Purpose: Drop-outs from programs are costly for programs as well as participants (Schreiner and Sherraden, 2002). As institutional factors affect savings performance of participants in the Individual Development Account (IDA) programs (Sherraden, 1991; Chang Keun, 2005), participants' drop-outs from the programs can be explained by institutional factors (Schreiner and Sherraden, 2002). This study examined the relationship between the institutional factors and the drop-outs from the programs using multi-level analysis.

Methods: This study used the data of the American Dream Demonstration (ADD), which is the first systematic study of IDA programs. IDAs are special accounts in which savings are matched for the poor (Schreiner, et. al., 2002). Sample (N=2,364) was drawn from 14 IDA programs in the United States. Participants who leave ADD without a matched withdrawal are defined as drop-outs. As of June 30, 2000, about 16 percent of ADD participants had dropped out (Schreiner and Schrraden, 2002). Based on the assumption of multi-level analysis, participants' drop-out was hypothesized to be not independent of the institutional characteristics of programs. SAS Proc mixed with logit link was used to test the effects of individual level (income, ethnicity, marital status, education, and debt), program level (matching rate, education for savings, and consulting about savings), and cross-level interaction effects between individual characteristics and institutional factors on the drop-outs (Bryk & Raudenbush, 2002).

Findings: To test the institutional effects on the drop-out from the IDA programs, three models are developed. Model A is the constant-only model and this merely partitions the variability into two levels around a fixed mean. In this model, the level 1 (Individual factors) variance has been constrained to 1 on the basis of an exact binominal distribution, while the level 2 (Institutional factors) variance was estimated from the data. The autocorrelation coefficient was calculated to measure the similarity of individual drop-out within programs. The coefficient, 0.35, suggests that there is a need to analyze these data in a multi-level analysis. Model A suggests that there are major differences between programs in their drop-out behaviors. Model B additionally includes individual level predictors. Age, race, and income are not strongly lined with drop-out. Interestingly, participants with debt are more likely to drop-out, perhaps because they had greater pressure on cash flows. Model C has both individual predictors and institutional ones. Among the institutional factors, matching rate is highly linked with drop-outs. In addition, participants with direct deposit to their IDA account are 15 percentage point less likely to drop out.

Implications: This study supports the hypotheses that institutional factors affect participants' drop-outs. This result suggests that programs can manage and reduce drop-out. Prevention efforts can be made, for example, by raising match rates, by counseling participants with large debts, or by working to decrease transaction costs. These efforts are costly, but they can be targeted to the most at-risk participants.