Tag Archives: visualisation

How to chase ERP monsters hiding behind bars

I think detailed and informative illustrations of results is the most important step in the statistical analysis process, whether we’re looking at a single distribution, comparing groups, or dealing with more complex brain imaging data. Without careful illustrations, it can be difficult, sometimes impossible, to understand our results and to convey them to an audience. Yet, from specialty journals to Science & Nature, the norm is still to hide rich distributions behind bar graphs or one of their equivalents. For instance, in ERP (event-related potential) research, the equivalent of a bar graph looks like this:

figure1

Figure 1. ERP averages in 2 conditions. Paired design, n=30, cute little red star indicates p<0.05.

All the figures in this post can be reproduced using Matlab code available on github.

Figure 1 is very much standard in the field. It comes with a little star to attract your attention to one time point that has reached the magic p<0.05 threshold. Often, the ERP figure will be complemented with a bar graph:

figure1b

Figure 1b. Bar graph of means +/- SEM for conditions 1 & 2.

Ok, what’s wrong with this picture? You might argue that the difference is small, and that the statistical tests have probably not been corrected for multiple comparisons. And in many cases, you would be right. But many ERP folks would reply that because they focus their analyses on peaks, they do not need to correct for multiple comparisons. Well, unless you have a clear hypothesis for each peak, then you should at least correct for the number of peaks or time windows of interest tested if you’re willing to flag any effect p<0.05. I would also add that looking at peaks is wasteful and defeats the purpose of using EEG: it is much more informative to map the full time-course of the effects across all sensors, instead of throwing valuable data away (Rousselet & Pernet, 2011).

Another problem with Figure 1 is the difficulty in mentally subtracting two time-courses, which can lead to underestimating differences occurring between peaks. So, in the next figure, we show the mean difference as well:

figure2

Figure 2. Mean ERPs + mean difference. The black vertical line marks the time of the largest absolute difference between conditions.

Indeed, there is a modest bump in the difference time-course around the time of the significant effect marked by the little star. The effect looks actually more sustained than it appears by just looking at the time-courses of the two original conditions – so we learn something by looking at the difference time-course. The effect is much easier to interpret by adding some measure of accuracy, for instance a 95% confidence interval:

figure3

Figure 3. Mean ERPs + mean difference + confidence interval.

We could also show confidence intervals for condition 1 and condition 2 mean ERPs, but we are primarily interested in how they differ, so the focus should be on the difference. Figure 3 reveals that the significant effect is associated with a confidence interval only very slightly off the zero mark. Although p<0.05, the confidence interval suggests a weak effect, and Bayesian estimation might actually suggest no evidence against the null (Wetzels et al. 2011). And this is why the focus should be on robust effect sizes and their illustration, instead of binary outcomes resulting from the application of arbitrary thresholds. How do we proceed in this case? A simple measure of effect size is to report the difference, which in our case can be illustrated by showing the time-course of the difference for every participant (see a nice example in Kovalenko et al. 2012). And what’s lurking under the hood here? Monsters?

figure4

Figure 4. Mean ERPs + mean difference + confidence interval + individual differences.

Yep, it’s a mess of spaghetti monsters!

After contemplating a figure like that, I would be very cautious about my interpretation of the results. For instance, I would try to put the results into context, looking carefully at effect sizes and how they compare to other manipulations, etc. I would also be very tempted to run a replication of the experiment. This can be done in certain experimental situations on the same participants, to see if effect sizes are similar across sessions (Bieniek et al. 2015). But I would certainly not publish a paper making big claims out of these results, just because p<0.05.

So what can we say about the results? If we look more closely at the distribution of differences at the time of the largest group difference (marked by a vertical line in Figure 2), we can make another observation:

figure5

Figure 5. Distribution of individual differences at the time of the maximum absolute group difference.

About 2/3 of participants show an effect in the same direction as the group effect (difference < 0). So, in addition to the group effect, there are large individual differences. This is not surprising. What is surprising is the usual lack of consideration for individual differences in most neuroscience & psychology papers I have come across. Typically, results portrayed in Figure 1 would be presented like this:

“We measured our favourite peak in two conditions. It was larger in condition 1 than in condition 2 (p<0.05), as predicted by our hypothesis. Therefore, when subjected to condition 1, our brains process (INSERT FAVOURITE STIMULUS, e.g. faces) more (INSERT FAVOURITE PROCESS, e.g. holistically).”

Not only this is a case of bad reverse inference, it is also inappropriate to generalise the effect to the entire human population, or even to all participants in the sample – 1/3 showed an effect in the opposite direction after all. Discrepancies between group statistics and single-participant statistics are not unheard of, if you dare to look (Rousselet et al. 2011).

Certainly, more subtle and honest data description would go a long way towards getting rid of big claims, ghost effects and dodgy headlines. But how many ERP papers have you ever seen with figures such as Figure 4 and Figure 5? How many papers contain monsters behind bars? Certainly, “my software does not have that option” doesn’t cut it; these figures are easy to make in Matlab, R or Python. If you don’t know how, ask a colleague, post questions on online forums, there are plenty of folks eager to help. For Matlab code, you could start here for instance.

Now: the final blow. The original ERP data used for this post are real and have huge effect sizes (check Figure A2 here for instance). However, the effect marked by a little star in Figure 1 is a false positive: there are no real effects in this dataset! The current data were generated by sampling trials with replacement from a pool of 7680 trials, to which pink noise was added, to create 30 fake participants and 2 fake conditions. I ran the fake data making process several times and selected the version that gave me a significant peak difference, because, you know, I love peaks. So yes, we’ve been looking at noise all along. And I’m sure there is plenty of noise out there in published papers. But it is very difficult to tell, because standard ERP figures are so poor.

How do we fix this?

  • make detailed, honest figures of your effects;
  • post your data to an online repository for other people to scrutinise them;
  • conclude honestly about what you’ve measured (e.g. “I only analyse the mean, I don’t know how other aspects of the distributions behave”), without unwarranted generalisation (“1/3 of my participants did not show the group effect”);
  • replicate new effects;
  • report p values if you want, but do not obsess over the 0.05 threshold, it is arbitrary, and continuous distributions should not be dichotomised (MacCallum et al. 2002);
  • focus on effect sizes.

References

Bieniek, M.M., Bennett, P.J., Sekuler, A.B. & Rousselet, G.A. (2015) A robust and representative lower bound on object processing speed in humans. The European journal of neuroscience.

Kovalenko, L.Y., Chaumon, M. & Busch, N.A. (2012) A pool of pairs of related objects (POPORO) for investigating visual semantic integration: behavioral and electrophysiological validation. Brain Topogr, 25, 272-284.

