TPC-Journal-V6-Issue2

The Professional Counselor /Volume 6, Issue 2 193 high school?”; and (3) “Among your close friends, how important is it that they continue their education past high school?” Cronbach’s alpha for the academic support scale was .72. The social support variable was also composed of three Likert-scaled items: (1) “Among your close friends, how important is it that they get together with friends?”; (2) “Among your close friends, how important is it that they go to parties?”; and (3) “How important is it to you to have strong friendships in your life?” Cronbach’s alpha for the social support scale was .49. All questions were asked of students in their sophomore year of high school and had three choices for answers: (1) not important, (2) somewhat important and (3) very important. Higher scores represented greater socialization. Criterion variable . The criterion variable measured student status 2 years after scheduled graduation and had three categories: (1) leaver (enrolled after high school but not enrolled in January of 2006), (2) still enrolled in a two-year institution, and (3) still enrolled in a four-year institution. This same criterion variable with four categories was used in a previous study (Sciarra & Ambrosino, 2011). Data Analysis Since the criterion variable has three categories (leaver, still enrolled in a two-year institution, still enrolled in a four-year institution), the appropriate method for analysis is a multinomial logistic regression (MLR; Norusis, 2004). The MLR models the relationship between a categorical criterion variable and predictor variables (Menard, 2010; Norusis, 2004; Pampel, 2000). In MLR, the effect size results from the odds ratios for each predictor. Odds ratios are ratios of the probability of being in a particular group compared to being in the baseline or reference group (Sciarra & Ambrosino, 2011). In the present analysis, the reference group was the first category (leaver), to which the other groups were compared along the predictor variables. Unlike linear regression, MLR employs categorical variables and cannot rely on traditional transformation methods to deal with missing data. The SPSS default position was employed, which excludes all cases with missing values on any of the indepen- dent variables. The analysis, more theory-testing than exploratory, utilized the forced entry method where all predictors are entered at the same time into the regression equation. In large data sets, there is a danger of overdispersion. To check for this, a dispersion parameter was calculated by di- viding the Pearson chi square goodness of fit by the degrees of freedom, which equaled 1.23. While any parameter greater than 1 indicates the presence of overdispersion, only a parameter approach- ing or greater than 2 suggests a problem (Field, 2009). Results The original MLR model had nine predictor variables (academic support, social support, talks with math teacher outside of class, talks with English teacher outside of class, has gone to counselor for college entrance information, performed volunteer/community service work, number of hours spent weekly on working, homework and extracurricular activities). From the sample of 7,271 who participated in all three waves (2002, 2004 and 2006) of ELS (U.S. Department of Education, 2008) and who enrolled in either a two-year or four-year institution upon graduation from high school, academic support [χ 2 (2, 3148) =.90, ρ=.64], social support [χ 2 (2, 3148) =.59, ρ=.74], talks with English teacher outside of class [χ 2 (2, 3148) =1.14, ρ=.57] , has gone to counselor for college entrance informa- tion [χ 2 (2, 3148) =1.44, ρ=.49], performed community/volunteer service [χ 2 (2, 3148) =.63, ρ=.73], and number of hours worked [χ 2 (6, 3148) =4.64, ρ=.59] were not significant and therefore were excluded from subsequent analyses. The revised model included the three remaining variables whose correlations were .066 (hours