The ‘Crit Day’ Study

Yadid Ayzenberg and Brian D. Mayton

Abstract. We set out to characterize skin conductance changes as a response to public speaking. A group of 11 graduate students participated in the study. These students were presenting their thesis proposals to the faculty and other graduate students on a given day. Each participant was given two wristband sensors. Each of the wristbands measured electrodermal activity, skin temperature and 3-axis accelerometery. The participants were required to wear the sensors for 72 hours. In addition, they were required to submit daily surveys at the end of each day. The presentations were videotaped.

1. Introduction

Public speaking is often associated with high levels of stress and anxiety. Many individuals fear being in the spotlight, being the center of attention, and the possibility of standing on stage and forgetting what they meant to say. Additional elements that may add to the anxiety are the possibility of being asked questions that they are not prepared to answer. We set out to characterize the skin conductance changes related to public speaking. Would it be possible to find a correlation between speakers perceived stress levels and their physiological response? Were the levels of stress highest before, during or after a talk? Were the physiological responses of individuals that perceived themselves as very stressed higher than the ones of those that perceived themselves as calm?

To gain insight into the answers to these questions, we conducted a study and collected a dataset including electrodermal activity (EDA) and survey responses from 11 participants for several days around a stressful public speaking event. In contrast to most studies concerned with EDA data, we also collected this data from both wrists for the duration of the study in an attempt to better understand what might cause an asymmetric EDA response.

This paper is organized as follows. Section 2 reviews previous studies on the subject of this work. Section 3 provides details about the experiment. Section 4 presents the resulting data set. Section 5 provides results and analysis.

2. Previous Work

It is possible to measure changes in stress level by using electro dermal activity sensors. Our “fight-or-flight” response that is controlled by the sympathetic nervous system is heightened under stressful situations. The sub-dermal sweat glands’ activity, which is controlled by the sympathetic nervous system (SNS), can be measured using biosensors that are placed on the wrist and measure skin conductivity. Boucsein showed that skin conductance is highly correlated to changes in sweat gland activity. Work done by Setz et al. showed that it is possible to distinguish between cognitive load and stress in electrodermal activity. By using EDA peak height and the instantaneous peak rate it is possible to determine the stress level of a person.

Dimberg et al. found that subjects that possessed a public speaking fear reacted with increased skin conductance activity when exposed to social stimuli compared to low-fear subjects.

3. Experiment

Each year, the MIT Media Lab organizes an event in which second year master's students present their thesis proposal to all of faculty, students, and researchers. Each proposal is evaluated in terms of depth, originality, and contribution. The purpose of these presentations is twofold. The first is that the student will receive constructive feedback on the proposed work. The second is that it compels the students to spend time to explore possible research topics while constraining their schedule. All master’s students are required to participate in a thesis preparation seminar that will aid them in solidifying their ideas. At the beginning of this seminar, the students are told of the importance of their ‘Crit Day’ performance that will be a factor that weighs heavily in their Ph.D. application. Naturally, this adds significant levels of stress to the students who realize that this speaking event may determine the future of their academic career.

We decided that ‘Crit Day’ was a valuable opportunity to measure the physiological effects of public speaking.

3.1. Study Design

We recruited 11 graduate students, 8 males and 3 females, who were designated to present on ‘Crit Day’. The youngest was aged 23, the oldest was aged 27 and the mean age was 24.6 (SD=1.2). The students would receive compensation in return for their participation in the study. We made sure beforehand that none of the participants were taking medication that could potentially have an arousing or calming affect, or ADHD medication.

Before the study began, we asked the participants to report their perceived level of stress during the last month. We used the Perceived Stress Scale (PSS) as defined by Cohen et al., which is the most widely used psychological instrument for measuring the perception of stress.

It is a measure of the degree to which circumstances in one’s life are considered as stressful. Items were designed to determine how unpredictable, uncontrollable, and overloaded subjects find their lives to be.

