Approach and avoidance tendencies: Using smartphones to study behavioral responses and food desires
Cristiana Cring, s1899589
Leiden University
Master of Science in Psychology Track: Economic and Consumer Psychology
Supervisor: Hilmar Zech
Secondary supervisors: Dr. Lotte van Dillen; Prof. Dr. Wilco van Dijk In collaboration with: Elsa van der Meer
Abstract
In the present experiment the approach and avoidance tendencies towards food stimuli were tested using the recently developed mobile version of the Approach-Avoidance Task (AAT). We hypothesized that people would show an approach bias towards food rather than objects, towards desirable food rather than undesirable, and that the measure of the approach bias would be predicted by the score on the Power of Food Scale, the healthy eating habits of
participants, their Body Mass Index, and their hunger level at the moment of testing. Forty eight participants completed demographic questions, the mobile AAT, the Power of Food Scale, a picture rating task, and a series of questions regarding eating habits. In the mobile AAT, images of food and objects were shown and participants had to either push or pull them. Our analysis revealed a significant approach bias towards food stimuli when compared to objects. However, we did not find the same when comparing the approach bias towards desirable and undesirable food. Moreover, neither one of the individual characteristics was a predictor of the approach bias. Even though our hypotheses were not confirmed in this study, we believe that the mobile AAT is a relevant method for analyzing approach and avoidance tendencies. Implications of the results and future research ideas are also discussed in this paper.
Introduction
Approaching or avoiding an object is an adaptive behavior, very prevalent in our
everyday lives. We often find ourselves in situations where such an action is needed and it has to be a fast and proper one (Phaf, Mohr, Rotteveel, & Wicherts, 2014). Making a decision to leave from an unsafe place or to get close to a food source is essential for our survival. However, there are many situations when these behaviors become maladaptive. For example, substance abusers have an automatic approach reaction towards drug-related stimuli which toughens their recovery process (Field, Caren, Fernie, & Houwer, 2011). Similarly, Hofmann, Friese and Wiers (2008) suggest that the mere perception of an attractive food triggers cognitive and motor functions directed towards obtaining that food which makes it difficult for people to abstain from over-eating. Such approach tendencies can have life-threatening consequences which is why it is important for psychologists to have a tool they can use to test them.
For the present study, the mobile version of the Approach-Avoidance task was used as a tool to test approach and avoidance tendencies towards food stimuli.
The Approach-Avoidance Task (AAT) was developed by Solarz in 1960 with the purpose of measuring approach and avoidance tendencies towards visual stimuli. Since then, the AAT has been used in many studies in different areas of psychology. Some examples of its use
include: in the clinical field it was used for studying phobias such as the fear of spiders (Rinck & Becker, 2007) as well as in training alcoholic people in resisting their addiction (Eberl et al., 2013); in the cognitive field it was used to test and understand cognitive functions (Lavender & Hommel, 2007); and in the field of social psychology it was used to understand the differences in reactions towards social crowds between socially anxious people and non-anxious people
(Lange, Keijsers, Becker, & Rinck, 2008).
In the original experiment, Solarz presented participants with cards with pleasant or unpleasant words and instructed them to either push or pull the cards. The participants had to complete two conditions: a congruent one in which they had to pull pleasant cards and push unpleasant ones and an incongruent condition where they had to push pleasant cards and pull the unpleasant cards. The dependent variable of this task was the reaction time measured in
milliseconds. Solarz found that people reacted faster when pulling positive stimuli (pleasant words) than when pulling negative ones (unpleasant words). Conversely, the reaction times were smaller when pushing images of unpleasant words than when pushing pleasant ones. This
suggests that positive stimuli trigger approach reactions while the negative stimuli trigger avoidance movements.
The task has been updated throughout time first by creating a computerized version using a lever for the reaction movements (Chen & Bargh, 1999) and later by replacing the lever with a joystick (Rinck & Becker, 2007). However, all these versions had to be used in a laboratory. In order to make it easier for researchers to use the task in the natural environment of participants, in 2015, Zech adapted the task for smartphones.
The mobile AAT conceptually follows the setup of the original experiment by Solarz but makes it possible to use the task outside the laboratory. Similarly to Solarz’s experiment, the task presents participants with a series of stimuli and requires them to make push or pull movements during which participants physically move the stimulus closer or further away from themselves. However, the mobile AAT differs from Solarz’s AAT by presenting stimuli on the phone instead of cards and by the way reaction times are measured. In the mobile AAT, reaction times are measured by a sensor in the phone, which records the acceleration and the direction of the movement. These two differences eliminate the need for an experimenter directly conducting the study which facilitates the use of the application in the field.
The mobile AAT has a series of advantages that can be considered improvements over previous versions of the AAT. First of all, using the mobile version of the task could increase the number of participants in studies. Sending an email with an invitation to download the
application can have a positive impact on potential participants such as young people or students. Moreover, the participants have the possibility to complete the task at any moment and in any place which reduces both the time and the effort they normally have to put into participating in a laboratory experiment. Secondly, the mobile AAT solves some of the issues reported for the previous versions. In the joystick version of the AAT, people have to push on the joystick in order to avoid and pull it in order to approach the stimulus on the computer (Rinck & Becker, 2007). However, these instructions are ambiguous and can confuse participants who could interpret the pull response as withdrawing their hand from the computer and the push response as moving their hand towards the computer (Seibt, Neumann, Nussinson, & Strack, 2008;
Krieglmeyer & Deutsch, 2010). The mobile AAT avoids this confusion as people physically act on the stimulus. Thus, with the approach movement the distance between the participant and the
stimulus is reduced, while the avoidance movement increases the distance (Dantzig, Pecher, & Zwaan, 2008).
The mobile AAT can be used for numerous purposes such as: measuring the reaction time to various stimuli, comparing reaction times to different types of stimuli or measuring approach and avoidance biases. An approach bias can be defined as a systematic tendency to approach a certain type of stimulus in favour of another. The approach bias has been previously studied mainly in relation to substance abuse. Substance users have been shown to have an approach bias for images of cannabis (Cousijn, Goudriaan, & Wiers, 2011), for alcohol (Peeters et al., 2012; Field, Caren, Fernie, & Houwer, 2011) and for smoking cues (Wiers, et al., 2013). However, few studies have examined the approach bias in relation to food which is disconcerting considering how prevalent eating disorders are. For this reason, the present research focuses on studying the approach biases towards food-related stimuli using the mobile AAT. The mobile AAT is especially suited for studies of food-related biases because people generally react
towards food with approach or avoidance movements. Before consuming food, approaching it is a necessary step which makes this a natural response to it.
