# Expressions to reconsider

By [Jonas Ranstam BSc PhD](https://paragraph.com/@jonasranstam) · 2022-11-18

---

Many commonly used phrases are methodologically problematic or indicate methodological confusion. It is a good idea to check the used terminology. Here are a few common and annoying examples.

**Expression**: _Chi-squared tests were used to compare variables_.

Make sure that you distinguish between description and inference. Statistical hypothesis tests such as the Chi-squared test are used to test hypotheses about a population using data from a sample representing this population. P-values and statistical significance are not useful for describing the characteristics of the sample itself.

**Expression**: _Chi-squared tests and Fisher's exact test were used as appropriate_.

What do you consider appropriate?

**Expression**: _Independent samples T-tests._

Several different t-tests have been developed for use with independent groups, e.g. Student's, Satterthwaite's, Welch's, Prien's t-tests and even Hotelling's T-test.

**Expression**: _Normality has been assessed using the Shapiro–Wilk test._

No, a variable's distribution in the population can be tested using a statistical hypothesis test but cannot be "assessed" until the entire population has been observed, which is impossible for unobservable populations.

**Expression**: _Nonparametric data._

A nonparametric hypothesis can perhaps be tested using a distribution-free test, but the phrase "nonparametric data" is just nonsense.

**Expression**: _Statistical difference._

Aren't all differences statistical in some sense?

**Expression**: _No difference._

The statement is ambiguous. Does it mean no clinically relevant difference or no statistically significant difference?

**Expression**: _Significant difference_

This statement is also ambiguous. Does the word significant refer to uncertainty (statistical significance) or relevance (clinical significance)?

**Expression**: _Independently associated variables_

The expression is common when estimating causal effects using a multiple regression model. However, these estimates can only be interpreted based on explicit assumptions regarding cause-effect relationships among the included variables. A multivariable analysis is unnecessary when all variables are assumed to be independent Schisterman EF, Cole SR, Platt RW. Overadjustment Bias and Unnecessary Adjustment in Epidemiologic Studies. Epidemiology. 2009 Jul;20(4):488 - 495.

---

*Originally published on [Jonas Ranstam BSc PhD](https://paragraph.com/@jonasranstam/expressions-to-reconsider)*