Development and Validation of the Robot Acceptance Questionnaire (RAQ)

Abstract

This study aimed to validate the Robot Acceptance Questionnaire (RAQ), a self-report instrument designed to assess user acceptance toward social robots. Originally structured around four theoretical domains—pragmatic, hedonic (identity and feelings), and attractiveness—the RAQ was empirically found to converge into two robust and inversely related dimensions: Positive Attitude (PA) and Negative Attitude (NA). A total of 208 participants (mean = 43.1; S.D. = 21.4) viewed a short video of a humanoid robot (Pepper) and completed the RAQ. Factorial structure (Principal Component Analysis), internal reliability (Cronbach’s alpha), and construct validity were assessed. Results showed excellent internal consistency for both PA and NA (α = 0.93), and intuitive associations with independent measures of ease of use, mastery, and willingness to interact. The RAQ thus offers a concise and reliable tool for assessing general robot acceptance, especially suitable for remote and large-scale studies.

1. Introduction

Social Assistive Robots (SARs) represent a class of technological devices designed to assist human users through socially oriented interactions [1,2]. To ensure effective human–robot interaction, these systems must demonstrate competencies such as natural communication, emotional responsiveness, and social adaptability [3,4]. Given the inherently social nature of these interactions, understanding users’ perspectives—including their needs, preferences, and demographic characteristics—is essential from the earliest stages of robot design. This calls for a user-centered approach throughout the development cycle [5]. In this context, assessing user acceptance becomes a key challenge. While existing instruments often target specific facets of acceptance, such as anxiety or usability, the Robot Acceptance Questionnaire (RAQ) was developed to offer a broader, integrative measure. Initially structured around four theoretical domains, the RAQ was empirically found to converge into two core dimensions—positive and negative attitudes—providing a concise yet reliable snapshot of user acceptance.

From this perspective, a concept as important as it is complex is that of robot acceptance. Robots’ acceptance can be affected by three main factors: robots’ functionality, robots’ social ability, and robots’ appearance [6]. Considering robots’ functionality, the potential role of service robots in domestic environments should be considered, particularly for individuals facing physical limitations or older adults, with a focus on tasks related to daily living and healthcare. A key consideration concerns the level of autonomy in robots, identified as a crucial aspect of human–robot interaction. The autonomy spectrum, ranging from full human control (teleoperation) to full robot control (fully autonomous agent), is discussed according to [7], underscoring the importance of aligning autonomy levels with user expectations for increased acceptance. Another factor influencing robot acceptance is how humans control and interface with them. Various roles that humans can assume in this context, such as supervisor, operator, teammate, mechanic/programmer, and bystander, are outlined based on [8]. It is fundamental to consider the appropriateness of control methods for specific tasks, ease of use, and user preferences when designing robots in order to enhance their acceptance. Concerning the social abilities of robots, social intelligence is a crucial element in the development of intelligent robotic agents as it profoundly influences interactions with humans. Equally vital are emotion expressive capabilities, particularly facial expressions, which are deemed essential for advanced intelligent technology [9], given the significant role of social cues, including emotions, in creating robots that can respond sensitively to human reactions [10]. The robot’s capacity to recognize and react to human emotions is viewed as indispensable for establishing effective social interaction and gaining acceptance. Furthermore, non-verbal social cues like nodding and eye movements have a notable impact on human–robot interaction, and robots that utilize natural dialog and gestures contribute to establishing a more cooperative relationship, as highlighted by research findings [11]. Lastly, robots’ acceptance is deeply affected by robots’ appearance. In order to ensure a positive interaction between humans and robots, it has been suggested that the design of the robot should be such that people can easily understand its behavior [12]. Furthermore, an appropriate match between the appearance of a robot and its assigned task has been indicated as a factor that can improve people’s acceptance of the robot [13].