MacCallum RC, Zhang S, Preacher KJ, Rucker DD. 2002. On the practice of dichotomization of quantitative variables. Psychological Methods 7: 19-40

Rousselet, G.A. & Pernet, C.R. (2011) Quantifying the Time Course of Visual Object Processing Using ERPs: It’s Time to Up the Game. Front Psychol, 2, 107.

Rousselet, G.A., Gaspar, C.M., Wieczorek, K.P. & Pernet, C.R. (2011) Modeling Single-Trial ERP Reveals Modulation of Bottom-Up Face Visual Processing by Top-Down Task Constraints (in Some Subjects). Front Psychol, 2, 137.

Wetzels, R., Matzke, D., Lee, M.D., Rouder, J.N., Iverson, G.J. & Wagenmakers, E.J. (2011) Statistical Evidence in Experimental Psychology: An Empirical Comparison Using 855 t Tests. Perspectives on Psychological Science, 6, 291-298.

the percentile bootstrap


Update: more in depth coverage is available in this tutorial, including R code:

A practical introduction to the bootstrap: a versatile method to make inferences by using data-driven simulations

For a very short introduction focused on the R implementation of the bootstrap, see this other tutorial:

The percentile bootstrap: a teaser with step-by-step instructions in R 

To make inferences about the population mean, the percentile bootstrap can perform poorly; instead use the bootstrap-t technique.


“The bootstrap is a computer-based method for assigning measures of accuracy to statistical estimates.” Efron & Tibshirani, An introduction to the bootstrap, 1993

“The central idea is that it may sometimes be better to draw conclusions about the characteristics of a population strictly from the sample at hand, rather than by making perhaps unrealistic assumptions about the population.” Mooney & Duval, Bootstrapping, 1993

Like all bootstrap methods, the percentile bootstrap relies on a simple & intuitive idea: instead of making assumptions about the underlying distributions from which our observations could have been sampled, we use the data themselves to estimate sampling distributions. In turn, we can use these estimated sampling distributions to compute confidence intervals, estimate standard errors, estimate bias, and test hypotheses (Efron & Tibshirani, 1993; Mooney & Duval, 1993; Wilcox, 2012). The core principle to estimate sampling distributions is resampling, a technique pioneered in the 1960’s by Julian Simon (particularly inspiring is how he used dice and cards to teach resampling in statistics classes). The technique was developed & popularised by Brad Efron as the bootstrap.

Let’s consider an example, starting with this small set of 10 observations:

1.2 1.1 0.1 0.8 2.6 0.7 0.2 0.3 1.9 0.4

To take a bootstrap sample, we sample n observations with replacement. That is, given the 10 original observations above, we sample with replacement 10 observations from the 10 available. For instance, one bootstrap sample from the example above could be (sorted for convenience):

0.4 0.4 0.4 0.8 0.8 1.1 1.2 2.6 2.6 2.6

a second one:

0.1 0.3 0.4 0.8 1.1 1.2 1.2 1.9 1.9 1.9

a third one:

0.1 0.4 0.7 0.7 1.1 1.1 1.1 1.1 1.9 2.6

etc.

As you can see, in some bootstrap samples, certain observations were sampled once, others more than once, and yet others not at all. The resampling process is akin to running many experiments.

fig1-bootstrap_philosophy

Figure 1. Bootstrap philosophy.

Essentially, we are doing fake experiments using only the observations from our sample. And for each of these fake experiments, or bootstrap sample, we can compute any estimate of interest, for instance the median. Because of random sampling, we get different medians from different draws, with some values more likely than other. After repeating the process above many times, we get a distribution of bootstrap estimates, let say 1,000 bootstrap estimates of the sample median. That distribution of bootstrap estimates is a data driven estimation of the sampling distribution of the sample median. Similarly, we can use resampling to estimate the sampling distribution of any statistics, without requiring any analytical formula. This is the major appeal of the bootstrap.

Let’s consider another example, using data from figure 5 of Harvey Motulsky’s 2014 article. We’re going to reproduce his very useful figure and add a 95% percentile bootstrap confidence interval. The data and Matlab code + pointers to R code are available on github. The file pb_demo.m will walk you through the different steps of bootstrap estimation, and can be used to recreate the figures from the rest of this post.

With the bootstrap, we estimate how likely we are, given the data, to obtain medians of different values. In other words, we estimate the sampling distribution of the sample median. Here is an example of a distribution of 1,000 bootstrap medians.

fig2-boot_median_est_density

Figure 2. Kernel density distribution of the percentile bootstrap distribution of the sample median.

The distribution is skewed and rather rough, because of the particular data we used and the median estimator of central tendency. The Matlab code let you estimate other quantities, so for instance using the mean as a measure of central tendency would produce a much smoother and symmetric distribution. This is an essential feature of the bootstrap: it will suggest sampling distributions given the data at hand and a particular estimator, without assumptions about the underlying distribution. Thus, bootstrap sampling distributions can take many unusual shapes.

The interval, in the middle of the bootstrap distribution, that contains 95% of medians constitutes a percentile bootstrap confidence interval of the median.

fig3-bootci_illustration

Figure 3. Percentile bootstrap confidence interval of the median. CI = confidence interval.

Because the bootstrap sample distribution above is skewed, it might be more informative to report a highest-density interval – a topic for another post.

To test hypotheses, we can reject a point hypothesis if it is not included in the 95% confidence interval (a p value can also be obtained – see online code). Instead of testing a point hypothesis, or in addition, it can be informative to report the bootstrap distribution in a paper, to illustrate likely sample estimates given the data.

Now that we’ve got a 95% percentile bootstrap confidence interval, how do we know that it is correct? In particular, how many bootstrap samples do we need? The answer to this question depends on your goal. One goal might be to achieve stable results: if you repeatedly compute a confidence interval using the same data and the same bootstrap technique, you should obtain very similar confidence intervals. Going back to our example, if we take a sub-sample of the data, and compute many confidence intervals of the median, we sometimes get very different results. The figure below illustrates 7 confidence intervals of the median using the same small dataset. The upper boundaries of the different confidence intervals vary far too much:

fig4-median_CI_rep

Figure 4. Repeated calculations of the percentile bootstrap confidence interval of the median for the same dataset.

The variability is due in part to the median estimator, which introduces strong non-linearities. This point is better illustrated by looking at 1,000 sorted bootstrap median estimates:

fig5-boot_median_est_sorted

Figure 5. Sorted bootstrap median estimates.

If we take another series of 1,000 bootstrap samples, the non-linearities will appear at slightly different locations, which will affect confidence interval boundaries. In that particular case, one way to solve the variability problem is to increase the number of bootstrap samples – for instance using 10,000 samples produces much more stable confidence intervals (see code). Using more observations also improves matters significantly.

