Obesity is a problem worldwide. There is an increasing number of obese individuals across age, sex, geographical location, ethnicity, and socioeconomic status (1). There are now more obese than underweight individuals (2). It is associated with numerous diseases (3) and is a significant economic burden for society (4). Furthermore, a large body of evidence suggests that the food environment impacts obesity (5). As a result, the food industry is now receiving pressure from governments worldwide to decrease sales of unhealthy food products. This may lead to stricter government policies, such as introducing nutritional warning labels on food products if retailers, food manufacturers, and marketers do not adapt. In addition, it may limit consumers’ product options. In contrast to this hard strategy, companies may nudge consumers to purchase healthier options without restricting their food choices by altering the purchase situation (6). One proposed strategy for increasing healthier food choices is simplified front-of-package food labels (7) that signal how healthy a food product is. However, such labels do not always increase healthy food purchases, although such labels do help consumers identify which products are healthy (8, 9). Further, such labels may impact people that are obese differently than people who are not obese (10). Hence, identifying possibilities of new healthy food labels may be one way to increase healthy food purchases, and this has academic, managerial, and societal value.

Technology-enabled labels that present specific information may help consumers to commit to healthier food options over unhealthier food options. Specifically, they may be presented to increase healthy food purchases. These technology-enabled healthy food labels may provide personalized, dynamic, and real-time based information regarding the healthfulness of products in point-of-purchase situations (11). For instance, Shin et al. (12) investigated the effects of dynamic displays of technology-enabled labels on healthy food purchases in an online grocery store setting. They found that these labels were effective in increasing healthy food purchases. Furthermore, Fuchs et al. (13) investigated the effects of tailored food labels on self-reported intention to use and performance expectancy. Specifically, different scores regarding healthy foods were given depending on gender, age, physical activity levels, and body-mass index of participants. They found that such labels were perceived as more helpful, relevant, and recommendable than non-tailored healthy food labels.

One may present different technology-enabled healthy food labels to consumers based on their behavior, and one may present different labels for impulsive and non-impulsive consumers in an online grocery store context. Research shows that some behaviors are associated with obesity (14), and one of these behavioral predictors may be impulsivity (15). Impulsivity can be viewed as a trans-disease, as impulsive behaviors may lead to obesity, substance abuse, and other behavioral problems. As proposed by Foxall (16), in the context of impulsivity, consumer behavior may be on a continuum from routine to extreme consumer choice. Furthermore, Foxall (17) suggests that consumer behavior models that incorporate environmental factors may provide more predictive power compared to models that do not take these into consideration. Building on this, one may use choice experiments to identify environmental variables that may increase healthy food choice (18), and examine whether some environmental factors are more effective for increasing healthy food choice for impulsive and non-impulsive consumers than others. There exists some research suggesting that the purchase of food products in an online grocery store context results in healthier choices compared to offline grocery stores (19). However, this effect may occur due to delivery time, as consumers have to wait after making the order before receiving the products. This effect may not occur if the delivery time is made shorter if online grocers become more effective in reducing delivery time. Hence, online grocers may create technology-enabled healthy food labels that use variables that increase self-control to increase healthy food purchases and provide personalized technology-enabled healthy food labels for impulsive and non-impulsive consumers.

There exist several knowledge gaps in the literature related to the effects of healthy food labels. For instance, few research articles exist on technology-enabled healthy food labels and how they impact consumer behavior despite existing theoretical literature on incorporating psychological variables in food labeling (20). Furthermore, there exist studies that have investigated how impulsivity impacts the effects of food labels on consumer behavior (21–23). However, there is little research on this in an online grocery store setting. Most of these studies have used participants’ self-reported measurements of impulsivity rather than using choice behavior. Impulsivity measured by self-reports may produce different results than choice behavior (24). In addition, implementing technology-enabled healthy food labels may provide several benefits for companies, consumers, and society. For companies, such labels may create a competitive advantage by increasing healthy food sales, build brand equity, and generating positive word-of-mouth that may attract new customers. For consumers, it may increase health benefits and well-being. For society at large, it may reduce obesity rates and the concomitant economic burden. Hence, research regarding technology-enabled healthy food labels has significant societal and academic value. This paper thus aims to contribute to the body of knowledge by providing such research. The research questions of this study are as follows:

Research question 1: What is the relative impact of (a) self-monitoring-based, (b) pre-commitment-based, and (c) social comparison-based technology-enabled healthy food labels on choice behavior in a hypothetical grocery shopping setting?

