Optimizing the EEG setup for
neuromarketing research: an exploratory
study
Abstract
Neuromarketing is an evolving and promising research field. This type of marketing uses neuroimaging to create strategies and improve sales. Functional Magnetic Resonance Imaging (fMRI) is the ‘holy grail’ when it comes to neuromarketing studies. The technique is based on the amount of (de)oxygenated blood flow in certain regions of the brain and creates images to represent brain activity. The equipment however is very expensive, and it comes with a long list of exclusion criteria for participants. An alternative is electro-encephalography (EEG). A method that is easy to use, portable and less expensive. One of the limitations of this alternative is that EEG only records tissue that is located close to the skull. And the installation of a participant in a MRI scanner only takes a couple of minutes, whereas EEG installation can take up to an hour or more. This study looks into optimizing the EEG setup, in order to create high quality data that is comparable to MRI data. A previously written paradigm was used, and adjustments to the setup were made in-between the groups of participants. Some of these adjustments include cleaning the electrodes, implementing new computer programs and adjusting the protocol. No significant result was found, but this does not mean that the data quality did not improve at all. There are several factors that may have influenced this outcome, and these should be investigated in future research.
Key words: Neuromarketing, neuroimaging, fMRI, EEG setup, optimalisation.
Bachelor Thesis Psychobiology
University of Amsterdam
Author
Anne Ebbink 11661216 22-01-2021
Amsterdam, The Netherlands
Supervisor
A. van der Leij – Neurensics
Second corrector
Introduction
Background: why neuromarketing?
A research field that is currently evolving and provides new and promising techniques is the field of neuromarketing. Neuromarketing is a type of marketing that is based on neuroimaging. Marketing researchers however seem unaware of the huge potential and are not yet sold on the idea of basing marketing strategies on brain imaging. Lee et al. (2007) discuss the fact that most psychological sciences have been quick to apply some of the techniques developed by neuroscientists. However, most social sciences have not yet included these methods in their research procedures. Marketers for example still use survey-based strategies, even though these show severe limitations. Larson (2018) discusses one of these limitations. He mentions that a social desirability bias (SDB) can change the outcome of a study. SDB occurs when participants change their answers for impression management (‘to look better than others’) and will provide answers that differ from their true personality. A study that looked into the effect of SDB showed that a difference can be found between subgroups (e.g., (non-)religious, culture, old/young, etc.) of participants as well (Sjöström & Holst, 2002). This problem could be solved with individual surveys; however, this would be a very time-consuming process and could provide problems with the data analysis. Another bias can be that participants interpret the questions wrong. These studies are conducted with participants that represent the entire population (e.g., low and high educated groups), and if they do not understand the questions this could potentially influence the data. Previously mentioned biases of survey-based studies can influence data negatively. The following paragraphs will therefore discuss different advantages of neuromarketing and elucidate different methods to illustrate the process and possible applications in the future.
Advantages of neuromarketing
Morin (2011) states that neuromarketing is a promising, cutting edge method that directly probes minds, without requiring demanding cognitive or conscious participation. His paper suggests that neuromarketing research can potentially significantly improve the effectiveness or marketing methods, such as commercial usage and advertising messages. One of the effects of not needing conscious participation is that participants cannot consciously influence the data. Which is something that often happens in questionnaire and survey-based studies. Besides that, other research has shown that some consumers cannot properly communicate their preferences when they are asked to do so. And that buyers’ neuronal activity can contain hidden data about their true preferences (Telpaz et al., 2015). Ariely & Berns (2010) mention that neuroimaging can be very attractive for marketers because it can provide (hidden) information about a product, that otherwise would be unobtainable. Another advantage is that neuromarketing studies can help understand complex purchase behaviour (Fortunato et al., 2014). A level of complexity that cannot be reached by survey studies alone. Overall, information provided by this research field can be very useful for the development and support of basic theories used in marketing studies.