During the three days of the study, we requested the participants to report their daily perceived stress levels at the end of each day. We used the Daily Stress Inventory (DSI) as defined by Brantley et al., which is a psychometrically sound self-report instrument for the daily assessment of the sources and the individualized effect of relatively minor stressful events. It was designed to evaluate causes of stress that are not typically evaluated by major life-event scales. We add 4 additional questions to the standard DSI survey:

1. If something that was unexpected or caused you a great deal of stress, please note the approximate time and duration of the event.

2. On a scale of 0=time seemed to take forever, to 4=time seemed to fly by, how would you say your day felt?

3. On a scale of 0=felt poorly prepared to 4=felt very prepared, how well prepared do you feel you were for today's events?

4. This question mostly applies to ‘Crit Day’ (Nov 14): On a scale of 0=unfair to 4=fair, in your opinion, how fair were the questions you were asked today?

The goal of question number 1 was to collect data that would be correlated to changes of the participants skin conductance. We hypothesized that unexpected events would cause a notable increase in skin conductance immediately following the event. We also collected similar data by performing exit interviews upon the completion of the study. However, since some time had passed from the occurrence of the events to the time of the interview, the results of these interviews may be biased.

We asked question 2 in order to find whether perceived stress had an impact on the perception of time duration. We hypothesized that the stringent deadline and sense of urgency would shorten the perceived duration of time, and that a day after the presentation, time would be perceived as having a longer duration.

Question 3 was asked in order to asses if the participants felt that they were well prepared. We hypothesized that well prepared participant would have a lower level of perceived stress and that this would also be evident in their skin conductance.

We asked question 4 to assess whether the participant felt in control during the presentation and whether they thought that they were asked fair questions. We hypothesized that during the question phase, participants who believed that they were asked “unfair” question would exhibit this physiologically as well.

3.2. Data Collection.

We collected the following data during the study:

1. Skin conductance, skin temperature, and 3-axis accelerometer – we used the Affectiva Q^TM wristband sensor. The data was recorded at a sampling rate of 8 Hz, and was recorded using dry Ag-AgCl electrodes. Each participant received a pair of sensors, one for each wrist, so we could collect bi-lateral data. Each sensor had a label that was used to note the participant's id and whether the sensor was to be placed on the left or right wrist. The participants were asked to wear the sensors for 72 hours starting on the morning of the day before their presentation.

2. Self-report measures – each participant was asked to fill in a PSS survey 2 days prior to the start of the study. During the study, participants were required to fill in a DSI survey at the end of each day.

3. In addition, we interviewed each participant a few days after the end of the measurements. During those interviews, we asked the participants to describe their experiences during the 72 hours. We asked them to note any unexpected events or events that caused them a great deal of anxiety or emotional strain.

4. Video – during the presentations we recorded a video of the participants. In some cases, we also recorded some of the participant's post-presentation question sessions.

3.2.2 IT Infrastructure and tools

For participant registration and conducting the surveys, we built custom Python-based web software that ran on a server in the lab. The participants could access the surveys by using a web browser from any location. The tool presented the surveys to the participants and collected their responses.

A MySQL server was used to store the survey responses. At the end of the study we exported all the results from the database into an Excel sheet.

The Q sensor data was downloaded from each unit and stored in a dedicated server, along with the survey responses and the video files.

For the signal processing we used MATLAB. We built a Python tool to convert all of the Q sensor data into MATLAB file format and filter out noise.

We also used a tool written in Processing by Miriam Zisook that enabled us to rapidly detect cases of asymmetry.

3.2.1 Privacy Measures

In order to protect the privacy of the participants, several measures were put in place. When the participants logged into the web system for the first time, they were randomly allocated a 5-digit PIN number. They were asked to use this number every time they logged into the system. In addition, they were asked to write this number on the labels of the sensors that they were given. This enabled us to correlate between the subjects sensor reading and their self-reports.

Results and Discussion

We have collected the following data set:

PSS	11 participants (all)
DSI	· 7 recipients filled in all surveys · 3 participants filled in the first 2 DSI surveys but not the 3^rd one · 1 participant did not fill in any of the DSI surveys
Video and audio during presentation	10 participants
Interviews	11 participants (all)

We calculated a PSS score per participant. The score is obtained by reversing the responses of the negatively stated items (e.g., 0 = 4, 1 = 3, 2 = 2, 3 = 1 & 4 = 0) to the four positively stated items (items 4, 5, 7, & 8) and then summing across all scale items. This score represented the perceived level of stress during the last month. The average PSS that was collected for a large population was 12.1 (SD=5.9) for males and 13.7 (SD=6.6) for females.

