Implicit Statistical Learning in Real-World Environments Leads to Ecologically Rational Decision Making

Abstract

Ecological rationality results from matching decision strategies to appropriate environmental structures, but how does the matching happen? We propose that people learn the statistical structure of the environment through observation and use this learned structure to guide ecologically rational behavior. We tested this hypothesis in the context of organic foods. In Study 1, we found that products from healthful food categories are more likely to be organic than products from nonhealthful food categories. In Study 2, we found that consumers’ perceptions of the healthfulness and prevalence of organic products in many food categories are accurate. Finally, in Study 3, we found that people perceive organic products as more healthful than nonorganic products when the statistical structure justifies this inference. Our findings suggest that people believe organic foods are more healthful than nonorganic foods and use an organic-food cue to guide their behavior because organic foods are, on average, 30% more healthful.

Keywords

decision making implicit statistical learning ecological rationality eye tracking field study open data open materials

Although it is certainly true that people sometimes behave irrationally, there are also plenty of examples of rational behavior in specific contexts. For instance, people’s behavior may be extremely shortsighted when they are in the grip of their urges (Ariely & Loewenstein, 2006), but the same instincts may save the day in other situations (Klein, 1998). This situational intelligence has been studied under different names, such as ecological rationality (Gigerenzer, Todd, & the ABC Research Group, 1999; Todd, Gigerenzer, & the ABC Research Group, 2012) and naturalistic decision making (Klein, 2008). To understand ecological rationality, it is essential to know when and why structures in the mind match structures in the environment (Todd & Gigerenzer, 2007). In Brunswik’s terms, such matching results from relying on cues with a high ecological validity (Brunswik & Kamiya, 1953). However, the most ecologically valid cue is not always available, so people may have to rely on other, less valid, cues. In Nordic countries, for instance, the Keyhole label is a 100% valid cue for a healthful product (Orquin, 2014), but this label appears on only 39% of healthful products. In contrast, the organic label appears on 100% of organic products (Orquin, 2014).

When cue availability varies, it is an advantage to know many cues because this reduces the number of times one must choose at random (Berretty, Todd, & Martignon, 1999). Imagine that you are searching for a good-quality watch. You know only that Swiss watches are of high quality, but during your search, you observe that Swiss watches are more expensive than other watches. When later presented with a watch of unknown origin, you may infer its quality from its price because the two are correlated. Although the learned cue (price) is valid in the learning context, it may mislead you if you apply it in other contexts; for example, inferring wine quality from prices may be misleading because the two are not correlated (Goldstein et al., 2008). Learning and exploiting ecologically valid cues should, therefore, be important to shaping ecologically rational behavior in uncertain environments, but how are such cues learned? One possibility might be implicit statistical learning.

Studies of implicit statistical learning have shown that adults and infants can learn the statistical properties of the environment merely through observation (Conway & Christiansen, 2006; Perruchet & Pacton, 2006). Such unsupervised learning allows one to infer distributional properties, correlations, and transition probabilities in the environment (Thiessen, Kronstein, & Hufnagle, 2013). Moreover, this learning happens fast (Saffran, Newport, Aslin, Tunick, & Barrueco, 1997), across sensory modalities (Conway & Christiansen, 2005), and in a variety of domains (Brady & Oliva, 2008; Kushnir, Xu, & Wellman, 2010; Xu & Garcia, 2008). Although studies of implicit statistical learning have been concerned mainly with language and visual learning, we believe that this mechanism offers an opportunity to understand ecological rationality in decision making.

In the three studies reported here, we tested implicit statistical learning in the context of organic foods. It is well known that people believe organic foods are more healthful than their conventional counterparts (Hughner, McDonagh, Prothero, Shultz, & Stanton, 2007) even though there is no conclusive scientific evidence behind this belief (Baranᑄski et al., 2014; Dangour et al., 2009; Smith-Spangler et al., 2012). Despite this lack of scientific evidence, we propose that the “organic = healthful” belief could be ecologically rational if the environment is structured such that organic foods are, in some way, more healthful than nonorganic foods. Although currently there is no evidence for such a claim, we hypothesized that organic foods are more prevalent in unprocessed, as opposed to processed, food categories—that is to say, that unprocessed foods (e.g., vegetables, fruit, milk, meat, eggs) are more likely to be organic than processed foods (e.g., frozen pizzas, candy, chips, prepackaged meals). If this is true, then the “organic = healthful” belief would be ecologically rational; a person who purchases primarily organic foods would have a higher likelihood of buying from healthful (unprocessed) food categories than would a person who purchases primarily conventional foods.

