AMND was regressed on program factors, participant characteristics, and interactions between the indicator for "education savers" and each of the other independent variables.
Table 2 presents the bivariate analysis of AMND by some subgroups of program and participant characteristics.
Table 3 displays the results from the regression analysis on AMND. The model as a whole was statistically significant (p = 0.01) and explained about 22 percent of the variance in the dependent variable.
Three of the four program factors were positively related to AMND. Participants who had higher monthly savings target and those who used direct deposit saved more.
Four demographic characteristics of participants were related to AMND: age, gender, race/ethnicity, and number of adults.
This association, however, was not strong; a $1 increase in monthly income was associated with about $0.003 more AMND. Home owners and bank-account owners also saved more than participants without such assets.
Analyses show that variables related to "education savers" (the stand-alone indicator and the interaction terms) explained about 2 percent of the variance in AMND. As a group, the variables related to "education savers" were statistically significant (p < 0.05) (based on the method of Pedhazur, 1997, p.
Table 3 indicates that being an "education saver" moderated the associations of several program and participant factors on AMND. Among program factors, the interaction with hours of education was positive and statistically significant.
Among participant demographic factors, savings for postsecondary education moderated the association of gender with AMND. While females in ADD on average saved $2.56 more than males, female "education savers" saved $9.05 less than female "non-education savers" and $6.49 ($9.05-$2.56) less than male participants.
Similarly, although married participants on the whole in ADD had higher AMND than not-married participants (p-value of 0.30), married "education savers" saved $11.88 less than married "non-education savers" and $10.02 ($11.08-1.86) less than non-married participants.
Because these two variables appear in multiple places in the regression model with interactions (see Table 3), evaluating how they affect AMND is not straightforward.