreting the differences in covariance structures.
|
|
A Quantile Regression Analysis of Incarcerated Youth’s Reading Achievement: Compare and Contrast the Ordinary Least Square Regression and Quantile Regression
|
| Presenter(s):
|
| Weijia Ren, The Ohio State University, ren.44@buckeyemail.osu.edu
|
| Ann O'Connell, The Ohio State University, aoconnell@ehe.osu.edu
|
| William Loadman, The Ohio State University, loadman.1@osu.edu
|
| Raeal Moore, The Ohio State University, moore.1219@osu.edu
|
| Abstract:
In current social science research, a lot of data do not meet the assumptions in ordinary least squares regression (i.e. non-normal, heteroscedastic) , and since the OLS approach is not robust to outliers, researchers can only choose to not to use the regression or continue to use the regression model without meeting the assumptions. In this case, ordinary least square (OLS) regression will be questionable and a new method is in need of providing better estimation to the data. Therefore, quantile regression is introduced because it can fit non-normal data, robust to outliers, and can provide a better capture to the distribution, relative to just the mean.
|
|
Evaluating Potential Impact of Intervention in Community Settings When No Comparison Data is Available: Mixture Latent Growth Modeling for Exploring Differential Change in Female Condom Use
|
| Presenter(s):
|
| Maryann Abbott, Institute for Community Research, maryann.abbott@icrweb.org
|
| Emil Coman, Institute for Community Research, comanus@netscape.net
|
| Peg Weeks, Institute for Community Research, weeks@icrweb.org
|
| Abstract:
This study illustrates the use of mixture analysis in evaluating preventive community interventions implemented in non-experimental framework. When no control group was designed and no matched comparison group is available as full panel data, the question of the potential impact of an intervention can be addressed by investigating model implied latent classes of participants that responded differently to an intervention. We show this approach with an intervention conducted in Hartford, CT aimed at increasing awareness and use of the female condom (FC) as a women-initiated HIV and sexually transmitted infections (STI) prevention method.
Overall then, the community intervention aimed at increasing FC use among community females seemed to have been very successful for 49% of the sample and successful for another 43%, while not impacting the rest of 8%. More detailed inquiries can be pursued regarding specific characteristics of the three groupings and the causal processes that may be responsible for the differential effects.
|
| | | |