We calculated DSI scores for each participant. Three daily scores are derived for each individual: (1) the number of events that are endorsed as having occurred (Freq) (2) the sum of the total of the impact rating of these events (Sum), and (3) the average impact rating of the events (sum divided by the number of events) (Avg).

The table below shows the PSS and DSI scores for each participant.

		Day 1			Day 2			Day 3
PIN	PSS Sums	Freq	Sums	Avg	Freq	Sum	Avg	Freq	Sum	Avg
54604	25	16	50	3.125	21	45	2.143
75049	22	14	62	4.429	20	67	3.350	17	34	2.000
34600	21	25	80	3.200	29	122	4.207
05972	19	15	46	3.067	9	25	2.778	11	34	3.091
18372	17	13	31	2.385	17	55	3.235	15	50	3.333
57607	17	19	73	3.842	14	48	3.429
41965	10	31	51	1.645	23	47	2.043	21	63	3.000
36237	9	13	63	4.846	17	55	3.235	4	9	2.250
13113	9	16	28	1.750	16	29	1.813	16	23	1.438
20052	8	10	26	2.600	8	16	2.000	0	0	0.000

Table 1 PSS and DSI scores

The first result is that according to the DSI the mean perceived stress across all participants was highest on the night before ‘Crit Day’ (3.09) and lowest on the night of the day after ‘Crit Day’ (2.16). On the night of ‘Crit Day’ the perceived stress was in between the two (3.09). The average DSI measured for a large population in Cohen’s was 2.36 (SD=0.82) for males and 2.68 for females (SD=0.97)

We assumed that there should be a correlation between the PSS and the first DSI as they were submitted within the same time frame. Initially we found that the correlation between them was 0.3122, which was statistically not significant. After viewing the plot we noticed that a single subject was an outlier. Further investigation reveled that the participant had a argument with his partner just before filling the survey. We suspected this resulted in a biased score. After removing this subject we found that the correlation between the PSS and first DSI was 0.71, which was statistically significant (p<0.05).

The DSI also revealed that participants perceived time as passing quickly the day before ‘Crit Day’ (mean = 2.4, SD = 1.5), and passing slower during ‘Crit Day’ (mean = 2.33, mean = 1.21), and yet slower the day after (mean = 1.83, SD = 1.17).

We also found that most participants felt that the questions they were asked were fair (mean = 3.8, SD = 0.42).

Initially, after viewing the EDA signals we found that some of the samples were extremely noisy. We estimate that the reason for this is loose contact between the sensor and the skin. We applied an infinite impulse response low pass filter (exponential smoothing) with different coefficients for rising signals and falling signals (alpha rise = 0.5, alpha fall = 0.95). This filter is an attempt to reduce artifacts where the signal briefly but sharply drops due to poor contact with the skin, which appeared in several subjects' data. We implemented this filter in Python.

The first result that arose from analyzing the EDA signals was that there is a notable difference within each subject between the presentation and the questions phase. In the question phase, the EDA level was higher, and there were more peaks per second. This is true across all participants. We conclude that either the fear of being asked a difficult question and evaluated by others, or just being asked a question results in a heightened physiological response

Above can be seen the left and right EDA signals for two subjects. The EDA during the presentation is colored in blue, and the EDA during the question phase is colored in red.

We analyzed all of the EDA signals for asymmetry. 2 participants out of 11 exhibited temporary asymmetry in their EDA. There was no correlation between perceived stress ratings (on DSI and PSS) and asymmetry in EDA. The snapshot below shows asymmetry that begins just prior to the participant’s presentation and lasts throughout.

Figure 1 EDA Asymmetry (18372) during presentation

The second EDA signal also shows asymmetry, as the participant is about to start his presentation.

