*2.6. Econometric Model*

To investigate how price-conscious consumers may behave differently in term of purchase patterns, we follow Long and Freese's [45] ordered logit model and post-estimation procedures to estimate predicted probabilities of participants' purchase likelihood. As shown in Figure 3, the purchase likelihood was measured using a 7-point Likert scale question, with 1 indicating very unlikely to purchase and 7 indicating very likely to purchase. The ordered logit model captures the nature that order of response matters. Let *yi* be the ordered rating scores of purchase likelihood, which is of interest to explain. *yi* is assumed to be generated by the underlying linear latent variable model:

$$y\_i^\* = x\_i \beta + \varepsilon\_i \tag{1}$$

where *<sup>y</sup>\** is varying from −<sup>∞</sup> to <sup>∞</sup>, *<sup>i</sup>* is the observation, and *<sup>ε</sup>* is a random error term. Our observed response categories (*yi*) are linked to the latent variable using the following subsequent measurement model:

$$y\_i = \begin{cases} 1 & \text{if } \qquad \kappa\_0 = -\infty \le y\_i^\* < \kappa\_1 \\ 2 & \text{if } \qquad \kappa\_1 \le y\_i^\* < \kappa\_2 \\ \vdots & \vdots & \vdots \\ 7 & \text{if } \qquad \kappa\_6 \le y\_i^\* < \kappa\_7 = \infty \end{cases} \tag{2}$$

where *κ* are thresholds that once crossed result in a category change. In the rest of the models, *i* is suppressed. Thus, the probability of observing *y* = *j* for given values of *x* is:

$$\Pr(y=j|\mathbf{x}) = \Pr(\kappa\_{j-1} \le y^\* < \kappa\_j|\mathbf{x}) \tag{3}$$

and *j* = 1 to *J* (purchase likelihood rating). Consequently, the predicted probability can be given as:

$$\Pr(y=j|\mathbf{x}) = F(\mathbf{x}\_{\circ} - \mathbf{x}\boldsymbol{\beta}) - F(\mathbf{x}\_{\circ -1} - \mathbf{x}\boldsymbol{\beta}) \tag{4}$$

where *F* indicates the cumulative distribution function of *ε*, and for ordered logit the *ε* is assumed to have a logistic distribution with a mean of 0 and variance of *π*2/3.

The dependent variable (purchase likelihood) is a rating score (1 = very unlikely; 7 = very likely) and the key independent variables of interest are the price-consciousness indicator and the FCs on the price sign. Other control variables include plant attributes (plant type, production method, origin) and individual socio-demographics, as well as visual data (fixation counts) on other non-price product attributes.

#### **3. Results and Discussion**

Prior to regression analysis, we first compare price conscious consumers' visual attention to price versus non-price attitudes, which were measured by FCs. With a mean FC of 2.6, price conscious consumers are typically less attentive to price than non-price attributes (compared to a mean FC of 3.3 across non-price attributes). The paired *t*-test statistic for each pair of price and non-price attributes (including pollinator friendly, production method, and origin) comparison is significant at 1% significant level except for when price and in-state attributes are compared. This result contradicts Hypothesis H1a that price conscious consumers would fixate more on price than non-price attributes. Further, a direct comparison of price-conscious and non-price conscious consumers' FCs is provided in Figure 5. Overall, price conscious consumers spend less time fixating on the total image, products, prices, origins, certified organic, and conventional signs than the non-price conscious group, except for the organically produced sign. The mean FC for non-price conscious consumers is 2.7, which is slightly more than that of the price-conscious group (2.6). Nonetheless, the difference is not statistically significant (pairwise *t*-test static is 1.20 with a *p*-value of 0.23). This result does not support Hypothesis H1b that price conscious consumers fixate more on price than non-price conscious consumers. Although there is no significant difference in terms of visual attention on price between price-conscious and non-price conscious groups, price conscious consumers tend to be more efficient (i.e., have fewer total fixations and fewer fixations on price and other attributes) than non-price conscious consumers when determining their purchase likelihood. Since price conscious consumers value price over other attributes [2,12], this may reduce their visual consideration time on different attributes because the attributes are less important than price. Alternatively, the price conscious consumers may have been quicker decision makers due to having preexisting reference prices and price cut-off values. Preexisting cut-off values streamlines the decision making process because if the product does not align with the reference prices, the product is eliminated from the choice set [46].

*Horticulturae* **2018**, *4*, 13

**Figure 5.** Mean Fixation Counts, by Price Consciousness. \* indicates the mean difference between price conscious and non-price-conscious consumers is significant (*p* < 0.05) based on pairwise *t*-test.