Methods used in neuromarketing studies
There are several techniques that can be used in neuromarketing research. One of which is functional Magnetic Resonance Imaging (fMRI). fMRI is currently the most commonly used brain imaging technique (Hüsing et al., 2006). It can be seen as the ‘holy grail’ of neuromarketing studies. A study done by Venkatraman et al. (2015) substantiates this. They compared six different neurophysiological methods and concluded that fMRI data shows the most variance in advertising elasticities. The technique is based on the amount of (de)oxygenated blood flow in certain regions of the brain and creates images to represent brain activity. The equipment used however is very expensive. A MRI scanner cannot be relocated and therefore researchers have to create an environment that is similar to realistic setups. Alternatives that are portable and can be used in realistic setups are functional near-infrared spectroscopy (fNIRS) and electroencephalography (EEG). fNIRS works with infrared light that is projected into the brain and absorbed by different brain areas. Optodes send light into the brain and detectors placed over the scalp receive information on how much light was absorbed by certain areas. In this way it can measure the changes of cortical (de-)oxygenated haemoglobin concentrations, which is used as an indication of brain activity in those areas (Ayaz et al., 2019). EEG is a method that uses electrodes closely placed to the scalp to record brain activation. The neural activity that is measured is a combination of excitatory and inhibitory postsynaptic potentials, produced by large groups of neurons firing simultaneously (Britton et al., 2016). There are several hardware configurations for conducting an EEG study. For example, Biosemi, this is a system where the signal amplifier is placed directly on the electrode, which leads to a better data quality. The equipment however can be very expensive, and the setup is very sensitive to gel bridges if researchers use too much gel during installation. Another system is Emotic Epoc, this is a portable and gel-free method. The setup uses dry electrodes, which makes it possible to install participants within a shorter timeframe. Low-cost EEG devices (e.g., Emotic Epoc) unfortunately come with some impairments, such as a lower signal quality and imprecise timing (Morán & Soriano, 2018). Even though there are several promising aspects (e.g., budget, portability and ease of handling), EEG data seems to be lacking when compared to MRI. A difference between EEG and MRI is the spatial resolution. MRI scanners can collect data from brain tissue that is located deep within the brain. Whereas EEG only records tissue that is located close to the skull. Another difference is that the installation of a participant in a MRI scanner only takes a couple of minutes. And EEG installation can
activity. To sum up, MRI provides a better quality of the data, but EEG is less expensive and can be used in realistic settings. See table 1 for a summary of these differences.
A multimethodological approach
A possibility that is currently looked into by different companies is combining some of these methods. In order to reproduce data with an MRI quality, but with lower costs and a portable setup. A study done by Ge et al. (2019) shows that fusing features from both EEG and fNIRS complex brain networks resulted in a classification of a 72.7% accuracy for different observations tasks, such as touching or grasping a cup. And Chiraelli et al. (2017) mention that the absence of electro-optical interference simplifies the integration of these two non-invasive recording procedures. They also state that EEG/fNIRS systems exploit the possibility of conducting experiments in a realistic setting. This is not possible for other neuroimaging techniques, such as MRI. There are researchers that have already used a combination of these different techniques in their studies. Balconi and Vanutelli (2017) use a multimethodological approach to study emotions in different interactions. And Holper et al (2014) used a combination of fNIRS and skin conductance response to investigate the relationship between sympathetic nervous activity and cerebral
hemodynamics and oxygenation. However, according to Ahn & Jun (2017) there are different limitations to the integration of EEG/fNIRS systems (e.g., the lack of computational methods to integrate the systems and an optimized sensor configuration). They also mention that the cost-effectiveness and portability are two major advantages, and more research has to be done.
Neurensics is currently working on a method that includes EEG and fNIRS. In order to create a low cost, high quality and fast multimethodological setup, that provides high quality data. However, there are certain limitations. The current issue with the method is that the data quality increases when time of preparation and the number of electrodes increase as well. This can be a problem within the field of neuromarketing, because time is money. The amount of preparation time has a significant impact on the business case of the setup, because it presses on the costs of carrying out an experiment (think of resources, people and capacity). This paper investigates the EEG aspect of the multimethodological approach, with the intention to provide a setup in which the quality-time-money triangle is optimized. This setup has to become the basis for a solid protocol that can easily be executed by (non-)experts within a given timeframe, to make it scalable. In order to do so, adjustments to the EEG setup will be made, and data quality will be compared in-between different groups of participants.