Figure 2 EDA Asymmetry (57607) during presentation

Finally, we tried to count peaks and correlate them with the percieved stress levels. However, we could not find any significant results. We suspect that our facility for counting peaks (the findpeaks function in MATLAB's signal processing toolkit) is confounded by noise in the data from poor contact with the skin, which varies greatly from participant to participant and, to a lesser extent, over the course of the day (perhaps due to participants re-adjusting the fit of the wristband.) A more careful definition of what constitutes a 'peak' feature in EDA data and is more robust to noise may yield better results.

PIN	13 Nov Left	13 Nov Right	14 Nov Left	14 Nov Right	15 Nov Left	15 Nov Right
5972	0.108931	0.042647	0.097395	0.091768	0.105573	0.091005
13113	0.006115	0.016222	0.115815	0.034953	0.112331	0.029933
18372	0.016222	0.00873	0.002861	0.031462	0.014959	0.055563
20052	0.09649	0.120339	0.013636	0.018172
34600	0.038732	0.245188	0.069173	0.106703	0.218192	0.086264
36237	0.036325	0.013221	0.015145	0.00938	0.033767	0.028134
41965	0.239856	0.206557	0.403685	0.236872	0.048534	0.032602
54604	2.868826	0.163433	0.010651	0.008393	0.562887	0.40959
57607	0.140879	0.238506	0.167272	0.438723	0.083897	0.124151
75049	0.119898	0.070873		0.269948	0.485653	0.129117
91405	0.113824	0.110514	0.125519	0.100129	0.148339	0.545966
Mean	0.344190727	0.112384545	0.1021152	0.122409364	0.1814132	0.1532325

Table 2 Number of Peaks per Minute using findpeaks and threshold of 0.001

Conclusion

We have collected a large dataset, consisting of bilateral EDA data, initial Perceived Stress Scale (PSS) responses, and daily Daily Stress Inventory (DSI) survey responses. Preliminary analysis shows some patterns that we expected, such as greater activation during Q&A after the presentation (which cannot be rehearsed) and less during the presentation (which participants likely rehearsed several times and felt prepared.) The dataset also contains several events where the participants' EDA is asymmetric, some of which we have associated with participants' descriptions of events at those times.

The preliminary analysis that we have done merely scratches the surface. We plan to continue investigating and to make the de-identified data available to other researchers for further study.

References

Boucsein, W.: Electrodermal Activity. Plenum Press, New York (1992)

Brantley P.J., Waggoner C.D., Jones G.N., Rappaport N.B.: A daily stress inventory: Development, reliability, and validity. J Behav Med;10:61–731, (1987)

Cacioppo, J.T., Tassinary, L.G., Berntson, G.G.: Handbook of Psychophysiology.

Cambridge University Press, Cambridge (2000)

Cohen, S; Kamarck T, Mermelstein .: "A global measure of perceived stress". Journal of Health and Social Behavior 24 (4): 385–396. PMID 6668417, R (December 1983)

Cohen, S. and Williamson, G.: Perceived Stress in a Probability Sample of the United States, The Social Psychology of Health, Newbury Park, CA: Sage (1988)

Dimberg, U., Fredrikson, M. and Lundquist, O.: Autonomic reactions to social and neutral stimuli in subjects high and low in public speaking fear, Journal Biological Psychology, Volume 23(3), p223 (1986).

Lykken, D.T., Venables, P.H.: Direct measurement of skin conductance: A proposal

for standarization. Psychophysiology 8(5), 656–672 (1971)

Mundy-Castle, A.C., McKiever, B.L.: The psychophysiological significance of the

galvanic skin response. Experimental Psychology 46(1), 15–24 (1953)

Poh, M., Swenson, N., Picard, R.: A wearable sensor for unobtrusive, long-term

assessment of electrodermal activity. IEEE Trans. Biomed. Eng. 57(5), 1243–1252 (2010)

Setz, C., Arnrich, B., Schumm, J., La Marca, R., Troster, G., Ehlert, U.: Discriminating stress from cognitive load using a wearable eda device. IEEE Transactions on Information Technology in Biomedicine 14(2), 410–417 (2010)