To fully explore price conscious consumers' purchasing decisions and test Hypotheses H2a, H2b, and H3, three different specifications of the ordered logit model are estimated. Baseline Specification 1 includes only the price-conscious indicator, plant attributes, and individual demographic information. Specification 2 and Specification 3 add visual attention variables (model 2) and interaction terms between price-conscious indicators to test H2a and H2b, and visual attention variables (model 3) to test H3, respectively. Recent studies have shown attention (i.e., visual attention) provides an additional explanation for how consumers selectively process product information and is a crucial aspect that should be considered when analyzing individual choice behavior, including purchasing decisions [24,29]. The interaction terms between the price-conscious indicator and visual attention variables, specifically, the interaction between the price conscious indicator and FCs on price (PC × FC price), further distinguishes price conscious consumers from non-price conscious consumers to test H3. Indicated by the lower Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC) values (Table 3), Specification 2 and Specification 3 have largely improved the model fit and model explanation power by incorporating visual attention data.

Regression results (Table 3) from the ordered logit model indicate that price conscious consumers are significantly less likely to purchase plants in comparison with non-price conscious consumers regardless of the model specification, supporting Hypothesis H2a. The average marginal effect based on Specification 1 indicates that a price conscious consumer, *ceteris paribus*, is 1.6 percentage points more likely to rate themselves as "very unlikely" to purchase a plant, while 4.4 percentage points less likely to rate themselves as "very likely" to purchase a plant. In addition, plant attributes (plant type, price, pollinator friendly, production method, and origin), respondents' social-demographic characteristics, and visual attention variables all influence the purchase likelihood. Respondents are more likely to purchase hibiscus and pentas plants than petunia plants. As expected, price is negatively associated with purchase likelihood. Consistent with previous empirical evidence [47–50], we also find that consumers value products "being green" or sustainable. Particularly, the pollinator friendly attribute increases consumers' purchase intention. Respondents are also more likely to purchase certified organic or organically produced plants than conventionally produced plants. Regarding origin, in-state and domestically grown plants are preferred to imported plants.

In terms of social-demographic characteristics, we find purchase likelihood increases with age. Male participants are more likely to purchase products than females as shown by the positive coefficient estimates across all specifications. Respondents with higher incomes are more likely to purchase products than respondents with lower incomes. Conversely, having a larger household size discourages purchase likelihood.

The visual attention data indicates there are statistically significant relationships between price consciousness, fixations, and purchase likelihood (Specification 2, Table 3). After controlling for consumers' visual attention, the negative impact of the price-conscious indicator on purchase likelihood remains statistically significant. Consistent with price theory and existing empirical evidence (e.g., Chen et al. [29]), increasing visual attention to the price sign discourages the likelihood of purchase, supporting Hypothesis H2b. Meanwhile, we find several positive relationships between consumers' visual attention to non-price attributes and their purchase likelihood. For example, more FCs on attribute signs, such as pollinator friendly, production method, and grown outside the United States, increases purchase likelihood. These results are in line with Van Loo et al. [23], finding that consumers fixate more on attributes that they value more and, thus, are more likely to purchase them.



*Horticulturae* **2018** , *4*, 13


#### *Horticulturae* **2018** , *4*, 13

**Table 3.** *Cont.*

The complete relationship between price consciousness, visual consideration, and purchase likelihood is captured by Specification 3 (Table 3). The impact of how increasing/decreasing visual attention to the price attribute may further affect price conscious consumers' purchase likelihood, which is our primary interest, is jointly determined by the coefficients in front of FCs of price (FC\_ price) and the interaction term between the price-conscious indicator and FCs of price (PC × FC\_price). Both coefficients are negative and statistically significant, suggesting that increasing visual attention on the price attribute will further reduce price conscious consumers' purchase likelihood. This result is in support of Hypothesis 3, which states that price conscious consumers' visual attention to price signs will inversely affect their purchase likelihood.

In addition, price conscious consumers who fixate on the product longer are less likely to purchase. Although FCs on the pollinator friendly attribute, in general, increases purchase likelihood, for price conscious consumers, more fixations corresponds with a decreased likelihood of purchase. The interaction terms between the price-conscious indicator and FCs on the three production methods (certified organic, organically produced, conventional) are not statistically significant, indicating that additional visual attention to production methods did not affect price conscious consumers' purchase decisions. In other words, visual attention does not differentiate the price-conscious group of consumers from their counterparts in terms of preferences for production methods. Nonetheless, we do find, interestingly, that price conscious consumers with increased visual consideration of the domestic and import origins are more likely to purchase the products. This result may be related to perceived price, since consumers are often willing to pay premiums for locally produced ('in-state') products [51,52]. Thus, domestic or import origins would likely be considered the less expensive options by price conscious consumers. The visual attention results indicate that product attributes, which are perceived as "less expensive", may improve price conscious consumers' visual consideration and, ultimately, purchase likelihood.
