Using Innovative Biometric Measurements in ...

Using Innovative Biometric Measurements in Consumer Decision Making Research Khachatryan, Hayk,* and Alicia L. Rihn University of Florida, Food and Resource Economics Department and Mid-Florida Research and Education Center, 2725 South Binion Road, Apopka FL 32703. * Corresponding author.

ABSTRACT Consumer decision making styles have been widely explored in marketing and psychology literature to segment consumers with similar characteristics. However, to date, only a few studies have investigated consumer decision making styles using the innovative biometric measurement eye-tracking. Previous literature suggests using eye-tracking technology to generate explicit measures of visual behavior in consumer decision making processes and to improve econometric model fit. This manuscript incorporates conjoint analysis and eye-tracking analysis in a case study of Florida consumers to assess differences in visual attention between two decision making styles – perfectionist and impulsive. Results indicate visual attention varies between the two decision making styles and that the incorporation of eye-tracking metrics improves the econometric model fit. Overall, the results have widespread implications for product design, marketing and promotion, and retail shopping behavior researchers. We conclude that eye-tracking metrics are a promising means of researching consumer behavior.

Key words: Conjoint analysis, eye-tracking, visual behavior, impulsive, marketing, perfectionist

1

1. Introduction By nature, consumers are heterogeneous in their decision making styles. As marketers, retailers and businesses attempt to position their products/services to attract consumers, consumers’ heterogeneous nature can deplete financial and labor resources. However, as Mihić and Kusan (2010) note, consumers can be clustered into segments that exhibit similar decision making behavior (i.e. styles). Businesses can then use targeted marketing strategies to efficiently reach specific consumer segments (Mihić & Kusan, 2010; Schimmenti et al., 2013). But what marketing strategies work best to reach consumers using different decision making styles? Does decision making style influence how consumers use in-store promotional materials? Although studies attempt to address these questions, very few incorporate non-hypothetical biometric measurements (i.e. visual attention metrics). In this manuscript, we argue that consumers’ decision making styles (i.e. perfectionist or impulsive) influence their use of in-store promotions. Previous studies have isolated typical characteristics of different decision making styles (Sproles & Kendall, 1987) with perfectionist consumers being more quality conscious and systematic in their behavior (Sproles & Kendall, 1987; Wesley et al., 2006) and impulsive consumers having a higher degree of impulsivity with little regard for amount spent (Hausman, 2000; Mihić &Kusan, 2010; Sproles & Kendall, 1987). However, few studies have investigated differences in decision making styles using eye-tracking measurements; although, research has shown a connection between visual attention and consumers’ overall decision making processes (Orquin & Loose, 2013). Specifically, visual attention is directed to important components of a task (Ares et al., 2013). In an attempt to better understand how visual attention and decision making styles relate, we provide a brief literature review of decision making styles and eye-tracking technology followed by a case study of Florida consumers which incorporates eye-tracking and decision making style measurements.

1.1. Decision Making Styles A consumer’s decision making style is defined as ‘a mental orientation characterizing a consumer’s approach to making consumer choices’ that reflects the consumer’s personality (Sproles & Kendall, 1987). Consumer decision making styles have been shown to be fairly consistent even though consumers often utilize more than one style while shopping (Wesley et al., 2006). Although there are several different decision making styles (Sproles & Kendall, 1987), this manuscript focuses on perfectionist and impulsive consumers due to space constraints and the dissimilarity between the styles (Wesley et al., 2006). There are distinct differences between perfectionist and impulsive decision making styles. Consumers who use the perfectionist style value high quality products and search ‘carefully and systematically’ for those items (Sproles & Kendall, 1987). Perfectionists are often willing to pay more because they use the rationale that ‘price equals quality’ to justify their purchases (Wesley et al., 2006). Perfectionist consumers also tend to have higher incomes (Wesley et al., 2006). Conversely, consumers using the impulsive style do not plan their purchases, are more careless, and often ignore the amount spent (Sproles & Kendall, 1987). Furthermore, impulsive consumers are often influenced by in-store promotions (Kalla & Aurora, 2011; Applebaum, 1951) and obtain hedonic rewards from their spontaneous purchases (Hausman, 2000). Consequently, impulsive decision making styles are not habitual or planned but rather ‘spur of the moment’ decisions based on in-store stimuli. Both styles have unique characteristics that may influence their visual search behavior which can be measured using eye-tracking technologies.