To investigate the impact that the aspects related to the robots may have on the users’ acceptance of the system, many technology acceptance models have been proposed. These models consider several factors such as the robot’s perceived effectiveness, its ease of use, its pleasantness, or its ability to emotionally engage the user and evoke positive feelings during the interaction [14]. Among them, there is the technology acceptance model (TAM) [15]; the Innovation Diffusion Theory [16]; and the Unified Theory of Acceptance and Use of Technology Model (UTAUT, UTAUT2) [17,18]. In addition to these models, various psychometrically validated scales have been implemented to assess both the robots’ acceptability and usability. A recent literature review [19] identified six questionnaires that attempt to measure robot acceptability and report their psychometric attributes: the Negative Attitudes toward Robots Scale (NARS) [20]; the Frankenstein Syndrome Questionnaire (FSQ) [21]; Multidimensional Robot Attitude Scale [22]; the Technology-Specific Expectations Scale (TSES) [23]; the Ethical Acceptability Scale [24]; and the Robotic Social Attribute Scale (RoSaS) [25]. Although these scales provide the possibility to measure different aspects related to acceptability of social robots, each of them focuses on specific aspects of the interaction (or potential one) by neglecting a more general evaluation of the system that simultaneously involves several aspects of the users’ acceptance. For example, the NARS and the FSQs assess the negative attitudes towards robots and users’ anxieties and apprehensions towards the use of this technology in public contexts. The Ethical Acceptability Scale concerns the ethical acceptability of social robots and was originally developed to specifically assess ethical issues in the use of this technology in therapy contexts with children with autism. Instead, the RoSaS questionnaire focuses its assessment on whether several adjectives can be associated or not with the proposed systems rather than providing statements assessing the degree of agreement. Moreover, this tool does not consider the willingness to interact with the robot. The TSES questionnaire was developed to assess the introduction of an empathic robot tutor in the educational context for adolescents and it served as a baseline measure to be integrate with another tool developed by the same authors, the Technology-Specific Satisfaction Scale (TSSS). The Multidimensional Robot Attitude Scale report 12 dimensions which capture a wider range of attitudinal aspects compared to the other tools. However, the validated tool is available only in the original languages (Japanese, Taiwanese, and Chinese) and it mainly focuses on domestic robots. A recent tool has been developed by Koverola and colleagues (2022) [26], the General Attitudes Towards Robots Scale (GAToRS), that measures fear, anxieties, hope, and expectations towards social technologies without, however, referring to any specific system. Instead, other questionnaires, such as the UTAUT [27,28] and the System Usability Scale [29,30], are formulated in a way that the assessment requires the actual physical encounter with the robot.

According to the brief review of the existing validated instruments, it emerged that in the user experience design and research fields there is a lack of tools capable of assessing simultaneously several information about the feelings the system may arouse, its engagement capability, the esthetic pleasantness, communication skills, the naturalness of a potential interaction, ease of use, and other aspects also related to user features. In other words, a tool capable of providing information about the needs and preferences of the end-users which have become necessary for developing an adequate autonomous system in order to be adopted in a variety of practical contexts, such as healthcare, education, and domestic environments [31]. Moreover, to our knowledge, there are no Italian psychometrically validated instruments built for these purposes.

To address the lack of integrative tools in the field, the present study introduces the Robot Acceptance Questionnaire (RAQ), originally designed to assess four dimensions of user acceptance—pragmatic utility, hedonic identity, hedonic feelings, and attractiveness. However, empirical validation revealed a two-factor structure, capturing users’ positive and negative attitudes toward the robot. This outcome supports the RAQ’s intended purpose: to offer a concise, psychometrically sound instrument for assessing general robot acceptance, particularly in remote and large-scale research settings.

The questionnaire has been developed by Esposito and colleagues [32] at Università degli Studi della Campania, “Luigi Vanvitelli”, Department of Psychology, in the context of the EMPATHIC Research and Innovation project (http://www.empathic-project.eu/ (accessed on 1 November 2017)) which aimed to develop a Personalized Virtual Coach to provide assistance to seniors in their daily life.

The proposed tool was inspired by the AttrakDiff Questionnaire developed by Hassenzahl [33], which allows the evaluation of a generally chosen technological product in terms of satisfaction of users’ psychological needs.

