Table 1.
Options for modelling education
| PCS | MCS | |||
|---|---|---|---|---|
| BIC | AIC | BIC | AIC | |
| Non-parametric | 507278.1a | 7.397166 | 390815.6a | 7.163119 |
| Continuous | 508023.1 | 7.394599 | 392894.8b | 7.163963b |
| Education credential |
510690.4b | 7.399685b | 392633.7 | 7.163696 |
| Montez spline | 507406.8 | 7.394471a | 391880.7 | 7.162639 |
| Data driven for PCS: flat relationship under 9 years, then linear1 |
507955.4 | 7.394478 | ||
| Data driven for MCS: flat relationship after 13 years2 |
391972.2 | 7.161882a | ||
Lower BICs and AICs indicate better model fit.
indicates the best model fit
indicates the worst model fit
For PCS, the AICs and BICs were similar across different operationalization’s of education. There was not persuasive evidence for or against any particular model; for this reason, we operationalized education continuously because it made the most sense in terms of interpretability.
For MCS, the AICs and BICs were similar across different operationalization’s of education, although both metrics indicated that continuous education had the worst model fit, so there was evidence not to use continuous. However, since results were not consistent on which model was best, we used the data driven approach for MCS since it makes sense in terms of the interpretability and it’s the best fit by AIC.