1.2. Eye-tracking Analysis Eye-tracking analysis allows researchers to accurately measure consumers’ visual attention to marketplace stimuli. Measuring visual attention in consumer behavior research is important because 83% of information is obtained through the eyes (Wästlund et al., 2010). However, only 2% of the visual field is projected into the brain for processing (Balcombe et al., 2013). Consequently, a lot of the visual field is never processed nor contributes to the decision making process. Eye-tracking provides a means of understanding what stimuli is viewed and processed and what stimuli are ignored. Previous studies have linked visual attention to choice behavior (Armel et al., 2008; Reutskaja et al., 2011), decision making (Arieli et al., 2011; Orquin & Loose, 2013; Reisen et al., 2008), purchase likelihood (Behe et al., 2014), and improved product sales (Wansink et al., 2001). Additionally, visual attention has been associated with task complexity and product familiarity (Aribarg et al., 2010; Arieli et al., 2011; Orquin & Loose, 2013). Although visual attention has been connected to many components of decision making, to date, it has not be paired with consumers decision making styles which we address in this manuscript. Eye-tracking measurements include eye fixations, saccades, or both. Fixations occur when the eye is relatively still and typically last 200500 milliseconds (Pieters & Wedel, 2008). Saccades are when the eye is moving (with a typical duration of 20-40 milliseconds) which is the fastest movement in the human body. Similar to other data collection methods, there are advantages and disadvantages to using eye-tracking technology. The advantages include obtaining real visual data, recording natural eye movements at 30-Hz (30 gaze points per second), high precision and accuracy, large amounts of data, and location flexibility (Khachatryan & Rihn, 2014; Reisen et al., 2008). Disadvantages include not all eyes can be calibrated to the eye-tracking camera, individual participation is required, only eye data is recorded, difficulty in measurement interpretation, and extensive resources requirements (i.e. time, monetary, labor, etc.) Researchers who support incorporating eye-tracking into consumer behavior research indicate the advantages outweigh the disadvantages, which allows for additional insights into consumer behavior (Agarwal et al., 2014). For instance, Van Loo et al. (2014) and Balcombe et al. (2013) combined eye-tracking technology with choice experiments to address attribute non-attendance. Both studies found that the inclusion of eye-tracking metrics improved model fit. Similarly, Behe et al. (2014) combined eye-tracking metrics and conjoint analysis to study consumer

2

preferences for ornamental plants. Their results demonstrated a correlation between trait importance and increased visual attention to relevant traits. Combined, these studies suggest eye-tracking technology is a reliable means of improving model fit and gaining additional insights on consumer behavior. However, to date, eye-tracking measurements are rarely used to generate explicit visual behavior data or improve model fit. There are numerous implications in the global marketplace for results obtained from experiments incorporating eye-tracking technology. For instance, retailers and other businesses could streamline their promotional and labeling strategies to better attract consumers, influence their choices, and convey their messages (Rahimi, 2012; Zhang et al., 2009). Labeling policies could benefit through knowing which designs best communicate with consumers (Piqueras-Fiszman et al., 2012). Marketers could also benefit through better aligning their promotional strategies with visuals that consumers notice and that influence their behavior (Aribarg et al., 2010). The following section is a case study of Florida consumers. In the case study, we assess combining eye-tracking technology, conjoint analysis, and decision making style measurements. A brief overview of the hypotheses is provided, followed by the methodology, data analysis, and the econometric results. The last section includes our conclusions.

