Volume 57, Issue 4 p. 1936-1944
Original article
Open Access

Experimental gluten-free biscuits with underexploited flours versus commercial products: Preference pattern and sensory characterisation by Check All That Apply Questionnaire

Maria Di Cairano

Maria Di Cairano

Scuola di Scienze Agrarie, Università degli Studi della Basilicata, Alimentari, Forestali ed Ambientali, viale dell’Ateneo Lucano, Potenza, 10 – 85100 Italy

Contribution: Conceptualization (equal), Data curation (equal), Formal analysis (equal), ​Investigation (equal), Writing - original draft (equal), Writing - review & editing (equal)

Search for more papers by this author
Nicola Condelli

Nicola Condelli

Scuola di Scienze Agrarie, Università degli Studi della Basilicata, Alimentari, Forestali ed Ambientali, viale dell’Ateneo Lucano, Potenza, 10 – 85100 Italy

Contribution: Conceptualization (equal), Formal analysis (equal), Methodology (equal), Resources (equal), Writing - review & editing (equal)

Search for more papers by this author
Fernanda Galgano

Corresponding Author

Fernanda Galgano

Scuola di Scienze Agrarie, Università degli Studi della Basilicata, Alimentari, Forestali ed Ambientali, viale dell’Ateneo Lucano, Potenza, 10 – 85100 Italy

*Correspondent: E-mail: [email protected]

Contribution: Conceptualization (equal), Methodology (equal), Project administration (equal), Supervision (equal), Writing - review & editing (equal)

Search for more papers by this author
Marisa C. Caruso

Marisa C. Caruso

Scuola di Scienze Agrarie, Università degli Studi della Basilicata, Alimentari, Forestali ed Ambientali, viale dell’Ateneo Lucano, Potenza, 10 – 85100 Italy

Search for more papers by this author
First published: 09 June 2021
Citations: 7


Acceptability and the sensory profile of experimental and commercial gluten-free biscuits were evaluated using a consumer test. Experimental biscuits contained flours little employed in commercial gluten-free products (buckwheat, sorghum and lentil) and without rice and maize flours and starches. Liking scores, answers to a Check All That Apply questionnaire and responses to a dietary habits survey were collected from consumers. Consumers were clustered through Hierarchical Clustering of Principal Components and CLUSCATA. Interestingly, the cluster that, based on dietary habits survey, was more used to the consumption of wholemeal biscuits preferred experimental formulations. On the contrary, liking scores expressed by consumers that consumed less often wholemeal biscuits were below the threshold of acceptance. Different groups of consumers had a diverse perception of the products and different drivers of liking and disliking. The information achieved in this study confirms the importance of consumers’ data.


The gluten-free (GF) market is expected to grow at a compound growth annual rate of 9.2% during 2020–2027 (Grand View Research, 2020) due to the increasing number of people suffering from celiac disease and gluten-related disorders and those who decide to follow a GF diet without any medical need. Nevertheless, there are still some issues related to GF products, such as technological challenges, nutritional deficiencies and the need for improvement of sensory properties (Stantiall & Serventi, 2017; Di Cairano et al., 2018).

GF biscuits generally present impaired sensory properties compared with conventional biscuits even if the development of gluten network is not fundamental in biscuit making (Schober et al., 2003; Mazzeo et al., 2014; Drabińska et al., 2016); it follows that the production of GF products with improved nutritional quality, and besides, sensory acceptance is still a current challenge.

Sensory descriptive methods are rarely intended to predict the success of a product in the market (Mello et al., 2019), and consumer tests represent a valid alternative to get information about food preference directly from the buyers. Moreover, through consumer tests, it is also possible to characterise food products. An increasingly used method for consumer-based sensory characterisation is Check All That Apply (CATA); it represents a rapid approach to collect information about the sensory characteristics of products directly from consumers (Valentin et al., 2012; Ares & Jaeger, 2015). They are asked to check the attributes they consider appropriate to describe the product, in a multiple-choice questionnaire. CATA answers have been proven to be able to discriminate among products and to be reproducible (Valentin et al., 2012; Ares & Jaeger, 2015).

