Table 9:

Recommendations for future research

General Recommendations
▪ Take chances on studies that attempt to go beyond merely supporting traditional views of emotion.
▪ Support papers that attempt to study facial movements in real life, measuring context, sampling across cultures even though these studies are often less well controlled than studies in the laboratory, or may use facial stimuli that are less familiar to reviewers than canonical stimulus sets.
▪ Prioritize multidisciplinary studies that combine classical psychology methods with cognitive neuroscience, machine learning, etc.
▪ Support larger scale studies that bridge the lab and the world, that study individual people across many contexts, and measure emotional episodes in high dimensional detail, including physical, psychological and social features; encourage multiple investigators with different areas of expertise to work together.
▪ Support the development of computational approaches.
▪ Create R&D teams that pair psychologists and cognitive scientists trained in the psychology of emotion with engineers and computer scientists.
▪ Increase opportunities to test innovative methods and novel hypotheses, with the acknowledgement that such approaches are likely to elicit resistance from established scientists in the field of emotion.
▪ Generate more studies to identify the underlying neural mechanisms of the production and perception of facial movements.
▪ Direct funding to thornier but necessary new questions and be critical of projects that perpetuate past errors in emotion research.
▪ Direct healthy skepticism to tests, measures, and interventions that rely upon assumptions about “reading facial expressions of emotion” that seem to ignore published evidence and/or ignore integration of contextual information along with facial cues.
▪ Develop systematic, precise ways to describe and/or manipulate the dynamics of specific facial actions.
Stimulus Selection Recommendations
Limitations in stimulus selection can bias results.	▪ For perception studies, incorporate images from the wild (e.g., from multiple internet sources) to capture the full range of facial movements that humans produce in their everyday lives. ▪ For both production studies (where stimuli are designated to evoke emotion) and perception studies, build variation into stimulus sets so conclusions about emotion categories are not inferred (or evoked) from limited stimuli. Consider randomly sampling a variety of stimuli for a given category and treating stimuli as a random variable. ▪ For production studies, ensure that multiple stimuli per emotion category are used to evoke an emotion.
Little is known about the dynamics of the production and perception of emotion signaling.	▪ For perception studies, use dynamic images rather than rely on still images. For production studies, code the temporal dynamics of facial movements. ▪ Attempt to determine full dynamics and the apex of an emotion signal, changes to AUs as signals emerge and recede, and whether the kinematics of distinct AUs are similar or different across sequences or phases of emotion signaling. ▪ Ensure sufficient temporal resolution to allow for event segmentation to be assessed in perception studies.
The role of context is hotly debated, but rarely measured.	▪ Manipulate (or at least measure) the context in which target stimuli are perceived to evaluate whether data are truly stimulus-specific or influenced by context features. ▪ Describe in a systematic way the differences in context, whether for production or perception studies. Theories about the effects of context cannot be resolved until we address how to measure and quantify context.
Sample Selection Recommendations
Cross-cultural studies can provide powerful insights, but are limited in number and scope.	▪ Quantify, as best as possible, participants’ degree of exposure to the west, as well as the amount and type of formal schooling made available to participants. ▪ Harness technology to collect larger numbers of images and video sequences of facial movements across cultures. Use unlabeled classification approaches to discover emotion categories and their expressive forms, rather than continuing to ask whether other cultures are similar to the US. Remember that emotions and mental inferences may be understood differently in different cultures.
Task and Method Design Recommendations
Measurement versus interpretation of emotion is often blurred in research studies.	▪ Contrast more than one “emotion” category with a baseline, so that conclusions about a specific emotion category are not drawn from a comparison of an emotion versus a no emotion condition. ▪ Compare multiple emotion categories to non-emotion categories in a given study.
New insights about emotion are constrained by reliance on, and assumptions about, traditional categories.	▪ Measure emotional episodes in a multimodal way and attempt to discover explicit criteria for when an emotion is present or absent. Such discovery may require within-person approaches. ▪ Sample broader categories of possible emotion states than the limited categories used in prior research (move beyond categories such happiness, anger, sadness, fear, etc.). Test for variations in intensity within these categories and similarity across categories. ▪ Unless a study design is completely data-driven, explicitly state the theoretical priors of the research team. The distinction is between whether you are seeking to discover versus verify emotion categories. Both approaches are valid, but should be clearly articulated.
Data Analysis Recommendations
Findings are limited by a failure to consider issues related to forward and reverse inference.	▪ Address issues of reliability and specificity when presenting data on emotion expression and emotion perception. ▪ Use formal signal detection analytics and information theoretic measures rather relying on frequency or levels of agreement. Consider using Bayesian methods so that the null hypothesis can be tested directly.