Table 1. Summary of (dis-)advantages of MRI, EEG & fNIRS.
MRI EEG FNIRS
COSTS €€€
- Machine is expensive and costs per participant are high. - Repair costs are high, machine
cannot be used after emergency (quench) procedures.
€
- Purchase of device is less expensive than fNIRS devices, electrodes however need to be replaced after a period of time.
€€
- Purchase of device is expensive, but it can be used for a long period of time.
USABILITY - Programmed settings.
- Participants have to stay in one position and hold a fixed posture.
- Participants are not able to touch the products during scanning. - A long list of exclusion criteria
for participants.
- Installation of participants takes a couple of minutes.
- Realistic settings.
- Participants are able to move around.
- Participants can ‘experience’ the products.
- A short list of exclusion criteria for participants.
- Installation can take up to an hour.
- Realistic settings.
- Participants are able to move around.
- Participants can ‘experience’ the products.
- A short list of exclusion criteria for participants.
- Interference of the signal depends on head size (the smaller the head, the more interference).
- Installation can take up to half an hour, and it is very hard to reach optimal signals.
ACCESS Different requirements to work as an
operator (e.g., exams and tutorials). No specific requirements to work as an operator and possible to master within a short time frame.
No specific requirements to work as an operator and possible to master within a short time frame.
LOCATION Located in a building where the rooms
enclose the magnetic field. The device cannot be relocated.
Devices stored in a suitcase. Relocation possible, no specific requirements for the room where the study is executed.
Devices stored in a suitcase. Relocation possible, no specific requirements for the room where the study is executed. CURRENT
Material and methodology
This study is based on a previously used paradigm (see ‘Overall design’) and participants’ brain activity is measured during their partaking. The study is divided into different blocks. Each part contains a group of participants (table 2) plus complementary data and is followed by data analyses. Adjustments were made after every analysis.
Overall design
The used paradigm was created by Neurensics and was previously used by other interns. The first paragraph of the adjustments section discusses their setup. The other three were performed by Saad Ashraf and myself. Between the different groups of participants the setup changed. The experiment itself did not change, the programs used and procedure prior and after the experiment however vary. The following paragraphs discuss the overall design, and the changes will be highlighted afterwards.
Participants
Group Number of participants M/F Age
1 21 8:13 19-28
2 16 7:10 18-24
3 12 10:2 21-26
4 10 2:8 19-28
Table 2. Overview of the number of participants per group. Paradigm
Prior to the experiment participants were screened, to ensure that they were consumers of chocolate, beers, toothpaste and shower gel. Every participant was asked to sign an informed consent before starting the experiment. They were told that they could stop at any time and received monetary compensation for their participation. Before starting the experiment participants had to determine the category of the neuropricing trials by pseudo-randomly picking a card. They got the opportunity to ‘experience’ these products by touching them as if they were in a store. The experiment itself lasted for a total of one hour and two minutes. During this period participants were shown four different blocks. The first and third contained different logos of brands/companies, e.g., insurance companies and food manufacturers. Alongside these trials came different combinations of words that could evoke feelings in the participants. Some of these words were ‘nietszeggend, sympathiek & vriendelijk’, and participants were instructed to choose the word that they found most compatible with the logo that was shown (fig. 1). During the second block, participants were asked to look at different television commercials (TVCs). A selection of 110 television commercials was made by Neurensics. These TVCs were picked from the Neurensics database. And the selection consisted TVCs that were released in the period between 1996 and 2019. A sample of these TVCs had additional requirements, such as winning a Dutch award (‘Gouden Loekie’, ‘EFFIE’ or ‘’Loden Leeuw’). A bigger subset was named ‘middle of the road’ commercials, these were used as ‘neutral stimuli’. They did not generate much notable emotional responses. The last number of TVCs did not have additional information on winning awards or not (table 3).
Figure 1. Overview of block 1 & 3. Between trials there was a 2 second period in which participants had to decide
which word was most compatible.
