OMB83 C Memo

Memorandum

To: Shelly Wilke-Martinez

Through: Kashka Kubzdela

CC: Tracy Hunt-White

From: Ted Socha, B&B COR

Date: 1/30/21

Re: 2008/12 Baccalaureate and Beyond Study (B&B:08/12) Survey Items Deletion and Responsive Design Update

Item Revisions

NCES would like to remove a selected set of items from the B&B:08/12 full-scale interview approved in June 2012 (OMB# 1850-0729 v.8). Based on recent testing, the approved interview has been taking an average of 45 minutes. To bring the interview time down to 35 minutes (per the estimated burden reported in the OMB package and in materials provided to sample members), we plan to remove the items presented in Attachment A. These items were selected for removal based on their relative lower importance for the survey as discussed at the November 2011 TRP meeting. The resulting interview is attached in Appendix G (to replace that originally approved in June 2012).

Responsive Design Update

As discussed in the previous OMB submission, we evaluated the impact of five additional variables on the propensity model used to calculate the Mahalanobis (M) distance values. The additional variables tested were:

• Whether the sample member applied for financial aid;

• The amount of financial aid loans the sample member received;

• Whether the sample member is a race other than White;

• Whether the sample member enrolled in post-graduate education; and

• Whether the sample member attained a post-graduate degree.

The race and enrollment indicators were statistically significant in the new model, but did not significantly change the overall model, suggesting that outcomes used in the simulations would not be affected. To ensure that no effects were overlooked, the simulations were rerun with the updated model.

 

The updated model changed the distribution of M values among the total sample, but not the outcomes. The average M value was 3.3 points higher with the new model and the maximum M value changed from 211.9 to 270.6. This suggests a greater spread in the M values from the mean to the maximum. However, even with the change to the distribution, the relationship between M and the R-indicator values for nonrespondents did not change, nor did the overall results of the two simulations. Simulation 1 still showed a convergence of the average M value for nonrespondents to that of respondents as the number of high distance cases were converted to respondents.  Further, Simulation 2 did not show any significant differences in the outcome measures when reducing the number of high distance respondents.  The evaluation of the NSC and NSLDS data showed significant differences among the same categories as seen in the initial simulations. Based on these results, we recommend proceeding with the more parsimonious initial model, as approved in June 2012 (OMB# 1850-0729 v.8).