Year: 2016
Type: article
Abstract: There are many changes necessary to improve the quality of neuroscience research. Suggestions abound to increase openness, transparency and reproducibility (Pernet & Poline, 2015; Gorgolewski & Poldrack, 2016; McKiernan et al., 2016; Spires-Jones et al., 2016; Weissgerber et al., 2016), to promote better experimental designs and analyses, and to educate researchers about statistical inferences (Kerr, 1998; Wagenmakers, 2007; Nieuwenhuis et al., 2011; Button et al., 2013; Head et al., 2015). These changes are necessary but will take time to implement. As part of this process, we would like to propose a few simple steps to improve the assessment of statistical results in neuroscience, by focusing on detailed graphical representations. Despite a potentially sophisticated experimental design, in a typical neuroscience experiment, raw continuous data tend to undergo drastic simplifications. As a result, it is common for the main results of a paper to be summarized in a few figures supported by a limited set of statistical tests. Unfortunately, graphical representations in many scientific journals, including neuroscience journals, tend to hide underlying distributions, with their excessive use of line and bar graphs (Allen et al., 2012; Weissgerber et al., 2015). This is problematic because common basic summary statistics, such as mean and standard deviation, are not robust and simply do not provide enough information about a distribution and can thus give misleading impressions about a dataset, particularly for the small sample sizes we are accustomed to in neuroscience (Anscombe, 1973; Wilcox, 2012). As a consequence of impoverished data representation, there can be a mismatch between the outcome of statistical tests, their interpretations and the information available in the data distributions. Let us consider a general and familiar scenario in which observations from two groups of participants are summarized using a bar graph and compared using a t-test on means. If the P-value is less than 0.05, we might conclude that we have a significant effect, with one group having larger values than the other; if the P-value is greater than 0.05, we might instead conclude that the two distributions do not differ. What is wrong with this description? In addition to the potentially irrational use of P-values (Gigerenzer, 2004; Wagenmakers, 2007; Wetzels et al., 2011), the situation highlights many limitations in current practices. Indeed, using bar graphs and an arbitrary P < 0.05 cut-off turns a potentially rich pattern of results into a simplistic binary outcome, in which effect sizes and individual differences are often ignored. For instance, a more fruitful approach to describing a seemingly significant group effect would be to answer the following questions as well: These questions can only be answered by using scatterplots or other detailed graphical representations of the results and by reporting quantities other than the mean and standard deviation of each group. Essentially, a significant t-test is neither necessary nor sufficient to understand exactly how two distributions differ (Wilcox, 2006). And because t-tests and anovas on means are not robust (for instance to skewness and outliers), failure to reach the 0.05 cut-off should not, indeed cannot, be used to claim that distributions do not differ. The reasons for this are first, the lack of significance (P < 0.05) is not the same as evidence for the lack of effect (Kruschke, 2013); second, unless appropriate robust statistical tests were employed, the lack of significance could be due to violations of the tests' assumptions and not to the lack of effect (Wilcox, 2012); third, distributions do not necessarily differ in central tendency and can potentially differ in their left or right tails only, for instance when only weaker animals respond to a treatment (Doksum, 1974; Doksum & Sievers, 1976; Wilcox, 2006; Wilcox et al., 2014). Essentially, if an article reports bar graphs and non-significant statistical analyses of the mean, not much can be concluded at all. Without detailed and informative illustrations of the results, it is impossible to tell if the distributions truly do not differ. Let us consider the example presented in Fig. 1, in which two groups of participants were tested in two conditions (2 independent × 2 dependent factor design). Figure 1A illustrates the results using a mean ± SEM bar graph. An anova on these data reveals a significant Group × Condition interaction (F1,20 = 8.345, P = 0.0091). Follow-up paired t-tests reveal a significant condition effect in group 1 (t(10) = 3.87, P = 0.003, difference = 1.24, difference's 95% confidence interval = [0.53, 1.95]), but not in group 2 (t(10) = 0.11, P = 0.92, difference = −0.03 [−0.71, 0.64]). These results do not seem well supported by the bar graph in Fig. 1A, given the overlapping error bars. Indeed, as we will see below, this visual representation is inappropriate to assess pairwise differences. Nevertheless, based on the t-test results, it is very common to conclude that group 1 is sensitive to the experimental manipulation, but not group 2. The discussion of the article might even pitch the results in more general terms, making claims about the brain in general. Although the scenario just described is very common in the literature, the conclusions are unwarranted. First, the lack of significance (P < 0.05) does not necessarily provide evidence for the lack of effect (Wetzels et al., 2011; Kruschke, 2013). Second, without showing the content of the bars, only limited conclusions can be drawn. So let us look inside the bars. Figure 1B shows the results from the two independent groups from Fig. 1A, but this time the pairwise differences are illustrated, so that we can appreciate effect sizes and their distributions across participants. The data show large individual differences and overlap between the two groups. In group 2, except for two potential outliers showing large negative effects, the remaining observations are within the range observed for group 1. Six participants from group 2 have differences suggesting an effect in the same direction as group 1, two are near zero, and three go in the opposite direction. So, clearly, the lack of a significant difference between conditions in group 2 is not supported by the data: yes, group 2 has overall smaller differences than group 1, but if group 1 is used as a control group, then most participants