Subcategory Number of TVCs
- Gouden Loekie 15
- EFFIE 20
- Loden Leeuw 11
- ‘Middle of the road’ commercials 43
- Other 21
The final part of the experiment contained the neuropricing trials. Participants were given a budget with which they had to buy either beer, chocolate, toothpaste or shower gel. Both the toothpaste and shower gel conditions came with a €19,99 budget, chocolate with a €6,49 budget and beer with a budget of €8,79. Every category contained four different products, and every trial one of these products was shown with a different price tag (fig. 2). The prices of these products differed from €0,20 to the maximum of the budget that came with that specific category. Participants were asked to decide on whether they would buy them for these prices or not. Afterwards one random trial was selected. This trial showed the participants purchase behaviour and if the participant had bought the product in this trial, they would receive the product. If they did not buy it, they received their total budget plus the fixed monetary compensation.
Figure 2. Overview of a single trial from block 4. Participants were presented a product and had to decide whether
they would buy it for that price or not.
Data acquisition
Artinis (Artinis; NIRx Medical Technologies, Elst, The Netherlands) provided several fNIRS caps, sizes ranging from small to large. It covered the entire scalp, and the material (neoprene) has the same thickness throughout (3 mm). The electrodes used for this experiment are the Ag/AgCl sintered ring electrodes, manufactured by EASYCAP GmbH. See figure 3 for the electrode configuration. The program used to acquire EEG data is Eventide (Okazolab Ltd. Delft, The Netherlands), with a sampling rate of 125 Hz.
Figure 3. EEG configuration. A total of 14 electrodes (13 on the scalp + 1 electrode on the ear as a ground
electrode). Pzref and POz electrodes were swapped on the cap, because of a dent on the back of the scull. (retrieved from Sky Coyote, 2001-2002)
Adjustments
Group 1
The first set of data collection was done by the previous group of interns. Their group of participants consisted of 8 males and 13 females, with ages between 19 and 28. The experiment itself did not change and can be found in the paragraph above. The following computer programs were used during this first collection; Open BCI Gui (OpenBCI GUI, v5.0.2), Oxysoft (Artinis; NIRx Medical Technologies, Elst, The Netherlands) and Eventide. The EEG setup, signal and noise levels were verified in Open BCI, an open-source brain-computer interface. Oxysoft was used to connect the fNIRS device and study the connection with certain brain areas, and Eventide displayed the experiment itself.
Group 2
The second period was executed by Saad and myself. Our first group of participants included 7 males and 10 females, with the ages 18 to 24. The experiment itself stayed the same. Before scanning the participants, the electrodes were thoroughly cleansed in an ultrasonic bath, because the previous interns did not wash them properly. The programs used in this second set of data collection changed. Open BCI Gui – and Oxysoft signals were
implemented in Eventide and were used as a verification of the brain signals. The program was now able to show the impedence levels of the EEG electrodes and the optical density of the fNIRS optodes. This change reduced the workload on the computer since less programs had to be turned on during scanning. During this period the application of the Signa gel slightly changed, which improved the impedence scores. Hair was removed by using the tip of the syringe and the gel was directly placed on the skin of the participant. The movement of the application was ‘J-shaped’. Underneath the electrode between the cap and the skin the syringe was slightly tilted and later it was held in a vertical position. If the impedence levels did not go down, the tip of the syringe was used to stir the gel and to increase contact with the skin.
Group 3
The screened group consisted of 2 males and 10 females, with ages 21 to 26. The experiment itself did not change, the TVCs however were updated (improved quality and the volume was normalized). Some changes were made in the instruction section of the experiment. The data from group 2 showed that participants did not buy low priced products, and that they only focussed on one specific product with the lowest possible price. This was not the behaviour we tried to provoke. Participants should try to buy a different number of products for the lowest possible price and not just focus on one specific product. The new set of instructions contained five products (toilet paper, detergent, gum, coke and chips), that were priced very low. Participants were asked if they would buy these products or not and explain their behaviour. Participants who did not buy these products received an additional explanation and were told that these prices were not trick questions. After finishing the entire experiment participants were asked some additional questions on their purchase behaviour and if they used a certain strategy, to see whether their behaviour matched the data. Besides that, the gel application improved. The timeframe that was used for installing participants went down from an hour or more to 45 minutes, and impedence levels decreased to 10-15Hz. The application method itself did not change; our skills however did improve after trying on several participants.