There are numerous mediators in the relation between food images and an actual
behavioural reaction, one of the most important being desire. According to Merriam Webster, a desire is “a strong wish for something”. Apart from feeling actual hunger, desiring a certain type of food is what drives most people to engage in eating behaviours. What is more, it is known that the subjective desire to eat a type of food can appear just by seeing it for a moment (Marcelino, Adam, Couronne, Koster, & Sieffermann, 2001; Wadhera & Capaldi-Phillips, 2014). Some of the factors that increase the desire for food are: the sensory qualities of food (e.g. good smell, attractive presentation) the emotional state of the person, or the state of hunger that person has (Hill, Weaver, & Blundell, 1991).
As far as the relation between desire and the reactions towards food cues (i.e. approach or avoidance) is concerned, previous research exists but is not comprehensive. Evolutionarily speaking, humans developed a preference for foods with a high level of fat, sugar, or salt as these have increased energetic values (Pinel, Assanand, & Lehman, 2000). Anticipating these tastes makes the food more attractive which leads to desire despite the fact that most of the time these items are unhealthy (Papies, 2013). Studies suggest that the mere perception of attractive food triggers cognitive and motor functions directed towards obtaining that food (Hofmann, Friese, &
Wiers, 2008). Furthermore, the level of desire for a certain type of food is a predictor for the approach bias, the latter increasing with the level of desire (Kemps, Tiggemann, Martin, & Elliott, 2013). Thus, we expect the mobile AAT to reveal the existence of an approach bias for desirable food.
Another variable that can influence the approach bias towards food is the level of
importance food has in an individuals’ life. This can be measured using the Power of Food Scale, a test that assesses peoples’ feelings of being controlled by food (Cappelleri, et al., 2009). It measures the appetite for tasty food at three levels of proximity: food available but not present, food present but not tasted and food tasted but not eaten (Lowe, et al., 2009). A high PFS score has been correlated to more daily snacking and it is a moderator between the internal or external food cues and snacking on a daily basis (Schüz, Schüz, & Ferguson, 2015). Thus, people with a high PFS score that go through an emotional moment or just have snacks available are more likely to eat more between meals than people with lower scores. The PFS score moderates the potentially unhealthy eating behaviour of people and we expect it to predict an approach bias towards food.
Healthy food habits are related to having a regular eating schedule and getting all the needed nutrients while limiting the consumption of highly processed foods. A study conducted by Yegiyan and Bailey (2016) shows that people with healthy eating habits feel more positive about images of healthy food than images of unhealthy food. Healthy food habits can also protect people from caving in to temptation (Lin, Wood, & Monterosso, 2016). People that were trained to choose a healthy food, when faced with a food choice, chose the healthy option, disregarding the unhealthy one. We therefore believe that the general level of health in food habits will predict the approach bias towards unhealthy food.
Another variable that is frequently correlated to the behaviour towards food is the body mass index (BMI). According to the Medical Dictionary, the BMI is a value calculated as weight in kilograms divided by height in meters squared (Medical Dictionary Online, 2017). Research shows that people with a higher BMI have less inhibited responses when exposed to food so they have an increased level of impulsivity toward food (Houben, Nederkoorn, & Jansen, 2014). When people with high BMI who are on a diet are exposed to attractive food, their wish to consume high-caloric food increases and they tend to forget about their goals (Ouwehand & Papies, 2010). Furthermore, people with high BMI values are more likely to associate positive
words to unhealthy food. People with high BMI had a shorter reaction time when pairing pictures of M&M candy with positive words than people with low BMI (Hofmann, Friese, & Roefs, 2009). Also, people with high BMI scores have shown reduced activation of inhibitory regions in the brain when confronted with food (Batterink, Yokum, & Stice, 2010). Consequently, the BMI should be a predictor of the approach bias for unhealthy and desirable food.
Lastly, as the physiological needs of humans have to be taken into consideration
whenever studying eating behaviour, the level of hunger at the moment of completion of the task needs to be analysed. Hunger is the normal state leading to food intake for people. Studies show that when hungry people are exposed to food-related stimuli, their response inhibition is lowered which leads to an approach bias towards food (Loeber, Grosshans, Herpertz, Kiefer, & Herpertz, 2013). When comparing hungry and satiated females, researchers found that only the hungry participants exhibited this type of automatic orientation to food, while satiated participants did not have the same reaction (Nijs, Muris, Euser, & Franken, 2010). From this information, we expect that when people are hungry, they will be faster in approaching and slower in avoiding all images of food compared to when they are satiated.
In accordance with the previously mentioned variables and expectations, the following hypotheses were formulated:
H1: The approach bias is bigger when reacting towards images of food than to images of objects.
H2: The approach bias is bigger when reacting towards images of desirable food than towards images of undesirable food.
H3: The approach bias towards images of food increases with the score on the Power of Food Scale.
H4: The approach bias towards images of unhealthy food decreases when people have healthy eating habits.
H5: The approach bias towards images of unhealthy desirable food stimuli increases with the BMI.
H6: The approach bias towards images of desirable food stimuli increases when the hunger level is high.
Method
Research design
The design of the study was a 2x2 within-subjects design. All participants were required to complete the same tasks: answer demographic questions, complete four blocks of the
Approach-Avoidance Task, rate the stimuli used in the AAT, complete the Power of Food Scale and answer questions regarding eating habits. For the mobile AAT, two counterbalancing conditions were used, changing the order of the blocks. In the first counterbalancing condition, the blocks from the congruent condition were first, while in the second counterbalancing condition, the incongruent instructions appeared first followed by the congruent ones.
Variables. The independent variables were the types of stimuli presented (food/objects, desirable/undesirable images of food, healthy/unhealthy food), the score on the Power of Food Scale, the level of health in eating habits, BMI and the level of hunger. The dependent variable was the approach bias. The approach bias was defined as the difference between the reaction time when pushing and the reaction time when pulling a certain category of stimuli (e.g.
positive/negative words, images of desirable/undesirable food). The mathematical expression for this is:
𝐴𝐵 = 𝑅𝑇𝑝𝑢𝑠ℎ− 𝑅𝑇𝑝𝑢𝑙𝑙
Participants
Participants were recruited from Leiden University through online application.
