Sensory Drivers of Consumer Acceptance, Purchase Intent and Emotions toward Brewed Black Coffee

The link between coffee aroma/flavor and elicited emotions remains underexplored. This research identified key sensory characteristics of brewed black coffee that affected acceptance, purchase intent and emotions for Thai consumers. Eight Arabica coffee samples were evaluated by eight trained descriptive panelists for intensities of 26 sensory attributes and by 100 brewed black coffee users for acceptance, purchase intent and emotions. Results showed that the samples exhibited a wide range of sensory characteristics, and large differences were mainly described by the attributes coffee identity (coffee ID), roasted, bitter taste, balance/blended and fullness. Differences also existed among the samples for overall liking, purchase intent and most emotion terms. Partial least square regression analysis revealed that liking, purchase intent and positive emotions, such as active, alert, awake, energetic, enthusiastic, feel good, happy, jump start, impressed, pleased, refreshed and vigorous were driven by coffee ID, roasted, ashy, pipe tobacco, bitter taste, rubber, overall sweet, balanced/blended, fullness and longevity. Contrarily, sour aromatic, sour taste, fruity, woody, musty/earthy, musty/dusty and molasses decreased liking, purchase intent and positive emotions, and stimulated negative emotions, such as disappointed, grouchy and unfulfilled. This information could be useful for creating or modifying the sensory profile of brewed black coffee to increase consumer acceptance.


Introduction
Coffee is one of the most widely consumed beverages in the world. In 2021, revenue in the coffee segment amounts to around $409 billion, and the global coffee market is anticipated to grow annually by 6.9% during 2021-2025 [1]. Coffee-drinking is driven by different motivations, such as the psychological stimulation induced by caffeine and the sensory enjoyment provided by coffee aroma and flavor [2]. Descriptive sensory analysis was used in several studies as an effective evaluation method to determine the effects of various factors (e.g., coffee origin, roast level, brewing method, serving temperature) on coffee aroma and flavor [3][4][5][6][7]. Chambers et al. [8] developed a sensory lexicon containing a set of 110 well-defined and referenced attributes, which could be used to describe the aroma and flavor of a wide range of brewed coffee samples.
Recently, the emotional experiences elicited by foods and beverages have increasingly gained interest from researchers, because they are known to affect consumer acceptance and consumption, apart from product sensory characteristics [9]. The EsSense Profile™ (ESP) with 39 emotion terms developed by King and Meiselman [10] has been widely used to measure emotional responses toward food products [11]. As different foods evoke different emotions [12], emotion scales were developed for specific food products, such as dark chocolate [13], chocolate and hazelnut spreads [14], blackcurrant squashes [15], wine [16], beer [17], ice cream [18], muffin [19], and entomophagy [20]. For coffee, an emotion lexicon A panel consisting of eight trained panelists (all females, age range 38-56 years) from the Kasetsart University Sensory and Consumer Research (KUSCR) Center participated in the test. The panelists had 120 h of training in descriptive analysis and a minimum of 2000 h of testing experiences with various food products and beverages, including brewed coffee. The number of panelists used in this study fell within the range of 8-12, as suggested by Heymann et al. [38] for descriptive analysis. Previously, Maximo-Gacula and Rutenbeck [39] and Drake [40] indicated that a sensory panel with 6 to 14 trained panelists was adequate for descriptive analysis.
Prior to testing, 3 days of orientation (6 h each day) were held, during which panelists identified terms for describing flavor characteristics of all eight black coffee samples. To assist in identifying attributes, terminology developed by Sanchez and Chambers [5] and Chambers et al. [8] for brewed coffee was provided as an initial list. Panelists tasted all samples, discussed possible terms, and compiled a final consensus list of attributes for testing. Panelists were allowed to add new sensory terms detected in the samples which were not in the initial list. Subsequently, they discussed the definition, references and reference intensities of each attribute. During the orientation, panelists also practiced scoring of each attribute on a 15 cm line scale, with 0 meaning none and 15 meaning extremely high. Scores were monitored to ensure that variations among panelists were low (standard deviation ± 1) prior to product testing [41].
Product testing was completed in five 3 h sessions, with three to four samples being tested during each session. Two replicates were evaluated for each sample, and the evaluation was completed by replication. A randomized complete block design was used to determine the serving order within each replication. Panelists received 80 mL of each coffee sample in a pre-heated white porcelain cup labelled with a three-digit random code and covered with a lid. After tasting the sample, panelists rated the intensity of all attributes on 15 cm line scales (0 = none and 15 = extremely high) before moving onto the next sample. References were provided during evaluations to anchor values on the scale. Unsalted crackers (Jacob's Original Cream Cracker, Kraft Foods Malaysia, Petaling Jaya, Malaysia), sliced apples and reverse osmosis deionized water were provided for panelists to cleanse their palates between samples. The panelists had at least a 15 min break after each sample evaluation.