The theoretical model underlying the AttrakDiff questionnaire focuses on the key elements of the user experience by considering both the designer and user perspective. More specifically, four essential aspects are identified, namely the quality of the product planned by the designer, the perception and subjective evaluation of quality, the independent pragmatic and hedonic qualities, and the resulting behavioral and emotional consequences. To assess these aspects, the AttrakDiff questionnaire presents 28 seven-step items whose poles are represented by opposite adjectives (e.g., “unruly-manageable”, “technical-human”, “rejective—inviting”). Each set of adjective items refers to a specific domain among pragmatic quality (PQ), hedonic quality (HQ—including stimulation and identification, respectively, HQ-S and HQ-I), and attractiveness (ATT), and it is ranked on an intensity scale. Therefore, the scale value for each domain is determined by the mean values of each group of items.

The domains identified by Hassenzahl in his theoretical model [33] provided the basis for the creation of 40 items that were purposefully developed for the RAQ to contribute to the field of user experience. Besides the development of original items devoted to assess the pragmatic, hedonic, and attractive attributes, in the proposed questionnaire, the domain of “HQ-Stimulation” has been replaced with the “HQ-Feelings”, with the aim of focusing on the users’ emotional responses to the robot, intended as the capability to elicit an affective response, rather than its stimulating features. Moreover, the label HQ-Identification has been changed in “HQ-Identity” since the Italian term (i.e., identità) more accurately describes the robot’s ability to provide its own communicative identity to the user.

Detailed information about the questionnaire’s sections is described in Section 2.3. The next section provides a review of the studies in which the RAQ was adopted before being validated, with the related results.

As previously described, the RAQ was developed within the Empathic Project which aimed to create an empathic virtual coach capable of improving seniors’ quality of life. Due to this, the first study [32] involving this tool was conducted to investigate seniors (>65 years) preferences towards five different robots which were presented through video clips: Roomba, Nao, Pepper, Geminoid HI-1, and Erica. The outcomes revealed a clear uncanny valley effect [34]. Indeed, participants strongly preferred the idea of being assisted by Pepper over the other robots. Seniors particularly appreciated Pepper for its effectiveness, originality, and ability to provide emotional support. Moreover, Pepper was judged the best to provide care to seniors, children, and people with disabilities, and for protection and security and front office duties. To further explore the potential dynamics underlying the uncanny effect that emerged in the previous study, Esposito and colleagues [35] planned a subsequent investigation in which participants were asked to evaluate female robots with different levels of human likeness (the study scheduled the administration of two androids, Erica and Sophia, and a humanoid robot, Pepper, gendered as female). Results showed that the RAQ was able to detect differences in participants’ preferences due to the human resemblance degree of the robots. In contrast to the previous study, seniors expressed a clear preference for female androids over the humanoid robot. Moreover, results revealed that among the tasks that participants would have entrusted to female robots, housework excelled over protection/security, welfare, and front office occupations.

Subsequently, the RAQ was adopted in another study [36] to investigate variables such as gender, ethnicity, and type of system to understand if, and possibly how, they influenced seniors’ preferences. To this aim, senior participants were asked to view and later to evaluate a set of stimuli including video clips of three robots: two androids and a humanoid. Each stimuli set included robots with two different ethnicities (Caucasian and Asian), gendered as female or male. The findings showed that older male participants were more inclined to be involved in a long-lasting interaction with the proposed robots and evaluated them more positively than female participants. Therefore, a clear participants’ gender difference in the robots’ evaluation emerged. Furthermore, it was also observed that regardless of their gender, androids were vastly preferred over humanoid robots. Lastly, regarding ethnicity, seniors showed a greater appreciation towards female android robots with Asian traits and male androids with Caucasian traits.