Throughout two weeks, 58 students participated in the laboratory experiment. Their participation was rewarded with 1 course credit or €3.50. Due to problems with the mobile application that led to loss of data, eight participants were eliminated from the final sample. Two more participants were eliminated because of an error rate higher than 20% in their results (See Appendix A for a detailed explanation). The final analysis was realized on a sample of 48 students (81.25% women, 18.75% men) with ages between 18 and 30 years (m = 22.47, SD = 2.78). Out of these 48 students, 27 (56.3%) completed the tasks in Dutch and 21 (43.8%) completed them in English meaning that they were from other countries than the Netherlands.
Instruments
The application was adapted so that the entire test can be done on the phone. The
was already in Dutch, the test would be in Dutch as well and if the phone was on any other language, the test would be in English.
Stimulus set for AAT. The stimulus set for the mobile AAT was formed using pictures from the food.pics database (Blechert, Meule, Busch and Ohla, 2014). The extended database contains a set of 896 pictures with various objects and types of food. When the set of pictures was realized by Blechert et al., it was tested on American and German people and a set of normative data was obtained. This set of normative data included attributes of the pictures such as brightness, contrast, complexity, familiarity, recognizability, craving, etc. The pictures included in the used stimulus set were chosen based on two main attributes: how easily recognizable the picture is and the level of craving, defined as the desire to eat the food presented. As far as the craving variable is concerned (M = 33.24, SD = 9.53) we considered images with one standard deviation over the mean to be desirable and images with one standard deviation below the mean to be undesirable. The level of recognizability was set over 85 in order to keep recognizability consistent while still having enough pictures for the stimuli sets. Based on these criteria a set of 135 pictures was obtained, that included 56 images with undesirable food and 75 images of desirable food. These images were then divided into healthy and unhealthy based on the way the food presented was cooked and general consensus on healthy alimentation.
Figure 1. Examples of images from the stimuli set: Image 1 – healthy undesirable; Image 2 –
healthy desirable; Image 3 – unhealthy desirable; Image 4 – unhealthy undesirable; Image 5 – object
Five categories, each with 20 pictures were created: healthy and desirable, healthy and undesirable, unhealthy and desirable, unhealthy and undesirable and objects (for examples, see Figure 1). A total of 120 pictures were used for the test. Furthermore, 30 additional pictures, different from the ones in the test, were chosen randomly from the database and used in the practice blocks of the application.
The mobile AAT. The mobile AAT had four blocks and before each block, instructions were shown. The four blocks were equally divided into two conditions: the congruent one where participants had to pull food and push objects and the incongruent one where they had to push food and pull objects. The participants were instructed to sit down, hold the phone with both hands at a 90 degree angle and make brisk pull or push movements.
Before each condition, a trial task including 15 pictures of food and objects was included in order to ensure the good understanding of instructions and the required movements. After each practice trial the participants got response feedback in the form of a green screen for correct responses and a red screen for incorrect ones. Each image was shown on the screen for 2 seconds and participants were asked to react to it as fast as possible. If participants gave no response after 2 seconds, am image of an alarm clock appeared on the screen. After each image, a white screen appeared which allowed participants to get their hands back to the neutral position.
The mobile AAT measures the reaction times in milliseconds for each stimulus. Furthermore, it measures the acceleration of the phone for each movement which shows the direction of each movement. Starting from the acceleration, the force can also be calculated.
The Power of Food Scale (PFS). The PFS was used to test the level of susceptibility of an individual to appetizing food (Lowe et al., 2009). The scale contains 21 statements regarding eating behavior and the reactions towards food. On a 5-point scale (1 - I don’t agree at all, 2 – I agree a little, 3 – I agree somewhat, 4 – I agree, 5 – I strongly agree), the participants were asked to indicate their level of agreement with each item. Some examples of items from the scale are: “If I see or smell a food I like, I get a powerful urge to have some”, “I love the taste of certain food so much that I can’t avoid eating them even if they’re bad for me”, “I feel like food controls me rather than the other way around”.
Procedure
Participants applied for this study on the SONA platform or by sending an email to the researchers. Before starting the study, the participants had to sign an informed consent form. For
the test, they had the possibility to use their own phones, if they had an Android operating
system, or to use a phone provided by the experimenters. Participants were alternatively assigned to one of the counterbalancing condition which only influenced the presentation of the AAT.
The task started with the demographic questions regarding age, gender and studies. The participants then called one of the experimenters in the room to explain the right way of doing the movements for the AAT. The experimenter remained in the room for the first trial run in order to offer further support to the participants and to make sure the instructions were well understood. After the trial block was completed, the experimenter left the room and did not interact with the participants any more until the end of the study. After completing the mobile AAT, participants had to rate the pictures, complete the PFS and answer a series of additional questions about themselves. In the picture rating section, each picture from the AAT was shown with the question “How attractive do you find this picture?” and a five-point Likert scale from which to choose the answer. The final questions regarding the participant included height, weight, hunger, being on a diet and the importance of healthy eating (e.g. “How hungry are you right now?”, “How long ago was your last meal?”, “Are you a vegetarian?”, “Are you on a diet?”, “How healthy do you usually eat?”).
The test did not take more than half an hour per participant. After finishing the
experiment, participants were debriefed with regards to the purpose of the study, gave feedback on the application and received their payment.
Results
For the data analysis the IBM SPSS Statistics 24 program was used. In order to test the hypotheses we needed to aggregate the reaction times so as to calculate the central tendency per participant. This aggregation was done based on the median. As reaction time data generally contains many outliers and has a distribution with a heavy positive skewness (Whelan, 2008), the median aggregation was preferred over the one based on the mean. See Appendix A for a more detailed presentation of the data before aggregation.
The practice trials, all cases with errors in response and cases with reaction time smaller than 268 ms (see Appendix A for explanation) were filtered out.