Research question 2: How does the relative impact of these technology-enabled healthy food labels on choice behavior differ for impulsive and non-impulsive consumers?

The rest of the paper is structured as follows. First, a literature review and hypotheses for this paper are provided. Second, the methodology and results of this paper are presented. Third, findings and discussion are given. At last, implications and further research directions are explored.

Impulsivity may be measured by delay discounting. Delay discounting refers to the phenomenon where the value of a reward decreases as a function of increasing the delay to receive the reward (25). This relationship can be expressed by the hyperbolic formula presented in Equation 1 for delay (26):

V is the subjective value of receiving a reward, A is the objective amount, D is the delay to receive the reward, and k is an empirically derived free parameter that determines the steepness of the subjective value. A higher k generates a steeper subjective value as a function of increasing delay than does a smaller k value. Typically, such functions are derived by asking individuals to make choices between receiving immediate and smaller or delayed and larger rewards, and then adjusting either the delay or amount. Participants’ indifference points between these two options are obtained and are used as a measure of empirical subjective value. Equation 1 has been shown to be more predictive of how the subjective value of a reward decreases as a function of delay than other models (e.g., traditional discounted utility model) and may describe preference reversals (27). Furthermore, some variables that moderate the effect of the probability of receiving a commodity on subjective value (probability discounting) may also be the same as variables that moderates the impact of the delay to receive a commodity on subjective value. However, evidence that these two constructs are the same phenomenon is small or moderates (28). In delay discounting, when the k-value is high, future events are discounted more than with lower k-values. Thus, impulsivity may be measured using k-values, as high k levels correspond to higher levels of impulsivity, while low k levels correspond to higher levels of non-impulsive (i.e., self-controlled) behaviors (for measurements of impulsivity see (29)).

High discounting rates are correlated with problematic health-related outcomes such as obesity and substance abuse (30), and discounting rewards depend on several factors. For instance, impulsivity may be due to genetic factors, as individuals who discount one commodity also tend to discount other commodities. However, it may also be influenced by current environmental factors. For instance, which type of reward is used (31, 32), cultural factors (33–35), and question framing (36, 37) may alter discounting rates. As exemplified by the Ainslie-Rachlin principle (38), there is a higher probability of choosing the immediate and smaller reward when the time between making a choice and receiving the reward is short. However, there is a higher probability of choosing the delayed and larger reward when both rewards are delayed by a constant. Using this knowledge, consumers may use external commitment devices to commit to choices that produce larger later rewards.

Delay and probability discounting have been used to investigate several factors influencing consumer behavior. For instance, it has been used to investigate the relationship between delivery fees and delay in e-commerce (39); rebates and for high and low-pricing products (40); online reviews and prices (41); hunger and discounting of food and non-food commodities (42). With regard to healthy food consumption, variables that may impact delay discounting may also impact healthy food choice (43). In accordance with this framework, there exists research that suggests that higher delay discounting of hypothetical momentary rewards is correlated with the purchase of unhealthy food products (44, 45) and that increasing delay for unhealthy foods may be used to increase the value of healthy food purchase (43).

Several systematic reviews and meta-analyses have identified that self-monitoring, pre-commitment, and social factors may increase non-impulsive behaviors (46–48). However, few studies have investigated how such strategies in the form of technology-enabled healthy food labels affect consumers at the point of purchase, and few have investigated their relative impact on choice behavior. For this study, the effects of technology-enabled healthy food labels that present self-monitoring of previous healthy food choice, pre-commitment options, and other consumers’ healthy food purchases on food choice behavior was investigated in point-of-purchase situations in a hypothetical online grocery store setting.