Over the years, gathering information on consumers is becoming increasingly common. It is in fact known that food preference is affected by different factors such as genetic, physiological and psycho-social; it follows that it is of interest for producers to understand whether consumers’ preferences are driven by sociodemographic or behavioural characteristics (Giacalone, 2017; Dinnella et al., 2020). This information represents a resource for targeted marketing and development of consumer-tailored products. In this context, dietary habits surveys represent a way to collect information that could be related to consumers’ preferences. Differences among consumers should be considered when analysing sensory data which otherwise could be misinterpreted, or rather specific trends, that may be drawn from segmented data, would go unnoticed. For example, consumer tests involve the use of untrained subjects; hence, their perception of the products can be variable and it is highly important to investigate the presence of segment of homogeneous subjects based on their perception of the products.

Analysing GF biscuits available on the Italian market, it emerged that rice and maize flours and starches are mostly employed (Di Cairano et al., 2018). Moreover, in the infrequent cases in which alternative flours are used, they only represent a small part of the recipe. Nevertheless, research involving the employ of flours still little exploited in commercial products exists (Molinari et al., 2018; Cannas et al., 2020; Giuberti et al., 2020; Silva et al., 2021). Indeed, in view of improving the nutritional profile of GF bakery, more nutrient-dense flours could be employed (Di Cairano et al., 2020). The same authors studied formulations of GF biscuits made with alternative flours, besides the replacement of other ingredients, such as sucrose (Di Cairano et al., 2021a). Buckwheat, sorghum, millet, lentil and chickpea flours were employed in laboratory trials (Di Cairano et al., 2021b), and then, two formulations were selected to be produced on an industrial scale, in a biscuits manufacturing plant. The aim of this paper was the comparison of preference scores and CATA analysis results relating to experimental and commercial GF biscuits. To the best of the authors’ knowledge, no biscuits without starches are available on the Italian market. Hence, the selection of commercial biscuits was based on the following criteria: having at least one flour in common, specifically buckwheat, with experimental formulations and a similar shape.

In particular, the paper focuses on the evaluation of data obtained by grouping consumers using two clustering approaches, CLUSCATA and Hierarchical Clustering of Principal Components (HCPC), alongside with the information gathered through an eating habits survey. CLUSCATA is a recent clustering approach that segments consumers according to their agreement on CATA answers (Llobell et al., 2019a), whereas HCPC clusters consumers according to the preference scores.

Materials and methods


Four types of GF biscuits were analysed (Fig. 1). They are two experimental samples (E1 and E2) and two commercial samples (C1 and C2). The experimental samples were produced in biscuits manufacturing plant thanks to the collaboration of ‘Di Leo Pietro spa’ (Matera, Italy). Commercial samples, on the other hand, were purchased in a local shop. They were chosen based on their ingredient list to share at least one flour with the experimental formulations. Furthermore, also their shape was considered. Table 1 reports the ingredient list of all the biscuits.

Details are in the caption following the image
Experimental (first column) and commercial (second column) gluten-free biscuits.
Table 1. Ingredients of experimental (E1&E2) and commercial (C1&C2) gluten-free biscuits.
Sample Ingredients
E1 Buckwheat flour (28.00%), sorghum flour (28.00%), sugar (19.00%) eggs (13.50%), sunflower oil (9.00%), ammonium bicarbonate (0.40%), sodium hydrogen carbonate (0.30%), salt (0.10%)
E2 Buckwheat flour (28.00%), sorghum flour (17.00%), lentil flour (11.00%) sugar (19.00%) eggs (13.50%), sunflower oil (9.00%), ammonium bicarbonate (0.40%), sodium hydrogen carbonate (0.30%), salt (0.10%)
C1 Potato starch, teff flour, cane sugar, sunflower oil, maize flour, diastased rice flour, glucose syrup, soy lectin, salt, leavening powder, natural flavour
C2 Buckwheat flour, sugar, palm oil, potato starch, rice flour, maize flour, butter, eggs, fibre, leavening powder, salt, flavour
  • For experimental samples, percentages reported in parenthesis are the amount employed in biscuit making, and the list reported for commercial formulations was taken from label of the products.

