Yawning is a stereotyped phylogenetically ancient phenomenon that occurs in almost all vertebrates. As an emotional behavior and an expressive movement, yawning has many consequences; nevertheless, it has so far been poorly addressed in medical research and practice. Bringing together the latest research from many fields, this volume integrates current insights within embryology, ethology, neurophysiology, psychology, fMRI and pathology. The phylogenetic and ontogenetic aspects of yawning offer an interesting perspective on human development, and its occurrence in neurological diseases - an area explored by only a few investigators - may provide useful clinical information. This book will make valuable and fascinating reading to neurologists, sleep specialists, psychologists, ethologists and pharmacologists, as well as to anybody interested in uncovering the mystery of yawning.
120 - 127: Methodological Problems in the Study of Contagious Yawning
The recent interest in contagious yawning has raised several challenges as the varied methods of testing have left some unresolved issues. We do not know how differences in key variables affect the observed rates of yawning, and we highlight these as being in need of direct testing. Different researchers analyze their results differently, and we make some recommendations for more rigorous, thorough and informative analyses. Ultimately, problems arise when authors compare studies that used different methods and different analyses without acknowledging how these differences may have affected the results. In these cases, authors make inappropriate comparisons, which lead to conclusions that add confusion to the literature. Our goal in raising awareness of these issues is to generate new experiments and improve the discussion of existing research. With its link to empathy, a more standardized study of contagious yawning may be a useful tool for a variety of disciplines.
Recently there has been renewed interest in the study of contagious yawning. More research on contagious yawning has been published in the span of 2000-2009 than in the previous 2 decades (and even longer) combined. The cause of this renewed interest is the theoretical link between contagious yawning and empathy [1, 2], which has been supported empirically [3, 4]. Since the early experiments on contagious yawning [5-8], the study of this phenomenon has employed developmental , comparative [10-14], neurological [15-18] and mental health [19-21] approaches in a pursuit that is beginning to formulate a varied body of research. However, the now numerous studies employed different methods of experimentation and analysis with different strengths and weaknesses, which complicates efforts to compare results across studies. Focusing on experimental manipulations of contagious yawning, which have been far more numerous than naturalistic observations , our goal of this chapter is to discuss some of these problems, propose some solutions, and highlight some unanswered questions to facilitate future experiments and discourse.
Problems in Experimental Design
When trying to compare results across studies, the first challenge is the different ways people test contagious yawning. The most fundamental aspects of design, what is presented and how, vary so much that no two studies by different sets of authors have employed the same methods (table 1).
Three variables in particular differ more frequently than others:
(1) The duration of the yawn shown ranges from 3 to 9 s (table 1), with some studies not reporting this detail. This magnitude of difference is small, and there is ambiguity in the length of the actual yawn within the stated length of a clip (i.e. how much neutral-expression lead-time and lag-time there was). As all researchers were attempting to show more or less ‘typical’ yawns, relatively small differences in duration may not influence the results greatly. Nonetheless, no one has expressly studied whether the duration of a stimulus yawn influences rates of contagion.
(2) Much more challenging is that the number of yawns displayed to the subjects ranges from 1 to 140. Of the 18 studies that presented yawns, there are 13 different total number of yawns shown. This is an enormous magnitude of variation, with the added wrinkle that in some experiments all of the clips were shown in 1 session, whereas in others they were shown in multiple sessions. We do not know if the number of yawns viewed by the subjects influences the rate of contagion. Does watching more yawns induce more individuals to yawn? Does watching more yawns induce more yawns from the same individuals? The obvious co-variable that may also contribute to yawning rates is the amount of time that the subjects are observed. The longer subjects are observed, the more likely one will observe yawning. These variables, time and the number of yawns presented, can be manipulated independently, and this awaits testing. In the meantime, we need to be cautious when comparing relative rates of contagious yawning between studies with even moderately different durations of exposure. Many studies make these comparisons without acknowledging that the methods differ in potentially important ways.
(3) Another critical aspect of design is the control. There is no consensus about what makes for the ideal control. Studies have used smiles, coughs, laughs, mouths opening and closing (also called gaping), still faces and species-specific expressions (for 3 comparative studies). The merits of different controls may be argued. Some may support gaping as it mimics much of the motor pattern of a yawn, yet is a meaningless expression (note: Nahab et al.  had a reason to use this expression specific to using fMRI). Conversely, it can be argued that gaping does not include, and therefore control for, motor activity in the eye region the way smiles, laughs and species-specific expressions do, which Provine  demonstrated was important for the perception of yawns. Since there has been no comparison of yawn rates in response to these different expressions, we have no data on whether any are better or worse controls than the others. However, significant differences in yawning rates when comparing yawns and controls have been detected using all of these expressions except coughs and still faces, which have only been used once in an fMRI study . Therefore, the specific control expression selected may not be important, as multiple expressions seem to turn up baseline levels of yawning. It would be interesting to know if yawning rates in response to any of these expressions differ from watching a neutral face, or even a blank screen. When selecting a control, it stands to reason that one should not use an extremely arousing expression, such as fear, but we reiterate that there has been no research comparing yawning rates to different controls.