The alpha level was set at .05 for all the statistical tests. Confirmatory results
Hypothesis 1. The first hypothesis of the experiment was that people have a bigger approach bias when reacting towards images of food than to images of objects. Reaction times from the mobile AAT were aggregated based on participant and groups of responses and types of pictures (push-object, push-food, pull-object, pull-food). A two by two repeated measures
ANOVA with type of picture (food and object) and response (push and pull) as factors was conducted. There was a significant main effect of type of picture, F(1, 47) = 74.08, p < .001, 𝜂𝑝𝑎𝑟𝑡𝑖𝑎𝑙2 = 0.612, with longer reaction times for pictures of objects (M = 485.13, SE = 8.26) than
for pictures of food (M = 454.55, SE = 8.48). There was also a main effect of response, F(1, 47) = 6.55, p = .014, 𝜂𝑝𝑎𝑟𝑡𝑖𝑎𝑙2 = 0.122 with slightly faster reaction times for the pull movement (M = 465.58, SE = 8.55) than for push movements (M = 474.10, SE = 8.14). A statistically significant interaction effect between the two factors can also be seen, F(1, 47) = 15.33, p < .001, 𝜂𝑝𝑎𝑟𝑡𝑖𝑎𝑙2 =
0.246. The table in Figure 2, shows the results of the post-hoc paired samples t-test. People pulled objects (M = 492.10, SD = 65.98) significantly slower than they pulled food (M = 439.06,
SD = 61.73) with t(47) = 7.69, p < .001, while the difference between the push responses was not
significant. Furthermore, people reacted faster towards food (M = 454.55) than objects (M = 485.13), regardless of the movement. Finally, in order to test whether there is a difference in approach bias, we computed it for food and objects stimuli and a paired samples t-test was used to compare the means. The test showed a significant difference between the mean approach biases for food (M = 30.97, SD = 43.71) and the one for objects (M = -13.73, SD = 48.06), t(47)
= 3.91, p < .001. Therefore, the first hypothesis is confirmed as the approach bias for food is significantly higher than the one for objects.
Figure 2. Reaction times for food and objects in the mobile AAT
Hypothesis 2. In order to test the second hypothesis reaction times from the mobile AAT only for food images were aggregated based on participant, response and the previously set qualification of the food as desirable or undesirable (i.e. push-desirable, push-undesirable, pull-desirable, pull-undesirable). A two by two repeated measures ANOVA with type of food
(desirable and undesirable) and response (push and pull) as factors was conducted and the results are presented in Table 1. Similarly to the first hypothesis, there was a main effect for response,
F(1, 47) = 14.52, p < .001, 𝜂𝑝𝑎𝑟𝑡𝑖𝑎𝑙2 = 0.236, with smaller reaction times for the pull response (M = 443.48, SE = 9.83) than for push (M = 470.03, SE = 8.99). Furthermore there was a significant main effect for temptation, F(1, 47) = 34.60, p < .001, 𝜂𝑝𝑎𝑟𝑡𝑖𝑎𝑙2 = 0.424 with slightly faster reactions to undesirable images (M = 445.05, SE = 9.00) than desirable ones (M = 468.47, SE = 8.96). The interaction effect was not statistically significant, F(1, 47) = 0.00, p = .987, suggesting that there is not a significant difference in reaction times in the four groups of interaction. For a follow-up analysis, the approach bias was computed for desirable and undesirable stimuli and a paired samples t-test was used to compare the means. The test showed no significant differences between the mean approach bias for desirable stimuli (M = 26.61, SD = 54.07) and the one for undesirable stimuli (M = 26.48, SD = 59.03), t(47) = -0.02, p = .987. We conclude that the
0 100 200 300 400 500 600 Push Pull Object Food
hypothesis is rejected and there is no difference in approach bias between desirable and undesirable stimuli.
Table 1
Results of repeated measures ANOVA for desirability and response
F Sig Mean Std. Error
Response Push 14.524 .000* 470.031 8.995
Pull 443.184 9.835
Desirability Desirable 34.599 .000* 468.469 8.965
Undesirable 445.047 8.996
Response*Desirability 0.000 .987
*Significant with p < .001; df1 = 1; dferror = 47
Hypothesis 3. To test the hypothesis according to which the score on the Power of Food scale predicted the approach bias, a simple linear regression was used. Only the results from images of food in the mobile AAT were aggregated based on participant and the response (push and pull) and the approach bias was computed. The results of the regression were
non-significant, R2 = .02, F(1, 46) = 0.87, p = .354 (see Figure 3). The score on the Power of Food Scale does not predict the approach bias towards images of food (β = -9.92, t(47) = -.94, p = .354). The hypothesis was rejected as the Power of Food score does not influence the approach bias towards food stimuli.
Figure 3. Scatterplot with the scores on the Power of Food Scale and the approach bias to images
Hypothesis 4. The fourth hypothesis was also tested with simple linear regression with the approach bias as dependent variable and the score on the healthy scale as predictor. How healthy people usually eat was measured on a 5-points Likert scale and the answers varied from 2 to 4 (60.4% - 4, 33.3% - 3, 6.3% - 2). Only the cases when the stimuli were images of
unhealthy food were selected from the database and aggregated based on participant and
response. The results of the regression were non-significant with R2 = .078, F(1, 46) = 3.92, p = .054. The approach bias towards unhealthy food images is not predicted by how healthy people usually eat (β = -23.85, t(47) = -1.98, p = .054). Therefore, the assumption is rejected and it cannot be supported that the approach bias towards unhealthy food decreases when people have healthy eating habits.
Hypothesis 5. For the fifth hypothesis, BMI was computed based on the weight and height of participants and the values ranged from 16.90 to 27.18. The cases were selected so that the stimuli would be only images of unhealthy and desirable food and aggregated based on participant and response. The results of the regression were non-significant, the predictor explaining 0.1% of the variance, R2 = .001, F(1, 46) = 0.06, p = .806. The relation between the two variables can be seen in Figure 4. The approach bias towards unhealthy and desirable food images is not predicted by the BMI (β = -0.88, t(47) = -0.25, p = .806).
Hypothesis 6. The last hypothesis was also tested using linear regression with the approach bias for food stimuli as independent variable and the level of hunger as predictor. The level of hunger was self-reported on a scale from 1 to 5 and the answers varied from 1 to 4. The results were not significant statistically R2 = .005, F(1, 46) = .24, p = .626. Since the approach bias towards food cannot be predicted by the level of hunger (β = 2.97, t(47) = .49, p = .627), this hypothesis is rejected.
