Can I create different groups to compare my results? (for example, based on gender or age)?
You can create different groups in the Group Results page by creating a filter with a specific selection of data. In the Group Results page select Compare and then select + Add Data Selection, in this window select edit filter. You can for example, create a filter with only males, and one with only females (you can save the filter for easy reuse later). Once you have selected at least two data sets, you can select whether you want to show a graph, table, or detect whether there are significant differences between the groups.
You can create different groups in the Group Results page by creating a filter with a specific selection of data. In the Group Results page select Compare and then select + Add Data Selection, in this window select edit filter. You can for example, create a filter with only males, and one with only females (you can save the filter for easy reuse later). Once you have selected at least two data sets, you can select whether you want to show a graph, table, or detect whether there are significant differences between the groups.
Why are there less participants in the group results page than there were recorded?
There is an automatic filter that excludes all recordings with a quality score lower than 4. These recordings contain very few data points that could be analyzed reliably, and would probably add noise and artifacts to the group results. You can change the quality filter decide which videos you want to include in the group analysis. It is advised to not include participants that received a 0 or 1 for the quality of the data, since the software was not able to reliably find the face. Even if the person seems to be clearly visible, sometimes the situation is still not ideal for the software (for example, when the light is strongly aimed at one side of the face). For tips to improve the quality of the data see the FAQ topic Participants recordings.
There is an automatic filter that excludes all recordings with a quality score lower than 4. These recordings contain very few data points that could be analyzed reliably, and would probably add noise and artifacts to the group results. You can change the quality filter decide which videos you want to include in the group analysis. It is advised to not include participants that received a 0 or 1 for the quality of the data, since the software was not able to reliably find the face. Even if the person seems to be clearly visible, sometimes the situation is still not ideal for the software (for example, when the light is strongly aimed at one side of the face). For tips to improve the quality of the data see the FAQ topic Participants recordings.
How is valence calculated?
The valence indicates whether the emotional status of the subject is positive or negative. 'Happy' is the only positive emotion, 'Sad', 'Angry', 'Scared' and 'Disgusted' are considered to be negative emotions. ‘Surprised’ can be either positive or negative. The valence is calculated
as the intensity of ‘Happy’ minus the intensity of the negative emotion with the highest
intensity. For instance, if the intensity of ‘Happy’ is 0.8 and the intensities of ‘Sad’, ‘Angry’,
‘Scared’ and ‘Disgusted’ are 0.1; 0.0; 0.05 and 0.05, respectively, then the valence. is 0.7.
The valence indicates whether the emotional status of the subject is positive or negative. 'Happy' is the only positive emotion, 'Sad', 'Angry', 'Scared' and 'Disgusted' are considered to be negative emotions. ‘Surprised’ can be either positive or negative. The valence is calculated
as the intensity of ‘Happy’ minus the intensity of the negative emotion with the highest
intensity. For instance, if the intensity of ‘Happy’ is 0.8 and the intensities of ‘Sad’, ‘Angry’,
‘Scared’ and ‘Disgusted’ are 0.1; 0.0; 0.05 and 0.05, respectively, then the valence. is 0.7.
How is arousal calculated?
Arousal indicates whether the test participant is active (+1) or not active (0). Arousal is based
on the activation of 20 Action Units (AUs) of the Facial Action Coding System (FACS). It is calculated with these steps:
1. The activation values (AV) of 20 AUs are taken as input. These are AU 1, 2, 4, 5, 6, 7, 9, 10,
12, 14, 15, 17, 18, 20, 23, 24, 25, 26, 27, and the inverse of 43. The value of AU43 (eyes closed)
is inverted because it indicates low arousal instead of high arousal like the other AUs.
2. The average AU activation values (AAV) are calculated over the last 60 seconds. During the
first 60 seconds of the analysis, the AAV is calculated over the analysis up to that moment.
AAV = Mean (AVpast 60 seconds)
3. The average AU activation values (AAV) are subtracted from the current AU activation
values (AV). This is done to correct for AUs that are continuously activated and might
indicate an individual bias. This results in the Corrected Activation Values (CAV).
CAV = Max(0, AV – AAV)
4. The arousal is calculated from these CAV values by taking the mean of the five highest
values.