The research assessing users’ acceptance of social robots through the RAQ was further broadened by considering the role of participants’ age in the evaluation process. This was justified by the researchers’ interest in caregivers’ preferences and opinions in addition to end-users’ expectations (seniors). To this end, six differently aged groups of participants (seniors, adolescents, and young adults) were required to assess two different sets of male (Yuri, Romeo, and Albert) and female (Erica, Sophia, and Pepper) android and humanoid robots. Results revealed that the proposed robots’ acceptance turned out to be a non-linear combination of the following factors: participants’ age, gender, and type of robot (android vs. humanoid). For further information concerning the study’s results, see [37].

The RAQ was developed in conjunction with another questionnaire, named Virtual Agent Acceptance Questionnaire (VAAQ), which aimed to assess the participants’ preferences towards virtual agents rather than robots. The two questionnaires have been used to investigate whether users and potential caregivers preferred to be assisted more by virtual agents or by robots [38,39]. In the study of Esposito and colleagues [38], a middle-aged group and a group of seniors were asked to express their opinions towards two female virtual agents (Giulia and Clara) while an additional two groups of middle-aged and senior participants were asked to evaluate two female android robots (Sophia and Erica). This study showed that seniors were more involved during potential interaction with virtual agents while middle-aged participants expressed a clear preference for robots. Participants’ gender also exerted an effect: compared to female users, their male counterparts expressed higher preferences towards female robots. The study of Greco and colleagues [39] is characterized by the same experimental design of the previously described one, with two exceptions: the sample was composed of seniors, young adults, and middle-aged people and the tele-presented robots were gendered as male. The outcomes underlined that, compared to virtual agents, robots were better evaluated by senior participants, whereas middle-aged ones slightly preferred virtual agents. Young adults seem to be more comfortable and confident with the idea of receiving support by both types of systems, suggesting a major proclivity. In addition, results revealed that male participants expressed a greater aptitude and propensity towards both types of administered technological devices than female ones. These studies, although involving different systems, were instrumental in refining the RAQ’s structure and item pool. Their inclusion serves to illustrate the questionnaire’s broader applicability and the empirical foundation upon which its final version was built.

Although the literature review showed that the RAQ could be an informative tool about several aspects of the acceptance of SARs, it has never been submitted to a validation process to test its psychometric qualities. However, the above-mentioned findings suggest that this questionnaire is able to detect differences due to important organismic variables concerning the user, such as age and gender. The RAQ allowed researchers to gather data about the role played by many features concerning the system and how they interact with those related to the user, providing information that can be useful in the designing process of the system. Due to its potential, it would be convenient to standardize the RAQ in order to provide a new tool capable of assessing several aspects of the user acceptance process, which may guide the development and design of SARs.

The above-mentioned studies [32,35,36,37,38,39] had two main objectives: on one hand, examining the preferences of differently aged potential users towards SARs by assessing many dimensions of the system; on the other hand, they also contributed to the development and revisions of the final version of the RAQ, which is presented in this paper to be validated. The psychometric properties of the RAQ have been assessed exploiting original data, specifically collected for research purposes and the assessment of the humanoid robot Pepper.

Conversely from the previous studies, in the current research, the aim is not to compare the assessment of different systems provided by users of varying ages. The participants enrolled in this work presenting the validation process of the RAQ intentionally belonged to different life stages (i.e., old age, adulthood, young adulthood, adolescence) with the aim to test the questionnaire’ psychometric properties on a sample which could be representative of different generations of potential end-users.

2. Materials and Methods

To achieve the above-mentioned purposes, an experiment assessing users’ degree of acceptance towards the humanoid robot Pepper was carried out (see Figure 1). Acceptance was evaluated by considering the participants’ willingness to interact with the robot and the scores attributed to different qualities (i.e., pragmatic, hedonic, and attractive), which are described in the detail in Section 2.3.

2.1. Participants

The total sample was composed by 208 volunteer participants (85 males; mean age: 43.1, SD = ± 21.4). In the wake of the previous studies adopting the proposed tool [32,35,36,37,38,39] and starting from the hypothesis that technology assessment may be affected by age [40], the recruitment process involved participants of varying ages (from 14 to 90 years old) that could represent distinct life stages (this is the reason why the standard deviation referring to participants’ mean age is relatively high).