In Study 1, we tested the hypothesis that organic foods tend to be more healthful (unprocessed) than conventional foods in the natural environment. Study 1 was a field study in which we surveyed the food products available at six Danish supermarkets. Using healthfulness ratings provided by a panel of food and nutrition experts, we calculated the correlation between the prevalence of organic foods and the healthfulness of the foods in a large number of food categories. In Study 2, we asked online participants to estimate the healthfulness and prevalence of organic foods within the food categories identified in Study 1, to test the hypothesis that people learn about the correlation we observed in our first study and therefore perceive organic products to be more prevalent in healthful food categories. Finally, in Study 3, an eye-tracking experiment, we investigated whether it is possible to experimentally reproduce this implicit statistical learning by manipulating the correlation between cues indicating organic foods and healthful foods. Specifically, we expected that a positive correlation between these two types of cues would increase participants’ attention to, and use of, an organic-food cue when estimating foods’ healthfulness.

Study 1

In Study 1, we tested our hypothesis that there is a correlation between the likelihood of a product being organic and the likelihood of that product being healthful. We obtained the true percentages of organic products in a large number of food categories at six supermarkets, as well as estimates of each category’s healthfulness from a panel of food and nutrition experts.

Method

To obtain estimates of the prevalence of organic food products, we manually counted the number of conventional and organic products at six supermarkets in Aarhus, Denmark; of these supermarkets, three were small, one was medium sized, and two were large. The inclusion criterion for the food products was whether they could be consumed independently of other products or ingredients. More specifically, we decided that raw ingredients included in other products (e.g., flour, salt, sugar) would not be taken into consideration. The initial coding scheme consisted of 17 superordinate categories and 54 subordinate categories. This initial scheme was revised two times, in the second and the fourth stores, respectively, as new products were encountered. The final coding scheme consisted of 17 superordinate and 59 subordinate categories. Organic products within those 59 food categories were identified by the presence of a Danish organic label or the European Union’s organic label. To ensure that our counting was unbiased, we asked an independent coder who was blind to our hypotheses to code the products in one of the supermarkets. Intercoder reliability was calculated using Krippendorff’s (2011) alpha and was very high, α = .93 and α = .88 for total food-product count and organic-food-product count, respectively.

To obtain objective estimates of the healthfulness of the 59 food categories, we asked 15 nutrition and food scientists to complete a short survey, indicating the healthfulness of each category on a 7-point Likert scale ranging from extremely unhealthful (1) to extremely healthful (7). Ten participants completed the survey, but 1 provided the same score for all 59 food categories and was excluded from further analysis. Thus, the final sample consisted of 9 experts.

Results

The field data showed that organic food products were more prevalent in food categories that required less processing. For instance, food categories such as whole-grain pasta, brown rice, milk, and eggs had a higher prevalence of organic food products compared with categories such as prepackaged meals, candy, chips, and canned meat. Table 1 provides an overview of the total number of food products, the percentage of organic products, and the experts’ estimates of the food’s healthfulness for each of the 59 categories. We found a medium-sized, positive correlation between the true percentage of organic food products and the experts’ estimates of healthfulness, r = .35, 95% confidence interval (CI) = [.10, .56] (see Fig. 1a).

Table 1.

Results From Studies 1 and 2: Average Total Number of Products, Percentage of Organic Products, and Estimates of Healthfulness for Each Food Category