Consumer Test
One hundred Thai consumers (62 females and 38 males, age range 18-64 years) participated in the test. They were recruited based on their weekly brewed black coffee consumption of at least three times a week. Of the 100 participants, 53 drank brewed black coffee 3-5 times a week, while 32 did once daily, and 15 did more than once daily. Bhumiratana et al. [21] classified participants who drank coffee 1-2 times a week, 3-5 times a week, and at least once daily as 'light', 'medium', and 'heavy' users, respectively. Thus, according to these authors, the consumer participants of this study consisted of 53 medium users and 47 heavy users of brewed black coffee.
Each consumer evaluated all eight black coffee samples in two sessions, four samples per session with a 5 min break between samples and 20 min break between sessions. Samples were served one at a time in a randomized and balanced order across consumers. After receiving 40 mL of each coffee sample in a paper cup, consumers were instructed to taste the sample and rate their overall liking on a 9-point hedonic scale (1 = dislike extremely, 9 = like extremely) and purchase intent on a 2-point category scale (purchase/not purchase). The consumers also checked all appropriate terms listed on a check-all that-apply (CATA) question to describe their emotions as evoked by each coffee sample. The CATA question included 20 emotion terms (Table 2) selected from our previous study [27], in which a consumer-defined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Of the 20 terms, 15 terms were similar to those listed in the EsSense Profile TM (ESP) [10], WellSense Profile TM [42] and/or Coffee-drinking Experience (CDE) Profile [21], while five terms were newly generated by Thai coffee drinkers. Terms were presented in a random order among consumers in accordance with other studies [15,26] to minimize the visual processing effect on participants [43]. Unsalted crackers (Jacob's Original Cream Cracker, Kraft Foods Malaysia, Petaling Jaya, Malaysia) and purified bottled water (Nestlé Pure Life ® , Nestlé Thai, Ayuthaya, Thailand) were provided for consumers to cleanse their palates between samples.

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co., Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).  * Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

Sensory Characteristics
Enthusiastic ( * Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

Sensory Characteristics
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

Results and Discussion
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

Sensory Characteristics
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

Sensory Characteristics
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

Sensory Characteristics
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

Sensory Characteristics
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

Sensory Characteristics
* Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

Sensory Characteristics
Good mood ( * Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

)
Jump start ( * Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.  * Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.  * Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.  * Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.  * Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.  * Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumerdefined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].

Data Analysis
Multivariate analysis of variance (MANOVA) and Duncan's multiple range test (DMRT) were performed to determine differences among eight brewed coffee samples based on attribute intensities, liking scores, and percentages of purchase intent at a 95% confidence level (p ≤ 0.05). Principal component analysis (PCA), with varimax rotation, was performed to uncover relationships between significant attributes and samples. Cochran's Q test was used to determine significant differences among the coffee samples based on the frequency counts of each emotion term included in the CATA questions at a 95% confidence level (p ≤ 0.05) [44]. When significant differences were found, the Marascuilo and McSweeney procedure [44] was used for multiple pairwise comparisons. Correspondence analysis (CA) was then performed to determine relationships between significant emotion terms and samples. Partial least square regression (PLSR) analysis was also performed to determine sensory drivers of liking, purchase intent, and emotions using sensory attribute intensities as x variables and liking scores, purchase intent (%) and emotions (proportion of frequency counts) as y variables. Insignificant sensory attributes and emotion terms were excluded from the data set prior to PLSR analysis. The statistical software used for MANOVA and DMRT was IBM SPSS Statistics version 28.0 (Thaisoftup Co. Ltd., Bangkok, Thailand) and that used for PCA, Cochran's Q test, CA and PLSR was XLSTAT statistical software version 19.6 (Addinsoft, New York, NY, USA).
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples.