Studies on yawning
There are many more methodological details that differ between studies, as can be seen in table 1, and the same comment applies: these details have not been studied for their effect on contagious yawning. We chose to highlight 3 aspects of methodology that we feel have the biggest potential to impact rates of contagion. The duration of the yawn clip may not be important, ultimately, and many different control expressions may do an equally good job of eliciting a baseline rate of yawning. However, the number of the yawn clips shown, with the co-variable of the amount of time the subjects are observed, has the potential to drastically impact the results and therefore interpretation. Research in this area is gravely needed, both for future experiments, and to more accurately compare existing studies that vary greatly in this domain.
Problems in Analysis
What qualifies as yawn contagion? This simple question has no clear answer in the literature. Whereas all authors agree on what yawn contagion is: a yawn stimulated by another yawn, there is no agreement on how to measure it. The most common analytical methods assess population-level differences between yawn and control conditions using either parametric (e.g. Student's t test) or non-parametric (e.g. Wilcoxon signed-rank test, ?2, McNemar's test) statistics. Buried within the parametric or non-parametric distinction is a subtle difference in what these tests measure. The t test and Wilcoxon examine whether there were more yawns in the yawn condition than the control; ?2 and the McNemar variant examine whether more subjects yawned in the yawn than the control condition, regardless of the magnitude of response. The ?2 and McNemar may be of more limited use as they require a lot of 0's in the control condition. Whether one type of test is preferable to the other is not clear, but experimenters and readers need to be aware of the differences.
While relatively rare, a few studies did not do a statistical test between their yawn and control responses, which is a significant oversight. Population-level tests tell us whether one group as a whole is different from another, but they do not inform us about any one individual. Less commonly used, a few studies (curiously, all studies of non-human animals) have also employed individual-level statistics in the form of binomials. Binomials compare one individual's rate of response between two conditions. This test can identify individuals who show a strong difference in yawning regardless of the population-level statistics. In short, they can identify high-performers. The limitation to binomials is that they are not sensitive to small differences that may be meaningful and consistent within a population, hence the need for population-level statistics. These two levels of analysis should be seen as complementary: population-level tests tell us whether one group as a group differs from another, whereas individual-level tests identify particularly strong performers within a population. One is not better than the other; rather, they test different questions. All experimental manipulations of contagious yawning should compare the response to the yawn stimulus to the control stimulus. This is mandatory. Our suggestion is that a thorough analysis should contain statistics at both population and individual levels.
One particular result that is frequently reported but difficult to interpret is the percent of subjects showing contagious yawning. The difficulty arises because first one must define operationally what qualifies a subject as showing contagious yawning. An operational definition of contagious yawning has not been made explicit in any study reporting this result. There are at least 3 different ways one could operationally define whether a subject shows contagious yawning: (1) if subject A yawns at all during the yawn condition, regardless of how many times or if the subjects yawns at all, or even more frequently, during the control; (2) if subject A yawns more in the yawn condition than the control; (3) if subject A yawns significantly more during the yawn than the control on a binomial test. Method 1 is the most liberal (and most commonly used), and method 3 is the most conservative. Using our own data on chimpanzees as an example , these three different methods yield percentages of contagiously yawning subjects of 65, 56.5 and 26%, respectively. Thus, when authors compare percentages calculated with different methods [calculation method not reported in: 10, 12, 13, 21] they end up drawing conclusions from heterogeneous comparisons.
Secondly, how many individuals are observed to yawn during the yawn condition will depend to some extent upon the duration of the experiment. The longer a subject is observed, the more likely one will see a yawn. This is not a problem when comparing results within a study since the yawn and control conditions will be the same, but when comparing results between studies, there are serious problems if one study watched subjects for 5 min, and another for 20 min. Comparing yawns per minute, as opposed to the absolute number of yawning individuals, is one way to control for this.
The last problem we raise with this comparison is that a population that naturally yawns at high rates will look like possessing more contagion when compared to a population that naturally yawns very little, regardless of how these populations perform between yawn and control conditions. One can imagine a population that yawns a lot spontaneously, with no significant difference between the yawn and the control condition, and another population that yawns very little, but does show a significant difference. Would it be accurate to conclude that the first population shows more contagious yawning than the second? It is the difference between the yawn and control conditions that indicates susceptibility to contagion. A better comparison would be to look at the magnitudes of the difference in response between yawn and control conditions. Therefore, we would look at which populations show strongly significant differences, moderately significant differences, or no differences at all. This allows each population to control for itself and its own natural rates of yawning, as well as controlling for different methodologies, as discussed above.
Our analysis yields the following broad recommendations:
new controlled experiments on variables of methodology, specifically how the number of yawns shown, the time duration of the experiment, and different control expressions affect rates of yawning;
demonstrating yawn contagion experimentally requires a statistically significant difference between the response to a yawn stimulus and a control;
calculations of percent of subjects showing yawn contagion are not informative;
when making comparisons between different studies, authors must acknowledge how different methods and analyses may influence the results.
The last 10 years has seen a resurgence in interest in contagious yawning. We are very excited about the varied directions research has taken, and we hope these trends continue. Studying key variables of presenting yawns will be beneficial for two reasons: (1) it will facilitate comparisons across studies, including clarifying which variables in past studies may be important; (2) it will bring us closer to a standardized methodology, which may be useful as a diagnostic test for some mental health conditions [3, 19-21]. Answering these basic questions and limiting confusion in the literature will facilitate using contagious yawning as a serious tool to better understand how empathy functions from developmental, comparative, neurological and mental health perspectives.