Skip to main content
Fig. 6 | BMC Medical Research Methodology

Fig. 6

From: Preventing bias from selective non-response in population-based survey studies: findings from a Monte Carlo simulation study

Fig. 6

Bias prevented when extreme cases were not totally missing (70% response-rate). Notes: × 1 to the left, and × 2 to the right. The x-patterned parts of the bars show relative bias that was present when extreme cases were totally missing from the sample, but that was not present in the situations where extreme values were under-represented but not totally missing. In other words, this is relative bias that was prevented by having persons with extreme values in the sample, even if they were under-represented. Dark grey parts of the bars show bias still present when extreme cases were in the sample. When there is no dark grey part of a bar, all of the bias (100%) was prevented by not totally excluding extreme scores. The sum of the x-patterned and dark grey parts of a bar, is the total amount of relative bias present in situations where extreme scores were totally missing from the sample. The percentages in the figure were calculated by dividing the difference in relative bias in the two situations by the amount of relative bias in the situations without extreme cases, and multiplied by 100

Percentage prevented bias \( =\frac{\mathbf{R1}-\mathbf{R}\mathbf{2}\ }{\mathbf{R1}}\ast \mathbf{100} \) where R1 and R2 are relative bias in situations without and with extreme scores present, respectively. bnon 1 is dependency of liability of non-response on predictors (× 1, × 2, and × 3).bnon 2 is dependency of liability of non-response on health outcome

Back to article page