Consumer test – Liking and CATA questionnaire

Biscuits were subjected to sensory evaluation by 111 consumers aged between 20 and 65 years old, recruited among students and staff of the University of Basilicata (Potenza, Italy). All participants gave informed consent before sensory evaluation. The evaluation was performed in individual boots under red light. All biscuits were served in a white plastic dish and each sample was associated with a random three-digit code. All biscuits were served in the same sensory session, and consumers were asked to respect at least one-minute rest whilst rinsing their mouth with water until the neutrality of the mouth was restored. Consumers were asked to express their liking score on a general labelled magnitude scale (gLMS). The labels of the conventional gLMS (Bartoshuk et al., 2003) were changed to collect hedonic ratings: 0 – ‘the most unpleasant you can imagine’ and 100 – ‘the most pleasant you can imagine’.

Biscuits were also evaluated using CATA questionnaire. A focus group, composed of three researchers who investigated literature and tasted the biscuits, generated the sensory terms. The list of terms comprised 18 attributes covering texture (easy to chew, easy to swallow, hard to chew, hard to swallow, hard, crispy, adhesive, sandy, mealy), flavour/taste, after taste and mouthfeel sensations (sweet, salty, crumbly, biscuit flavour, off-flavour, roasted flavour, cereal flavour, legume flavour, persistent). Consumers were asked to select the attributes considered appropriate to describe the sample they were tasting. All samples were completely randomised. Sensory sessions were conducted by using FIZZ software (BioSystem, France).

Alternative bakery consumption survey

A dietary habits questionnaire including questions about:
  1. demographics (gender, age);
  2. consumption frequency of wholemeal baked goods and biscuits and baked goods and biscuits made with alternative flours to wheat, measured on a 5 point category scale [never (0), rarely (1), a few times a month (3), a few times a week (4), daily (5)];

was submitted to consumers. The questionnaire was created on Google Forms (Google Docs Suite, Google, United States) and was sent to consumers via e-mail or instant messaging.

Statistical analysis

Two clustering methods were employed to cluster subjects; HCPC was used to identify groups of consumers with similar acceptance patterns as reported by Alencar et al., (2019). Alongside HCPC, a new clustering approach called CLUSCATA (Llobell et al., 2019a) was adopted to identify groups of consumers according to their answers to CATA questionnaire. Data analysis was carried both on un-clustered and clustered data. The permutation test was also applied to check the consistency of the panel and the attributes (Llobell et al., 2019b). Preference data were subjected to one-way ANOVA followed by Tuckey’s post hoc test (P < 0.05).

Through CATA data analysis, it was first established if consumers had detected differences between samples based on proposed attributes with Cochran’s Q test followed by multiple pairwise comparisons by using a critical difference (Sheskin) procedure. Then, correspondence analysis (CA) was made to see how the whole panel or each cluster described the biscuits. Penalty lift analysis was made to evaluate the impact of each attribute on the liking score (P < 0.05). Answers to the dietary habits questionnaire were subjected to chi-square test (P < 0.05).

One-way ANOVA, HCPC and CATA data analysis were made by using XLSTAT Premium version (2020.4.1). R software v. 3.6.0 (R Core Team, 2020), and specifically the ClustBlock package (Llobell et al., 2020) was used to perform CLUSCATA analysis and permutation test. Excel (Microsoft Office Professional 2013, Microsoft) was used to make a chi-square test.

Results and discussion

Clusters of consumers and attribute consistency

Generally, preferences are driven by different factors as genetic, physiological and psycho-social (Dinnella et al., 2020), and a dietary habits survey was included to see whether consumers’ behaviours in bakery consumption affected sensory data. Since they are not trained, consumers can have a variable perception of the attributes (Llobell et al., 2019a). Considering these premises, alongside sensory data analysis on the whole group of consumers involved in the test; two clustering approaches, HCPC and CLUSCATA, were adopted to better understand the behaviour of the panel. HCPC segmented consumers into two clusters made by 71 and 40 consumers, respectively, whereas a cluster made by 53 and one by 58 consumers was obtained by CLUSCATA.