2. Case study – Florida consumers 2.1. Overview and Hypotheses Eye-tracking technology provides a unique opportunity to explore consumers’ decision making processes by studying consumers’ use of visual stimuli (Orquin & Loose, 2013). Previously, consumer decision making styles have been well defined in psychology and marketing literature (Applebaum, 1951; Hausman, 2000; Sproles & Kendall, 1987; Wesley et al., 2006). However, they have not been explored using eye-tracking analysis. In this case study, we investigate if visual search behavior aligns with decision making style characteristics. Based on previous research linking visual behavior to choice decisions, we hypothesize that consumers’ decision making styles influences their visual search behavior. Specifically, since impulsive consumers are characterized as “spur of the moment” decision makers that they will spend less time visually inspecting the products before reaching a purchasing decision (H1). Conversely, perfectionist consumers tend to carefully search for high quality products using a systematic method. Therefore, we hypothesize that perfectionist consumers will be more thorough in their visual search behavior, characterized by increased time spent visually attending to the images and stimuli (H2). Lastly, we hypothesize that eyetracking metrics (here, total visit duration) improves econometric model fit by providing additional information on consumer information acquisition behavior (H3).

2.2. Experimental Design In this study, a conjoint analysis, eye-tracking analysis, and questionnaire were used to assess the relationship between Florida consumers’ decision making styles, visual attention and purchase likelihood. Landscape plants were used as the product of interest due to wide availability and year around use. Product attributes included plant type, price, production method, pollinator friendly, and origin. Table 1 summarizes the attributes and attributes levels. Three different types of landscape plants were used to capture variance based on plant type. Price points were based on retail observations within the study area and represented low price points (as found at mass discount stores) and high price points (from specialty outlets). Production method, pollinator friendly and origin were included because each of these attributes are currently being explored in the green industry as a means of improving sustainability and attracting consumers (Breeze et al., 2015; Schimmenti et al., 2013; Wollaeger et al., 2015; Yue et al., 2011). All attributes were defined for participants at the beginning of the experiment. Additionally, to reduce external product variance, participants were instructed to assume that all non-target attributes (i.e. plant size, care requirements, etc.) were consistent across the plant types. Table 1. Scenario attributes and attribute levels. Attribute

Definition

Attribute levels

Plant type

Type of plant shown in the scenario image

Pricea

Price per plant

Production method

How the plants were produced

Pollinator

Describes if the plant benefits pollinators

Origin

Where the plants were produced

Petunias Pentas Hibiscus $10.98 $12.98 $14.98 Certified organicb Organic productionc Conventional Pollinator friendly Not rated In-state (‘Fresh from Florida’) Domestic (‘Grown in U.S’) Import (‘Grown outside U.S.’)

a

Price points were based on retail observations in Florida.

3

b c

Certified organic was defined as ‘the plants are certified as organically produced.’ Organic production was defined as ‘the plants are produced in an organic manner but are not certified organic.’

Cumulatively, the attribute levels result in a total of 162 possible scenarios (3 plant types x 3 price points x 3 production methods x 2 pollinator levels x 3 origins). A fractional factorial design generated 16 scenarios which reduced participant fatigue and improved experimental efficiency (Wollaeger et al., 2015). Scenario images were shown to participants on a computer monitor (58.4cm; 1920 x 1080 pixel resolution) with a Tobii X1 Light Eye Tracker camera installed at the bottom (Figure 1). The eye-tracking camera was used to record eye movements throughout the conjoint experiment. Attributes were presented to participants using above plant signs which is often the case in real retail outlets. Attribute order was randomized to minimize order effect. All signs had consistent font (color, size, style) and dimensions (1.78 x 3.89cm). Five plants were shown on a bench to represent the products being evaluated. For each scenario, participants rated their purchase likelihood using a 7 point Likert scale (1=very unlikely; 7=very likely). After the conjoint analysis, participants completed a questionnaire that included decision making style scales (adopted from Sproles & Kendall (1987) and Rook (1987)) and standard socio-demographic questions.

Figure 1. Photo of the experimental set-up showing the computer monitor, scenario image, and Tobii X1 Light Eye Tracker (as indicated by the arrow).