They were required to watch a video clip depicting the humanoid robot Pepper. To be enrolled in the study, participants had to report no vision and/or hearing problems. The recruitment occurred in the Campania region (south of Italy). Before starting the experiment, participants signed an informed consent formulated according to the current Italian and European laws about privacy and data protection (D. Lgs. 196/2003). The research was approved, with the protocol number 25/2017, by the ethical committee of the Università degli Studi della Campania “Luigi Vanvitelli”, Department of Psychology.

2.2. Stimuli

Pepper is a humanoid robot developed by SoftBank Robotics. A video clip depicting Pepper available on the website “YouTube” was selected; the video clip displayed the selected robot’s upper half in including the torso, in a forward position. The video clip lasted between 4 and 7 s and the robot was provided with a female synthetic voice, created through the Natural Reader synthesizer (www.naturalreaders.com (accessed on 14 February 2003)). The voice was recorded through the Audacity free audio software (www.audacityteam.org (accessed on 12 October 2003)) and implemented in the robot’s video clip by using the “Videomomenti” software, which is available on the Windows10 operating system. The voice recording consisted in the Italian sentence “Ciao sono Tina. Se vuoi posso aiutarti nelle tue attività quotidiane” (Hi, my name is Tina. If you want, I would like to assist in your daily activities). Pepper was renamed Tina in order to be contextualized into Italian culture.

2.3. RAQ

An ad hoc questionnaire was developed and structured into sections, taking inspiration from the domains investigated in the AttrakDiff questionnaire described in the introduction paragraph. The RAQ was used to assess the participants’ preferences toward the proposed robot (Pepper).

Before administering the RAQ, 7 preliminary questions devoted to collecting participants’ sociodemographic information (respectively, age, gender, years of schooling, and job) and evaluating the participants’ degree of experience and ease of use of three different technological devices (smartphones, tablets, and laptops) were purposely created. The experience degree was measured on a scale from 1 to 4, where 1 = I have no experience with technology and 4 = I frequently use a technological device. The ease of use was measured on a 5-point Likert scale, where 1 = Very Hard and 5 = Very Easy. These seven preliminary questions were followed by another single one assessing the probability participants were willing to be involved in a potential long-lasting interaction with the robot, assessed on a 5-point Likert Scale (1 = impossible; 5 = very likely).

Therefore, the RAQ was originally conceived as a four-dimensional questionnaire. In its developmental and piloting phase, the questionnaire presented 4 sections with ten items each, assessing, respectively, the pragmatic, hedonic (feelings and identity), and attractive qualities an interactive system should be endowed with in order to be accepted by their potential users.

Specifically, the investigated qualities were as follows:

• Pragmatic Qualities (PQ): these refer to the usefulness, practicality, and ease of use of the proposed robot.
• Hedonic Qualities—Identity (HQI): these are associated with originality, creativity, and esthetic pleasantness attributed by users to the proposed robot.
• Hedonic Qualities—Feeling (HQF): these assess to what extent the proposed robot is capable of arousing either positive or negative emotions.
• Attractiveness (ATT): this is devoted to evaluating whether the proposed robot can engage its users in an increasing usage.

These sections were assessed through a 5-point Likert scale, where 1 = strongly disagree, 2 = disagree, 3 = I do not know, 4 = agree, 5 = strongly agree, and involve both positive and negative statements.

In addition to sociodemographic information, three specific measures were derived from the preliminary section of the RAQ:

Frequency of Use (FoU): Participants were asked to indicate how frequently they use three technological devices—smartphone, tablet, and laptop—by selecting one of the following options for each device: “I never use it” (coded as 1), “I use it often but not every day” (coded as 2), and “I use it every day” (coded as 3). The FoU score was calculated as the sum of the responses across the three devices, yielding a total score ranging from 3 to 9.