Category	Total number of products	Organic products (%)	Estimate of healthfulness
Category	Total number of products	Organic products (%)	Experts	Consumers
Whole-grain pasta	17.33	84.39	5.38	5.06
Nondairy milk	9.17	78.94	4.50	4.61
Brown rice	3.83	64.58	5.00	5.27
Milk	15.33	53.47	5.50	5.09
Unprocessed breakfast cereals	28.17	50.89	5.88	5.21
Eggs	9.00	36.82	6.38	5.22
Oil	30.33	31.46	4.88	4.16
Plain yogurt and plain-yogurt products	19.50	30.67	6.13	4.99
Syrups	32.17	29.67	1.88	2.94
Crispbread and rice crackers	37.00	27.15	4.75	4.35
Dried fruits, nuts, and seeds	100.33	25.97	5.25	4.84
Vegetables	136.33	25.45	6.63	6.16
Butter	14.67	24.46	2.75	3.34
White rice	15.83	22.92	3.50	3.57
Fruit	39.50	20.61	6.13	5.80
Honey	8.50	20.40	3.38	4.36
Juices	57.83	19.39	3.63	4.02
Processed meat	25.17	18.79	2.38	2.87
Marmalade	51.17	16.37	2.88	3.15
Chocolate spreads	14.00	16.26	2.00	2.74
Savory crackers	11.17	15.48	2.50	3.09
Fruit yogurt	42.00	14.86	3.00	4.07
Frozen meat	13.33	14.35	4.50	4.49
Fresh meat	62.17	13.08	5.00	4.96
Whole-grain bread	25.33	12.37	6.25	5.52
Canned vegetables	91.50	12.23	4.88	4.50
Cream	12.33	11.42	2.63	3.04
Frozen fruit	6.50	11.21	5.38	4.82
Frozen bread	19.67	10.78	4.13	3.67
Cheese	153.67	10.56	4.50	4.57
Frozen vegetables	33.33	9.90	6.25	5.27
Canned fruit	15.83	9.89	3.50	3.96
Sauces (tomato, pesto)	42.83	9.58	4.63	4.05
Cold cuts	123.00	9.29	3.50	4.02
Refined-wheat-flour pasta	36.00	8.21	2.75	3.20
Ice cream	39.50	7.49	2.25	2.68
Dressings (salad dressings, mayonnaise, ketchup, mustard)	76.50	6.38	3.38	2.86
Cakes and cookies	80.67	5.49	2.13	2.24
Muesli and protein bars	16.00	5.28	4.00	4.09
Frozen prepackaged meals	64.67	5.11	2.50	3.22
Processed breakfast cereals	25.33	4.71	1.75	2.66
Refined-wheat-flour bread	42.00	4.19	2.88	2.77
White wine	44.83	4.18	4.00	3.36
Sodas	109.50	3.51	1.50	2.06
Alcoholic beers and other alcoholic beverages	131.33	3.32	3.00	2.45
Mayonnaise-based salads	36.83	3.03	3.00	3.09
Chips	56.33	2.67	1.50	1.92
Red wine	121.17	2.59	4.38	3.69
Candy	382.83	2.19	1.38	1.92
Soups	12.83	1.31	4.00	3.97
Frozen fish	15.17	1.04	6.00	5.21
Dry prepackaged meals	17.33	0.83	2.88	3.16
Processed fish (refrigerated)	49.17	0.76	5.25	4.99
Prepackaged meals: sauces	38.00	0.29	2.13	2.95
Take-out meals	4.67	0.00	3.00	3.18
Canned fish	35.50	0.00	5.63	4.63
Canned meat	5.83	0.00	3.25	3.56
Fresh fish	5.17	0.00	6.63	5.99
Refrigerated prepackaged meals	11.17	0.00	2.75	3.27

Fig. 1.

Results from Studies 1 and 2: scatterplots showing the relationships between (a) the true percentages of organic foods in the 59 food categories and experts’ estimates of the healthfulness of those categories, (b) experts’ and consumers’ estimates of the healthfulness of the food categories, (c) the true percentages of organic foods in the categories and consumers’ perception of those percentages, and (d) consumers’ perception of the percentages of organic foods in the categories and their estimates of the healthfulness of the categories. The trend line in each plot represents the best-fitting linear regression line, and the shaded area around the trend line is its 95% confidence interval.

Next, we compared the estimated healthfulness of conventional and organic foods. We found that organic foods (M = 4.47, SD = 1.48) were, on average, 30% more healthful than conventional foods (M = 3.44, SD = 1.59), d = 0.65.

Discussion

Study 1 confirmed our hypothesis that more healthful food categories have a higher prevalence of organic foods. We suggest that this happens for two reasons. First, multiple ingredients must be organic in order for a processed food to be organic, so it is likely rare for highly processed foods with many ingredients to be organic. Foods that are highly processed (e.g., prepackaged meals, candy, and chips) tend to be unhealthful foods. Second, it appears that some organic-food producers target health-conscious consumers, which leads to an overrepresentation of organic foods in relatively healthful subcategories; for example, whole-grain pasta is more likely to be organic than is refined-wheat-flour pasta.

Study 2

In Study 1, we found a correlation in the environment between the likelihood of a product being organic and the likelihood of that product being healthful. We hypothesized that people learn this statistical structure and therefore perceive organic products to be more prevalent in more healthful food categories. In Study 2, we tested this hypothesis.

Method

Participants

Seven hundred seventy-three participants representative of the Danish population were recruited through an online consumer-panel provider. Six hundred thirty-seven participants completed the study. Their age ranged from 17 to 81 years (M = 42.95, SD = 16.09), and the numbers of men and women were approximately equal (315 women). The sample captured a broad spectrum of the population with regard to age, gender, and organic-food purchasing behavior as well as psychographic dimensions (attitudes toward organic food products). Each participant received approximately €1 for completing the study. The sample size was the maximum achievable given budgetary constraints.