2.3. Experimental Procedure All experimental procedures were approved by the Institution Review Board. Florida consumers were recruited by Internet advertisements, in local newspapers, and through printed fliers distributed at public locations and garden centers. A total of 108 people participated in the study from June 23-29, 2014 in Orlando, Florida. A sample size of 108 was considered acceptable since similar studies incorporating eye-tracking often included less than 50 participants (Reisen et al., 2008; Reutskaja et al., 2011). Upon arrival, participants were greeted and asked to read and sign a consent form before starting the experiment. Next, they sat approximately 21 inches (53cm) in front of the eye-tracking computer. Then the eye-tracking camera was calibrated to their eyes using a standard 5 point calibration system (Behe et al., 2013). Participants were then shown instruction slides which defined the attributes, explained the purchase likelihood scale, and provided an example scenario image using a non-target plant (i.e. tomato). Participants then evaluated the 16 scenario images, followed by a questionnaire including decision making style scales and socio-demographic questions. Participation lasted approximately 30 minutes and participants were compensated $30. Table 2 presents a summary of participants’ socio-demographic statistics. The mean age of participants was 52.8 years, 38.5% were male, and the average household size was 1.9 people. Their 2013 mean household income was between $51,000-60,000 (USD) and most participants had completed some college at the time of the study. Florida Census statistics are provided for comparison purposes (U.S. Census Bureau, 2014). Overall, the study sample overrepresented older consumers and female consumers. Likely these differences occurred due to the study topic since older women are the core consumers of plants (Mason et al., 2008). The sample’s household size was also smaller than the state statistics. Regarding decision making styles, the majority of participants (89.5%) exhibited perfectionist decision making strategies while 21.6% were impulsive (Table 2). Some participants were included in both the perfectionist and impulsive groups which is consistent with Wesley et al. (2006) who found consumers tend to use several decision making styles rather than just one.

4

Table 2. Participants’ socio-demographic statistics (n=108). Definition Age

Age of participant

Gender Household Income

Perfectionistbc

1=male; 0=female Household size 2013 household income 1=$100K 1=some high school or less, 2=high school diploma/GED, 3=some college, 4=associate’s degree, 5=bachelor’s degree, 6=some graduate school, 7=graduate degree 1=not characteristic; 7=very characteristic

Impulsivebc

1=not characteristic; 7=very characteristic

Education

0.385 (0.487) 1.870 (1.395) 5.010 (3.025)

Floridaa Mean 20.6% < 18 years 18.7% > 65 years 0.499 2.58 $47,309

3.880 (1.649)

26.2% ≥ Bachelor’s degree

5.867 (0.715) 89.5% of sample 4.702 (0.601) 21.6% of sample

na

Total Mean (Std. Dev.) 52.782 (16.330)

na

a

Source: U.S. Census Bureau (2014) Decision making style was determined from responses to decision making scale derived from Sproles and Kendall (1987) and Rook (1987). bc The percent of sample indicates the percentage of the total sample that indicated greater than 4 (indicating the statement was ‘characteristic’ of the consumer) on the decision making style statements. Some participants were included in both groups, hence the percent of sample does not sum to 100%. b

2.4. Visual Data Preparation After the experiment, researchers defined areas of interest (AOI) within the scenario images. AOI are geometric outlines around the visual stimuli (i.e. attributes) that researchers are interested in obtaining visual data on. Here, AOI were constructed around the plant image and attribute signs, including price, production method, pollinator friendly, and origin. Researchers were then able to extract visual attention metrics from each participant for each scenario. Specifically, total visit duration (TVD) metrics were obtained for further analysis. TVD is the total amount of time (in seconds) the participants spent visually within each AOI. TVD includes both eye fixations and saccades. Previously, duration has been used in research on decision complexity (Arieli et al., 2011) and impact of task studies (Ares et al., 2013). Specifically, consumers spent more time viewing stimuli if the decision was complex (Arieli et al., 2011) or if the stimuli were relevant to the task (Ares et al., 2013).