Sensory Characteristics
) * Emotion terms were selected from a previous study of Pinsuwan et al. [27] in which a consumer-defined emotion lexicon for coffee-drinking was developed and further refined by Thai coffee drinkers. Some of these terms were similar to those listed in other existing emotion profiles. a Terms that were also present in the EsSense Profile TM (ESP) [10]. b Terms that were also present in the WellSense Profile TM [42]. c Terms that were also present in the Coffee-drinking Experience (CDE) Profile [21].
Although sample description was given (Table 1), sample identification was not revealed when reporting the results to avoid conflicts of interest. The three-digit random codes were used to denote the samples. Table 3 shows the final list of 26 attributes detected in the coffee samples along with their definitions, references and reference intensities for sensory evaluation. Most of the terms were consistent with those in the coffee lexicons developed by Chambers et al. [8] and Sanchez and Chambers [5] (n = 22 and 2, respectively), while two terms (bitter aromatic and creosote/tar) were newly added to the list. The definitions of the pre-existing attributes were consistent with those of previous studies; however, references for most of these attributes (n = 17) were modified because the ones used in the previous studies were not available in Thailand. Newly developed references were selected through a long discussion process among all panelists to fit with the attribute definitions [41]. Chambers et al. [8] introduced the concept of 'living' lexicons, that is, lexicons must be allowed to grow, change, or adapt as new samples are tested or new understandings of the attribute dimensions and new references arise. Thus, the current research expanded the pre-existing lexicons for brewed coffee by providing new sensory terms and references. A lexicon with well-defined and referenced descriptors helps to facilitate accurate and precise communication among the sensory panelists [45].

Sensory Characteristics
Mean intensities of sensory attributes for eight brewed black coffee samples are shown in Table 4. Most attributes were present in all samples, but at varying intensities. The attributes fruity and rubber were present only in a few samples (samples 562 and 187 for fruity and sample 712 for rubber), and thus described the unique sensory characteristics of the samples. MANOVA results revealed that almost all attributes were significantly different (p ≤ 0.05) among samples, except nutty. Large differences across the samples were mainly described by coffee ID, roasted, bitter taste, balance/blended and fullness. Table 3. Sensory attributes, definitions and references for brewed black coffee evaluation.

Attribute Definition References and Intensities
Coffee identity (Coffee ID) a A distinctly roasted brown, slightly bitter aromatic characteristic of brewed coffee. Additional descriptors may/may not include woody, oily, acidic, and full bodied, and these notes may occur at varying intensities Bon Aroma gold instant coffee 1 = 3.5 Giovanni Caffé American roasted & ground 2 = 7.0 Suzuki Coffee Arabica special blend 2 = 8.5

Roasted b
Dark brown impression characteristic of products cooked to a high temperature by dry heat. It does not include bitter or burnt notes Medium roasted peanuts 3 = 6.5 Dark roasted peanuts 4 = 9.5 Over roasted peanut 5 = 15.0 The dark brown carbon impression of an over-cooked or over-roasted product that can be sharp, bitter and sour Over roasted peanuts 5 = 7.5

Acrid c
The sharp, pungent, bitter, acidic aromatics associated with products that are excessively roasted or browned Nutty c A combination of slightly sweet, brown, woody, oily, musty, astringent, and bitter aromatics commonly associated with nuts, seeds, beans, and grains Dr. Green wheat germ = 7.5 Dark chocolate c A high-intensity blend of cocoa and cocoa butter that may include dark roast, spicy, burnt, musty notes which include increased astringency and bitterness Lindt Excellence dark chocolate bar (85% cocoa) = 11.0