in group 2 appear to have effects similar to those of controls. Or so it seems, until we explore the nature of the difference scores by visualizing paired observations in each group (Fig. 1C). In group 1, as already observed, results in condition 2 are overall larger than in condition 1. In addition, participants with larger scores in condition 1 tend to have proportionally larger differences between conditions 1 and 2. Such relationship seems to be absent in group 2, which suggests that the two groups differ not only in the overall sensitivity to the experimental manipulation, but that other factors could be at play in group 1 and not in group 2. Thus, the group differences might actually be much more subtle than suggested by our first analyses. The group dichotomy is easier to appreciate in Fig. 1D, which shows a scatterplot of the paired observations for the two groups. In group 1, the majority of paired observations are above the unity line, demonstrating an overall group effect; there is also a positive relationship between the scores in condition 2 and the scores in condition 1. Again, no such relationship seems to be present in group 2. In particular, the two larger negative scores in group 2 are not associated with participants who scored particularly high or low in condition 1, giving us no clue as to the origin of these seemingly outlier scores. At this stage, we have learnt a great deal more about our dataset using detailed graphical representations than relying only on a bar graph and an anova. However, we would need many more than n = 11 participants in both groups to quantify the effects and understand how they differ across groups. We have also not exhausted all the representations that could help us make sense of the results. There is also potentially more to the data, because we have not considered the full distribution of single-trials/repetitions. For instance, it is very common to summarize a reaction time distribution of potentially hundreds of trials using a single number, which is then used to perform group analyses. An alternative is to study these distributions for each participant, to understand exactly how they differ between conditions (Pernet et al., 2011). This single-participant approach would be necessary here to understand how the two groups of participants respond to the experimental manipulation. In sum, there is much more to the data than what we could conclude from the bar graphs, the anova and t-tests. Once bar graphs are replaced by scatterplots (or boxplots etc.), the story can get much more interesting, subtle, convincing or the opposite! It depends what surprises the bars are holding and hiding. Showing scatterplots is the start of a discussion about the nature of the results, an invitation to go beyond the significant vs. non-significant dichotomy. For the particular results presented in Fig. 1, it is rather unclear what is gained by the anova compared with detailed graphical representations. It would, of course, be beneficial to properly model the data to make predictions (Kuhn & Johnson, 2013) and to allow integration across subsequent experiments and replication attempts – a critical step that requires Bayesian inferences (Verhagen & Wagenmakers, 2014). Finally, the problems described so far are not limited to relatively simple, one-dimensional data: they are present in more complex datasets as well, such as electroencephalographic (EEG) and magnetoencephalographic (MEG) time series. For instance, it is common to see EEG and MEG evoked responses illustrated using solely the mean across participants (Fig. 2A). Although common, this representation is equivalent to a bar graph but without error bars/whiskers. It is high time that such impoverished representations were a thing of the past. At a minimum, some measure of accuracy should be provided, such as confidence or credible intervals (Fig. 2B). Furthermore, because it can be difficult to mentally subtract two time courses, it is also important to illustrate the time course of the difference (Fig. 2C). In particular, showing the difference helps to consider all the data, not just large peaks, and helps to avoid underestimating potentially large effects occurring before or after the main peaks. In addition, Fig. 2C illustrates event-related potential (ERP) differences in each participant – an ERP version of a scatterplot. This more detailed illustration is essential to allow readers to assess effect sizes and inter-participant differences and ultimately to interpret significant and non-significant group results in relation to results from individual participants. For instance, in Fig. 2C, there is a non-significant group negative difference at ~100 ms and a large positive difference at ~120–280 ms. What do they mean? The individual traces, more clearly illustrated in Fig. 2D, reveal a small number of participants with relatively large differences at ~100 ms despite the lack of a significant group effect, and all participants have a positive difference ~120–250 ms post-stimulus. There are also large individual differences at most time points. So Fig. 2C,D, although certainly not the ultimate representations, presents a much richer and compelling story than the group averages on their own; Fig. 2C,D also suggests that more detailed group analyses would be beneficial, as well as single-participant analyses (Pernet et al., 2011; Rousselet & Pernet, 2011). For instance, Fig. 2E illustrates the time course of the difference, with a confidence interval, for participant 6. This representation confirms large and early differences in this participant. We can extend this approach by performing statistical analyses, controlled for multiple comparisons, for every participant (Fig. 2F). To conclude, we urge authors, reviewers and the neuroscience community, in general, to promote and implement the following guidelines to achieve higher standards in reporting neuroscience research. For publication in EJN, we suggest the following: Finally, we cannot ignore that using detailed illustrations for potentially complex designs, or designs involving many group comparisons, is not straightforward: research in that direction, including the creation of open-source toolboxes, is of great value to the community and should be encouraged by funding agencies.
Source: European Journal of Neuroscience
DOI: 10.1111/ejn.13400
Language: English
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,560
OpenAlex ID: W2515271233