2.5. Econometric Model Since the dependent variable (purchase likelihood) was discrete and ordinal, two ordered probit models were used to determine the impact of product attributes, decision making styles, visual attention (TVD), and socio-demographic variables on consumers’ purchasing decisions. When rating their purchase likelihood, consumers will choose the level that provides him/her with the most utility. Therefore, let Uij represent the utility consumer i derives from attribute j. Therefore, Uij can be expressed as: (1)

𝑈𝑖𝑗 = 𝛼∅ + 𝛼1 ℎ𝑖𝑏𝑖𝑠𝑐𝑢𝑠𝑖 + 𝛼2 𝑝𝑒𝑡𝑢𝑛𝑖𝑎𝑖 + 𝛼3 𝑝𝑒𝑛𝑡𝑎𝑖 + 𝛼4 𝑝𝑟𝑖𝑐𝑒𝑖 + 𝛼5 𝑝𝑜𝑙𝑙𝑖𝑛𝑎𝑡𝑜𝑟𝑖 + 𝛼6 𝑐𝑒𝑟𝑡𝑜𝑟𝑔𝑖 + 𝛼7 𝑜𝑟𝑔𝑝𝑟𝑜𝑑𝑖 + 𝛼8 𝑐𝑜𝑛𝑣𝑒𝑛𝑡𝑖𝑜𝑛𝑎𝑙𝑖 + 𝛼9 𝑖𝑛𝑠𝑡𝑎𝑡𝑒𝑖 + 𝛼10 𝑑𝑜𝑚𝑒𝑠𝑡𝑖𝑐11 + 𝛼11 𝑖𝑚𝑝𝑜𝑟𝑡𝑖 + 𝛼12 𝑎𝑔𝑒𝑖 + 𝛼13 𝑔𝑒𝑛𝑑𝑒𝑟𝑖 + 𝛼14 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑖 + 𝛼15 𝑖𝑛𝑐𝑜𝑚𝑒𝑖 + 𝛼16 𝑒𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛𝑖 + 𝛼17 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 + 𝛼18 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 + 𝜀𝑖𝑗 ; 𝑖 = 1, … , 89 (𝑛).

Model 2 incorporates the interaction effects between the decision making style and attribute TVD. Model 2 can be written as follows: (2)