If we get back to the question of the number of bootstrap samples needed, another goal is to achieve accurate probability coverage. That is, if you build a 95% confidence interval, you want the interval to contain the population value 95% of the time in the long run. Concretely, if you repeat the same experiment over and over, and for each experiment you build a 95% confidence interval, 95% of these intervals should contain the population value you are trying to estimate if the sample size is large enough. This can be achieved by using a conjunction of 2 techniques: a technique to form the confidence interval (for instance a percentile bootstrap), and a technique to estimate a particular quantity (for instance the median to estimate the central tendency of the distribution). The only way to find out which combo of techniques work is to run simulations covering a lot of hypothetical scenarios – this is what statisticians do for a living, and this is why every time you ask one of them what you should do with your data, the answer will inevitably be “it depends”. And it depends on the shape of the distributions we are sampling from and the number of observations available in a typical experiment in your field. Needless to say, the best approach to use in one particular case is not straightforward: there is no one-size-fits-all technique to build confidence intervals; so any sweeping recommendation should be regarded suspiciously.

The percentile bootstrap works very well, and in certain situations is the only (frequentist) technique known to perform satisfactorily to build confidence intervals of or to compare for instance medians and other quantiles, trimmed means, M estimators, regression slopes estimates, correlation coefficients (Wilcox 2012). However, the percentile bootstrap

does not perform well with all quantities, in particular with the mean (Wilcox & Keselman, 2003). You can still use the percentile bootstrap to illustrate the variability in the sample at hand, without making inferences about the underlying population. We do this in the figure below to see how the percentile bootstrap confidence interval compares to other ways to summarise the data.

Figure 6. Updated version of Motulsky’s 2014 figure 5.

This is a replication of Motulsky’s 2014 figure 5, to which I’ve added a 95% percentile bootstrap confidence interval of the mean. This figure makes a critical point: there is no substitute for a scatterplot, at least for relatively small sample sizes. Also, using the mean +/- SD, +/- SEM, with a classic confidence interval (using t formula) or with a percentile bootstrap confidence interval can provide very different impressions about the spread in the data (although it is not their primary objective). The worst representation clearly is mean +/- SEM, because it provides a very misleading impression of low variability. Here, because the sample is skewed, mean +/- SEM does not even include the median, thus providing a wrong estimation of the location of the bulk of the observations. It follows that results in an article reporting only mean +/- SEM cannot be assessed unless  scatterplots are provided, or at least estimates of skewness, bi-modality and complementary measures of uncertainty for comparison. Reporting a boxplot or the quartiles does a much better job at conveying the shape of the distribution than any of the other techniques. These representations are also robust to outliers. In the next figure, we consider a subsample of the observations from Figure 6, to which we add an outlier of increasing size: the quartiles do not move.

fig7-outliers_quartiles

Figure 7. Outlier effect on the quartiles. The y-axis is truncated.

Contrary to the quartiles, the classic confidence interval of the mean is not robust, so it provides very inaccurate results. In particular, it assumes symmetry, so even though the outlier is on the right side of the distribution, both sides of the confidence interval get larger. The mean is also  pulled towards the outlier, to the point where it is completely outside the bulk of the observations. I cannot stress this enough: you cannot trust mean estimates if scatterplots are not provided.

fig8-outliers_classic_ci

Figure 8. Outlier effect on the classic confidence interval of the mean.

In comparison, the percentile bootstrap confidence interval of the mean performs better: only its right side, the side affected by the outlier, expends as the outlier gets larger.

fig9-outliers_pbci_mean

Figure 9. Outlier effect on the percentile bootstrap confidence interval of the mean.

Of course, we do not have to use the mean as a measure of central tendency. It is trivial to compute a percentile bootstrap confidence interval of the median instead, which, as expected, does not change with outlier size:

fig10-outliers_pbci_median

Figure 10. Outlier effect on the percentile bootstrap confidence interval of the median.

Conclusion

The percentile bootstrap can be used to build a confidence interval for any quantity, whether its sampling distribution can be estimated analytically or not. However, there is no guarantee that the confidence interval obtained will be accurate. In fact, in many situations alternative methods outperform the percentile bootstrap (such as percentile-t, bias corrected, bias corrected & accelerated (BCa), wild bootstraps). With this caveat in mind, I think the percentile bootstrap remains an amazingly simple yet powerful tool to summarise the accuracy of an estimate given the variability in the data. It is also

the only frequentist tool that performs well in many situations – see Wilcox 2012 for an extensive coverage of these situations.

Finally, it is important to realise that the bootstrap does make a very strong & unwarranted assumption: only the observations in the sample can ever be observed. For this reason, for small samples the bootstrap can produce rugged sampling distributions, as illustrated above. Rasmus Bååth wrote about the limitations of the percentile bootstrap and its link to Bayesian estimation in a blog post I highly recommend; he also provided R code for the bootstrap and the Bayesian bootstrap in another post.

References

Efron, B. & Tibshirani Robert, J. (1993) An introduction to the bootstrap. Chapman & Hall, London u.a.

Mooney, C.Z. & Duval, R.D. (1993) Bootstrapping : a nonparametric approach to statistical inference. Sage Publications, Newbury Park, Calif. ; London.

Motulsky, H.J. (2014) Common misconceptions about data analysis and statistics. J Pharmacol Exp Ther, 351, 200-205.

Wilcox, R.R. (2012) Introduction to robust estimation and hypothesis testing. Academic Press, Amsterdam ; Boston.

Wilcox, R.R. & Keselman, H.J. (2003) Modern Robust Data Analysis Methods: Measures of Central Tendency. Psychological Methods, 8, 254-274.

Robust effect sizes for 2 independent groups

When I was an undergrad, I was told that beyond a certain sample size (n=30 if I recall correctly), t-tests and ANOVAs are fine. This was a lie. I wished I had been taught robust methods and that t-tests and ANOVAs on means are only a few options among many alternatives. Indeed, t-tests and ANOVAs on means are not robust to outliers, skewness, heavy-tails, and for independent groups, differences in skewness, variance (heteroscedasticity) and combinations of these factors (Wilcox & Keselman, 2003; Wilcox, 2012). The main consequence is a lack of statistical power. For this reason, it is often advised to report a measure of effect size to determine, for instance, if a non-significant effect (based on some arbitrary p value threshold) could be due to lack of power, or reflect a genuine lack of effect. The rationale is that an effect could be associated with a sufficiently large effect size but yet fail to trigger the arbitrary p value threshold. However, this advise is pointless, because classic measures of effect size, such as Cohen’s d, its variants, and its extensions to ANOVA are not robust.