Exploratory results
In addition to our hypotheses, we performed some exploratory analyses that gave us a better understanding of the data.
The stimulus set used was tested on German and American people (Blechert, Meule, Busch & Ohla, 2014), thus the normative data was not based on our main group which is Dutch people. In order to check if this fact influenced the found results in the second hypothesis, an additional analysis was performed on the approach bias towards desirable food but based on the reported level of temptation from the participants. A two by two repeated measures ANOVA with type of food (desirable and undesirable) and response (push and pull) as factors was conducted. The images of food rated with 4 and 5 on the 5 points scale, were considered desirable and the ones rated 3 or lower than 3 were considered undesirable. There was a main effect for response, F(1, 47) = 16.93, p < .001, 𝜂𝑝𝑎𝑟𝑡𝑖𝑎𝑙2 = 0.265, with smaller reaction times for the pull response (M = 442.60, SE = 9.54) than for push (M = 470.42, SE = 9.32). Furthermore there was a significant main effect for desirability, F(1, 47) = 9.84, p = .003, 𝜂𝑝𝑎𝑟𝑡𝑖𝑎𝑙2 = 0.173 with faster reactions to desirable food images (M = 450.60, SE = 9.11) than undesirable ones (M = 462.43, SE = 8.90). The interaction effect was also statistically significant, F(1, 47) = 6.40, p = .015, 𝜂𝑝𝑎𝑟𝑡𝑖𝑎𝑙2 = 0.119. Participants pulled images of desirable food (M = 430.71, SE = 9.34)
faster than undesirable ones (M = 454.50, SE = 10.45) while the results showed no such effect for the push responses (Mdesirable = 470.49, SEdesirable = 10.50 vs. Mundesirable = 470.35, SEundesirable = 9.25).
Figure 5. Boxplot with the approach bias for desirable and undesirable food
Similarly to the procedure followed in Hypothesis 2, the approach bias was computed for desirable and undesirable stimuli and a paired samples t-test was used to compare the means. A boxplot of the two variables can be seen in Figure 5. The test revealed a statistically significant difference with t(47) = 2.53, p = .015, between the mean approach bias for desirable stimuli (M = 39.78, SD = 55.03) and the one for undesirable stimuli (M = 15.85, SD = 59.20).
Additional analyses were conducted based on the PFS scores and the reaction time. No type of approach bias can be predicted by the PFS (i.e. not towards desirable food or healthy food). However, when another simple linear regression was used to check how the scores predicted the median reaction time per participant towards food images, regardless of the
response, the results were highly significant. The predictor explained 19.5% of the variance, R2 = .19, F(1, 46) = 11.13, p = .002. It was found that the score on the Power of Food Scale
significantly predicted the reaction time (β = -41.52, t(47) = -3.34, p = .002). As it can be seen in Figure 6, when the PFS scores are high, the reaction times towards food are smaller.
Figure 6. left - Scatterplot with the aggregated median reaction time for food and PFS scores; right – Scatterplot with the approach bias towards food stimuli and BMI.
Several analyses of the BMI and the approach bias towards food images were realized. There were no statistically significant results when analyzing the approach bias towards healthy/unhealthy food stimuli or towards desirable/undesirable food. The results were only significant when the analysis used the approach bias calculated regardless of the type of food shown in the image. The results were significant with the predictor explaining 13.5% of the variance, R2 = .13, F(1, 46) = 7.16, p = .010. Therefore, the BMI value significantly predicted the approach bias towards images of food (β = -5.98, t(47) = 2.68, p = .010). As Figure 6 shows, when the BMI scores are high, the approach bias towards food slightly increases.
Discussion
The aim of this study was to test the approach bias towards food stimuli, using the mobile AAT. We found that people had an approach bias towards images of food when compared to objects. Contrary to what we expected participants did not show an approach bias for more desirable food when the rating of it was based on normative data. These results might have been influenced by the way the instructions were formulated. Participants were instructed to react (push or pull) to pictures of food and objects so the separation between desirable and undesirable food was not specified. Thus when looking at the picture, the participants only focused on
whether it was food or object and might not have actually processed what type of food it was. However, in our exploratory analysis we found a significant approach bias to images of desirable food when the rating was based on the participants’ opinions. This might suggest that the ratings based on normative data were not suitable for our sample which can explain the results for the first hypothesis.
Also contrary to what we expected, no individual characteristics of the participants predicted an approach bias. The score on the Power of Food Scale, the healthy eating habits of participants or their hunger levels showed no statistically significant relation with any type of approach bias. The initial assumption that the BMI would predict an approach bias to unhealthy desirable food was rejected. However, exploratory analysis revealed that the BMI was a
predictor of an approach bias towards food in general. Limitations
Due to the fact that the study of food desires using a mobile application is relatively new and not much previous research has been done on this subject, some design related decisions have been made which might have affected the results. Thus it should be kept in mind that the present study has a series of limitations which should be addressed in future research.
One possible limitation of this research is the lack of generalizability. First of all, the results cannot be generalized to populations outside of the laboratory. As the mobile application is still new, we chose to test it in an experimental environment where potential problems with the application could be found and solved. However, this impedes us from indicating that the
findings of the study would be the same outside the laboratory or in another type of environment. Secondly, all the participants in the study were students from the Faculty of Social Sciences in the University of Leiden. Although they were enrolled in various faculties and fields of study,
they shared some similarities in personality and the results may have been different for students in other fields or for non-students. Finally, the sample used was a relatively small one. All in all, this research is not enough to determine whether different populations have an approach bias towards food stimuli.
It is also appropriate to note some potential shortcomings of the stimulus set used. The database with the pictures was tested on German and North American subjects (Blechert, Meule, Busch, & Ohla, 2015). Although some of the participants in our research were from these
countries, the majority of them were from The Netherlands and other European countries. Despite the fact that food related cultural differences were identified in previous research
(Jaeger, Andani, Wakeling, & MacFie, 1998; Brunsø, Grunert, & Bredahl, 1996), it is difficult to generalize individual food preferences to an entire culture. However, both cultural and individual differences might have influenced the results.