𝑈𝑖𝑗 = 𝜇∅ + 𝜇1 ℎ𝑖𝑏𝑖𝑠𝑐𝑢𝑠𝑖 + 𝜇2 𝑝𝑒𝑡𝑢𝑛𝑖𝑎𝑖 + 𝜇3 𝑝𝑒𝑛𝑡𝑎𝑖 + 𝜇4 𝑝𝑟𝑖𝑐𝑒𝑖 + 𝜇5 𝑝𝑜𝑙𝑙𝑖𝑛𝑎𝑡𝑜𝑟𝑖 + 𝜇6 𝑐𝑒𝑟𝑡𝑜𝑟𝑔𝑖 + 𝜇7 𝑜𝑟𝑔𝑝𝑟𝑜𝑑𝑖 + 𝜇8 𝑐𝑜𝑛𝑣𝑒𝑛𝑡𝑖𝑜𝑛𝑎𝑙𝑖 + 𝜇9 𝑖𝑛𝑠𝑡𝑎𝑡𝑒𝑖 + 𝜇10 𝑑𝑜𝑚𝑒𝑠𝑡𝑖𝑐11 + 𝜇11 𝑖𝑚𝑝𝑜𝑟𝑡𝑖 + 𝜇12 𝑎𝑔𝑒𝑖 + 𝜇13 𝑔𝑒𝑛𝑑𝑒𝑟𝑖 + 𝜇14 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑖 + 𝜇15 𝑖𝑛𝑐𝑜𝑚𝑒𝑖 + 𝜇16 𝑒𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛𝑖 + 𝜇17 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 + 𝜇18 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 + 𝛽1 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑝𝑙𝑎𝑛𝑡𝑖 + 𝛽1 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑝𝑙𝑎𝑛𝑡𝑖 + 𝛽1 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑝𝑙𝑎𝑛𝑡𝑖 + 𝛽2 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑝𝑟𝑖𝑐𝑒𝑖 + 𝛽3 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑝𝑜𝑙𝑙𝑖𝑛𝑎𝑡𝑜𝑟𝑖 + 𝛽4 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑐𝑒𝑟𝑡𝑜𝑟𝑔𝑖 + 𝛽5 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑜𝑟𝑔𝑝𝑟𝑜𝑑𝑖 + 𝛽6 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑐𝑜𝑛𝑣𝑒𝑛𝑡𝑖𝑜𝑛𝑎𝑙𝑖 + 𝛽7 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑖𝑛𝑠𝑡𝑎𝑡𝑒𝑖 + 𝛽8 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑑𝑜𝑚𝑒𝑠𝑡𝑖𝑐𝑖 + 𝛽9 𝑝𝑒𝑟𝑓𝑒𝑐𝑡𝑖𝑜𝑛𝑖 × 𝑇𝑉𝐷𝑖𝑚𝑝𝑜𝑟𝑡𝑖 + 𝛽10 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 × 𝑇𝑉𝐷𝑝𝑙𝑎𝑛𝑡𝑖 + 𝛽11 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 × 𝑇𝑉𝐷𝑝𝑟𝑖𝑐𝑒𝑖 + 𝛽12 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 × 𝑇𝑉𝐷𝑝𝑜𝑙𝑙𝑖𝑛𝑎𝑡𝑜𝑟𝑖 + 𝛽13 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 × 𝑇𝑉𝐷𝑐𝑒𝑟𝑡𝑜𝑟𝑔𝑖 + 𝛽14 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 × 𝑇𝑉𝐷𝑜𝑟𝑔𝑝𝑟𝑜𝑑𝑖 + 𝛽15 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 × 𝑇𝑉𝐷𝑐𝑜𝑛𝑣𝑒𝑛𝑡𝑖𝑜𝑛𝑎𝑙𝑖 + 𝛽16 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 × 𝑇𝑉𝐷𝑖𝑛𝑠𝑡𝑎𝑡𝑒𝑖 + 𝛽17 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 × 𝑇𝑉𝐷𝑑𝑜𝑚𝑒𝑠𝑡𝑖𝑐𝑖 + 𝛽18 𝑖𝑚𝑝𝑢𝑙𝑠𝑖𝑣𝑒𝑖 × 𝑇𝑉𝐷𝑖𝑚𝑝𝑜𝑟𝑡𝑖 + 𝜀𝑖𝑗 ; 𝑖 = 1, … , 86 (𝑛).

In model 1 and 2 αj and μj are the attribute, socio-demographic and decision making style coefficients, while in model 2 βj are the interaction effect coefficients between the decision making style (i.e. perfectionist, impulsive) and attribute TVD variables. εij represents the residual error term uncaptured by the explanatory variables. Due to incomplete questionnaire answers a total of 89 participants (82.4% of the total sample) were included in model 1. Additionally, not all participants were able to have their eye movements recorded, thus only 86 participants (79.6% of

5

the total sample) were included in model 2. Both sample sizes were considered acceptable since they were substantially more than previous eyetracking experiments (Reisen et al., 2008; Reutskaja et al., 2011). In ordered probit models, a “base variable” from each attribute category must be excluded to prevent perfect multicollinearity (Greene, 2008). Base variables included penta for plant type, not rated for pollinator friendly, conventional for production method, and import for origin. A positive significant coefficient indicates a greater likelihood of purchase when compared to the base level. Conversely, a negative coefficient indicates a lower likelihood of purchase than the base variable. Not significant estimates are comparable to the base variables.

