Random vs. systematic sampling from administrative databases involving human subjects

Two sampling techniques, simple random sampling (SRS) and systematic sampling (SS), were compared to determine whether they yield similar and accurate distributions for the following four factors: age, gender, geographic location and years in practice.

Any point estimate within 7 yr or 7 percentage points of its reference standard (SRS or the entire data set, i.e., the target population) was considered "acceptably similar" to the reference standard. The sampling frame was from the entire membership database of the Canadian Chiropractic Association. The two sampling methods were tested using eight different sample sizes of n (50, 100, 150, 200, 250, 300, 500, 800). From the profile/characteristics, summaries of four known factors [gender, average age, number (%) of chiropractors in each province and years in practice], between- and within-methods chi 2 tests and unpaired t tests were performed to determine whether any of the differences [descriptively greater than 7% or 7 yr] were also statistically significant. The strengths of the agreements between the provincial distributions were quantified by calculating the percent agreements for each (provincial pairwise-comparison methods). Any percent agreement less than 70% was judged to be unacceptable.

Our assessments of the two sampling methods (SRS and SS) for the different sample sizes tested suggest that SRS and SS yielded acceptably similar results. Both methods started to yield "correct" sample profiles at approximately the same sample size (n > 200).

SS is not only convenient, it can be recommended for sampling from large databases in which the data are listed without any inherent order biases other than alphabetical listing by surname.