Table 5.
Results of KHB Mediation after Propensity Score Matching, Showing Logistic Regression Coefficients (Standard Errors in Parentheses) for Association between Paternal Incarceration and Early Suspension or Expulsion (Children with Residential Fathers)
| Mediation by Matching Method | A Mediators: Child Behavior Problems |
B Mediators: Weakened Family Bonds |
C Mediators: Weakened Employment Bonds |
D Mediators: All Mediators |
|---|---|---|---|---|
| Nearest-Neighbor Matching | ||||
| Total effect | 0.667 *** (0.170) | 0.586 *** (0.155) | 0.591 *** (0.155) | 0.687 *** (0.173) |
| Direct effect | 0.524 ** (0.168) | 0.529 ** (0.157) | 0.544 ** (0.156) | 0.461 ** (0.171) |
| Indirect effect | 0.143 * (0.057) | 0.056 (0.038) | 0.047 * (0.024) | 0.226 ** (0.075) |
| Percent explained | 21 | 10 | 8 | 33 |
| Treatment N | 566 | 566 | 566 | 566 |
| Control N | 1,152 | 1,152 | 1,152 | 1,152 |
| Kernel Matching | ||||
| Total effect | 0.638 *** (0.162) | 0.560 *** (0.150) | 0.565 *** (0.149) | 0.660 *** (0.164) |
| Direct effect | 0.499 ** (0.161) | 0.499 ** (0.154) | 0.513 ** (0.152) | 0.431 ** (0.166) |
| Indirect effect | 0.139 * (0.056) | 0.061 (0.035) | 0.052 * (0.024) | 0.229 ** (0.073) |
| Percent explained | 22 | 11 | 9 | 35 |
| Treatment N | 566 | 566 | 566 | 566 |
| Control N | 1,655 | 1,655 | 1,655 | 1,655 |
Notes: Fragile Families and Child Wellbeing Study. Sample limited to observations with nonmissing values for survey items about child’s experience with suspension/expulsion by Y9. Children with deceased/unknown fathers by Y5 or whose fathers were not living with them at Y1 are excluded. Observations in region outside common support are also excluded. Child behavior problems include parent-reported externalizing behaviors and self-reported delinquency. Family bonds include low father involvement, mother harsh parenting, mother and father separated, mother repartnered. Employment bonds include mother income-to-poverty ratio and father unemployment. For nearest-neighbor matching, neighbors = 10. For kernel matching, kernel = Epanechnikov; bandwidth = .06. Results based on first of 25 imputed datasets. KHB = Karlson, Holm, and Breen (2012); SE = robust standard error; Y1 = one-year follow-up survey; Y5 = five-year follow-up; Y9 = nine-year follow-up.
p<.001
p<.01
p<.05 (two-tailed tests)