CODES 2000 User Forum -- Data Network Note #3
Test Data for Multiple Imputation
Applies to: CODES 2000 Version 2.1.314.
Last updated: Monday November 12, 2001.
SUMMARY
The first step for multiple imputation of linked record pairs is to run and merge several match passes in CODES 2000 at a very low cutoff probability, say 0.001. This produces a Match Pairs In Sets table that contains all record pairs with match probabilities greater than 0.001. Representative CODES data are now available for testing multiple imputation steps requiring such data. The data consist of three tables: Crash, Hospital, and Match Pairs In Sets. Crash records were extracted from a complete table containing 56,689 records. Hospital inpatient records were extracted from a complete table containing 12,877 records. The estimated total number of true matched record pairs for the matches was 1,200.
PROCEDURE
- Download the test data.
The test data are available in an Access 2000 database that you can download from this site. Download TestData.mdb database now (1.02 MB).
- Import the linkage Imputation Wizard form.
Open the TestData.mdb database.
Select Forms in the database window. Right click on the unused area and select Import... Navigate to the ImputeForm.mdb database and import frmImpute.
Open the Visual Basic environment by typing CTRL+G. Select Tools on the menu bar and click on References. If the CODES2000Lib.mde library database is not listed as an available reference, browse to the CODES2000\Library folder and add it:
Close the Visual Basic environment by typing ALT+Q.
- Open and use the Linkage Imputation Wizard to impute multiple complete links.
Select Forms in the database window and double-click on the form named frmImpute:
Enter the number of records in Table A (Crash), 56689.
Enter the number of records in Table B (Hospital), 12877.
Enter the estimated total number of true matched record pairs for Table A and Table B, 1200.
Select the imputation option. Normally this will be Impute missing links.
Select the number of imputations. Normally this will be 5.
Click on Run. CODES 2000 will read the test Match Pairs In Sets table and impute multiple complete sets of one-to-one linked pairs in tables named LinkedPairs1, LinkedPairs2, ...
- Link the test Crash and Hospital tables using the imputed tables of linked pairs.
Create a join query to link Crash to Hospital.
Run the join query to produce imputed links for SAS MI or other multiple imputation software for missing values.