Materials and procedure

Participants were recruited online, and all gave informed consent before commencing the study. First, participants were asked to estimate the percentage of organic foods in each of the 59 food categories identified in Study 1. Subsequently, they were asked to estimate the healthfulness of each food category on a 7-point Likert scale ranging from extremely unhealthful (1) to extremely healthful (7). We also collected demographic and psychographic information about the sample, as well as information about their purchases of organic foods. Purchasing behavior was measured with two items. The first item asked participants how often they purchased organic foods; the 7-point unipolar response scale ranged from never (1) to always (7; Magnusson, Arvola, Hursti, Åberg, & Sjödén, 2001). The second item asked participants to use a visual analogue scale ranging from 0 to 100 to indicate the percentage of the food they purchased that was organic. Attitudes toward purchasing organic foods were measured by asking participants to indicate how “good,” “important,” and “wise” they thought it was to purchase organic food products. The 7-point bipolar response scales ranged from very bad (1) to very good (7), from very unimportant (1) to very important (7), and from very foolish (1) to very wise (7), respectively (Magnusson et al., 2001). Beliefs about organic foods were measured by asking participants to rate, on 7-point Likert scales, whether they thought organic products are “healthier,” are “tastier,” have “fewer calories,” are of “better quality,” are “fresher,” and are “safer” than conventional products. (See Figs. S1–S3 in the Supplemental Material available online for results for the food-shopping and psychographic questions.)

Results

Combining the data from Study 1 and Study 2, we found a strong, positive correlation between the true and perceived percentages of organic food products across the food categories, r = .65, 95% CI = [.45, .77] (see Fig. 1c). This result suggests that participants had accurately learned the prevalence of organic foods in the various categories. The results also showed a strong, positive correlation between experts’ and consumers’ estimates of healthfulness, r = .95, 95% CI = [.91, .97] (see Fig. 1b), which suggests that consumers’ healthfulness estimates were very accurate. Finally, we found a strong, positive correlation between consumers’ perceptions of the prevalence of organic foods in the categories and consumers’ estimates of the categories’ healthfulness, r = .72, 95% CI = [.55, .81] (see Fig. 1d). An overview of the consumer estimates can be found in Table 1.

Discussion

Study 2 supports our hypothesis that people learn the statistical structure of their environment. Consumers accurately estimated the prevalence of organic foods in different food categories and made very accurate estimates of the categories’ healthfulness (i.e., estimates similar to those of food and nutrition experts). Interestingly, there was a stronger correlation between consumers’ perceptions of the prevalence of organic products and their estimates of healthfulness, r = .72, than between the true prevalence of organic products and experts’ estimates of healthfulness, r = .35. This result could have been due to the consumers’ “organic = healthful belief” influencing either their perception of the prevalence of organic products in the food categories or their perception of the healthfulness of the categories.

Study 3

Although Studies 1 and 2 provided evidence supporting our hypothesis that consumers learn the statistical structure of the environment through implicit statistical learning, these studies were correlational in nature. In Study 3, we therefore conducted a lab-based, eye-tracking study, manipulating the correlation between organic-food and healthful-food cues. We asked participants to select the most healthful of eight alternative food products on each trial. As an objective health cue, we use the Nordic Keyhole label, which indicates healthful alternatives within a product category (Ministry of Food, Agriculture and Fisheries, 2013). Because the Keyhole is present only on some healthful products (Orquin, 2014), it is useful to rely on other cues as well when judging a product’s healthfulness. We therefore expected that participants would be more likely to attend to an organic-food cue when this cue was positively correlated with the Keyhole than when there was no or a negative correlation between these two types of cues.

Method

Participants

Seventy-eight Danish participants were recruited through a consumer-panel provider. Seven participants were excluded after the experiment because of insufficient data quality (calibration-related errors). Thus, the final sample consisted of 71 participants, all of whom gave informed consent. The participants ranged in age from 18 to 74 years (M = 45.73, SD = 15.12), and the majority were men (19 women). Only participants with normal, or corrected-to-normal, and full color vision were included in the study. Each participant received a gift card worth approximately €34 for completing the study. The sample size was the maximum achievable given budgetary constraints and provided more than the minimum of 20 participants per cell suggested by Simmons, Nelson, and Simonsohn (2011).