Another possible limitation of this study is the way the approach bias was defined. The method to calculate the approach bias followed the example of numerous other studies (e.g. Schumacher, Kemps, & Tiggemann, 2016; Kemps & Tiggemann, 2015; Wiers, et al., 2013). However, another possible method would be to use standardized scores which reduce the possibility of biases appearing because of differences in average response time (Sriram,
Greenwald, & Nosek, 2010). This method was created initially for the Implicit Association Test (Greenwald, Nosek, & Banaji, 2003) and it “standardizes the difference in response latencies by dividing an individual’s difference in response times by a personalized standard deviation of these response latencies” (Wiers, Eberl, Rinck, Becker, & Lindenmeyer, 2011). Furthermore, the method in this research, does not correct for differences using the control variable (i.e. the pull and push movements towards objects). In other studies, the approach bias was measured by combining the responses to the main variable with the ones to the control variable (Cousijn, Goudriaan, Ridderinkhof, Brink, Veltman, & Weirs, 2012). Based on the results of the previously mentioned studies, we can consider the fact that the our results might have been influenced by the way we defined the approach bias.
Future research
Additional replications and extensions of this study are encouraged in order to increase the confidence in these results and to further develop the technique.
In order to increase the generalizability of the results, the mobile AAT should be used to test a larger population, within a greater age range and from different studies, cultural
backgrounds as well as people outside the university. Furthermore, when testing food desires, the stimulus set should be pre-tested on a sample of the relevant population or the analysis should be based on rating from the participants. This would help avoid cultural differences in food
preferences and would increase the validity of the results. In addition to this, we believe that including the results for the control variable in the formula for the approach bias would give a clearer image of the analyzed phenomenon.
This research aimed at testing the mobile application and solving its potential issues. Since most of these have been solved, future research should continue the development of this form of testing. First of all, the application should be developed for any type of operating system and smartphone. Otherwise, by selecting for the sample only people with Android phones a sampling bias could appear as they share some personality traits (Shaw, David, Kendrick, Ziegler, & Wiseman, 2016). Secondly, the results of this study should be replicated outside of the laboratory. The participants could be invited to participate via email, download the
application on their personal smartphone and complete the tasks in their own environment. The results should then be compared with the ones in the laboratory and checked for similarities and differences.
Finally, future research might also directly test the mechanism underlying the approach-avoidance effect in the mobile AAT. Numerous studies show that approach and approach-avoidance tendencies are driven by emotions and affective evaluations (Roelofs, Elzinga, & Rotteveel, 2005). However the mechanisms and motives for the movements have not been clearly identified yet. The theory of event coding supports the idea that actions are represented by feature codes and both actions and stimuli are formed in the same representational domain (Hommel,
Musseler, Aschersleben, & Prinz, 2001). Consequently, responses are faster when the stimulus and the representation of the action have more common features, which would explain the reaction time differences in congruent and incongruent situations of the AAT. Following this reason, a subsequent theory was developed that suggested that people intentionally label responses with positive valence for approach and negative for avoidance. When the valence of the stimulus overlaps with the one from the movement, the latter one will be faster, thus explaining the approach-avoidance effect (Lavender & Hommel, 2007). Motivational theories
suggest that stimuli with a certain affective valence trigger physical responses in order to change the distance from one to the stimulus (Krieglmeyer, Deutsch, Houwer, & Raedt, 2010). Future research could study exactly what mechanisms create the approach-avoidance effects in the mobile AAT and confirm the previous theories or develop new ones.
Implications of the results
The results of our study support the functionality of the mobile AAT and its ability to measure approach and avoidance tendencies or biases. These findings have many implications in psychological testing, clinical psychology and treatment as well as in market research.
The use of the mobile AAT could change the way the approach-avoidance tendencies are measured. Since it can be downloaded from the app stores or from an email, the application could reach thousands of people from various cultures, countries and continents. During our debriefing talks with the participants of the study, we learned that they found it much more “exciting” and “interesting” to be part of an unconventional study. They also thought it is a great idea to develop more mobile applications for psychological studies as it will increase the
participants’ enthusiasm for the research. If presented in the correct way, the mobile AAT could attract many participants. Additionally, in 2017, 2.32 billion people worldwide with ages
between 16 and 50 years old will have a smartphone and the number of users is constantly increasing (Statista, 2016). A big percentage of these people could be reached easily for research purposes, which would increase both the generalizability of results and the statistical power of the studies. Furthermore, the mobile AAT could be used to test approach and avoidance tendencies in longitudinal studies or when repeated measures are needed. For example in our subject, desires for food, the application could be tested to see if approach biases differ in the morning compared to evening, before and after eating, in stressful situations or in calm ones, at home or outside the house, when the weather is good or bad, etc. It can also more easily test people for a longer period of time, for example to study how approach-avoidance tendencies develop with age or throughout a year. Once installed, the application could send notifications whenever the task needs to be completed thus reducing the possibility for participants to forget. All of these developments ideas would improve the psychological testing and would increase the amount of information on the topic of approach and avoidance tendencies.
Many previous studies show that the AAT can be used in trainings in order to modify substance or food consumption (Kakoschke, Kemps, & Tiggemann, 2017; Schumacher, Kemps,
& Tiggemann, 2016; Cousijn, Goudriaan, & Wiers, 2011; Becker, Wiers, & Holland, 2015). If such trainings would be developed for the mobile application they would be much more useful to participants and might have results on a longer term. By developing mobile trainings, the
participants could use the task in their everyday life where they meet temptations. Thus the trainings would help them resist the desire to consume those substances or foods whenever they encounter them which could increase the success rate of the training programs.
Finally, as far as the market research is concerned, in the future, the mobile AAT should be used in order to reach consumers and test them. The application could test their reaction times and approach and avoidance tendencies towards products, packaging, posters, colors or fonts. This would make it easier for businesses to gain a real image of the market and their consumers’ reactions to various objects. For example, if when asked to approach a product, participants have longer reaction times, it can be inferred that they had to solve an internal conflict and they actually do not like that product.
In conclusion, although the majority of our assumptions were rejected, this study showed the level of usefulness of the mobile AAT. After the limitations of this study will be corrected, the mobile AAT will be an easy method of testing approach and avoidance tendencies. It can be used in many fields and applied on bigger populations which would be a great improvement for psychological testing. With the constant development of technology and its prevalence among people worldwide, the use of smartphones should be the future of psychological testing and interventions.
