Using the right terms matters

From 50 psychological and psychiatric terms to avoid to 10 cogneuro/stats terms to use.

There is a great review of the misuse of psychological terms titled ‘Fifty psychological and psychiatric terms to avoid: a list of inaccurate, misleading, misused, ambiguous, and logically confused words and phrases’ by Lilienfield et al. 2015 (doi: 10.3389/fpsyg.2015.01100). The review points out ‘errors’ that can be often found in the literature and is definitively worth reading. Here are 10 terms I took from this paper and for which I thought we have good alternatives.

Biomedical

(1)   A gene for. As the authors point out, this is extremely rare that a single gene has a causal relationship to behaviour. When testing for the role or involvement of a gene, we could use something like: ‘We found that gene X participates to the control of ’

(2) Genetically determined. When it comes to brain morphology, patterns of activation or behaviour there is no genetic determination. Best is to stick to a statistical conclusion like: ‘Between X and Y % of the variance of X was genetically explained.’

(3)   Comorbidity: Here I think we should stick to the standard definition of simultaneous presence of a disease or conditions within individuals. As the authors points out it is sometimes used to refer to covariations within a sample but this is simply wrong.  

Imaging

(4)  Brain region X lights up. This is a typical trait of activation studies, areas light up, but only if you use hot colours! I can only concur with the author of the article that we need to add the contrast under which it ‘lights up’, without that context ‘activations’ become meaningless. Area X shows significant activations in condition A relative to baseline or it shows a significantly stronger signal for condition A than B.

(5)  Neural signature. As the author put it ‘identifying a genuine neural signature would necessitate the discovery of a specific pattern of brain responses that possesses nearly perfect sensitivity and specificity for a given condition or other phenotype’ so simply never ever use that term – Most of the time we don’t know how sensitive is a pattern of activation, but if a pattern is seen in one condition only then we can say we found a specific neural pattern.

Statistics

(6)  No difference. Most often, biomedical/neuroscience paper use null hypothesis testing. In this framework we can only fail to reject the null hypothesis and it is therefore impossible to conclude that there is no difference. To conclude this alternative analyses (being frequentist or Bayesian) must be used.

(7)   p = 0.000. Clearly we cannot write many 0 as decimals since that simply equivalent as writing p=0 and the probability under the null that the observed effect differ from the null, cannot be 0 for a given sample. The authors suggest using p < 0.01 or p < 0.001, but it is possible to observe p=0 in a sample, using randomization procedures, in that case we can write p~0.

(8)  Interaction. The point made by the authors is that interaction used in the general sense and in the statistical term are different. While the former implies that multiple factors play a role in something, the statistical interaction implies that changes in one factor leads to changes in another. I guess the simple solution is to systematic use ‘statistical interaction’ if this is what we mean. 

(9)   Validity. It is pointed out in the paper that it is often use in ‘validity of the hypotheses’ or ‘validity of the test’. Validity refers to the accuracy of measures, and it thus should not be used otherwise. I concur here with the authors to use alternatives when describing results and hypotheses such as saying this is ‘empirically supported’ and when referring to a behavioural test say there is ‘evidence for construct validity’

(10) Reliability. The most common and horrific way to use that term appears in sentences like ‘our results were very reliable with a p value of 0.001’. I am not going to explain here what p values are but this has nothing to do with reliability. This term must be used only in the specific contexts of i) inter-rater (a measure is reliable if different raters give the same result), ii) test–retest (a measure is reliable if replications give the same result) and, iii) internal consistency (a test is reliable if the different items measure the same construct).