Group 4
During the final period there were 2 male and 8 female participants, with ages from 19 to 28. The overall setup of the experiment stayed the same. The EEG electrodes were cleaned in an ultrasonic batch, because the impedence levels increased during this period.
Data preparation and statistical analysis
In-between the different groups a Rstudio (Rstudio team, 2020) script was used to study improvements and to locate flaws. This script was written by Neurensics. Changes made to the protocol were based on this data and can be found in the paragraphs above.
The data was automatically acquired by Eventide and transferred to Excel. Data processing was done with R (R 4.0.3 GUI 1.73 Catalina build (7892)) in Rstudio (version 1.3.1093). The first step was checking the quality and usability of the data, using a script made by Neurensics. This was followed by implementing a double notch filter (30-32Hz and 49-51 Hz). To filter the noise produced by the impedence checker (31.25Hz) and the noise produced by the electrical outlet (50Hz). In order to compare the different data sets a small part of the data was extracted. Radüntz (2018) stated that a significant decrease in the parietal alpha power band should be obtained when more and less demanding cognitive tasks are compared. Based on this study we selected the data from the second and last block, watching TVCs and neuropricing trials. Watching TVCs was used as our less demanding task and the neuropricing trials were labelled more demanding. This extraction of data was necessary for the comparison of the different groups of participants. A Fast Fourier transform was executed to extract the different brain waves. With this data a file was created that contained the frequency differences between power bands. These differences were calculated with a subtraction of the first two minutes of each block of both categories. Only data with frequencies ranging from 8 to 12 Hz (Alpha waves) was used for further analysis. The distribution of these differences between the two conditions within the alpha wave group can be found in figure 4. For the statistical significance the alpha level was set on 0.05. The data was transformed with a Z-transformation to limit differences between participants. A Shapiro-Wilk test was performed to test the assumption of normality. The assumption of equal variances was tested with the Bartlett test. And in the end a Kruskal-Wallis test was executed to determine the p-values.
Figure 4. The y-axis shows the distribution of frequency differences between the two cognitive tasks (watching TVCs
and neuropricing trials). And the x-axis shows the different groups of participants. Group 1 (M = 0.0839651, SD = 0.393), Group 2 (M = 0.1700919, SD =0.3487212), Group 3 (M = 0.1226603, SD = 0.2269477) and Group 4 (M = -0.1172139, SD = 0.2283409).
Exclusion of datasets
Not all data was applicable after performing the Fast Fourier transform. For the analysis a total of 45 datasets was used. Group 1 contained 19, the second group contained 11, group 3 contained 8 and the last group contained 7 datasets.
Results
A total of 45 datasets was used for the statistical analysis. Differences between the frequencies of both conditions were determined, with group 1 (0.0839651 ± 0.393, n=19), group 2 (0.1700919 ± 0.3487212, n=11), group 3 (0.1226603 ± 0.2269477, n=8) and the last group (-0.1172139 ± 0.2283409, n=7). To investigate whether the data was normally distributed a Shapiro-Wilk test was executed. This showed a normal distribution of the data in group 2 (W = 0.88357, p-value = 0.1154) and group 3 (W = 0.93753, p-value = 0.587). Group 1 (W = 0.82309, p-value = 0.002524) and group 4 (W = 0.76935, p-value = 0.0306) however did not show a normal distribution of the data. Therefore, a Kruskal-Wallis test was conducted. To test the assumption of equal variances a Bartlett test was
performed, and the data failed to meet the assumption (Bartlett's K-squared = 4.2134, df = 3, p-value = 0.2393). The Kruskal-Wallis test showed that the datasets were not significant (Kruskal-Wallis chi-squared = 6.2036, df = 3, p-value = 0.1021). Figure 5 shows a distribution of the data.
figure 5. The distribution of frequency differences in alpha waves, when two conditions (watching TVCs and
neuropricing) are subtracted, are shown on the y-axis. The x-axis shows the different groups, and between these groups some changes were made (as mentioned in the methodology section). The Kruskal-Wallis test showed a p>0.05, which means that the mean ranks of the groups are not the same.
