09.08.2013 Views

Copyright 2012 Aileen M. Echiverri-Cohen - University of Washington

Copyright 2012 Aileen M. Echiverri-Cohen - University of Washington

Copyright 2012 Aileen M. Echiverri-Cohen - University of Washington

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

demonstrated appropriate skewness given the expected skewed nature <strong>of</strong> the U-shaped curve <strong>of</strong><br />

the AB effect. In contrast, PPI showed significant positive skewness and was corrected using<br />

natural log transformations, consistent with other PPI studies (Ornitz et al., 1989). To reduce<br />

outlier status, univariate and multivariate outliers were examined separately at pre- and post-<br />

treatment to isolate the correlated observations that <strong>of</strong>ten result in consecutive measurements in<br />

longitudinal studies (Tasca & Gallop, 2009). Univariate outliers were evaluated by running<br />

scatterplots <strong>of</strong> standardized residuals <strong>of</strong> measures, in addition to the inspection <strong>of</strong> z-scores where<br />

cases with standardized residual values exceeding 3.29 in absolute value were identified as<br />

potential outliers (Tabachnick & Fidell, 2001). The outliers for the repeated measures, were<br />

determined by getting the absolute value <strong>of</strong> the correlation <strong>of</strong> the observed values and the<br />

subject-specific predicted values. The lower 5 th percentile <strong>of</strong> the correlation coefficients were<br />

linked back to the subjects corresponding to potential multivariate outliers. Where there were<br />

potential outliers, variable transformation or changing scores on the variable for the outlier so<br />

that it is less deviant, were considered to minimize variability within groups (Tabachnick &<br />

Fidell, 2001). Missing data was addressed using restricted maximum likelihood (REML) within a<br />

random effects modeling and mixed model regressions analytic approach assuming that the<br />

missing data is independent <strong>of</strong> outcome variables (Gallop & Tasca, 2009).<br />

Evaluating Possible Covariates<br />

A bivariate correlation matrix was conducted across all variables to evaluate possible<br />

covariates. Covariates were included in the analysis if it was theoretically correlated with the<br />

outcome variables (Tabachnick et al., 2001). Age and education are demographic variables<br />

known to be related to executive functioning (Hasher & Zacks, 1988; Posner & Snyder, 1975;<br />

28

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!