To illustrate the problem, first, let’s consider a simple situation in which we compare 2 independent groups of 20 observations, each sampled from a normal distribution with mean = 0 and standard deviation = 1. We then add a constant of progressively larger value to one of the samples, to progressively shift it away from the other. As illustrated in Figure 1, as the difference between the two groups increases, so does Cohen’s d. The Matlab code to reproduce all the examples is available here, along with a list of matching R functions from Rand Wilcox’s toolbox.

fig1-cohend_3ex

Figure 1. Examples of Cohen’s d as a function of group differences. For simplicity, I report the absolute value of Cohen’s d, here and in subsequent figures.

We can map the relationship between group mean differences and d systematically, by running a simulation in which we repeatedly generate two random samples and progressively shift one away from the other by a small amount. We get a nice linear relationship (Figure 2).

fig2-cohend_sysmap

Figure 2. Linear relationship between Cohen’s d and group mean differences.

Cohen’s d appears to behave nicely, so what’s the problem? Let’s consider another example, in which we generate 2 samples of 20 observations from a normal distribution, and shift their means by a fixed amount of 2. Then, we replace the largest observation from group 2 by progressively larger values. As we do so, the difference between the means of group 1 and group 2 increases, but Cohen’s d decreases (Figure 3).

fig3-cohend_outliers

Figure 3. Cohen’s d is not robust to outliers.

Figure 4 provides a more systematic illustration of the effect of extreme values on Cohen’s d for the case of 2 groups of 20 observations. As the group difference increases, Cohen’s d wrongly suggests progressively lower effect sizes.

fig4-cohend_sysout

Figure 4. Cohen’s d as a function of group mean differences in the presence of one outlier. There is an inverse and slightly non-linear relationship between the two variables.

What is going on? Remember that Cohen’s d is the difference between the two group means divided by the pooled standard deviation. As such, neither the numerator nor the denominator are robust, so that even one unusual value can potentially significantly alter d and lead to the wrong conclusions about  effect size. In the example provided in Figure 4, d gets smaller as the mean difference increases because the denominator of d is composed of a non-robust estimator of dispersion, the variance, such that the outlier increases variability, which leads to an increase of the denominator, and thus a lower d. The outlier also has a strong effect on the mean, which leads to an increase of the numerator, and thus larger d. However, the outlier has a stronger effect on the variance than the mean: this imbalance explains the overall decrease of d with increasing outlier size. I leave it as an exercise to understand the origin of the non-linearity in Figure 4. It has to do with the differential effect of the outlier on the mean and the variance.

One could argue that the outlier value added to one of the groups could be removed, which would solve the problem. There are 3 objections to this argument:

  • there are situations in which extreme values are not outliers but expected and plausible observations from a skewed or heavy tail distribution, and thus physiologically or psychologically meaningful values. In other words, what looks like an outlier in a sample of 20 observations could well look very natural in a sample of 200 observations;
  • for small sample sizes, relatively small outliers could go unnoticed but still affect effect size estimation;
  • outliers are not the only problem: skewness & heavy tails can affect the mean and the variance and thus d.

For instance, in some cases, two groups can differ in skewness, as illustrated in Figure 5. In the left panel, the two kernel density estimates illustrate two samples of 100 observations from a normal distribution. The two groups overlap only moderately, and Cohen’s d is high. In the right panel, group 1, with a mean of zero, is the same as in the previous panel; group 2, with a mean of 2, is almost identical to the one in the left panel, except that its largest 10% observations were replaced with slightly larger observations. As a result, the overlap between the two distributions is the same in the two panels – yet Cohen’s d is quite smaller in the second example.

fig5-cohend_mixed

Figure 5. Cohen’s d for normal & skewed distributions.

The point of this example is to illustrate the potential for discrepancies between a visual inspection of two distributions and Cohen’s d. Clearly, in Figure 5, a useful measure of effect size should provide the same estimates for the two examples. Fortunately, several robust alternatives have this desirable property, including Cliff’s delta, the Kolmogorov-Smirnov test statistic, Wilcox & Muska’s Q, and mutual information.

Robust versions of Cohen’s d

Before going over the 4 robust alternatives listed above, it is useful to consider that Cohen’s d is part of a large family of estimators of effect size, which can be described as the ratio of a difference between two measures of central tendency (CT), over some measure of variability:

(CT1 – CT2) / variability

From this expression, it follows that robust effect size estimators can be derived by plugging in robust estimators of central tendency in the numerator and robust estimators of variability in the denominator. Several examples of such robust alternatives are available, for instance using trimmed means and Winsorised variances (Keselman et al. 2008; Wilcox 2012). R users might want to check these functions from Wilcox for instance:

  • akp.effect
  • yuenv2
  • med.effect

There are also extensions of these quantities to the comparison of more than one group (Wilcox 2012).

Robust & intuitive measures of effect sizes

In many situations, the robust effect sizes presented above can bring a great improvement over Cohen’s d and its derivatives. However, they provide only a limited perspective on the data. First, I don’t find this family of effect sizes the easiest to interpret: having to think of effects in standard deviation (or robust equivalent) units is not the most intuitive. Second, this type of effect sizes does not always answer the questions we’re most interested in (Cliff, 1996; Wilcox, 2006).

The simplest measure of effect size: the difference

Fortunately, effect sizes don’t have to be expressed as the ratio difference / variability. The simplest effect size is simply a difference. For instance, when reporting that group A differs from group B, typically people report the mean for each group. It is also very useful to report the difference, without normalisation, but with a confidence or credible interval around it, or some other estimate of uncertainty. This simple measure of effect size can be very informative, particularly if you care about the units. It is also trivial to make it robust by using robust estimators, such as the median when dealing with reaction times and other skewed distributions.

Probabilistic effect size and the Wilcoxon-Mann-Whitney U statistic

For two independent groups, asking by how much the central tendencies of the two groups differ is useful, but this certainly does not exhaust all the potential differences between the two groups. Another perspective relates to a probabilistic description: for instance, given two groups of observations, what is the probability that one random observation from group 1 is larger than a random observation from group 2? Given two independent variables X and Y, this probability can be defined as P(X > Y). Such probability gives a very useful indication of the amount of overlap between the two groups, in a way that is not limited to and dependent on measures of central tendency. More generally, we could consider these 3 probabilities:

  • P(X > Y)
  • P(X = Y)
  • P(X < Y)

These probabilities are worth reporting in conjunction with illustrations of the group distributions. Also, there is a direct relationship between these probabilities and the Wilcoxon-Mann-Whitney U statistic (Birnbaum, 1956; Wilcox 2006). Given sample sizes Nx and Ny:

U / NxNy = P(X > Y) + 0.5 x P(X = Y)

In the case of two strictly continuous distributions, for which ties do not occur:

U / NxNy = P(X > Y)

Cliff’s delta

Cliff suggested to use P(X > Y) and P(X < Y) to compute a new measure of effect size. He defined what is now called Cliff’s delta as:

delta = P(X > Y) – P(X < Y)

Cliff’s delta estimates the probability that a randomly selected observation from one group is larger than a randomly selected observation from another group, minus the reverse probability (Cliff, 1996). It is estimated as:

delta = (sum(x > y) – sum(x < y)) / NxNy

In this equation, each observation from one group is compared to each observation in the other group, and we count how many times the observations from one group are higher or lower than in the other group. The difference between these two counts is then divided by the total number of observations, the product of their sample sizes NxNy. This statistic ranges from 1 when all values from one group are higher than the values from the other group, to -1 when the reverse is true. Completely overlapping distributions have a Cliff’s delta of 0. Because delta is a statistic based on ordinal properties of the data, it is unaffected by rank preserving data transformations. Its non-parametric nature reduces the impact of extreme values or distribution shape. For instance, Cliff’s delta is not affected by the outlier or the difference in skewness in the examples from Figure 3 & 5.