Conclusion and discussion
The purpose of this study was to determine whether the EEG setup could be improved, in order to use it on a larger scale and write a solid protocol for (non-)experts. However, the data analysis did not show a significant result. This means that the EEG setup did not improve significantly. Nevertheless, there is a possibility that the setup did improve, but that these changes could not be detected by this type of data analysis. The following paragraphs highlight several explanations for this outcome and suggest some ideas and adjustments for future research.
Data analysis
To analyse whether there was a growth in data quality, differences between the second and the final block (watching TVCs and neuropricing) were studied. It is possible that this growth was present, but that it was not the right method to use to compare the data from the different groups. The study done by Radüntz (2018), did show a significant effect, but no other studies confirm nor deny this. For future research it is important to extract a certain signal from the EEG data, that is ‘steady’ and can be used to compare data quality and analyse improvement.
EEG
The electrodes had already been used before by the previous interns and have a different age, because some were replaced during the study. The lifespan of these electrodes is ±200-250 participants. This number was reached during the measurement of the final group of participants. The electrodes were treated in an ultrasonic bath, before
measuring group 2 and during the measurement in group 4, to cleanse them thoroughly. This did make a difference for group 2 measurements. During the measurements of group 4 however the electrodes seemed to keep producing instable signals with a lot of noise. This problem can easily be solved by replacing the electrodes in an earlier stadium in future research. Another problem with this setup was the combined EEG/fNIRS cap. Both fNIRS and EEG come with their own setup and use different caps. In this study however only the fNIRS cap was used and holes for EEG electrodes were implemented in this cap. fNIRS caps are made from a different material and are thicker than original EEG caps. This means that the distance between the skull and the EEG electrodes increased, which led to a noisy signal and a lot of Signa gel had to be used to create a signal with low impedence levels. A recommendation for future research is trying to implement the fNIRS optodes in an EEG cap to see whether the signal improves or not. In order to monitor the EEG signals prior to the experiment an impedence checker was implemented. This gadget however, caused a significant amount of noise in the data and showed a peak around 31.25 Hz. A 30-32Hz notch filter was implemented to filter the data, but this led to a decrease in quality. For future studies it is important to turn off the impedence checker before starting the experiment. Furthermore, the used setup still included several loose wires and was very fragile. Neurensics is currently working with an engineer from the University of Amsterdam that is building a more robust and compact amplifier that includes ports for EEG electrodes and GSR- & HR equipment. When this product is finished it will be possible to carry it around in a small bag, which will make the setup portable. The current placement of the participant is behind a screen with curtains to close off the area. These curtains however do not block sound. It is possible that participants got distracted during the experiment, which may have caused noise in the data. Some of the participants mentioned that because of these curtains they got a little sleepy during the experiment, which led to decreased attention.
Implications due to COVID-19 measures
This study was executed in a period running from October until December. During this period a different number of measures was implemented. One of which was, that if possible, people had to work from home. This limited our time in the lab and caused a significant decrease in the number of available participants. The (low) number of participants may have influenced the statistical analysis, because most statistical techniques are sensitive to sample size and it reduces the chance of detecting a significant effect (Siddiqui (2013); Button et al., 2013). It is possible that when this study is conducted on a bigger group of participants in the future a significant result will be found.
