6  Vibration of Effects

Vibration-of-effects analysis makes one target association the focus and asks how much it moves as the adjustment set changes (Patel et al. 2015).

6.1 The summary

voe() reduces the per-specification estimates to a compact summary: the median effect, the first and ninety-ninth percentiles, the relative ratio, the Janus fraction, and the proportion of specifications that reach significance.

voe(results)

The relative ratio is the ninety-ninth percentile of the exponentiated effect divided by the first percentile, so for a hazard or odds ratio it reports how many fold the association ranges across models. The Janus fraction is the share of estimates above zero; a value near zero or one means the sign is stable, and a value near one half means the association changes direction across defensible models.

6.2 The volcano

voe_volcano() plots the exponentiated effect against the negative base-ten logarithm of the p-value for every specification, with a dashed line at no effect and colour mapped to the number of adjusters.

voe_volcano(results, colour = n_adjusters)

The cloud shows the joint spread of effect size and significance. A colour gradient that separates left from right reveals that model size itself moves the association, a common and instructive pattern.

6.3 At scale

For phenome-wide or exposome-wide work with very many exposures, the quantvoe package computes vibration summaries across thousands of associations, extending the logic used here to a screening setting.

Patel, Chirag J., Belinda Burford, and John P. A. Ioannidis. 2015. “Assessment of Vibration of Effects Due to Model Specification Can Demonstrate the Instability of Observational Associations.” Journal of Clinical Epidemiology 68 (9): 1046–58.