For an MEEG application, we’ve used Cliff’s delta to quantify effect sizes in single-trial ERP distributions (Bieniek et al. 2015). We also used Q, presented later on in this post, but it behaved so similarly to delta that it does not feature in the paper.

An estimate of the standard error of delta can be used to compute a confidence interval for delta. When conditions differ, the statistical test associated with delta can be more  powerful than the Wilcoxon-Mann-Whitney test, which uses the wrong standard error (Cliff, 1996; Wilcox, 2006). Also, contrary to U, delta is a direct measure of effect size, with an intuitive interpretation. There are also some attempts at extending delta to handle more than two groups (e.g. Wilcox, 2011). Finally, Joachim Goedhart has provided an Excel macro to compute Cliff’s delta.

Update: Cliff’s delta is also related to the later introduced “common-language effect size” – see this post from Jan Vanhove.

All pairwise differences

Cliff’s delta is a robust and informative measure of effect size. Because it relies on probabilities, it normalises effect sizes onto a common scale useful for comparisons across experiments. However, the normalisation gets rid of the original units. So, what if the units matter? A complementary perspective to that provided by delta can be gained by considering all the pairwise differences between individual observations from the two groups (Figure 6). Such distribution can be used to answer a very useful question: given that we randomly select one observation from each group, what is the typical difference we can expect? This can be obtained by computing for instance the  median of the pairwise differences. An illustration of the full distribution provides a lot more information: we can see how far away the bulk of the distribution is from zero, get a sense of how large differences can be in the tails…

fig6-all_pairwise_differences

Figure 6. Illustration of all pairwise differences. Left panel: scatterplots of the two groups of observations. One observation from group 1 (in red) is compared to all the observations from group 2 (in orange). The difference between all the pairs of observations is saved and the same process is applied to all the observations from group 1. Right panel: kernel density estimate of the distribution of all the pairwise differences between the two groups. The median of these differences is indicated by the continuous vertical line; the 1st & 3rd quartiles are indicated by the dashed vertical lines.

Something like Figure 6, in conjunction with Cliff’s delta and associated probabilities, would provide a very useful summary of the data.

When Cohen’s d & Cliff’s delta fail

Although robust alternatives to Cohen’s d considered so far, including Cliff’s delta, can handle well situations in which 2 conditions differ in central tendency, they fail completely to describe situations like the one in Figure 7. In this example, the two distributions are dramatically different from each other, yet Cohen’s d is exactly zero, and Cliff’s delta is very close to zero.

fig7-vardiffexample

Figure 7. Measures of effect size for two distributions that differ in spread, not in location. Cd = Cohen’s d, delta = Cliff’s delta, MI = mutual information, KS = Kolmogorov-Smirnov test statistics, Q = Wilcox & Muska’s Q.

Here the two distributions differ in spread, not in central tendency, so it would wise to estimate spread instead. This is indeed one possibility. But it would also be nice to have an estimator of effect size that can handle special cases like this one as well. Three estimators fit the bill, as suggested by the title of Figure 7.

The Kolmogorov-Smirnov statistic

It’s time to introduce a powerful all-rounder: the Kolmogorov-Smirnov test statistic. The KS test is often mentioned to compare one distribution to a normal distribution. It can also be used to compare two independent samples. In that context, the KS test statistic is defined as the maximum of the absolute differences between the empirical cumulative distribution functions (ecdf) of the two groups. As such KS is not limited to differences in central tendency; it is also robust, independent of the shape of distributions, and provides a measure of effect size bounded between 0 and 1. Figure 8 illustrates the statistic using the example from Figure 7. The KS statistic is quite large, suggesting correctly that the two distributions differ. More generally, because it is robust and sensitive to differences located anywhere in the distributions, the KS test is a solid candidate for a default test for two independent samples. However, the KS test is more sensitive to differences in the middle of the distributions than in the tails. To correct this problem, there is also a weighted version of the KS test which provides increased sensitivity to differences in the tails of the distributions – check out the ks R function from Wilcox.

fig8-vardiff_ks_illustration

Figure 8. Illustration of the KS statistic for two independent samples. The top panel shows the kernel density estimates for the two groups. The lower panel shows the matching empirical cumulative distribution functions. The thick black line marks the maximum absolute difference between the two ecdfs – the KS statistic. Figure 8 is the output of the ksstat_fig Matlab function written for this post.

The KS statistic non-linearly increases as the difference in variance between two samples of 100 observations progressively increases (Figure 9). The two samples were drawn from a standard normal distribution and do not differ in mean.

fig9-vardiff_map

Figure 9. Relationship between effect sizes and variance differences. The 3 measures of effect size illustrated here are sensitive to distribution differences other than central tendency, and are therefore better able to handle a variety of cases compared to traditional effect size estimates.

Wilcox & Muska’s Q

Similarly to KS, the Q statistic is also a non-parametric measure of effect size. It ranges from 0 to 1, with chance level at 0.5. It is the probability of correctly deciding whether a randomly selected observation from one of two groups belongs to the first group, based on the kernel density estimates of the two groups (Wilcox & Muska, 1999). Essentially, it reflects the degree of separation between two groups. Again, similarly to KS, in situations in which two distributions differ in other aspects than central tendency, Q might suggest that a difference exists, whereas other methods such as Cohen’s d or Cliff’s delta would not (Figure 9).

Mutual information

In addition to the KS statistic and Q, a third estimator can be used to quantify many sorts of differences between two or more independent samples: mutual information (MI). MI is a non-parametric measure of association between distributions. As shown in Figure 9, it is sensitive to group differences in spread. MI is expressed in bits and is quite popular in neuroscience – much more so than in psychology. MI is a powerful and much more versatile quantity than any of the tools we have considered so far. To learn more about MI, check out Robin Ince’s tutorial with Matlab & Python code and examples, with special applications to brain imaging. There is also a clear illustration of MI calculation using bins in Figure S3 of Schyns et al. 2010.

