As science progresses, methods used to obtain knowledge are improving and becoming more complicated. Boisgontier & Cheval (2016) note that there is a transition to using mixed models for statistical analysis in various areas of science. They offer a better basis for analyzing variables than other more familiar tools. In neuroscience, where the use of Analysis of Variance (ANOVA) prevails, the transition began slightly later than in other fields (Boisgontier & Cheval, 2016). At the same time, the accuracy of neuroscience is questioned, and there is a need to use more reliable tools within its framework. Although the application of ANOVA may be simpler than linear mixed models (LMM), its disadvantages have a significant harmful impact, which justifies the importance of accelerating the transition.
The widespread use of ANOVA in the majority of neuroscience studies may be the reason why there are suspicions about the unreliability of the area. ANOVA is a statistical method used to study relationships applicable in various fields of science to analyze the results of experimental research – from sociology to neuroscience. In particular, it helps to determine the effect of one or more independent qualitative variables on a dependent variable. Independent variables in ANOVA are factors, and the purpose of the analysis is to find out the values of the dependent change concerning various factors and their conditions.
The tool has gained wide popularity due to the assumption that its use implies the independence of observations. However, in the field of neuroscience, observations of single subjects/animals are usually used, which cannot be independent and requires analysis through “models controlling for variability within units of observations” (Boisgontier & Cheval, 2016, p. 1004). Even another ANOVA model – repeated measures ANOVA, that considers observations’ non-independence, is limited since to be independent observations must be made in several conditions different from fixed ones (Boisgontier & Cheval, 2016). In neuroscience, it is impossible to consider all conditions, and only limited observations are possible. Ignoring the variability of conditions will result in errors, namely false positives.
Using LMM can solve the problem that arises in ANOVA. According to Boisgontier & Cheval (2016), mixed models take into account both multiple observations for a single subject/animal under a specific condition and observations under numerous conditions. Moreover, LMM solves the problem of incomplete data and eliminates the loss of information that may happen because of averaging. Other areas already use these tools widely; for example, there is a 260% increase in LMM applications in medicine (Boisgontier & Cheval, 2016). To reach other fields, neuroscience representatives need to raise awareness about the transition from ANOVA to mixed models.
My position concerning the article is that continual improvement of analysis tools is indeed crucial for all areas. Modifying the old and applying new models is part of science development. In terms of making ethical decisions that may affect other people’s lives, the article’s usefulness may be reflected in the fact that it will recall the importance of using reliable tools. Moreover, the text also pushes to question the effectiveness and fairness of all arguments that affect decision-making. For example, a study of the idea’s origins may show that an outdated tool was used to prove it, and there is a need for additional verification.
Thus, there is a transition from the use of ANOVA to LMM in scientific analysis. Some spheres as neuroscience carry out the transition longer than others due to the habit of using ANOVA. LMM is more flexible and accurate than ANOVA, but their application is much more complex, making the transition to their use in research difficult. However, mixed models solve fundamental problems of older tools, which helps to make analysis more accurate and avoid errors.
Reference
Boisgontier, M. P., & Cheval, B. (2016). The anova to mixed model transition. Neuroscience & Biobehavioral Reviews, 68, 1004-1005.