Rank-Based Inverse Normal Transformations are Increasingly Used, But are They Merited?

By T. Beasley, Stephen Erickson, D. B. Allison. Behavior Genetics, 2009

319 citations recorded by the source corpus.

Many complex traits studied in genetics have markedly non-normal distributions. This often implies that the assumption of normally distributed residuals has been violated. Recently, inverse normal transformations (INTs) have gained popularity among genetics researchers and are implemented as an option in several software packages. Despite this increasing use, we are unaware of extensive simulations or mathematical proofs showing that INTs have desirable statistical properties in the context of genetic studies. We show that INTs do not necessarily maintain proper Type 1 error control and can also reduce statistical power in some circumstances. Many alternatives to INTs exist. Therefore, we contend that there is a lack of justification for performing parametric statistical procedures on INTs with the exceptions of simple designs with moderate to large sample sizes, which makes permutation testing computationally infeasible and where maximum likelihood testing is used. Rigorous research evaluating the utility of INTs seems warranted.

Subfields: Genetics, Management Science and Operations Research, Plant Science

View Paper: https://www.semanticscholar.org/paper/b09edca8c8a80a591de2a5a5901f610c27cda834

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Genetics

In Management Science and Operations Research

In Plant Science

Citations in the corpus

Testing the significance of a correlation with nonnormal data: comparison of Pearson, Spearman, transformation, and resampling approaches. (2012, methodology): “When heteroscedasticity or factorial designs are present, rankbased methods in general often demonstrate inferior performance (T. M. Beasley et al., 2009; Salter & Fawcett, 1993; Zimmerman, 1996).”
A statistical framework for neuroimaging data analysis based on mutual information estimated via a gaussian copula (2016, methodology): “Transformations of this type are sometimes referred to as inverse normal transformations and have been used in ﬁelds such as genetics [Beasley et al., 2009].”