Stimuli and apparatus

The experimental stimuli consisted of 50 choice sets of processed food products, each with eight alternatives positioned in a 4 × 2 array on a computer monitor; within each array, neighboring items were separated by 5.1° of visual angle horizontally and 10.3° of visual angle vertically. Each alternative was represented by a product picture, name, brand, price, and weight. In addition, the presence of two features—the Keyhole label and the organic label—was manipulated. The rate of co-occurrence of the two labels varied across three conditions (25%, 50%, and 75% co-occurrence). More specifically, the number of Keyhole labels and the number of organic labels were constant across conditions (four Keyhole and four organic labels). Therefore, in the –.5 condition (i.e., r = −.5), there was a negative correlation between the labels; in the .0 condition (i.e., r = 0), there was no correlation between the labels; and in the .5 condition (i.e., r = .5), there was a positive correlation between the labels. An example of a stimulus set is shown in Figure 2. The Keyhole and organic labels were randomly distributed across alternatives in each choice set, and the presentation order of the choice sets was randomized across participants.

Fig. 2.

Example of a choice set used in Study 3. Each set consisted of eight alternatives, four of which had the organic label and four of which had the Keyhole label. The example shown here is from the −.5 condition, in which the two labels co-occurred 25% of the time. Note that the spacing between the images has been reduced because of space constraints.

Eye movements were recorded using a Tobii (Stockholm, Sweden) T60 XL eye tracker with a temporal resolution of 60 Hz and a screen resolution of 1,920 × 1,200 pixels. Average viewing distance was 60 cm from the screen, and a chin rest was used to stabilize head position. Areas of interest (AOIs) were determined by defining the pixel positions of the potential labels in each choice set (16 possible positions). Fixations were identified using a velocity based algorithm (I-VT algorithm) with default settings (Salvucci & Goldberg, 2000). Specifically, the maximum length of the gap between fixations was set to 75 ms. A noise-reduction function was not applied, and we used averaged data from the left and right eyes. The velocity threshold was set to 30° per second. Fixations with a duration less than 60 ms were discarded. The AOIs’ margins around the actual labels were set to approximately 0.15° to take into account inaccuracy in the recording of fixation locations. There have been several attempts to define the most suitable AOI margins (Orquin, Ashby, & Clarke, 2016). More specifically, we tested AOI margin sizes of 0°, 0.15°, and 0.5° of visual angle by comparing results obtained with those margin sizes with results obtained using hand-coded fixations. For this comparison, we used a total of 432 hand-coded fixations drawn from six trials for each of 3 participants in each condition. Given the number of hand-coded fixations on each AOI as our criterion, we counted the number of false negatives and false positives in the fixation counts produced by the software. We found that different AOI margin sizes influenced the number of false negatives and false positives registered. The AOI margin size of 0.15° of visual angle had the most acceptable rates of false negatives and false positives.

Procedure

The study was conducted in a light-controlled laboratory environment. Upon arrival, participants were greeted and seated in front of the eye tracker. We adjusted the height of the chin rest and proceeded with calibration using the Tobii Studio nine-point calibration procedure. After calibration, each participant was randomly assigned to one of the three conditions. The instructions told participants to select the most healthful alternative among the eight food products in each set and to indicate the choice with a mouse click. A fixation cross lasting 1,000 ms appeared before each choice set. Participants used as much time as needed to make their choices.

Results

Eye-movement analysis

To test whether participants attended more to the organic label when it co-occurred with the Keyhole label more frequently, we analyzed the eye-tracking data by means of a generalized linear mixed model. The model was fitted using the lme4 package in R (Bates, Mächler, Bolker, & Walker, 2015). Fixation selection (AOI fixated or not) was the dependent variable, and condition and label type were independent variables. The best-fitting model had a binomial response distribution, a logit link function, and two random intercepts grouped by participant and choice set. The analysis revealed no significant main effect of condition, χ²(2, N = 71) = 0.63, p = .73; a significant main effect of label type, χ²(1, N = 71) = 24.58, p < .001; and a significant interaction between condition and label type, χ²(2, N = 71) = 24.13, p < .001.

To interpret the direction of the interaction effect, we plotted fixation likelihood for each combination of condition and label type (see Fig. 3). The graph shows that participants attended to the organic label more frequently, at the expense of the Keyhole label, as the two labels’ rate of co-occurrence increased.

Fig. 3.

Eye-movement results from Study 3: likelihood of fixation on the Keyhole and organic labels in each condition. Error bars represent 95% confidence intervals.

Follow-up analysis

One potential problem with the fixation-likelihood analysis is that fixations on the organic label in the .5 condition could have been an artifact. Specifically, the pattern in Figure 3 could have resulted if participants searched for the Keyhole label and then fixated the remaining information on Keyhole-labeled products. If this were the case, the Keyhole should have driven fixations elsewhere on the products; that is, participants should have been faster to fixate the Keyhole label than the organic label. To exclude this possibility, we inspected the trials in which participants fixated both labels. As Table 2 shows, participants who fixated both labels on a product were equally likely to fixate the Keyhole label first and to fixate the organic label first. We take this to imply that the Keyhole label did not drive fixations and hence that the results of the fixation-likelihood analysis are not artifactual.