According to the permutation test, the whole panel resulted to be consistent, meaning that consumers showed agreement in their answers. The permutation test was also applied, in turn, to each attribute to assess whether the respondents were (in)consistent in their evaluation (Llobell et al., 2019b). Consistency was found only for easy to swallow, sweet, mealy, legume flavour, hard, roasted flavour, biscuit flavour and sandy. All other attributes resulted to be inconsistent. The permutation test was applied also to each cluster. Regarding HCPC clusters, cluster 1 resulted to be consistent, whereas cluster 2 was not consistent. Subjects grouped according to preferences showed no agreement on CATA attributes. For cluster 1, consistent attributes were easy to swallow, sweet, salty, mealy, hard, adhesive, sandy and persistent, whereas for cluster 2 only hard to swallow and persistent resulted to be consistent. Both clusters obtained with CLUSCATA resulted to be consistent; attributes resulted to be consistent too. This result was expected since consumers were grouped according to their CATA answers.

Clusters of consumers and dietary habits survey

Table 2 reports the answers to the dietary habits survey of the clusters obtained by HCPC. Two out of 111 consumers did not answer the survey; hence only 109 answers are reported. According to chi-square test, the only significant association between groups of consumers and their answers regarded the consumption of wholemeal biscuits in clusters obtained by HCPC. Answers grouped according to clusters obtained by CLUSCATA method are not shown since no significant associations were found.

Table 2. Answers to dietary habits survey of the two cluster of consumers identified with hierarchical clustering on principal components.
Question Cluster 1 (n = 69) Cluster 2 (n = 40) P-value*
n % n %
How often do you eat wholemeal bakery? Daily 16 23.19 6 15.00 0.453
Sometimes a week 25 36.23 12 30.00
Sometimes a month 15 21.74 10 25.00
Very rarely 13 18.84 12 30.00
Never 0 0 0 0.00
How often do you eat wholemeal biscuits? Daily 7 10.14 5 12.50 0.028
Sometimes a week 24 34.75 9 22.50
Sometimes a month 24 34.75 7 17.50
Very rarely 13 18.41 16 40.00
Never 1 1.15 3 7.50
How often do you eat bakery made with flours other than wheat? Daily 4 5.78 1 2.50 0.251
Sometimes a week 15 21.74 4 10.00
Sometimes a month 18 26.09 10 25.00
Very rarely 30 43.78 21 52.50
Never 2 2.90 4 10.00
How often do you eat biscuits made with flours other than wheat? Daily 7 10.45 1 2.78 0.645
Sometimes a week 10 14.92 6 16.67
Sometimes a month 29 43.28 12 33.33
Very rarely 20 29.85 16 44.44
Never 1 1.49 1 2.78
Do you eat or have you ever eaten gluten-free products? Yes 38 55.07 21 47.50 0.445
No 31 44.93 19 52.50
  • * Chi-square test (P < 0.05), P-values in bold are significant.

Preference patterns

Different preference patterns were found for clustered and un-clustered data and are reported in Fig. 2. Mean liking scores expressed by un-clustered consumers ranged between 54.67 and 65.53. Sample C1 and E2 had similar liking scores; sample C2 was the most preferred. C2 was the only sample containing butter, unlike others made with sunflower oil. This may have had a positive effect on biscuit preference. It has been seen that buttery flavour is a characteristic welcomed by consumers (Tarancón et al., 2015). Regarding preference scores of HCPC clusters, cluster 1 attributed higher scores to experimental samples and sample E2 (buckwheat, sorghum and lentil flours) was the most liked, whereas other formulations were not dissimilar between each other. On the contrary, cluster 2 showed a strong preference for commercial samples. Cluster 2 deemed experimental formulations below the threshold of acceptance. Interestingly, the dietary habit survey gave some clues about this trend. Indeed, it resulted that cluster 1, which preferred experimental samples, was more used to the consumption of wholemeal baked goods and wholemeal biscuits, besides bakery products made with alternative flours to wheat. Nevertheless, only the frequency of consumption of wholemeal biscuits was significantly different (P < 0.05) between the two clusters. However, the trend was pretty clear and gave important suggestions for eventual targeted marketing of the experimental biscuits. Even if no wholemeal flours were included in the recipes, the flours employed in experimental formulations had sensory properties closer to wholemeal flours compared with the milder-taste flours used in commercial formulations. Wholemeal flours and so as buckwheat, sorghum and lentil are richer in phenolic compounds, which can exert an effect on product taste (Huang & Hoseney, 1999). In addition, the two clusters obtained through CLUSACATA were characterised by two different preference patterns. Cluster 1 expressed similar liking scores for all samples, whereas cluster 2 showed a slight preference for commercial biscuits. E1 (buckwheat and sorghum) was the least preferred, but the linking score was still above the threshold of acceptance. In this case, differences between the two clusters based on their dietary habits were not evident and no significant associations were found between the two groups and consumers’ eating behaviour.

