Statistical Analysis Missing Data
The first two paragraphs provide a general plan for dealing with missing data for the two types of data collection activities that will be undertaken in the evaluation.
In order to minimize missing and inaccurate data in the family finding web-based database, Child Trends provides ongoing training and technical support for data entry, and conducts regular audits of the data to ensure the completeness and accuracy of the information program staff are entering. In addition, we developed a system to extract the data from the case management system and conduct regular data exports so analyses can be conducted using SAS programming. Child Trends currently provides reports to program staff to monitor program implementation. These reports include numbers of children served, number and types of kin discovered, number of interactions that family finding staff have with kin, number of family meetings held, and reasons for family finding case closure.
For the in-person interviews, the majority of children are relatively easy to locate at the 12 month time period and more difficult at the 24 month time period as some have emancipated from foster care and no longer locatable through the child welfare agency resources. This issue is of lesser importance for youth in the treatment group given that family finding workers will have discovered and engaged a large number of relatives and other adult supports, making tracking of runaway or otherwise disconnected youth far more easier and our estimated response rates take this into consideration.
This paragraph addresses potential sources of missing data and how the evaluation team will attempt to mitigate them.
As noted previously, for the full sample, data on the key outcome for the impact analysis (as well as for other child welfare outcomes) will come from North Carolina state administrative data. One potential source for missing data could be North Carolina’s inability to find records for specific children for whom we request data. This could occur, for example, if a case identifier has been recorded incorrectly in the Family Finding case management database. To address this problem, we have carried out “pre-tests” involving requesting preliminary data from the North Carolina in order to ascertain how many records we are able to obtain. Results indicate that we are able to obtain administrative data records for all children who have undergone random assignment. Another potential issue could be that, even though we are able to obtain data records, relevant information may be missing from records. During the coming months, we will be examining the extent of missing data.
This paragraph explains the imputation methods that will be used to deal with missing data.
Similarly, with the in-person interviews, respondents may decline to answer some questions. In the case of missing data on covariates, we will carry out multiple imputation. While listwise deletion generally yields less biased parameter estimates than single imputation, multiple imputation is an improvement over both methods by producing less biased results than single imputation and by maximizing the use of available data (Allison, 2009). In multiply imputing the data, we will include a set of auxiliary variables in our multiple imputation model in order to increase the plausibility of the assumption necessary for this approach, that data are missing at random (Allison, 2009). Additionally, we will include variables used as regressands in our multiple imputation model in order to increase efficiency and potentially reduce bias (Allison, 2009). However, if records are missing on outcome variables, they will be omitted from analyses of impacts using the multiply imputed data (i.e., listwise deletion).
There is no content in this SEP example that identifies with the currently selected SEP Guidance Checklist element.