Table 2.

Results From Study 3: Number of Trials on Which Each Cue Label Was Fixated First Given That Both Labels Were Fixated

Condition	Keyhole label fixated first	Organic label fixated first
–.5	18	15
.0	29	31
.5	42	44

Choice analysis

To examine the effect of condition on participants’ choice of products, we fitted individuals’ choice data by means of multinomial logit models using the mlogit package in R (Croissant, 2013). Each individual’s data were fitted with a null model, including only intercepts for the eight product alternatives, and a full model, including a term for product type (i.e., whether the alternative had the Keyhole label, the organic label, both labels, or neither label). We calculated the difference between Akaike’s information criterion (AIC) for the full model and the null model (AIC_full – AIC_null). Out of 71 participants, 42 were identified as label users (AIC difference > 0), and 29 were identified as non–label users (AIC difference ≤ 0). We then calculated the standardized mean difference (SMD) between the choice likelihood in the .5 and −.5 conditions for the products with the organic label and products with both labels, correcting for chance level:

S M D = \frac{(M_{. 5} - M_{- . 5}) - (M_{. 5 chance} - M_{- . 5 chance})}{S D_{p o o l e d}}

We found a medium-size increase in the corrected likelihood of choosing products with the organic label, SMD = .42, and a large increase in the corrected likelihood of choosing products with both labels, SMD = .83, for label users in the .5 condition, relative to the −.5 condition. For non–label users, we found that choices were close to chance level for products with the organic label, SMD = −.08, and products with both labels, SMD = .06. Figure 4 shows choice likelihood as a function of condition for products carrying the organic label, the Keyhole label, both labels, and neither label, separately for label users and non–label users.

Fig. 4.

Choice results from Study 3: likelihood of a product being chosen as a function of condition, separately for label users and non–label users and for products with the organic label, the Keyhole label, both labels, and neither label. Error bars represent 95% confidence intervals. Also shown are the likelihoods that would be expected if choices were made according to chance.

Discussion

In Study 3, we experimentally investigated whether people are capable of learning the statistical structure of a natural environment. We found that participants responded to the statistical structure created by our task, both in their eye movements and in their choices. When there was a positive correlation between the organic label and the Keyhole label, participants were more likely to fixate on the organic label. This gaze bias suggests that participants in this condition considered the organic label as relevant to the health judgment task (Orquin & Mueller Loose, 2013). We also found that the majority of participants incorporated labels in their judgments, and these participants were more likely to choose products with the organic label when there was a positive correlation between the two types of labels than when there was no correlation or a negative correlation, after correcting for chance. Figure 4 shows that participants chose products with both labels more often than would be expected by chance in all three conditions. This means that participants generally preferred products with both labels to products with either label or no label. The preference for foods with both labels increased when there was a positive correlation between the labels. Overall, the findings support our hypothesis that people are, without explicit instructions, capable of learning the statistical structure of the environment and applying a learned cue in their decision making.

General Discussion

We hypothesized that people learn statistical structures in their environment and use this information to shape ecologically rational decision making. We tested this hypothesis in the context of organic foods. In Study 1, we found that in the environment, the prevalence of organic foods within a category correlates with the category’s healthfulness. In Study 2, we found that people are familiar with this statistical structure and therefore perceive organic products to be more prevalent in more healthful food categories, as indicated by their highly accurate perceptions of the prevalence of organic products in different food categories. In Study 3, we found that a positive correlation between organic-food and healthful-food cues, compared with a negative or no correlation, leads people to attend more to the organic-food cue when they are judging foods’ healthfulness. Specifically, participants’ likelihood of fixating on the organic-food cue and their likelihood of choosing products with the organic label were higher in the positive-correlation condition. We take this to imply that people are capable of learning the statistical structure of the environment and applying a learned cue correctly when making decisions. Our findings contribute to a better understanding of ecological rationality by showing that implicit statistical learning can lead to accurate beliefs about correlational structures in the environment. These beliefs translate into decisions that match the environment and produce ecologically rational behavior.