Details are in the caption following the image
Liking score of experimental (E1&E2) and commercial (C1&C2) gluten-free biscuits. (a) Un-clustered data; (b) clustered data according to HCPC, (c) clustered data according to CLUSCATA. Data are expressed as mean ± dev. st. Different letters indicate different means (P < 0.05).

CATA data analysis

Cochran’s Q test allowed seeing how consumers discriminated biscuits by attributes. Almost all groups of consumers differentiated commercial and experimental biscuits on the basis of the attribute sweet. Even if almost the same amount of sugar was present in the recipes (on the basis of the nutritional label calculated for commercial biscuits, and nutritional label calculated for experimental biscuits, data are not shown), the attribute sweet was associated with a higher frequency to commercial biscuits than to experimental ones. Flours employed in the experimental biscuits could have masked the sensation of sweetness given by sucrose, whereas ingredients with a milder taste such as rice flour and starches mainly constituting commercial samples did not exert this effect. Experimental biscuits were perceived harder than commercials, whereas, regarding the most of other attributes related to the texture of the biscuits, they did not discriminate products. Sandy and mealy, which are generally associated with GF biscuits, were perceived differently in all the samples, without a specific trend. The non-discrimination of the products could be explained by the disagreements among consumers. It has been reported that it is very probable that inconsistent attributes do not discriminate products (Llobell et al., 2019b). In short, consumers probably had a different interpretation of the meaning of the attributes and they were not able to discriminate homogenously samples according to the attributes.

The results obtained after permutation tests performed on the different groups of data (Section 7) confirmed what previously reported (Llobell et al., 2019b), that is the consistency test can be complemental to Cochran’s and draws the same conclusions. Indeed, attributes resulting consistent were discriminant attributes for biscuit characterisation according to Cochran’s Q test. For example, considering un-clustered data, exactly the same attributes resulted to be consistent after permutation test and significant according to Cochran: easy to swallow, sweet, mealy, legume flavour, hard, roasted flavour, biscuit flavour and sandy. On the contrary, inconsistent attributes were not able to discriminate products. The same trend was found for all other groups of consumers. The only exception was represented by the attributes selected by cluster 2 (HCPC); no one of the consistent attributes identified by the permutation test was significant according to Cochran’s Q test. However, in this case, the whole panel was not consistent. CATA data were used to carry a CA, whose outcome is reported in Fig. 3. The perceptual maps do not include the same attributes, because some of them were not selected a relevant number of times. CATATIS method was applied to un-clustered data; despite classical CA, it reduces the weight of data from consumers whose agreement deviates from the general point of view of the panel to reduce their impact on the result of the analysis (Llobell et al., 2019a). Samples were relatively different. Perceptual maps obtained through CA, even if different among clusters, showed some common results. In principle, it can be said that commercial samples were more often associated with sweet and biscuit flavour as positive attributes and sandy and mealy as negative attributes. Experimental samples, instead, were more frequently associated with peculiar flavours such as roasted, legume and cereal. Hence, specific attributes resulted to be commonly perceived by different groups of consumers regardless of the clustering method employed. Among attributes, it was possible to identify the drivers of liking and disliking. Penalty lift analysis allowed detecting attributes with a significant impact on liking scores. This information may be relevant in developing new products since it allows to understand attributes exerting pleasant or unpleasant sensations to consumers. Through mean drop graphs (Fig. 4), it has been seen that the effect of attributes varied based on the cluster considered. Many factors play a role in sensory analysis. This test allowed us to see the effect that each perceived attribute plays on the consumer. Unlike descriptive tests, a simulation closer to reality is achieved combining CATA answers and liking, seeing how much each perceived attribute may affect the liking. Sweet was the only attribute with a significant impact on liking scores in all groups of consumers. On average, it contributed to a 10 points increase in liking score. Biscuits are expected to be sweet, and it was not unexpected that all clusters were homogeneous in considering it a positive attribute; indeed, sweet as a driver of liking is also reported in other studies (Mello et al., 2019). Attributes hard and legume flavour also appeared in all mean drop graphs, but their effect was not significant. Despite that, it is interesting to observe that in some clusters they potentially could have a big impact on liking, and in others could contribute minimally. This could depend on the individual preferences of clustered subjects. For example, the presence of legume flavour has a potential ability to reduce the liking score of almost 15 points in cluster 2 obtained with HCPC method, whereas it contributed to a 2 point reduction for cluster 1. Cluster 2 was the cluster containing consumers less used to the consumption of wholemeal bakery and bakery made with alternative flours; this is the reason explaining the big mean drop value. Texture generally contributes to the overall acceptability of biscuits, and it was seen that some textural attributes such as crispy, easy to swallow and easy to chew had a positive effect on liking score, whereas hard, hard to chew, mealy and sandy were negative drivers of acceptance.