Ease of Use (EoU): Participants rated the perceived difficulty of using each of the three devices on a 5-point Likert scale, where 1 = Very Easy and 5 = Very Difficult. For analytical purposes, scores were reversed so that higher values indicated greater ease of use. The EoU score was computed as the sum of the reversed ratings across the three devices, resulting in a score ranging from 3 to 15.

Mastery of Use (MoU): Participants were asked to self-assess their general experience with technology by selecting one of four statements: “I frequently use a technological device” (coded as 4), “I use it sometimes” (coded as 3), “I have tried it a few times” (coded as 2), and “No experience” (coded as 1). This single-item score was used to represent the participant’s perceived mastery of technology.

These measures were included in the construct validity analyses to examine their associations with the RAQ dimensions—Positive Attitude (PA) and Negative Attitude (NA)—toward the robot. As reported in

Table 1. Spearman’s correlation coefficients between RAQ subscales and construct validity measures.

	FoU	EoU	MoU	WtI
RAQ-PA	rs = 0.07	rs = −0.23 *	rs = −0.19 *	rs = 0.51 *
RAQ-NA	rs = −0.02	rs = −0.21 *	rs = −0.18	rs = −0.33 *

Notes. * = significant at α_adjusted = 0.006. RAQ = Robot Acceptance Questionnaire; PA = Positive Attitude; NA = Negative Attitude; FoU = Frequence of Use; EoU = Ease of Use; MoU = Master of Use; WtL = Willingness to interact with the robot.

, PA scores were positively correlated with both EoU and MoU, while NA scores were negatively correlated with these measures, supporting the expected relationships between technological familiarity and robot acceptance.

2.4. Procedure

The experiment was implemented with “Lab.js” software and exported on a platform named “JATOS” (Just Another Tool for Online Studies) which was used to generate the links through which participants carried out the experiment. Participants were required to perform the experiment through a laptop. Before starting, they were briefed on the aims of the study, and required to sign an informed consent. After that, they were asked to answer the preliminary questions. Once this first task was completed, participants watched and listened to the video clip depicting Pepper, and after the presentation, they were asked to respond to the single item aimed at assessing how willing they were to interact with the proposed robot and then to fill in the sections of questionnaire (for a total of 40 items) concerning the evaluation of the robot.

3. Statistics

The factorial structure of the RAQ was exploratively assessed via a Principal Component Analysis (PCA), whilst its internal consistency was assessed via Cronbach’s α. Pursuantly to current guidelines, sample sizes of N = 100 and of N = 20 were deemed as sufficient in order to perform a PCA [41] and to compute Cronbach’s α coefficients [42], respectively.

PCA assumptions were tested via Bartlett’s sphericity test and the Kaiser–Meyer–Olkin (KMO) statistic (this last measure being judged as acceptable if >0.60). The number of components to be extracted was defined according to Horn’s parallel analysis. Oblique rotations were deemed as justified if at least one between-component correlation coefficient was ≥|0.20|. A cut-off of ≥|0.35| was addressed as indexing substantial loadings. Item removal was pursued in the occurrence of either not substantial or equivocal loadings. An interstitial item was dropped if (1) there was substantial loading on more than one component and provided that (2) the ratio between its primary and secondary loading was <|2|.

Construct validity of the RAQ was assessed by means of Bonferroni-corrected correlational analyses against those items assessing participants’ perceived frequency, ease and mastery of use of digital devices (i.e., FoU, EoU, and MoU, respectively), as well as willingness to interact with the robot (WtI). The methodological choice to use these preliminary questions as a measuring method of construct validity derived from the absence in the literature of further, gold-standard standardized measures of the target construct to be used as comparison for the proposed instrument.

Due to FoU, EoU, MoU, and WtI measures not distributing normally (i.e., skewness and kurtosis values > |1| and |3|, respectively) [43], Spearman’s technique was employed to such an aim. The minimum sample size required for construct validity testing was set at N = 80 according to current guidelines [42]. The effect of demographic confounders (i.e., age, education, and sex) was simultaneously tested on the RAQ through regressions analyses; within such models, Bonferroni’s correction was applied for selecting significant predictors (i.e., α_adjusted = 0.05/k, with k being equal to the number of independent variables). Finally, the empirical percentiles for the two obtained measures were calculated.