Supplemental Material

sj-doc-2-pss-10.1177_0956797617733831 – Supplemental material for Implicit Statistical Learning in Real-World Environments Leads to Ecologically Rational Decision Making

Supplemental material, sj-doc-2-pss-10.1177_0956797617733831 for Implicit Statistical Learning in Real-World Environments Leads to Ecologically Rational Decision Making by Sonja Perkovic and Jacob Lund Orquin in Psychological Science

Supplemental Material

sj-docx-3-pss-10.1177_0956797617733831 – Supplemental material for Implicit Statistical Learning in Real-World Environments Leads to Ecologically Rational Decision Making

Supplemental material, sj-docx-3-pss-10.1177_0956797617733831 for Implicit Statistical Learning in Real-World Environments Leads to Ecologically Rational Decision Making by Sonja Perkovic and Jacob Lund Orquin in Psychological Science

Supplemental Material

sj-pdf-1-pss-10.1177_0956797617733831 – Supplemental material for Implicit Statistical Learning in Real-World Environments Leads to Ecologically Rational Decision Making

Supplemental material, sj-pdf-1-pss-10.1177_0956797617733831 for Implicit Statistical Learning in Real-World Environments Leads to Ecologically Rational Decision Making by Sonja Perkovic and Jacob Lund Orquin in Psychological Science

Footnotes

Acknowledgements

The authors thank Nicola Bown, Wandi Bruine de Bruin, Edward Cokely, Morten H. Christiansen, Blazenka Divjak, Mirta Galesic, Klaus G. Grunert, Gulbanu Kaptan, Brandi S. Morris, Michael Morris, Yasmina Okan, and Matthew Robson for their helpful comments on an earlier version of the manuscript. The authors also thank Valon Buxhovi and Linda Redere for their contribution to Study 1 and Study 3, respectively.

Action Editor

Steven W. Gangestad served as action editor for this article.

Declaration of Conflicting Interests

The authors declared that they had no conflicts of interest with respect to their authorship or the publication of this article.

Funding

This research was supported by the Aarhus University Research Foundation (AUFF).

Supplemental Material

Additional supporting information can be found at

Open Practices

All data and materials have been made publicly available via the Open Science Framework and can be accessed at https://osf.io/zpjbg/. The complete Open Practices Disclosure for this article can be found at https://journals-sagepub-com.web.bisu.edu.cn/doi/suppl/10.1177/0956797617733831. This article has received badges for Open Data and Open Materials. More information about the Open Practices badges can be found at .

References

Ariely

Loewenstein

(2006). The heat of the moment: The effect of sexual arousal on sexual decision making. Journal of Behavioral Decision Making, 19, 87–98. doi:10.1002/bdm.501

Baranᑄski

Sᑄrednicka-Tober

Volakakis

Seal

Sanderson

Stewart

G. B.

, . . . Leifert

(2014). Higher antioxidant and lower cadmium concentrations and lower incidence of pesticide residues in organically grown crops: A systematic literature review and meta-analyses. British Journal of Nutrition, 112, 794–811. doi:10.1017/S0007114514001366

Bates

Mächler

Bolker

B. M.

Walker

S. C.

(2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67, 1–48. doi:10.18637/jss.v067.i01

Berretty

P. M.

Todd

P. M.

Martignon

(1999). Categorization by elimination: Using few cues to choose. In Gigerenzer

Todd

P. M.

& the ABC Research Group (Eds.), Simple heuristics that make us smart (pp. 235–254). Oxford, England: Oxford University Press.

Brady

T. F.

Oliva

(2008). Statistical learning using real-world scenes: Extracting categorical regularities without conscious intent. Psychological Science, 19, 678–685. doi:10.1111/j.1467-9280.2008.02142.x

Brunswik

Kamiya

(1953). Ecological cue-validity of ‘proximity’ and of other Gestalt factors. The American Journal of Psychology, 66, 20–32. doi:10.2307/1417965

Conway

C. M.

Christiansen

M. H.

(2005). Modality-constrained statistical learning of tactile, visual, and auditory sequences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31, 24–39. doi:10.1037/0278-7393.31.1.24

Conway

C. M.

Christiansen

M. H.

(2006). Statistical learning within and between modalities: Pitting abstract against stimulus-specific representations. Psychological Science, 17, 905–912.

Croissant

(2013). mlogit: Multinomial logit model (R package version 0.2-4) [Computer software]. Retrieved from https://cran.r-project.org/package=mlogit

10.

Dangour

A. D.

Dodhia

S. K.

Hayter

Allen

Lock

Uauy

(2009). Nutritional quality of organic foods: A systematic review. American Journal of Clinical Nutrition, 90, 680–685. doi:10.3945/ajcn.2009.28041

11.

Gigerenzer

Todd

P. M.

& the ABC Research Group. (1999). Simple heuristics that make us smart. Oxford, England: Oxford University Press.