References
Batterink, L., Yokum, S., & Stice, E. (2010). Body mass correlates inversely with inhibitory control in response to food among adolescent girls: an fMRI study. Neuroimage, 52(4), 1696-1703.
Becker, N. B., Wiers, R. W., & Holland, R. W. (2015). Approach avoidance training in the eating domain: Testing the effectiveness across three single session studies. Appetite, 85, 58-65.
Blechert, J., Meule, A., Busch, N. A., & Ohla, K. (2015). Food-pics: an image database for experimental research on eating and appetite. Food cravings, 26, 40.
Body Mass Index (BMI) [Def.1]. (n.d.). In Medical Dictionary Online, Retrieved March 17, 2017 from http://medical-dictionary.thefreedictionary.com/BMI.
Brebner, J. T. (1980). Reaction time in personality theory. Reaction times, 309-320.
Brunsø, K., Grunert, K. G., & Bredahl, L. (1996). An analysis of national and cross-national
consumer segments using the food-related lifestyle instrument in Denmark, France, Germany and Great Britain. Retrieved May 28, 2017, from Danish National Research
Database: http://pure.au.dk//portal/files/32299546/wp35.pdf
Cappelleri, J., Bushmakin, A., Gerber, R., Leidy, N., Sexton, C., Karlsson, J., et al. (2009). Evaluating the Power of Food Scale in obese subjects and a general sample of
individuals: development and measurement properties. International Journal of Obesity,
33, 913-922.
Chen, M., & Bargh, J. (1999). Consequences of automatic evaluation: Immediate behavioral predispositions to approach or avoid the stimulus. Personality and social psychology
bulletin, 25(2), 215-224.
Cousijn, J., Goudriaan, A. E., & Wiers, R. W. (2011). Reaching out towards cannabis: approach-bias in heavy cannabis users predicts changes in cannabis use. Addiction, 106(9), 1667-1674.
Cousijn, J., Goudriaan, A. E., Ridderinkhof, K. R., Brink, W. v., Veltman, D. J., & Weirs, R. W. (2012). Approach-Bias Predicts Development of Cannabis Problem Severity in Heavy Cannabis Users: Results from a Prospective FMRI Study. PloS one, 7(9), e42394.
Dantzig, S. v., Pecher, D., & Zwaan, R. A. (2008). Approach and avoidance as action effects.
The Quarterly Journal of Experimental Psychology, 61(9), 1298-1306.
Desire [Def. 1]. (n.d.). In Merriam Webster Online, Retrieved May 16, 2017, from https://www.merriam-webster.com/thesaurus/desire.
Eberl, C., Wiers, R., Pawelczack, S., Rinck, M., Becker, E., & Lindenmeyer, J. (2013). Approach bias modification in alcohol dependence: do clinical effects replicate and for whom does it work best? Developmental cognitive neuroscience, 4, 38-51.
Eckner, J. T., Kutcher, J. S., & Richardson, J. K. (2010). Pilot evaluation of a novel clinical test of reaction time in National Collegiate Athletic Association Division I football players.
Journal of athletic training, 45(4), 327-332.
Field, M., Caren, R., Fernie, G., & Houwer, J. D. (2011). Alcohol approach tendencies in heavy drinkers: comparison of effects in a Relevant Stimulus-Response Compatibility task and an approach/avoidance Simon task. Psychology of addictive behaviors, 25(4), 697. Galton, F. (1899). On instruments for testing perception of differences of tint and for
determining reaction time. Journal of the Anthropological Institute, 27-29.
Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (2003). Understanding and using the Implicit Association Test: I. An improved scoring algorithm. Journal of Personality and Social
Psychology, 85, 197-216.
Hill, A. J., Weaver, C. F., & Blundell, J. E. (1991). Food craving, dietary restraint and mood.
Appetite, 17(3), 187-197.
Hofmann, W., Friese, M., & Roefs, A. (2009). Three Ways to Resist Temptation: The Independent Contributions of Executive Attention, Inhibitory Control, and Affect Regulation to the Impulse Control of Eating Behavior. Journal of Experimental Social
Psychology, 45(2), 431-435.
Hofmann, W., Friese, M., & Wiers, R. W. (2008). Impulsive versus reflective influences on health behavior: A theoretical framework and empirical review. Health Psychology
Review, 2(2), 111-137.
Hommel, B., Musseler, J., Aschersleben, G., & Prinz, W. (2001). The Theory of Event Coding (TEC): A framework for perception and action planning. Behavioral and brain sciences,
Houben, K., Nederkoorn, C., & Jansen, A. (2014). Eating on impulse: The relation between overweight and food-specific inhibitory control. Obesity, 22(5), E6-E8.
Jaeger, S. R., Andani, Z., Wakeling, I. N., & MacFie, H. J. (1998). Consumer preferences for fresh and aged apples: a cross-cultural comparison. Food quality and preference, 9(5), 355-366.
Kakoschke, N., Kemps, E., & Tiggemann, M. (2017). Approach bias modification training and consumption: A review of the literature. Addictive Behaviors, 64, 21-28.
Kemps, E., & Tiggemann, M. (2015). Approach bias for food cues in obese individuals.
Psychology & Health, 30(3), 370-380.
Kemps, E., Tiggemann, M., Martin, R., & Elliott, M. (2013). Implicit approach-avoidance associations for craved food cues. Journal of Experimental Psychology: Applied, 19(1), 30.
Kosinski, R. J. (2010, September). A literature review on reaction time. Retrieved 04 22, 2017, from archive.is: http://archive.is/jw9W#selection-1237.0-1241.36
Krieglmeyer, R., Deutsch, R., Houwer, J. D., & Raedt, R. D. (2010). Being moved: Valence activates avoidance behavior independent of evaluation and approach-avoidance intentions. Psychological Science, 21(4), 607-613.
Lange, W. G., Keijsers, G., Becker, E., & Rinck, M. (2008). Social anxiety and evaluation of social crowds: Explicit and implicit measures. Behaviour Research and Therapy, 46(8), 932-943.
Lavender, T., & Hommel, B. (2007). Affect and action: Towards an event-coding account.
Cognition and Emotion, 6, 1270-1296.