Attribute Definition References and Intensities
Bitter taste b The fundamental taste factor associated with a caffeine solution 0.5 g/L caffeine solution = 6.5 0.6 g/L caffeine solution = 8. Suzuki Coffee Arabica special blend 2 = 10.0 a Attribute from a study of Sanchez and Chambers [5] with modification of reference samples. b Attribute from a study of Chambers et al. [8]. c Attribute from a study of Chambers et al. [8] with modification of reference samples. d Newly added attribute in the current study. e Attribute from a study of Sanchez and Chambers [5]. 1 Prepared by dissolving instant coffee (30 g) in hot (74 • C) water (1420 mL). 2 Prepared by brewing ground coffee (30 g) with water (1420 mL) using a drip coffee maker machine. 3,4,5 Prepared by roasting raw peanut in an oven at 218 • C for 12, 17 and 22 min, respectively.
Results from the PCA map ( Figure 1) indicated that significant sensory attributes formed two key dimensions that explained 72.9% of the total variability (41.6% and 31.3%, respectively). For principal component (PC) 1, heavily loaded attributes with absolute loading values greater than 0.6 included burnt, acrid, bitter aromatic, smoky, ashy, musty/dusty, musty/earthy, creosote/tar, rubber, bitter taste, overall impact and longevity in the positive dimension and overall sweet and dark chocolate in the negative dimension. For PC2, the heavily loaded attributes in the positive dimension were coffee ID, roasted, balance/blended and fullness, while those in the negative dimension were woody, fruity, sour aromatic and sour taste. Spreading of the samples over the PCA map indicated that the selected samples represented a wide range of sensory characteristics as intended.
Sample 712, located on the far-right of the map (Figure 1), was differentiated from the rest of the samples by having the highest (p ≤ 0.05) intensities in burnt, acrid, ashy, creosote/tar, bitter taste and overall impact (Table 4). Additionally, it tended to be rated higher in bitter aroma, smoky, pipe tobacco, musty/dusty, musty/earthy, longevity, coffee ID, roasted, balance/blended and fullness, but lower in overall sweet, woody, sour aromatic and sour taste than other samples. Sample 712 was the only sample with a rubber note, even though such a flavor was detected at very low intensity. It was also the only sample with no detectable molasses flavor. Rubber 0.00 ± 0.00 b 0.00 ± 0.00 b 0.00 ± 0.00 b 0.00 ± 0.00 b 0.00 ± 0.00 b 0.00 ± 0.00 b 0.00 ± 0.00 b 1. 33  aromatic and sour taste. Spreading of the samples over the PCA map indicated that the selected samples represented a wide range of sensory characteristics as intended. Sample 712, located on the far-right of the map (Figure 1), was differentiated from the rest of the samples by having the highest (p ≤ 0.05) intensities in burnt, acrid, ashy, creosote/tar, bitter taste and overall impact (Table 4). Additionally, it tended to be rated higher in bitter aroma, smoky, pipe tobacco, musty/dusty, musty/earthy, longevity, coffee ID, roasted, balance/blended and fullness, but lower in overall sweet, woody, sour aromatic and sour taste than other samples. Sample 712 was the only sample with a rubber note, even though such a flavor was detected at very low intensity. It was also the only sample with no detectable molasses flavor.
Samples 139, 562 and 745 shared some common characteristics, as shown by a closer grouping on the PCA map (Figure 1). These samples were rated as high as (p > 0.05) sample 712 in coffee ID (Table 4). In addition, their intensities in burnt, bitter aromatic, smoky, ashy, creosote/tar and bitter taste were second to those of sample 712 and were higher (p ≤ 0.05) than those of other samples. However, differences still existed among samples 139, 562 and 745. Sample 139 was different from the other two samples by having higher (p ≤ 0.05) smoky, ashy, roasted and fullness intensities, but lower (p ≤ 0.05) woody, sour aromatic and sour taste intensities. Sample 562 was rated higher (p ≤ 0.05) in fruity, sour aromatic and sour taste, but lower (p ≤ 0.05) in bitter taste, dark chocolate and fullness than the others, while sample 745 was higher (p ≤ 0.05) in musty/dusty and dark chocolate, but lower (p ≤ 0.05) in pipe tobacco and astringent than the others.