In the lab, we use MI to quantify the relationship between stimulus variability and behaviour or brain activity (e.g. Rousselet et al. 2014). This is done using single-trial distributions in every participant. Then, at the group level, we compare distributions of MI between conditions or groups of participants. We thus use MI as a robust measure of within-participant effect size, applicable to many situations. This quantity can then be illustrated and tested across participants. This strategy is particularly fruitful to compare brain activity between groups of participants, such as younger and older participants. Cliff’s delta for instance could then be used to quantify the MI difference between groups.

Comparisons of effect sizes

We’ve covered several useful robust measures of effect size, with different properties. So, which one should be used? In statistics, the answer to this sort of questions often is “it depends”. Indeed, it depends on your needs and on the sort of data you’re dealing with. It also depends on which measure makes more sense to you. The code provided with this post will let you explore the different options using simulated data or your own data. For now, we can get a sense of the behaviour of delta, MI, KS and Q for relatively large samples of observations from a normal distribution. In Figure 10, two distributions are progressively shifted from each other.

fig10-escomp_kde

Figure 10. Examples of effect size estimates for different distribution shifts.

Figure 11 provides a more systematic mapping of the relationship between effect size estimates and the difference between the means of two groups of 100 observations. The KS statistic and Q appear to have similar profiles, with a linear rise for small differences, before progressively reaching a plateau. In contrast, Cliff’s delta appears to be less variable and to reach a maximum earlier than KS and Q. MI differs from the other 3 quantities with its non-linear rise for small mean differences.

fig11-escomp_diffmean

Figure 11. Relationship between effect sizes and mean differences.

To more clearly contrast the 4 effect sizes, all their pairwise comparisons are provided in Figure 12. From these comparisons, it seems that KS and Q are almost completely linearly related. If this is the case, then there isn’t much advantage in using Q given that it is much slower to compute than KS. Other comparisons reveal different non-linearities between estimators. These differences would certainly be worth exploring in particular experimental contexts… But enough for this post.

fig12-escomp_sys

Figure 12. Relationship between effect sizes.

Final notes

Given that Cohen’s d and related estimators of effect size are not robust suggests that they should be abandoned in favour of robust methods. This is not to say that Cohen’s d is of no value – for instance in the case of single-trial ERP distributions of 100s of trials, it would be appropriate (Bieniek et al. 2015). But for typical group level analyses, I see no reason to use non-robust methods such as Cohen’s d. And defending the use of Cohen’s d and related measures for the sake of continuity in the literature, so that readers can compare them across studies is completely misguided: non-robust measures cannot be compared because the same value can be obtained for different amounts of overlap between distributions. For this reason, I am highly suspicious of any attempt to perform meta-analysis or to quantify effect sizes in the literature using published values, without access to the raw data. To allow true comparisons across studies, there is only one necessary and sufficient step: to share your data.

In the literature, there is a rampant misconception assuming that statistical tests and measures of effect size are different  entities. The Kolmogorov-Smirnov test and Cliff’s delta demonstrate that both aspects can be combined elegantly. Other useful measures of effect size, such as mutual information, can be used to test hypotheses by combining them with a bootstrap or permutation approach.

Which technique to use in which situation is something best worked out by yourself, given your own data and extensive tests. Essentially, you want to find measures that are informative and intuitive to use, and that you can trust in the long run. The alternatives described in this post are not the only ones on the market, but they are robust, informative, intuitive, and they cover a lot of useful situations. For instance, if the fields of neuroscience and psychology were to use the Kolmogorov-Smirnov test as default test when comparing two independent groups, I would expect a substantial reduction in the number of false negatives reported in the literature. The Kolmogorov-Smirnov test statistic is also a useful measure of effect size on its own. But because the KS test does not tell us how two distributions differ, it requires the very beneficial addition of detailed illustrations to understand how two groups differ.  This comment applies to all the techniques described in this post, which, although useful, do not provide a full picture of the effects. Notably, they do not tell us how two distributions differ. But detailed illustrations can be combined with robust estimation to compare 2 entire distributions.

References

Bieniek, M.M., Bennett, P.J., Sekuler, A.B. & Rousselet, G.A. (2015) A robust and representative lower bound on object processing speed in humans. The European journal of neuroscience.

Birnbaum ZW. 1955. On a use of the Mann-Whitney statistic

Cliff N. 1996. Ordinal methods for behavioral data analysis. Mahwah, N.J.: Erlbaum

Keselman HJ, Algina J, Lix LM, Wilcox RR, Deering KN. 2008. A generally robust approach for testing hypotheses and setting confidence intervals for effect sizes. Psychol Methods 13: 110-29

Rousselet, G.A., Ince, R.A., van Rijsbergen, N.J. & Schyns, P.G. (2014) Eye coding mechanisms in early human face event-related potentials. J Vis, 14, 7.

Wilcox RR. 2006. Graphical methods for assessing effect size: Some alternatives to Cohen’s d. Journal of Experimental Education 74: 353-67

Wilcox, R.R. (2011) Inferences about a Probabilistic Measure of Effect Size When Dealing with More Than Two Groups. Journal of Data Science, 9, 471-486.

Wilcox RR. 2012. Introduction to robust estimation and hypothesis testing. Amsterdam ; Boston: Academic Press

Wilcox RR, Keselman HJ. 2003. Modern Robust Data Analysis Methods: Measures of Central Tendency. Psychological Methods 8: 254-74

Wilcox RR, Muska J. 2010. Measuring effect size: A non-parametric analogue of omega(2). The British journal of mathematical and statistical psychology 52: 93-110

Simple steps for more informative ERP figures

I read, review and edit a lot of ERP papers. A lot of these papers have in common shockingly poor figures. Here I’d like to go over a few simple steps that can help to produce much more informative figures. The data and the code to reproduce all the examples are available on github.

Let’s first consider what I would call the standard ERP figure, the one available in so many ERP papers (Figure 1). It presents two paired group averages for one of the largest ERP effect on the market: the contrast between ERP to noise textures (in black) and ERP to face images (in grey). This standard figure is essentially equivalent to a bar graph without error bars: it is simply unacceptable. At least, in this one, positive values are plotted up, not down, as can still be seen in some papers.

fig1_standard

Figure 1. Standard ERP figure.

How can we improve this figure? As a first step, one could add some symbols to indicate at which time points the two ERPs differ significantly. So in Figure 2 I’ve added red dots marking time points at which a paired t-test gave p<0.05. The red dots appear along the x-axis so their timing is easy to read. This is equivalent to a bar graph without error bars but with little stars to mark p<0.05.

fig2_standard_with_stats

Figure 2. Standard figure with significant time points.

You know where this is going: next we will add confidence intervals, and then more. But it’s important to consider why Figure 2 is not good enough.