Lin, P.-Y., Wood, W., & Monterosso, J. (2016). Healthy eating habits protect against temptations. Appetite, 103(1), 432-440.
Loeber, S., Grosshans, M., Herpertz, S., Kiefer, F., & Herpertz, S. C. (2013). Hunger modulates behavioral disinhibition and attention allocation to food-associated cues in normal-weight controls. Appetite, 71, 32-39.
Lowe, M. R., Butryn, M. L., Didie, E. R., Annunziato, R. A., Thomas, J. G., Crerand, C. E., et al. (2009). The Power of Food Scale. A new measure of the psychological influence of the food environment. Appetite, 53(1), 114-118.
Marcelino, A. S., Adam, A. S., Couronne, T., Koster, E. P., & Sieffermann, J. M. (2001). Internal and external determinants of eating initiation in humans. Appetite, 36, 9-14. Nijs, I., Muris, P., Euser, A., & Franken, I. (2010). Differences in attention to food and food
intake between overweight/obese and normal-weight females under conditions of hunger and satiety. Appetite, 54(2), 243-254.
Ouwehand, C., & Papies, E. k. (2010). Eat it or beat it. The differential effects of food temptations on overweight and normal-weight restrained eaters. Appetite, 55, 56-60. Papies, E. K. (2013). Tempting food words activate eating simulations. Frontiers in Psychology,
4, 838.
Peeters, M., Wiers, R. W., Monshouwer, K., Schoot, R., Janssen, T., & Vollebergh, W. A. (2012). Automatic processes in at-risk adolescents: the role of alcohol-approach
tendencies and response inhibition in drinking behavior. Addiction, 107(11), 1939-1946. Phaf, R. H., Mohr, S. E., Rotteveel, M., & Wicherts, J. M. (2014). Approach, avoidance, and
affect: a meta-analysis of approach-avoidance tendencies in manual reaction time tasks.
Frontiers in psychology, 5, 378.
Pinel, J., Assanand, S., & Lehman, D. (2000). Hunger, eating, and ill health. American
Psychologist, 55(10), 1105.
Rinck, M., & Becker, E. (2007). Approach and avoidance in fear of spiders. Journal of behavior
therapy and experimental psychiatry, 38(2), 105-120.
Roelofs, K., Elzinga, B. M., & Rotteveel, M. (2005). The effects of stress-induced cortisol responses on approach-avoidance behavior. Psychoneuroendocrinology, 30, 665-677. Schumacher, S. E., Kemps, E., & Tiggemann, M. (2016). Bias modification training can alter
approach bias and chocolate consumption. Appetite, 96, 219-224.
Schüz, B., Schüz, N., & Ferguson, S. G. (2015). It's the power of food: individual differences in food cue responsiveness and snacking in everyday life. International Journal of
Behavioral Nutrition and Physical Activity, 12(1), 1.
Seibt, B., Neumann, R., Nussinson, R., & Strack, F. (2008). Movement direction or change in distance? Self-and object-related approach-avoidance motions. Journal of Experimental
Shaw, H., David, E. A., Kendrick, L. R., Ziegler, F., & Wiseman, R. (2016). Predicting smartphone operating system from personality and individual differences.
Cyberpsychology, Behavior, and Social Networking, 19(12), 727-732.
Solarz, A. K. (1960). Latency of instrumental responses as a function of compatibility with the meaning of eliciting verbal signs. Journal of Experimental Psychology, 59(4), 239-245. Sriram, N., Greenwald, A. G., & Nosek, B. A. (2010). Correlational biases in mean response
latency differences. Statistical Methodology, 7, 277-291.
Statista. (2016). Number of smartphone users worldwide from 2014 to 2020. Retrieved May 30, 2017, from Statista. The statistics Portal:
https://www.statista.com/statistics/330695/number-of-smartphone-users-worldwide/. Wadhera, D., & Capaldi-Phillips, E. (2014). A review of visual cues associated with food on
food acceptance and consumption. Eating Behaviors, 15(1), 132-143.
Whelan, R. (2008). Effective analysis of reaction time data. The Psychological Record, 58(3), 475.
Wiers, C. E., Kuhn, S., Javadi, A. H., Korucuoglu, O., Wiers, R. W., Walter, H., et al. (2013). Automatic approach bias towards smoking cues is present in smokers but not in ex-smokers. Psychopharmacology, 229(1), 187-197.
Wiers, R. W., Eberl, C., Rinck, M., Becker, E. S., & Lindenmeyer, J. (2011). Retraining Automatic Action Tendencies Changes Alcoholic Patients’ Approach Bias for Alcohol and Improves Treatment Outcome. Psychological science, 22(4), 490-497.
Yegiyan, N. S., & Bailey, R. L. (2016). Food as Risk: How Eating Habits and Food Knowledge Affect Reactivity to Pictures of Junk and Healthy Foods. Health communication, 31(5), 635-642.
Appendix A
The mobile AAT measured the error rate of each participant in percentages. The analysis of these results revealed two outliers which can be seen in Figure 7. These outliers have values of 26% and 49%. As we believed the percentage of errors could influence our results, we decided to have an error cutoff of 20% and delete all participants with a higher percentage. From the initial 50 participants, two were excluded from the analysis which resulted in a final sample of 48 participants.
Figure 7. Boxplot with the distribution of the error rate
Figure 8 shows the histogram and boxplot for the reaction times before aggregation. As expected, the data had a slight skewness to the right and the boxplot shows multiple outliers. Although the boxplot does not show asymmetry, there were many outliers and extreme values in the distribution. This distribution included 48 participants, 5430 cases with a minimum of 10 ms and a maximum of 1792 ms (m = 481.95, SD = 140.93).
Figure 8. Distribution of reaction times from the mobile AAT before aggregation
The analysis of this distribution showed some unlikely reaction times. Although there is no known mean reaction time for people, many researchers believed it to be 190 ms for light stimuli (Brebner, 1980; Galton, 1899). More recent studies believe the mean reaction time to be 268 ms for simple visual stimuli or when testing on the computer (Kosinski, 2010; Eckner, Kutcher, & Richardson, 2010). The small reaction times in our distribution could have been caused by either lucky mistakes from the participants or the malfunction of the application. Thus, the cases with reaction times under 268 ms were excluded from the analysis. There were 67 such cases which represented approximately 1.23% of all cases.