12.

Goldstein

Almenberg

Dreber

Emerson

Herschkowitsch

Katz

(2008). Do more expensive wines taste better? Evidence from a large sample of blind tastings. Journal of Wine Economics, 3, 1–9. doi:10.1017/S1931436100000523

13.

Hughner

R. S.

McDonagh

Prothero

Shultz

C. J. I.

Stanton

(2007). Who are organic food consumers? A compilation and review of why people purchase organic food. Journal of Consumer Behaviour, 6, 94–110. doi:10.1002/cb.210

14.

Klein

G. A.

(1998). Sources of power: How people make decisions. Cambridge, MA: MIT Press.

15.

Klein

G. A.

(2008). Naturalistic decision making. Human Factors: The Journal of the Human Factors and Ergonomics Society, 50, 456–460. doi:10.1518/001872008X288385

16.

Krippendorff

(2011). Computing Krippendorff’s alpha-reliability. Retrieved from http://repository.upenn.edu/cgi/viewcontent.cgi?article=1043&context=asc_papers

17.

Kushnir

Wellman

H. M.

(2010). Young children use statistical sampling to infer the preferences of other people. Psychological Science, 21, 1134–1140. doi:10.1177/0956797610376652

18.

Magnusson

M. K.

Arvola

Hursti

U.-K. K.

Åberg

Sjödén

P.-O.

(2001). Attitudes towards organic foods among Swedish consumers. British Food Journal, 103, 209–227. doi:10.1108/00070700110386755

19.

Ministry of Food, Agriculture and Fisheries. (2013). Lov om Fødevarer [Law on foodstuffs]. Lovtidende A, 250, 1–10.

20.

Orquin

J. L.

(2014). A Brunswik lens model of consumer health judgments of packaged food. Journal of Consumer Behaviour, 13, 270–281. doi:10.1002/cb.1465

21.

Orquin

J. L.

Ashby

N. J. S.

Clarke

A. D. F.

(2016). Areas of interest as a signal detection problem in behavioral eye-tracking research. Journal of Behavioral Decision Making, 29, 103–115. doi:10.1002/bdm.1867

22.

Orquin

J. L.

Mueller Loose

(2013). Attention and choice: A review on eye movements in decision making. Acta Psychologica, 144, 190–206. doi:10.1016/j.actpsy.2013.06.003

23.

Perruchet

Pacton

(2006). Implicit learning and statistical learning: One phenomenon, two approaches. Trends in Cognitive Sciences, 10, 233–238. doi:10.1016/j.tics.2006.03.006

24.

Saffran

J. R.

Newport

E. L.

Aslin

R. N.

Tunick

R. A.

Barrueco

(1997). Incidental language learning: Listening (and learning) out of the corner of your ear. Psychological Science, 8, 101–105. doi:10.1111/j.1467-9280.1997.tb00690.x

25.

Salvucci

D. D.

Goldberg

J. H.

(2000). Identifying fixations and saccades in eye-tracking protocols. In Proceedings of the Eye Tracking Research & Applications Symposium 2000 (pp. 71–78). New York, NY: ACM Press.

26.

Simmons

J. P.

Nelson

L. D.

Simonsohn

(2011). False-positive psychology: Undisclosed flexibility in data collection and analysis allows presenting anything as significant. Psychological Science, 22, 1359–1366. doi:10.1177/0956797611417632

27.

Smith-Spangler

Brandeau

M. L.

Hunter

G. E.

Bavinger

J. C.

Pearson

Eschbach

P. J.

, . . . Bravata

D. M.

(2012). Are organic foods safer or healthier than conventional alternatives?: A systematic review. Annals of Internal Medicine, 157, 348–366. doi:10.7326/0003-4819-157-5-201209040-00007

28.

Thiessen

E. D.

Kronstein

A. T.

Hufnagle

D. G.

(2013). The extraction and integration framework: A two-process account of statistical learning. Psychological Bulletin, 139, 792–814. doi:10.1037/a0030801

29.

Todd

P. M.

Gigerenzer

(2007). Environments that make us smart: Ecological rationality. Current Directions in Psychological Science, 16, 167–171.

30.

Todd

P. M.

Gigerenzer

& the ABC Research Group. (2012). Ecological rationality: Intelligence in the world. New York, NY: Oxford University Press.

31.

Garcia

(2008). Intuitive statistics by 8-month-old infants. Proceedings of the National Academy of Sciences, USA, 105, 5012–5015. doi:10.1073/pnas.0704450105

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.59 MB

0.29 MB

0.33 MB

0.02 MB

0.29 MB