Arousal = Mean(5 max values of CAV)
Arousal indicates whether the test participant is active (+1) or not active (0). Arousal is based
on the activation of 20 Action Units (AUs) of the Facial Action Coding System (FACS). It is calculated with these steps:
1. The activation values (AV) of 20 AUs are taken as input. These are AU 1, 2, 4, 5, 6, 7, 9, 10,
12, 14, 15, 17, 18, 20, 23, 24, 25, 26, 27, and the inverse of 43. The value of AU43 (eyes closed)
is inverted because it indicates low arousal instead of high arousal like the other AUs.
2. The average AU activation values (AAV) are calculated over the last 60 seconds. During the
first 60 seconds of the analysis, the AAV is calculated over the analysis up to that moment.
AAV = Mean (AVpast 60 seconds)
3. The average AU activation values (AAV) are subtracted from the current AU activation
values (AV). This is done to correct for AUs that are continuously activated and might
indicate an individual bias. This results in the Corrected Activation Values (CAV).
CAV = Max(0, AV – AAV)
4. The arousal is calculated from these CAV values by taking the mean of the five highest
values.
Arousal = Mean(5 max values of CAV)
What does the graph with the circumplex model depict?
The circumplex model of affect describes emotions in a two-dimensional circular space, containing arousal on the vertical axis and valence on the horizontal axis. The center of the circle represents a neutral valence and a medium level of activity. The circumplex model of affect in FaceReader is based on the model described by Russel (1980). The color of the heatmap is based on the percentage of time a certain affect was found in a video or camera analysis.
The circumplex model of affect describes emotions in a two-dimensional circular space, containing arousal on the vertical axis and valence on the horizontal axis. The center of the circle represents a neutral valence and a medium level of activity. The circumplex model of affect in FaceReader is based on the model described by Russel (1980). The color of the heatmap is based on the percentage of time a certain affect was found in a video or camera analysis.
How is the emotion intensity scored?
Each emotion is scored between 0-1. So it can be the case that the total of the emotion scores is higher than 1. Sometimes people express different emotions at the same time, so these are all scored independently. If you want to see the emotions relative to each other, you can select the pie chart, this converts each emotion (that you specify) to a percentage, with a total of 100%.
Each emotion is scored between 0-1. So it can be the case that the total of the emotion scores is higher than 1. Sometimes people express different emotions at the same time, so these are all scored independently. If you want to see the emotions relative to each other, you can select the pie chart, this converts each emotion (that you specify) to a percentage, with a total of 100%.
How are the comparison insights calculated?
The comparison insights are based on a 'Student's T-test' with unequal sample sizes and unequal variances between two means. This test can indicate whether two averages are significantly different (you can set the p-value). For all tests that are significant, the effect size is calculated with Cohen’s D. These effect sizes indicate whether the difference is small or large.
Note, the comparison insights give rough estimates of whether a difference is potentially interesting. It does not check if all assumptions are met, since these differ depending on each individual design and data. The tests are not corrected for multiple comparisons. The tests also do not account for within subject effects and always use the more stringent version of the t-test. For marketing researchers the insights give relevant information. We advise scientists to do additional statistic tests, based on a-priori hypothesis.
When there are more than 5 significant effects, only the most significant ones are given.
The comparison insights are based on a 'Student's T-test' with unequal sample sizes and unequal variances between two means. This test can indicate whether two averages are significantly different (you can set the p-value). For all tests that are significant, the effect size is calculated with Cohen’s D. These effect sizes indicate whether the difference is small or large.
Note, the comparison insights give rough estimates of whether a difference is potentially interesting. It does not check if all assumptions are met, since these differ depending on each individual design and data. The tests are not corrected for multiple comparisons. The tests also do not account for within subject effects and always use the more stringent version of the t-test. For marketing researchers the insights give relevant information. We advise scientists to do additional statistic tests, based on a-priori hypothesis.
When there are more than 5 significant effects, only the most significant ones are given.
Do you have a benchmark to compare the results to?
Emotion scores depend on many variables such as: the type of video, the intended audience, and the country. It is therefore difficult to compare the results to other commercials. We advise to create your own comparison! Do you have a commercial that performed particularly well, or did not perform well? Then include it in your assessment with your new commercial. It is then easy to build your own database within your own branch and with your own public.
For an example of the reach of scores that can be expected see the question below.
Emotion scores depend on many variables such as: the type of video, the intended audience, and the country. It is therefore difficult to compare the results to other commercials. We advise to create your own comparison! Do you have a commercial that performed particularly well, or did not perform well? Then include it in your assessment with your new commercial. It is then easy to build your own database within your own branch and with your own public.
For an example of the reach of scores that can be expected see the question below.
What are the average emotion scores that I can expect?