Samples 311, 849 and 968 were located closer to each other on the left of the PCA map ( Figure 1) so they shared some common characteristics. These samples tended to be rated Samples 139, 562 and 745 shared some common characteristics, as shown by a closer grouping on the PCA map (Figure 1). These samples were rated as high as (p > 0.05) sample 712 in coffee ID (Table 4). In addition, their intensities in burnt, bitter aromatic, smoky, ashy, creosote/tar and bitter taste were second to those of sample 712 and were higher (p ≤ 0.05) than those of other samples. However, differences still existed among samples 139, 562 and 745. Sample 139 was different from the other two samples by having higher (p ≤ 0.05) smoky, ashy, roasted and fullness intensities, but lower (p ≤ 0.05) woody, sour aromatic and sour taste intensities. Sample 562 was rated higher (p ≤ 0.05) in fruity, sour aromatic and sour taste, but lower (p ≤ 0.05) in bitter taste, dark chocolate and fullness than the others, while sample 745 was higher (p ≤ 0.05) in musty/dusty and dark chocolate, but lower (p ≤ 0.05) in pipe tobacco and astringent than the others.
Samples 311, 849 and 968 were located closer to each other on the left of the PCA map ( Figure 1) so they shared some common characteristics. These samples tended to be rated lower in burnt, acrid, bitter aromatic, smoky, musty/dusty, musty/earthy, creosote/tar, bitter taste, overall impact and longevity than the rest of the samples (Table 4). Differences among these three samples existed, especially between samples 311 and 968. Of the three samples, woody and sour aromatic were the lowest (p ≤ 0.05) for sample 311, while coffee ID, roasted, bitter aromatic, ashy, molasses dark chocolate, musty/dusty, bitter taste and balance/blended were the lowest (p ≤ 0.05) for sample 968.
Sample 187, located at the bottom of the map (Figure 1), was differentiated from the rest of the samples by having the highest (p ≤ 0.05) intensities in fruity, sour aromatic and sour taste, but the lowest (p ≤ 0.05) intensities in coffee ID, balance/blended and fullness (Table 4). Additionally, it tended to be rated higher in woody and must/dusty, but lower in roasted, burnt, ashy and bitter taste than other samples.
Differences in sensory characteristics across the eight brewed black coffee samples were mainly due to differences in the roast level of the coffee beans. Generally, intensities of coffee ID, roasted, burnt, acrid, bitter aromatic, smoky, ashy, dark chocolate, creosote/tar, bitter taste and fullness tended to increase with an increased roast level from light to dark (Table 4). On the other hand, sour aromatic and sour taste intensities tended to decrease with the degree of roasting. The increases in coffee ID, roasted, burnt, acrid, smoky and ashy notes with an increased roast level have been reported by other researchers [3,33,46,47], while the decreases in sour characters with an increased roast level were in agreement with the studies of Bhumiratana et al. [33] and Akiyama et al. [47]. Roasting is one of the important factors that affects the sensory properties of coffee [48]. During the roasting process, coffee beans are exposed to a high temperature, and sugars in the beans undergo a caramelization reaction. Once the beans are heated to 205 • C, thermal decompositions and chemical changes occur, resulting in development and degradation of various volatile compounds, such as carbon dioxide, aldehydes, ketones, ethers, acetic acid, methanol, oils, and glycerol [3]. Thus, the temperature and condition during the roasting process are among the main factors that control the complexity of coffee aroma and flavor. An exception was observed for sample 562. Although it was labelled as a light roasted coffee, its coffee ID, roasted and burnt intensities were as high as those of other medium and dark roasted coffees. Apart from a roast level, other factors, such as the growing region/condition and processing methods from coffee cherries to green coffee beans [33] could have an impact on the aroma and flavor of coffee.