Self-monitoring refers to the recording and presentation of one’s own previous behavior to promote behavior change. Self-monitoring can function as a form of soft commitment (49). Specifically, observing one’s own previous patterns of choices may moderate the effects of long-term consequences on choice behavior without altering the immediate consequences of individual choices. Self-monitoring can be done actively, where individuals are required to record their behavior manually, or passively, where individuals may be presented with their own behavior history that is automatically recorded by a device. Research suggests that instructing individuals to actively record their choices may promote an increase in healthy food choices, and this has been investigated by using different technologies. For instance, Teasdale et al. (50) conducted a meta-analysis on remotely delivered strategies that used self-monitoring and tailored feedback and their effect on eating behavior. The strategies were delivered using paper reports, letters, booklets, and computers, and their results suggest that such strategies had a positive impact on eating behavior. Furthermore, Bartels et al. (51) conducted a systematic review of the effects of digital self-monitoring on improving health in middle-aged or older adults. The strategies were delivered using interactive voice response through using dials on telephones, personal digital assistants, short message services (SMS), smartphone apps, and computers. Their results show that most of the studies across behaviors lead to a change in at least one outcome measurement, including food and water consumption. Lim et al. (52) conducted a systematic review of the effects of technology apps to promote healthy food purchases and consumption. The devices that provided the strategies were mostly smartphones, and some used personal digital assistants. Their results show modest evidence for the efficacy of such strategies in improving healthy food purchase and consumption. These authors suggest that further research should explore passive automatic and personal feedback, that such digital health strategies could be incorporated into supermarket loyalty cards, and that real-time self-monitoring, feedback, and social incentives may increase healthy food choices. Hence, passive self-monitoring may be more effective in increasing the effects of long-term healthy food choices than active self-monitoring. One possible mechanism for this effect is that the presentation of previous higher values of non-impulsive behaviors may increase the probability of current non-impulsive behaviors. In addition, one may assume that non-impulsive individuals are more likely to be impacted by the presentation of their patterns of previously healthy food choice compared to impulsive individuals. This assumption is based on that non-impulsive behavior may be under the influence of temporally extended contingencies (49), such as environmental events that occur as a function of patterns of choices. Based on this, the following hypotheses are proposed:

Pre-commitment may refer to the voluntary act of changing the immediate consequences of individual choice to set the occasion for choosing larger-later rewards. Specifically, a commitment response that removes future available choices or that imposes a cost for certain choices (53, 54) in order to promote behavior change may be one way of defining pre-commitment. For instance, when consumers prefer healthy over unhealthy food when the time between making a choice and receiving the reward is large, then they can use hard commitment devices that provide additional consequences of their future individual choices. There exist studies that have investigated the effects of pre-ordering healthy food purchases and choice. For instance, Stites et al. (55) investigated the combined effects of pre-ordering lunch online, mindful eating training, fat information, and price reductions on healthy food purchases by employers in a hospital. Their results show that individuals allocated to the treatment condition purchased on average fewer calories and fat content and had a higher degree of mindful eating than individuals in the control condition. Similarly, Miller et al. (56) investigated the effects of pre-ordering compared to pre-ordering with a behavioral nudge. The nudge consisted of messages suggesting that all the components of a healthy meal or messages stating that the participants had selected a balanced meal if they selected all the healthy components. They found that participants in the pre-ordering condition had a higher average selection of fruit, vegetables, and milk products than individuals in the control condition. Furthermore, participants in the pre-ordering and behavioral nudge condition chose on average healthier products than the participants in the pre-ordering-only condition. Schwartz et al. (57) examined healthy food purchases as a function of pre-commitment by self-imposed aversive consequences. Specifically, households were enrolled in an incentive program that gave discounts on food products. The strategy consisted of an increase in the price of food products if they did not increase their prior healthy food purchase. Their results show that roughly one-third of the recruited households agreed to participate in the study. These households had higher healthy food purchases than the control group (and households that declined to participate). These studies suggest that pre-commitment may increase healthy food purchases. However, little research exists on the relationship between pre-commitment and the choice of healthy products for impulsive and non-impulsive consumers. In addition, one may assume that immediate environmental variables that may alter choice are more impactful for impulsive consumers than non-impulsive consumers. This assumption is based on that impulsive behavior may be under the influence of temporally narrow contingencies (49). Based on this, the following hypotheses are proposed:

Social proof refers to the phenomenon whereby individuals tend to copy other people’s behavior when they are uncertain regarding what choices are correct in a given situation (58). Research has examined how social proof in the context of information sources, social identity, and self-control may impact healthy food choices. However, few articles have examined the effects of personalized healthfulness information on food basket choice when it is low or high compared to other consumers’ choices. Sigurdsson et al. (59) investigated the effects of different sources of social proof on the hypothetical choice and purchase of fresh fish. Specifically, the quality of the product was based on other consumers’ ratings by using a “Top Seller” label or authoritative sources by using a “Store’s Choice” label. Their first and second study found that other consumers’ ratings had more impact on choice behavior in hypothetical online grocery and brick-and-mortar store settings. Their third study found that both labels were effective in increasing sales of fresh fish and ground beef. Furthermore, Liu et al. (60) investigated the effectiveness of social norms on eating behavior as moderated by social identity. They found that social-proof messages regarding healthy foods were effective in increasing self-reports regarding healthy eating behavior for individuals who identified with the social group that the message referred to. Furthermore, Salmon et al. (61) investigated the effects of social proof on low-fat cheese purchases of consumers with high or low self-reported self-control. Their study induced high or low self-control by using an ego-depletion. Their results show that social proof increased the average percentage of low-fat cheese purchases consumers allocated to the ego-depletion task compared to controls. However, individuals who did not perform the ego-depletion task purchased, on average less low-fat cheese. The authors suggest that high self-control individuals may have purchased other healthy food products and that these results do not necessarily show a negative effect of purchase behavior for highly self-controlled consumers. However, these results have been produced by another similar study. Gonçalves et al. (62) investigated the effects of social proof on fruits and vegetables purchases of soft, medium, and hard buyers of fruits and vegetables. Their results show that social proof increases healthy food purchases for all consumers except hard buyers. These articles indicate that social comparison presented by other consumers’ purchases increases food choices in impulsive consumers. In addition, consumers who already purchase healthy food may be assumed to have higher self-control than individuals who do not. Based on this assumption, social comparison may be more effective for impulsive consumers and may not be effective for non-impulsive consumers. Based on these studies, the following hypotheses are proposed:

Four hundred and twenty-three participants were invited to perform a study regarding consumer choice. Eighteen did not complete the survey, and their responses were removed from the analysis. The analysis was performed based on the remaining 405 participants in total. The average participant completed the study by in 526.47 s (8.77 min), with a range of 153–2,822 s (2.55–47.03 min), and a standard deviation of 290.77 (4.84 min).

The results from the adjusting delay task are shown in Figure 3. The figure shows that the category with the most participants was the 21st category (k-value = 0.0047), with a total of 62 participants. Based on these results, impulsive participants were defined as participants who completed the adjusting delay task and had a k-value of 24 to 0.0067 (from the 1st to the 20th category). Similarly, non-impulsive participants were defined as participants who completed the task and had a k-value of 0.0047–0.00011 (from the 21st to 32nd category.) As a result, 193 participants were classified as impulsive, and 212 were classified as non-impulsive.

The results of the demographic questions are shown in Table 1. Regarding all participants, the majority were males, and the most common age category was 25–34 years old. Most participants lived in a couple-household with children and had a personal annual income between £25,000 and £49,999. The majority shopped online once a week. Clothing and footwear were the most common items that were bought online, the majority of the participants bought groceries online at least once in a year, and the majority had no allergies. Regarding the impulsive participants, the majority were females, were between 25 and 34 years old, lived in a couple-household, had a personal annual income between £25,000 and £49,999, shopped online once every 2 weeks, bought online, majority of the participants bought groceries online at least once in a year, and had no allergies. Clothing and footwear were the most common type of products that were bought online. Regarding the non-impulsive participants, the majority were males, between 35 and 44 years old, lived in a couple-household, had a personal income between £25,000 and £49,999, and shopped online once a week. Books, music, movies, and games were the most common type of products bought online. Most participants bought groceries online at least once a year, and the majority had no allergies.

