Contradictory subject response in longitudinal research

This study investigated the prevalence of logical and estimation errors in a 4-year longitudinal survey of substance abuse for elementary and junior high school students. Logical errors occur when a subject reports having used a substance on one measurement occasion, but reports never having used the substance on a second measurement occasion. Estimation errors occur when a subject reports a lifetime frequency of use on a later survey that is less than the lifetime frequency of use reported for an earlier survey.

Substance abuse surveys were completed every year for four years by predominantly black District of Columbia students, beginning when the students were in 4th/5th grade, and ending when they were in 7th/8th grade. (Ns ranged from 5,671 to 6,421). Errors for alcohol, cigarette and marijuana use were calculated for every pair of years. Background characteristics such as demographic and academic performance variables were also assessed for each of the four years.

Logical errors decreased for all substances as student got older, whereas estimation errors remained stable. Marijuana had the highest frequency of logical errors in every pair of years except one, and had the lowest frequency of estimation errors. Students who made errors in Years 1-2 were 1.64 times more likely to make errors again in Years 3-4. For some of the years, students who made any kind of error had lower school grades, were judged untrustworthy by their peers and were more likely to be males.

The alarming prevalence of errors should be considered when conducting longitudinal substance abuse research. Several methods of dealing with these errors are reviewed. In most cases, the best a researcher can do is assess the error rates and report them.