The emotions are scored within a range of 0 to 1. For a single individual, scores around 0.2 mean the emotion is slightly present (e.g. a smile), 0.5 moderately (e.g. an open smile), 0.8 strongly (e.g. laughter). Some people in general are not very expressive; there are also differences in the timing of emotions. Therefore, the average emotion ratings and a certain point will always be lower than what is expected for a single individual.
Also, note that commercials are usually not expected to elicit very strong emotions. For example, for funny commercials, we often see average peak scores for happiness from 0.1 to 0.2 (the total mean depends on the length of the video). Meaningful and subtle differences can easily be found if a good comparison is made between, for example, two videos, two episodes, or two audiences.
The emotions are scored within a range of 0 to 1. For a single individual, scores around 0.2 mean the emotion is slightly present (e.g. a smile), 0.5 moderately (e.g. an open smile), 0.8 strongly (e.g. laughter). Some people in general are not very expressive; there are also differences in the timing of emotions. Therefore, the average emotion ratings and a certain point will always be lower than what is expected for a single individual.
Also, note that commercials are usually not expected to elicit very strong emotions. For example, for funny commercials, we often see average peak scores for happiness from 0.1 to 0.2 (the total mean depends on the length of the video). Meaningful and subtle differences can easily be found if a good comparison is made between, for example, two videos, two episodes, or two audiences.
How do I automatically create independent variables (for example, to create groups for comparison)?
You can specify independent variables that you know in advance by adding them to your recording link.
You can also manually add/edit independent variables in the Recording Results page.
You can specify independent variables that you know in advance by adding them to your
recording link.
You can also manually add/edit independent variables in the Recording Results page.
Why are not all emotions visible in the graph?
You can select which emotions to visualize in the right corner of the graph.
You can select which emotions to visualize in the right corner of the graph.
Why do participants appear to be angry at the video or website?
The emotions are scored as prototypical basic emotional expressions, however, that doesn't have to mean that it is also the emotion people are feeling. It is also important to look at the context of the test and function of the expression, e.g. see this relevant blog by Noldus.
It is also important to realize that some people may look more angry or sad when they are looking neutral. It is always a good idea to look at changes in expressions over time or to include a neutral control stimulus, so you can compare the results to the baseline expression of your sample.
The emotions are scored as prototypical basic emotional expressions, however, that doesn't have to mean that it is also the emotion people are feeling. It is also important to look at the context of the test and function of the expression, e.g. see
this relevant blog by Noldus.
It is also important to realize that some people may look more angry or sad when they are looking neutral. It is always a good idea to look at changes in expressions over time or to include a neutral control stimulus, so you can compare the results to the baseline expression of your sample.
Can I export the raw data?
Yes, you can download a csv file with the raw data (under tab Export). This file includes the frame-by-frame (15 fps) scores of each emotion per participant and an average per stimulus/event for each participant.
Yes, you can download a csv file with the raw data (under tab Export). This file includes the frame-by-frame (15 fps) scores of each emotion per participant and an average per stimulus/event for each participant.
Is it easy to use the desktop version of FaceReader on data gathered with FaceReader Online?
Yes, if you own FaceReader, you can also export the whole FaceReader Online project and reanalyze the videos to include more output (e.g. AUs, head pose, eye direction) and more analysis options.
Yes, if you own
FaceReader, you can also export the whole FaceReader Online project and reanalyze the videos to include more output (e.g. AUs, head pose, eye direction) and more analysis options.
What eye tracking metrics are available?
When implementing eye tacking it can give several gaze statistics. First it is important to create areas for your stimulus (see Media file on how to create areas). During the analysis the algorithm determines whether a fixation occurs based on the gaze speed. During fixations visual content is processed (opposed to saccades, the rapid movements of the eye). In the gaze statistics you can find the average total duration of fixation time on user defined areas of interests, reflecting the total time spend looking at the area. It also calculates average duration of fixation, total number of fixations (average number of times people looked at the area), and time to first fixation (average time it took people to look at the area).
When implementing eye tacking it can give several gaze statistics. First it is important to create areas for your stimulus (see Media file on how to create areas). During the analysis the algorithm determines whether a fixation occurs based on the gaze speed. During fixations visual content is processed (opposed to saccades, the rapid movements of the eye). In the gaze statistics you can find the average total duration of fixation time on user defined areas of interests, reflecting the total time spend looking at the area. It also calculates average duration of fixation, total number of fixations (average number of times people looked at the area), and time to first fixation (average time it took people to look at the area).