Consumer Liking, Purchase Intent and Emotions
Sample 712 received the highest overall liking score; however, it was not significantly different (p > 0.05) from samples 139 and 311 (Table 5). These three samples were scored higher than six (like slightly) on a 9-point hedonic scale. On the contrary, sample 187 received the lowest (p ≤ 0.05) liking score that was in the range of 'dislike slightly' to 'neither like nor dislike', while scores of samples 849, 745, 968 and 562 were in the middle range of the scale (5.6-5.7) and were not significantly different (p > 0.05) from one another. The purchase intent also was significantly (p ≤ 0.05) different among samples and followed a similar trend of liking scores. Results indicated that liking scores and purchase intent tended to increase as roast levels increased from light to dark. However, an exception was observed for sample 745. Although the sample was a dark roasted coffee, it was rated lower in liking and purchase intent than other medium roasted coffee samples. Previously, Bhumiratana et al. [21,33] found that liking scores of one consumer cluster (n = 10) toward brewed coffee tended to decrease with increased roast levels (n = 10); however, no consistent trend was observed for other consumer clusters (n = 84) regarding the effect of roast levels.
Results based on Cochran's Q test revealed that differences existed across the eight coffee samples for 17 out of 20 emotion terms. The non-discriminating emotion terms were bored, joyful and relaxed. A symmetrical correspondence analysis (CA) map ( Figure 2) shows the positioning of each coffee sample in the emotion space that explains 75.2% of total variability. The top three most liked samples (712, 139 and 311) were on the left quadrants and were explained by the emotion terms active, alert, awake, energetic, enthusiastic, feel good, good mood, happy, impressed, jump start, pleased, refreshed, vigorous and wistful. Next, on the right of these samples were samples 745, 968 and 562, which received liking scores in a neutral range. Although sample 849 was also rated in a neutral range for overall liking, its position in the CA map was closer to the top three most liked sample. By contrast, sample 187 that was disliked by consumers was on the right quadrant and was anchored by the terms disappointed, grouchy and unfulfilled. Results clearly showed that positive emotions were elicited by coffee samples that received liking scores in positive and neutral ranges of the scale, with generally higher frequency counts of positive emotions for liked samples than for neutrally liked samples, while negative emotions were evoked by disliked coffee samples. These results were supported by previous works on coffee [21,26,33] and other food categories, such as sweetener [49], salty snack, yogurt, cheese [29], beef [50] and fruit and vegetable juices [15,51] that samples with higher liking scores were more strongly associated with positive emotions, while less liked samples were more strongly associated with negative emotions. Product liking sometimes does not correlate well with emotions [52]. Even though liking scores of samples 712, 139 and 311 were not significantly different (P>0.05), sample 712 was described to stimulate positive, high-energy emotions such as alert, awake and energetic, more frequently (p ≤ 0.05) than the other two samples. Additionally, the feeling of good mood was mentioned more frequently (p ≤ 0.05) when drinking sample 849 compared to sample 745, although both samples received similar (p > 0.05) liking scores. Thus, the measurement of emotions elicited using a consumer-defined lexicon provided more discrimination across coffee samples than the hedonic measure, despite the fact that CATA emotion measures were only monitored in terms of presence/absence. The results concurred with those observed in the literature [

