MR Dictionary

The F-statistic tests whether a model (e.g., a linear or logistic regression model) with independent variables explains variation in the dependent variable, compared to a model with no independent variables.

In MR analyses, the first-stage F-statistic (or just "F-statistic") can be used to test the strength of the association between the genetic instrumental variable (IV) and exposure of interest and the extent of the relative bias that is likely to occur in estimating the causal effect of the exposure on the outcome using the genetic IV. As a general rule, an F-statistic >10 indicates that the level of weak instrument bias is likely to be small. F-statistics should not be used to select IVs to avoid overfitting the estimation model. For example, an F-statistic of <10 does not indicate that an IV should not be used but, instead, it should be noted in the analysis that weak instrument bias should be a considered limitation. See Conditional F-statistic.

References

Other terms in 'Sources of bias and limitations in MR':