2.6. Results – Total Visit Duration Means Table 4 summarizes the TVD means when considering participants over all 16 scenarios. Not surprisingly, participants spent the most time viewing the plant image likely due to the image being larger and more colorful than the attribute signs (Zhang et al., 2009). Regardless of decision making style, participants spent the least amount of time on the in-state origin and price signs. Conversely, they spent more time on attributes that required additional processing, including pollinator friendly, production methods, domestic and import origins. These results are potentially due to ease of interpretation and familiarity. Prices are frequently displayed in retail settings and are very easy for consumers to interpret (Ares et al., 2013; Arieli et al., 2011). Similarly, since all participants were Florida residents, they are likely familiar with the in-state promotional campaign (i.e. Fresh from Florida). In actuality, 52% indicated having noticed the campaign on ornamental plants prior to the study. Therefore, the in-state sign required less time to interpret due to familiarity. Results are consistent with previous research which has shown complexity increases visual attention and cognitive effort (Arieli et al., 2011; Orquin & Loose, 2013) while familiarity decreases visual attention (Aribarg et al., 2010). In terms of the influence of decision making style on visual attention (i.e. TVD), perfectionist consumers spent slightly more time evaluating the different product attributes with the exception of the product image (Table 4). The decision making style TVD results support hypotheses 1 and 2. Additionally, the results are consistent with the literature that perfectionist consumers are more systematic and careful when making purchasing decisions (Sproles & Kendall, 1987; Wesley et al., 2006). Figure 2 shows a pictorial representation of the visual attention differences between the two decision making groups. Table 4. Total visit duration means, by decision making style. Total Mean (Std. Dev.) Total 6.467 (2.980) Product image 1.847 (1.474) Price 0.621 (0.395) Pollinator 0.740 (0.502) Certified organic 0.742 (0.526) Organic production 0.829 (0.602) Conventional 0.742 (0.574) In-state 0.584 (0.460) Domestic 0.699 (0.523) Import 1.093 (0.745)

Perfectionist Mean (Std. Dev.) 6.427 (2.019) 1.823 (1.431) 0.621 (0.409) 0.730 (0.505) 0.727 (0.541) 0.804 (0.612) 0.746 (0.588) 0.582 (0.474) 0.689 (0.534) 1.076 (0.755)

Impulsive Mean (Std. Dev.) 6.029 (3.156) 1.949 (1.540) 0.569 (0.451) 0.731 (0.645) 0.682 (0.582) 0.719 (0.563) 0.654 (0.373) 0.520 (0.443) 0.599 (0.395) 1.048 (0.757)

6

Original Image

Perfectionist Consumers

Impulsive Consumers

Figure 2. Heat maps demonstrating variance between perfectionist and impulsive consumers’ visual attention to attributes. Note: Color represents visual attention with red indicating a higher concentration of eye fixations and green a lower concentration of eye fixations.

2.7. Results – Ordered Logit Models Although the TVD means provide evidence in support of hypotheses 1 and 2, additional quantitative analysis is needed in order to assess the influence of consumer’s decision making styles on TVD and model fit (hypothesis 3). Table 5 reports the ordered probit model results. The Adjusted Pseudo/McFadden R2, Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC) were used to assess model fit. The increased value of the Adjusted Pseudo/McFadden R 2 in model 2 indicates superior fit when compared to model 1. Additionally, the lower AIC and BIC scores in model 2 indicate better model fit. In this instance, all three measurements support hypothesis 3 that eye-tracking

