Merging Incomplete Tertiary Datasets: The Case of 2-1-1 Information and Referral Missing Data

This study proposes a method to handle missing data when merging large tertiary datasets. Combining 25 datasets of 2-1-1 Texas Information & Referral Network's call records to analyze unmet needs during Hurricanes Katrina and Rita highlighted a considerable bias problem due to missing data of a key variable in some of the 25 datasets. First, extensive literature about existing techniques for handling missing data was reviewed but determined not applicable for this type of missing data problem. Next, a systematic algorithm was developed to calculate missing data types and strategies in tertiary datasets. Last, this method was applied to the 2-1-1 datasets to test its effectiveness on bias due to previous missing data. Using this approach, the volume of cases available for analysis was increased approximately 30 percent, hence greatly improving validity of the findings. In terms of social service research, minimizing bias of missing data in existing tertiary data resources would help policymakers make more appropriate decisions and provide more effective and timely social support and disaster services to residents. This new method could be applied to using tertiary data with a similar dilemma and contribute to increasing potential use of available public datasets.