Future research
Several adjustments have already been mentioned in the previous paragraphs. There are however some other adjustments that can be made in the near future to optimize the setup. Mathewson et al. (2017) for example
mentioned that dry electrodes are a good alternative when it comes to moving around in ‘real world research’, due to the setup being wireless. And because dry electrodes do not require skin abrasion or gel application, they might be useful for EEG recordings that are not executed by professionals that lack these skills. Besides that, they will be able to execute studies within a shorter amount of time. This flexibility however does come with an increment in noise as well, but they concluded that this level of noise does not affect the ability to measure classic EEG and ERP signals. It can be concluded that the different changes made to the EEG setup did not show significant improvement. However, making these changes did help us understand the method. And the gained knowledge can be used when instructing new researchers and it provides a good basis for writing a solid protocol. As figure 4 shows, there is a
EEG setup did cause an improvement of the data quality, but that during measurement of group 4 the quality decreased due to the ‘age-limitations’ of the electrodes. It would be interesting to see if the current setup shows improvement when the electrodes are exchanged for a set of new electrodes. Based on all the previously mentioned information it can be concluded that further research has to be done to optimize the EEG setup for neuromarketing studies.
References
Ahn, S., & Jun, S. C. (2017). Multi-Modal Integration of EEG-fNIRS for Brain-Computer Interfaces – Current Limitations and Future Directions. Frontiers in Human Neuroscience, 11. https://doi.org/10.3389/fnhum.2017.00503 Ariely, D., & Berns, G. S. (2010). Neuromarketing: the hope and hype of neuroimaging in business. Nature Reviews
Neuroscience, 11(4), 284–292. https://doi.org/10.1038/nrn2795
Ayaz, H., & Dehais, F. (2018). Neuroergonomics: The Brain at Work and in Everyday Life (1ste editie). Academic Press. Balconi, M., & Vanutelli, E. (2017). Empathy in Negative and Positive Interpersonal Interactions. What is the Relationship Between Central (EEG, fNIRS) and Peripheral (Autonomic) Neurophysiological Responses? Advances in
Cognitive Psychology, 13(1), 105–120. https://doi.org/10.5709/acp-0211-0
Brenner, P. S. (2011). Identity Importance and the Overreporting of Religious Service Attendance: Multiple Imputation of Religious Attendance Using the American Time Use Study and the General Social Survey. Journal for
the Scientific Study of Religion, 50(1), 103–115. https://doi.org/10.1111/j.1468-5906.2010.01554.x
Britton JW, Frey LC, Hopp JLet al., authors; St. Louis EK, Frey LC, editors. (2016) Electroencephalography (EEG): An
Introductory Text and Atlas of Normal and Abnormal Findings in Adults, Children, and Infants [Internet]. Chicago: American
Epilepsy Society. Available from: https://www.ncbi.nlm.nih.gov/books/NBK390346/
Button, K. S., Ioannidis, J. P. A., Mokrysz, C., Nosek, B. A., Flint, J., Robinson, E. S. J., & Munafò, M. R. (2013). Power failure: why small sample size undermines the reliability of neuroscience. Nature Reviews Neuroscience, 14(5), 365–376. https://doi.org/10.1038/nrn3475
Chiarelli, A. M., Zappasodi, F., Di Pompeo, F., & Merla, A. (2017). Simultaneous functional near-infrared spectroscopy and electroencephalography for monitoring of human brain activity and oxygenation: a review. Neurophotonics, 4(04), 1. https://doi.org/10.1117/1.nph.4.4.041411
Eser, Z., Isin, F. B., & Tolon, M. (2011). Perceptions of marketing academics, neurologists, and marketing professionals about neuromarketing. Journal of Marketing Management, 27(7–8), 854–868.
https://doi.org/10.1080/02672571003719070
Fortunato, V. C. R., Giraldi, J. D. M. E., & De Oliveira, J. H. C. (2014). A Review of Studies on Neuromarketing: Practical Results, Techniques, Contributions and Limitations. Journal of Management Research, 6(2), 201.
