The Professional Counselor | Volume 11, Issue 4 391 Bivariate Correlations A bivariate correlational analysis of interval and ratio variables in the study allowed for preliminary examination of collinearity and provided information on relationships between the variables of interest. The bivariate correlation matrix indicated no concerns regarding multicollinearity. The correlations contain indications of relationships to school counseling access. For example, school counseling caseload and percentage of time spent on college-readiness counseling were inversely related (r = −.181, p < .01). School counselor caseload was negatively significantly correlated to SES, STEM GPA, and math selfefficacy. School counselor percentage of time spent on college-readiness counseling was positively significantly correlated with SES, STEM GPA, math self-efficacy, and science self-efficacy. See Table 3 for the full results of the bivariate correlations. Table 3 Bivariate Correlations Variables 1 2 3 4 5 6 1. SES - 2. STEM GPA .398** - 3. Math self-efficacy .152** .302** - 4. Science self-efficacy .15** .223** .395** - 5. School counseling caseload −.152** −.105** −.045** −.015 - 6. Percentage of time spent on college-readiness counseling .150** .104** .042** .027** −.181** - Note. SES = socioeconomic status; STEM = science, technology, engineering, mathematics; GPA = grade point average. **p < .01. Primary Analysis Next, the results of the sequential logistic regression are presented (see Table 4). The outcome variable is a dichotomous variable of STEM major persistence and attainment and indicated if a student either is or is not enrolled as a declared STEM major in a postsecondary institution or has or has not attained a degree in a STEM field from a postsecondary institution. Statistical assumptions of the model were assessed. Tolerance (0.26) and VIF values (mean VIF = 1.34) indicated no concerns regarding multicollinearity. The Box-Tidwell test indicated the assumption of a linear relationship between continuous predictors and the logit transform of the outcome variable was met, with nonsignificant p values. Utilizing the balanced repeated replication variance estimation method, 16,007 observations were included in the regression model, with a population size of 1,540,118 and 192 replications.