The results of the conjoint experiment based on all participants are shown in Figure 4. The results were the same when using logistic regression and Hierarchical Bayes estimation. Regarding the Streak label, the blue square with the number 3 was chosen more often than the blue square with the number 2, and the blue square with the number 2 was chosen more often than the absence of the Streak label. Regarding the Incentive label, the white circle was estimated to be chosen more often than the absence of Incentive labels. With regard to the Comparison label, the triangle with +15% was chosen more often than triangle with −15%, and the latter was chosen more often compared to the absence of the Comparison label. Regarding the delivery time, 30 min was chosen more often than 6 h, and the latter was chosen more often than 24 h. With regard to price, £60 was chosen more often than $70, and the latter was chosen more often than £80. The log-likelihood for the null model was −6,737.39, and the log-likelihood for the estimated model was −4,594.24, with a total difference of 2,143.15. In addition, the results from the logistic regression coefficients of the Comparison label based on impulsive participants were as follows: absent = −0.36, the triangle with −15% = −0.03, and the triangle with +15% = 0.39. The Hierarchical Bayes estimations for the same participants were as follows: absent = −0.70 (SD = 0.54), the triangle with −15% = −0.06 (SD = 0.91), and the triangle with +15% = 0.76 (SD = 0.74.) The logistic regression coefficients of the Comparison label based on non-impulsive participants were as follows: absent = −0.38, the triangle −15% = 0.03, and the triangle with +15% = 0.41. The Hierarchical Bayes estimations for the same participants were as follows: absent = −0.80 (SD = 0.67), the triangle with −15% = −0.11 (SD = 0.85), and the triangle with +15% = 0.92 (SD = 0.86.) The relative impact of the Streak label, Incentive label, Comparison label, delivery time, and price and Latent Class analyses based on these for all participants, impulsive participants, and non-impulsive participants are shown in Figure 5.

When comparing each group with itself, the results show a similar relative impact for all participants, including impulsive and non-impulsive participants. Specifically, price, Streak label, Incentive label, Comparison label, and delivery time had the most to least impact on choice in that order, using logistic regression and Hierarchical Bayes estimation. When comparing across the groups, the Streak label and incentive label had more impact on choice for impulsive participants than non-impulsive participants. Similarly, delivery time had more impact on impulsive participants compared to non-impulsive participants. In addition, price had less impact on choice for impulsive participants than non-impulsive participants. The log-likelihood for the null model based on impulsive participants was −3,210.66, and the log-likelihood model for the estimated model was −2,158.80, with a total difference of 1,051.85. The log-likelihood for the null model based on non-impulsive participants was −3,526,73, and the log-likelihood for the estimated model was −2,426.46, with a total difference of 1,100.26. When using three latent classes, the largest class shows that the Streak label and Incentive label had the most impact on choice, and the second largest shows that price and Incentive label had the most impact on choice for all participants, impulsive participants, and non-impulsive participants.

The results presented here support H1, H3, and H5, while the they do not support H2, H4, and H6. Specifically, the results show that higher values of prior healthy food choice, pre-commitment to healthy foods, and higher social comparison increase the probability of choice behavior compared to the absence of these labels. Furthermore, the latent class analysis and relative impact of these three independent variables (presented in Figure 6) did not identify segments that differed with regard to impulsive and non-impulsive participants. When using logistic regression coefficients and Hierarchical Bayes estimations of the impact of the Comparison label, the results showed no negative impact of the triangle with +15% on choice behavior for non-impulsive participants.

The main purpose of this study was to investigate whether choice behavior impacted by technology-enabled healthy food labels differed from impulsive and non-impulsive participants. Specifically, the relative impact of self-monitoring, pre-commitment and social comparison when presented as technology-enabled healthy food labels on choice behavior in a conjoint experiment was used. Impulsivity was measured through choice behavior by using an adjusting delay task.

This research contributes to two research fields. First, it relates to the emerging online grocery store and healthy food choice literature. Second, it relates to the general self-control literature and variables impacting healthy food choice. To the best of our knowledge, this is the first study to do so.

The datasets presented in this article are not readily available because the data that has been used in this study is confidential. Requests to access the datasets should be directed to NL, nikola.kristiania@kristiania.no.

Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. The patients/participants provided their written informed consent to participate in this study.

NL came up with the idea and performed data analysis. NL and AF planned the study. VS and EA provided input on data presentation and refinement of concepts. All authors contributed to the discussion of the results and final manuscript.