https://doi.org/10.5296/jmr.v6i2.5446
Ge, S., Wang, P., Liu, H., Lin, P., Gao, J., Wang, R., Iramina, K., Zhang, Q., & Zheng, W. (2019). Neural Activity and Decoding of Action Observation Using Combined EEG and fNIRS Measurement. Frontiers in Human
Neuroscience, 13. https://doi.org/10.3389/fnhum.2019.00357
Hüsing, B., Jäncke, L., & Tag, B. (2006). Impact Assessment of Neuroimaging. Amsterdam University Press. Avaliable from
https://books.google.nl/books?hl=nl&lr=&id=SZjsrqkMiIkC&oi=fnd&pg=PP4&dq=H%C3%BCsing,+B.,+J%C3 %A4ncke,+L.,+%26+Tag,+B.+(2006).+Impact+Assessment+of+Neuroimaging.+Amsterdam+University+Press. &ots=VU30K1BRfL&sig=b2OUAN5V6djEwnnmijM_X_O9r3w#v=onepage&q&f=false
Kim, K., Hayes, J., Avant, J., & Reid, L. (2014). Trends in Advertising Research: A Longitudinal Analysis of Leading Advertising, Marketing, and Communication Journals, 1980 to 2010. Journal of Advertising, 43, 296–316.
https://doi.org/10.1080/00913367.2013.857620
Larson, R. B. (2018). Controlling social desirability bias. International Journal of Market Research, 61(5), 534–547. https://doi.org/10.1177/1470785318805305
Lazar, J., Feng, J. H., & Hochheiser, H. (2017). Research Methods in Human Computer Interaction. Elsevier Gezondheidszorg.
Lee, N., Broderick, A. J., & Chamberlain, L. (2007). What is ‘neuromarketing’? A discussion and agenda for future research. International Journal of Psychophysiology, 63(2), 199–204. https://doi.org/10.1016/j.ijpsycho.2006.03.007 MacKenzie, S. B., & Podsakoff, P. M. (2012). Common Method Bias in Marketing: Causes, Mechanisms, and Procedural Remedies. Journal of Retailing, 88(4), 542–555. https://doi.org/10.1016/j.jretai.2012.08.001
Mathewson, K. E., Harrison, T. J. L., & Kizuk, S. A. D. (2016). High and dry? Comparing active dry EEG electrodes to active and passive wet electrodes. Psychophysiology, 54(1), 74–82. https://doi.org/10.1111/psyp.12536
Morán, A., & Soriano, M. C. (2018). Improving the quality of a collective signal in a consumer EEG headset. PLOS
ONE, 13(5), e0197597. https://doi.org/10.1371/journal.pone.0197597
Morin, C. (2011). Neuromarketing: The New Science of Consumer Behavior. Society, 48(2), 131–135. https://doi.org/10.1007/s12115-010-9408-1
Radüntz, T. (2018). Signal Quality Evaluation of Emerging EEG Devices. Frontiers in Physiology, 9. https://doi.org/10.3389/fphys.2018.00098
Siddiqui, K. (2013). Heuristics for Sample Size Determination in Multivariate Statistical Techniques. World Applied
Sciences Journal, 27. 285-287. Doi: 10.5829/idosi.wasj.2013.27.02.889.
Sjöström, O., & Holst, D. (2002). Validity of a questionnaire survey: response patterns in different subgroups and the effect of social desirability. Acta Odontologica Scandinavica, 60(3), 136–140.
https://doi.org/10.1080/000163502753740133
Sky Coyote (2001-2002). Acquiring a simple EEG signal [Photo]. http://www.skycoyote.com/eeg/may01/eeg.html Telpaz, A., Webb, R., & Levy, D. J. (2015). Using EEG to Predict Consumers’ Future Choices. Journal of Marketing
Research, 52(4), 511–529. https://doi.org/10.1509/jmr.13.0564
Vecchiato, G., Astolfi, L., De Vico Fallani, F., Cincotti, F., Mattia, D., Salinari, S., Soranzo, R., & Babiloni, F. (2009). Changes in Brain Activity During the Observation of TV Commercials by Using EEG, GSR and HR
Measurements. Brain Topography, 23(2), 165–179. https://doi.org/10.1007/s10548-009-0127-0
Venkatraman, V., Dimoka, A., Pavlou, P. A., Vo, K., Hampton, W., Bollinger, B., Hershfield, H. E., Ishihara, M., & Winer, R. S. (2015). Predicting Advertising success beyond Traditional Measures: New Insights from
Neurophysiological Methods and Market Response Modeling. Journal of Marketing Research, 52(4), 436–452. https://doi.org/10.1509/jmr.13.0593