7

measurements improve model fit. These results are consistent with Van Loo et al. (2014) and Balcombe et al. (2013) who suggest using eyetracking metrics to improve model fit. Both models indicate that plant type, price, pollinator friendly, production method, and origin influence consumers’ purchase likelihood (Table 5). In model 1, compared to pentas, participants were more likely to purchase hibiscus plants and less likely to purchase petunias. As price increased, participants were less likely to purchase the products. Participants were more likely to purchase pollinator friendly plants than those not rated. Compared to conventional production methods, participants were more likely to purchase certified organic or organically produced plants. In-state and domestic plants were preferred over imported plants. The pollinator friendly, production method, and origin results are consistent with previous literature suggesting the results are robust (Schimmenti et al., 2013; Wollaeger et al., 2015; Yue et al., 2011). Regarding socio-demographic results, males were more likely to purchase the products which is inconsistent with previous studies (Mason et al., 2008). People with higher incomes were also more likely to purchase the products. Larger household size and higher level of education decreased purchase likelihood. Neither of the decision making styles were significant. Model 2 incorporates interaction effects between TVD variables and the two decision making styles (Table 5). The attribute (i.e. plant type, price, pollinator, production method, and origin) and socio-demographic results were comparable to model 1 in terms of sign/directionality and statistical significance. However, in model 2, perfectionist consumers were less likely to purchase the products while impulsive consumers were more likely to purchase. This may reflect that the participants were unaware of the products they were evaluating prior to the study. Thus reducing the perfectionist consumers’ opportunity to systematically consider which products they were interested in purchasing beforehand. The interaction terms also revealed differences between the two decision making groups, supporting hypotheses 1 and 2. Perfectionists’ TVD to the plant image, domestic and import signs positively influenced their purchase likelihood. Plant aesthetics are often positively associated with plant quality (Brand & Leonard, 2001). The results align with perfectionist consumers’ focus on product quality (Sproles & Kendall, 1987; Wesley et al., 2006). On the other hand, impulsive consumers’ TVD to price negatively impacted their purchase likelihood while their TVD to pollinator friendly improved their purchase likelihood. Interestingly, the impulsive consumer’s price TVD results are counterintuitive to the ‘little regard to what is spent’ characteristic (Sproles & Kendall, 1987). Table 5. Ordered probit model coefficient estimates. Attributes Plant type - Hibiscus Plant type - Petunia Plant type - Penta Price Pollinator Certified organic Organic production Conventional In-state Domestic Import

Model 1 Coeff. (Std. Err.) 0.179 (0.079) -0.218 (0.069) Base -0.106 (0.017) 0.188 (0.057) 0.330 (0.068) 0.452 (0.078) Base 0.676 (0.071) 0.535 (0.075) Base

Socio-demographics Age Gender Household Income Education

-0.000 (0.002) 0.178 (0.059) -0.067 (0.024) 0.054 (0.011) -0.084 (0.020)

Decision making style Perfectionist Impulsive

0.002 (0.030) -0.031 (0.028)

-0.171 (0.055) 0.167 (0.077)

**

-----------------------

0.036 (0.012) 0.063 (0.085) -0.125 (0.075) 0.072 (0.084) 0.051 (0.050) -0.073 (0.063) -0.076 (0.070) 0.149 (0.060) 0.078 (0.040) -0.041 (0.022) -0.387 (0.160)

**

TVD interaction terms Perfectionist x TVD_plant Perfectionist x TVD_price Perfectionist x TVD_pollinator Perfectionist x TVD_certified organic Perfectionist x TVD_organic production Perfectionist x TVD_conventional Perfectionist x TVD_in-state Perfectionist x TVD_domestic Perfectionist x TVD_import Impulsive x TVD_plant Impulsive x TVD_price

* **

*** *** *** ***

*** ***

** ** *** ***

Model 2 Coeff. (Std. Err.) 0.214 (0.081) -0.229 (0.070) Base -0.118 (0.017) 0.197 (0.059) 0.331 (0.070) 0.464 (0.080) Base 0.723 (0.072) 0.594 (0.077) Base 0.001 (0.002) 0.166 (0.069) -0.093 (0.028) 0.061 (0.013) -0.138 (0.025)

** ***

*** *** *** ***

*** ***

* *** *** ***

*

* *

*

8

Impulsive x TVD_pollinator Impulsive x TVD_certified organic Impulsive x TVD_organic production Impulsive x TVD_conventional Impulsive x TVD_in-state Impulsive x TVD_domestic Impulsive x TVD_import Threshold parameters a 1 2 3 4 5 6

---------------

0.297 (0.131) -0.048 (0.156) -0.055 (0.085) 0.218 (0.117) -0.051 (0.119) -0.157 (0.122) -0.039 (0.064)

-2.577 (0.353) -1.924 (0.350) -1.485 (0.348) -1.259 (0.348) -0.624 (0.348) 0.074 (0.348)

-2.184 (0.374) -2.098 (0.370) -1.634 (0.369) -1.389 (0.368) -0.685 (0.368) 0.070 (0.368)

*

Number of obs.b 1424 1376 Degree of freedom 15 33 LR chi2 259.83 452.82 Prob > chi2