Sensory Drivers of Liking, Purchase Intent and Emotions
Results from PLSR analysis (Figure 3) revealed sensory drivers of liking, purchase intent and emotions toward brewed black coffee based on 100 Thai coffee users. The first Product liking sometimes does not correlate well with emotions [52]. Even though liking scores of samples 712, 139 and 311 were not significantly different (p >0.05), sample 712 was described to stimulate positive, high-energy emotions such as alert, awake and energetic, more frequently (p ≤ 0.05) than the other two samples. Additionally, the feeling of good mood was mentioned more frequently (p ≤ 0.05) when drinking sample 849 compared to sample 745, although both samples received similar (p > 0.05) liking scores. Thus, the measurement of emotions elicited using a consumer-defined lexicon provided more discrimination across coffee samples than the hedonic measure, despite the fact that CATA emotion measures were only monitored in terms of presence/absence. The results concurred with those observed in the literature [15,21,32,[53][54][55][56].

Sensory Drivers of Liking, Purchase Intent and Emotions
Results from PLSR analysis (Figure 3) revealed sensory drivers of liking, purchase intent and emotions toward brewed black coffee based on 100 Thai coffee users. The first two components generated by PLSR explained 72.8% of total variability found in sensory attributes (x, explanatory variables) and 77.2% of total variability found in overall liking, purchase intent (%) and emotions (proportion of frequency counts) (y, dependent variables). Only the attributes with standardized β-coefficients in the PLSR models (data not shown) greater than 0.05 in absolute value were interpreted as the sensory drivers of these parameters [57]. It was found that liking, purchase intent and positive emotions including active, alert, awake, energetic, enthusiastic, feel good, good mood, happy, jump start, impressed, pleased, refreshed and vigorous were mainly driven by coffee ID, roasted, balanced/blended and fullness. Ashy and pipe tobacco also stimulated liking, purchase intent and almost all positive emotions except good mood. The bitter taste of brewed black coffee not only promoted liking, but also evoked an impressed feeling and positive high-energy feelings, including alert, awake, energetic, jump start, refreshed and vigorous. Unexpectedly, rubber was positively associated with liking, purchase intent and the feelings of alert, awake, energetic, impressed, jump start, pleased, refreshed and vigorous. This attribute was present in only one sample (sample 712) at very low intensity ( Table 4). The consumers also related an overall sweet note to a happy feeling. Longevity of coffee flavor during tasting and after swallowing made the consumers feel alert, awake and impressed. awake, energetic, impressed, jump start, pleased, refreshed and vigorous. This attribute was present in only one sample (sample 712) at very low intensity ( Table 4). The consumers also related an overall sweet note to a happy feeling. Longevity of coffee flavor during tasting and after swallowing made the consumers feel alert, awake and impressed. On the contrary, sour aromatic, sour taste, fruity, woody, musty/earthy, musty/dusty and molasses negatively affected liking and purchase intent. In addition, the presence of these attributes generally stimulated negative emotions and/or suppressed positive emotions. Specifically, sour aromatic, sour taste, fruity and woody evoked disappointed, grouchy and On the contrary, sour aromatic, sour taste, fruity, woody, musty/earthy, musty/dusty and molasses negatively affected liking and purchase intent. In addition, the presence of these attributes generally stimulated negative emotions and/or suppressed positive emotions. Specifically, sour aromatic, sour taste, fruity and woody evoked disappointed, grouchy and unfulfilled feelings and suppressed all positive feelings. The presence of the musty/earthy flavor induced grouchy, and decreased active, enthusiastic, feel good, good mood, pleased and happy emotions. Similarly, a musty/dusty note elicited grouchy and decreased feel good, good mood and happy feelings. Although molasses did not evoke any negative emotions, it decreased the feelings of pleased, feel good, refresh and jump start. Acrid was also associated with the decrease of happy feeling. Therefore, these attributes were not desirable in the view of brewed black coffee users.
The positive impact of Coffee ID found in this study agreed with common knowledge that the coffee aroma/flavor generally induces positive feelings and is a key driver of coffee consumption [48,58]. Moreover, our findings on the positive effects of roasted, pipe tobacco, and bitter taste and the negative effects of sour aromatic and sour taste on emotions were consistent with those reported by Bhumiratana et al. [32] who determined sensory drivers for emotions toward brewed coffee using sensory descriptive data evaluated by trained panelists and emotion data evaluated by 94 US consumers. They found that roasted aroma and flavor aroused the feelings of jump start, satisfied, boosted and special, while tobacco flavor made the consumers feel jolted and content. In addition, bitter taste was reported to evoke energetic and productive feelings, while acidity, which corresponded to sour aromatic and sour taste in the current study, was found to arouse an off-balance feeling. However, a contradictory result was observed for the coffee aroma. Bhumiratana et al. [32] found the coffee aroma to associate with negative emotions (bored, disgusted, annoyed and disappointed) and attributed that to different understandings toward the term 'coffee' aroma in the views of consumers and trained panelists. Other attributes found to play an important role on consumer acceptance, purchase intent and emotions toward brewed black coffee in the current study (i.e., balanced/blended, fullness, rubber, overall sweet, fruity, woody, musty/earthy, musty/dusty, molasses and acrid) were not evaluated in the study of Bhumiratana et al. [32]. Another study by Hu and Lee [26] found that Korean and Chinese consumers did not like coffee with a strong bitter taste, which was contradictory to our findings. Disagreement between the two studies could be due to differences in terms of coffee sample context and consumer groups. In the current study, all samples being tested were black coffee, and all of the consumers participating in the test were black coffee users. Thereby, these consumers preferred coffee samples with a strong bitter taste, while in the study of Hu and Lee [26], various types of coffee, including the sweetened all-in-one type were evaluated, and participants were general coffee users.
It seems that most of the sensory attributes that drive acceptance, purchase intent and positive emotions while suppressing negative emotions for Thai consumers who drink brewed black coffee are the characteristics of dark roast coffees, which are the results of the roasting process. Product developers and coffee manufacturers could use the information obtained from this research in creating or modifying the sensory profile of coffee to increase consumer acceptance.

Conclusions
This research determined the sensory characteristics of brewed black coffee that affected consumer acceptance, purchase intent and emotions using data obtained from descriptive analysis and a consumer test. Results indicated that coffee ID, roasted, ashy, pipe tobacco, bitter taste, rubber, overall sweet, balanced/blended, fullness and longevity were the key sensory attributes driving liking, purchase intent and positive emotions, such as active, alert, awake, energetic, enthusiastic, feel good, happy, jump start, impressed, pleased, refreshed and vigorous. While sour aromatic, sour taste, fruity, woody, musty/earthy, musty/dusty and molasses decreased liking, purchase intent and positive emotions and stimulated negative emotions, such as disappointed, grouchy and unfulfilled, thus they were undesirable attributes. An increased roast level from light to dark tended to increase the intensities of desirable attributes (e.g., coffee ID, roasted, ashy, bitter taste and fullness), while decreasing the intensities of undesirable attributes (e.g., sour aromatic and sour taste). This information could be valuable for product developers and the coffee industry for creating or modifying the sensory profile of brewed black coffee to increase consumer acceptance. This research also expanded the pre-existing lexicons for brewed coffee by providing new sensory terms and references that could be used in sensory research.