Are reconstructed self-reports of drinking reliable?

When follow-up interviews are missed, researchers sometimes try to reconstruct the data that would have been obtained by asking clients to recall the missed interval when they are interviewed at a later point. Are such data reliable? The reliability of remote reconstruction was estimated by asking 57 participants in a clinical trial to recall their drinking for the 12-month follow-up interval when interviewed, on average, 33 weeks later. These reports were obtained after delays averaging 231 days. These reconstructed reports were compared with the same clients' self-reports obtained during the 12-month interview. Reconstructed data were found to be reasonably accurate estimates of clients' reports at the time of original interview on global alcohol use variables including percentage of drinking days and total volume of consumption. No systematic bias was found for over-reporting or under-reporting at the point of reconstruction. However, on some variables (e.g. total drinks consumed), clients on average reported more drinking at the reconstruction period than during the initial interview. Discrepancies between initial and reconstructed reports were found to be unrelated to the length of delay in the second interview or to client characteristics.