First, are significant effects that interesting? We can generate noise in Matlab or R for instance, perform t-tests, and find significant results – doesn’t mean we should write papers about these effects. Although no one would question that significant effects can be obtained by chance, I am yet to see a single paper in which an effect is described as potential false positive. Anyway, more information is required about significant effects:

  • do they make sense physiologically? For instance, you might find a significant ERP difference between 2 object categories at 20 ms, but that does not mean that the retina performs object categorisation;

  • how many participants actually show the group effect? It is possible to get significant group effects with very few individual participants showing a significant effect themselves. Actually, with large enough sample sizes you can pretty much guarantee significant group effects;

  • what is the group effect size, e.g. how large is the difference between two conditions?

  • how large are effect sizes in individual participants?

  • how do effect sizes compare to other known effects, or to effects observed at other time points, such as in the baseline, before stimulus presentation?

Second, because an effect is not statistically significant (p<0.05), it does not mean it is not there, or that you have evidence for the lack of effect. Similarly to the previous point, we should be able to answer these questions about seemingly non-significant effects:

  • how many participants do not show the effect?

  • how many participants actually show an effect?

  • how large are the effects in individual participants?

  • is the group effect non-significant because of the lack of statistical power, e.g. due to skewness, outliers, heavy tails?

Third, most ERP papers report inferences on means using non-robust statistics. Typically, results are then discussed in very general terms as showing effects or not, following a p<0.05 cutoff. What is assumed, at least implicitly, is that the lack of significant mean differences implies that the distributions do not differ. This is clearly unwarranted because distributions can differ in other aspects than the mean, e.g. in dispersion, in the tails, and the mean is not a robust estimator of central tendency. Thus, interpretations should be limited to what was measured: group differences in means, probably using a non-robust statistical test. That’s right, if you read an ERP paper in which the authors report:

“condition A did not differ from condition B”

the sub-title really is:

“we only measured a few time-windows or peaks of interest, and we only tested group means using non-robust statistics and used poor illustrations, so there could well be interesting effects in the data, but we don’t know”.

Some of the points raised above can be addressed by making more informative figures. A first step is to add confidence intervals, which is done in Figure 3. Confidence intervals can provide a useful indication of the dispersion around the average given the inter-participant variability. But be careful with the classic confidence interval formula: it uses mean and standard deviation and is therefore not robust. I’ll demonstrate Bayesian highest density intervals in another post.

fig3_standard_with_ci

Figure 3. ERPs with confidence intervals.

 

Ok, Figure 3 would look nicer with shaded areas, an example of which is provided in Figure 4 – but this is rather cosmetic. The important point is that Figures 3 and 4 are not sufficient because the difference is sometimes difficult to assess from the original conditions.

fig4_standard_with_ci2

Figure 4. ERPs with nicer confidence intervals.

 

So in Figure 5 we present the time-course of the average difference, along with a confidence interval. This is a much more useful representation of the results. I learnt that trick in 1997, when I first visited the lab of Michele Fabre-Thorpe & Simon Thorpe in Toulouse. In that lab, we mostly looked at differences – ERP peaks were deemed un-interpretable and not really worth looking at…

fig5_difference

Figure 5. ERP time-courses for each condition and their difference.

 

In Figure 5, the two vertical red lines mark the latency of the two difference peaks. They coincide with a peak from one of the two ERP conditions, which might be reassuring for folks measuring peaks. However, between the two difference peaks, there is a discrepancy between the top and bottom representations: whereas the top plot suggests small differences between the two conditions around ~180 ms, the bottom plot reveals a strong difference with a narrow confidence interval. The apparent discrepancy is due the difficulty in mentally subtracting two time-courses. It seems that in the presence of large peaks, we tend to focus on them and neglect other aspects of the data. Figure 6 uses fake data to illustrate the relationship between two ERPs and their difference in several situations. In row 1, try to imagine the time-course of the difference from the two conditions, without looking at the solution in row 2 – it’s not as trivial as it seems.

fig6_erp_differences

Figure 6. Fake ERP time-courses and their differences.

 

Because it can be difficult to mentally subtract two time-courses, it is critical to always plot the time-course of the difference. More generally, you should plot the time-course of the effect you are trying to quantify, whatever that is.

We can make another important observation from Figure 5: there are large differences before the ERP peaks ~140-180 ms shown in the top plot. Without showing the time-course of the difference, it is easy to underestimate potentially large effects occurring before or after peaks.

So, are we done? Well, as much as Figure 5 is a great improvement on the standard figure, in a lot of situations it is not sufficient, because it does not portray individual results. This is essential to interpret significant and non-significant results. For instance, in Figure 5, there is non-significant group negative difference ~100 ms, and a large positive difference ~120 to 280 ms. What do they mean? The answer is in Figure 7: a small number of participants seem to have clear differences ~100 ms despite the lack of significant group effect, and all participants have a positive difference ~120 to 250 ms post-stimulus. There are also large individual differences at most time points. So Figure 7 presents a much richer and compelling story than the group averages on their own.

 

fig7_full

Figure 7. A more detailed look at the group results. In the middle panel, individual differences are shown in grey and the group mean and its confidence interval are superimposed in red. The lower panel shows at every time point the proportion of participants with a positive difference.

 

Given the presence of a few participants with differences ~100 ms but the lack of significant group effects, it is interesting to consider participants individually, as shown in Figure 8. There, we can see that participants 6, 13, 16, 17 and 19 have a negative difference ~100 ms, unlike the rest of the participants. These individual differences are wiped out by the group statistics. Of course, in this example we cannot conclude that there is something special about these participants, because we only looked at one electrode: other participants could show similar effects at other electrodes. I’ll demonstrate  how to assess effects potentially spread across electrodes in another post.

fig8_every_participant

Figure 8. ERP differences with 95% confidence intervals for every participant.

 

To conclude: in my own research, I have seen numerous examples of large discrepancies between plots of individual results and plots of group results, such that in certain cases group averages do not represent any particular participant. For this reason, and because most ERP papers do not illustrate individual participants and use non-robust statistics, I simply do not trust them.

Finally, I do not see the point of measuring ERP peaks. It is trivial to perform analyses at all time points and sensors to map the full spatial-temporal distribution of the effects. Limiting analyses to peaks is a waste of data and defeats the purpose of using EEG or MEG for their temporal resolution.

References

Allen et al. 2012 is a very good reference for making better figures overall and with an ERP example, although they do not make the most crucial recommendation of plotting the time-course of the difference.

For one of the best example of clear ERP figures, including figures showing individual participants, check out Kovalenko, Chaumon & Busch 2012.

I have discussed issues with ERP figures and analyses here and here. And here are probably some of the most detailed figures of ERP results you can find in the literature – brace yourself for figure overkill.