Details are in the caption following the image
Perceptual maps of experimental (E1&E2) and commercial (C1&C2) GF biscuits. (a) un-clustered data, (b) and (c) data referring to cluster 1 and 2 obtained by hierarchical clustering of principal components; (c) and (d) data referring to cluster 1 and 2 obtained by CLUSCATA.
Details are in the caption following the image
Mean drop graph of (a) un-clustered data, (b) and (c) data referring to cluster 1 and 2 obtained by hierarchical clustering of principal components; (c) and (d) data referring to cluster 1 and 2 obtained by CLUSCATA. Attributes in bold had a significant impact on linking score according to penalty lift analysis (P < 0.05).


This work showed that biscuits with underexploited GF flours, potentially able to enhance their nutritional profile, were accepted by consumers. In particular, it was observed that specific segments of consumers had different preference patterns, highlighting that the use of consumer data could be pivotal for the production of consumer-tailored food products and the development of targeted marketing actions. Moreover, it has also been seen that the perception of attributes and their impact on liking varies according to the cluster.

The presence of flours less commonly employed in commercial biscuits might turn out a double-edged sword. In effect, celiac patients are used to the mild taste of commercial formulation; stronger tastes as in experimental formulations could be unpleasant for them. In the future, it would be interesting to propose to celiac consumers these experimental biscuits, in order to verify their preferences and to find out whether they share the drivers of liking and disliking with healthy consumers.


The authors are grateful to Di Leo Pietro spa (Matera, Italy) for biscuit production and to all consumers participating in the test. Icons employed in graphical abstract are made by Freepik from www.flaticon.com.Open Access Funding provided by Universita degli Studi della Basilicata. [Correction added on 24 May 2022, after first online publication CRUI funding statement has been added.]

    Author contribution

    Maria Di Cairano: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Writing-original draft (equal); Writing-review & editing (equal). Nicola Condelli: Conceptualization (equal); Formal analysis (equal); Methodology (equal); Resources (equal); Writing-review & editing (equal). Fernanda Galgano: Conceptualization (equal); Methodology (equal); Project administration (equal); Supervision (equal); Writing-review & editing (equal). Marisa C Caruso: Conceptualization (equal); Data curation (equal); Project administration (equal); Supervision (equal); Writing-review & editing (equal).

    Declaration of interest


    Ethical guidelines

    Ethics approval was not required for this research.


    This research does not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors.

    Peer review

    The peer review history for this article is available at https://publons.com/publon/10.1111/ijfs.15188.

    Data availability statement

    The data that support the findings of this study are available on request from the corresponding author.