One simple step to improve statistical inferences

There are many changes necessary to improve the quality of neuroscience & psychology research. Suggestions abound to increase science openness, promote better experimental designs, and educate researchers about statistical inferences. These changes are necessary and will take time to implement. As part of this process, here, I’d like to propose one simple step to dramatically improve the assessment of statistical results in psychology & neuroscience: to ban bar graphs.

banbargraphs
[https://figshare.com/articles/Ban_bar_graphs/1572294]

The benefits of illustrating data distributions has been emphasised in many publications and is often the topic of one of the first chapters of introductory statistics books. One of the most striking example is provided by Anscombe’s quartet, in which very different distributions are associated with the same summary statistics:

990px-Anscombe's_quartet_3.svg
[https://en.wikipedia.org/wiki/Anscombe%27s_quartet]

Moving away from bar graphs can achieve a badly needed departure from current statistical standards. Indeed, using for instance scatterplots instead of bar graphs can help shift the emphasis from the unproductive significant vs. non-significant dichotomy to a focus on what really matters: effect sizes and individual differences. By effect size, here, I do not mean Cohen’s d and other normalised non-robust equivalents (Wilcox, 2006); I mean, literally how big the effect is. Why does it matter? Say you have a significant group effect, it would be (more) informative to answer these questions as well:

  • how many participants actually show an effect in the same direction as the group?
  • how many participants show no effect, or an effect in the opposite direction as the group?
  • is there a smooth continuum of effects across participants, or can we identify sub-clusters of participants who appear to behave differently from the rest?
  • exactly how big are the individual results? For instance, what does it mean for a participant to be 20 ms faster in one condition than another? What if someone else is 40 ms faster? Our incapacity to answer these last questions in many situations simply reflects our lack of knowledge and the poverty of our models and predictions. But this is not an excuse to hide the richness of our data behind bar graphs.

Let’s consider an example from a published paper, which I will not identify. On the left is the bar graph alone representation, whereas the right panel contains both bars and scatterplots. The graphs show results from two independent groups: participants in each group were tested in two conditions, and the pairwise differences are illustrated here. For paired designs, illustrating each condition of the pair separately is inadequate to portray effect sizes because one doesn’t know which points are part of a pair. So here the authors selected the best option: to plot the differences, so that readers can appreciate effect sizes and their distributions across participants. Then they performed two mixed linear analyses, one per group, and found a significant effect for controls, and a non-significant effect in patients. These results seem well supported by the bar graph on the left, and the authors concluded that unlike controls, patients did not demonstrate the effect.

bargraphexample

We can immediately flag two problems with this conclusion. First, the authors did not test the group interaction, which is a common fallacy (Nieuwenhuis et al. 2011). Second, the lack of significance (p<0.05) does not provide evidence for the lack of effect, again a common fallacy (see e.g. Kruschke 2013). And obviously there is a third problem: without showing the content of the bars, I would argue that no conclusion can be drawn at all. Well, in fact the authors did report the graph on the right in the above figure! Strangely, they based their conclusions on the statistical tests instead of simply looking at the data.

The data show large individual differences and overlap between the two distributions. In patients, except for 2 outliers showing large negative effects, the remaining observations are within the range observed in controls. Six patients have results suggesting an effect in the same direction as controls, 2 are near zero, 3 go in the opposite direction. So, clearly, the lack of significant group effect in patients is not supported by the data, and arises from the use of a statistical test non-robust to outliers.

Here is what I would conclude about this dataset: both groups show an effect, but the effect sizes tend to be larger in controls than in patients. There are large individual differences, and in both groups, not all participants seem to show an effect. Because of these inter-participant differences, larger sample sizes need to be tested to properly quantify the effect. In light of the current data, there is evidence that patients do show an effect. Finally, the potential lack of effect in certain control participants, and the rather large effects in some patients, questions the use of this particular effect as a diagnostic tool.

I will describe how I would go about analysing this dataset in another post. At the moment, I would just point out that group analyses are highly questionable when groups are small and heterogenous. In the example above, depending on the goals of the experiment, it might suffice to report the scatterplots and a verbal description, as I provided in the previous paragraph. I would definitely favour that option to reporting a single statistical test of central tendency, whether it is robust or not.

The example of the non-significant statistical test in patients illustrate a critical point: if a paper reports bar graphs and non-significant statistical analyses of the mean, not much can be concluded! There might be differences in other aspects than the mean; central tendency differences might exist, but the assumptions of the test could have been violated because of skewness or outliers for instance. Without informative illustrations of the results, it is impossible to tell.

In my experience as reviewer and editor, 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. So show your data, and ask others to do the same.

“But what if I have clear effects, low within-group dispersion, and I know what I’m doing? Why can’t I use bar graphs?”

This is rather circular: unless you show the results using, for instance, scatterplots, no one knows for sure that you have clear effects and low within-group dispersion. So, if you have nothing to hide and you want to convince your readers, show your results. And honestly, how often do we get clear effects with low intra-group variability? 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.

“But scatterplots are ugly, they make my results look messy!”

First, your results are messy – scatterplots do not introduce messiness. Second, there is nothing stopping you from adding information to your scatterplots, for instance lines marking the quartiles of the distributions, or superimposing boxplots or many of the other options available.

References

Examples of more informative figures

Wilcox, R.R. (2006) Graphical methods for assessing effect size: Some alternatives to Cohen’s d. Journal of Experimental Education, 74, 353-367.

Allen, E.A., Erhardt, E.B. & Calhoun, V.D. (2012) Data visualization in the neurosciences: overcoming the curse of dimensionality. Neuron, 74, 603-608.

Weissgerber, T.L., Milic, N.M., Winham, S.J. & Garovic, V.D. (2015) Beyond bar and line graphs: time for a new data presentation paradigm. PLoS Biol, 13, e1002128.

Overview of robust methods, to go beyond ANOVAs on means

Wilcox, R.R. & Keselman, H.J. (2003) Modern Robust Data Analysis Methods: Measures of Central Tendency. Psychological Methods, 8, 254-274.

Wilcox, R.R. (2012) Introduction to robust estimation and hypothesis testing. Academic Press.

Extremely useful resources to go Bayesian

Kruschke, J.K. (2015) Doing Bayesian data analysis : a tutorial with R, JAGS, and Stan. Academic Press, San Diego, CA.

http://doingbayesiandataanalysis.blogspot.co.uk

Understanding Bayes

Other references

Kruschke, J.K. (2013) Bayesian estimation supersedes the t test. J Exp Psychol Gen, 142, 573-603.

Nieuwenhuis, S., Forstmann, B.U. & Wagenmakers, E.J. (2011) Erroneous analyses of interactions in neuroscience: a problem of significance. Nat Neurosci, 14, 1105-1107.