4. Results

Participants’ background and psychometric measures are shown in

Table 2. Participants’ demographic descriptive variables.

Descriptive Variables (Mean, S.D., Min–Max.)
N	208
Sex (male/females)	Male: 85–Female: 123
Age (ys)	43.1 ± 21.4 (14–91)
Education (ys)	11.6 ± 4.5 (3–19)
MoU	3.4 ± 1.1 (1–4)
FoU	5.9 ± 1.7 (3–9)
EoU	11.3 ± 3.3 (3–15)
WtL	3.7 ± 1.2 (1–5)
RAQ NA	38.8 ± 10.1 (19–66)
RAQ PA	56.2 ± 11.4 (20–85)
Questionnaire original items (mean, S.D. Min–Max)
PQ1	3.1 ± 1.1 (1–5)	HQF1	3.2 ± 1 (1–5)
PQ2	2.7 ± 1 (1–5)	HQF2	2.8 ± 1.1 (1–5)
PQ3	3.5 ± 0.9 (1–5)	HQF3	3.7 ± 0.8 (1–5)
PQ4	3.1 ± 1.1 (1–5)	HQF4	2.8 ± 0.9 (1–5)
PQ5	3.6 ± 0.9 (1–5)	HQF5	3.1 ± 1 (1–5)
PQ6	3.6 ± 1.1 (1–5)	HQF6	2.8 ± 0.9 (1–5)
PQ7	3.5 ± 0.9 (1–5)	HQF8	2.6 ± 1 (1–5)
PQ8	2.7 ± 1 (1–5)	HQF7	3.4 ± 0.9 (1–5)
PQ9	3.1 ± 0.9 (1–5)	HQF9	3.1 ± 1 (1–5)
PQ10	3 ± 1 (1–5)	HQF10 HQF10	2.8 ± 1 (1–5)
HQI1	3.5 ± 1.1 (1–5)	ATT1	3.4 ± 1.1 (1–5)
HQI2	2.8 ± 1.1 (1–5)	ATT2	3.2 ± 1 (1–5)
HQI3	3.4 ± 1 (1–5)	ATT3	3.3 ± 1 (1–5)
HQI4	2.6 ± 0.9 (1–5)	ATT4	2.8 ± 1 (1–5)
HQI5	2 ± 1 (1–5)	ATT5	3.2 ± 1 (1–5)
HQI6	2.3 ± 0.9 (1–5)	ATT6	2.7 ± 1 (1–5)
HQI7	3.1 ± 0.9 (1–5)	ATT7	3.6 ± 1 (1–5)
HQI8	3.9 ± 0.9 (1–5)	ATT8	2.9 ± 1.1 (1–5)
HQI9	3.3 ± 0.9 (1–5)	ATT9	3.2 ± 1 (1–5)
HQI10	2.8 ± 0.9 (1–5)	ATT10	2.7 ± 1.1 (1–5)

Notes. RAQ = Robot Acceptance Questionnaire; PA = Positive Attitude; NA = Negative Attitude; FoU = Frequence of Use; EoU = Ease of Use; MoU = Master of Use; WtL = Willingness to interact with the robot; PQ = Pragmatic Quality; HQI = Hedonic Quality—Identity; HQF = Hedonic Quality—Feeling; ATT = Attractiveness. Continuous data are shown as M ± SD (range), whereas categorical ones are shown as percentages.

. The PCA initially revealed, after successfully applying an OBLIMIN rotation (i.e., Component 1 correlating with Component 2), a three-component, complex structure (i.e., >10% of interstitial and/or non-substantially loading items). Thus, six items were dropped in the following order (prioritizing the removal of non-substantially loading items): HQF2, PQ3, PQ7, HQI5, HQF8, and HQF4. After such a removal, a three-component, OBLIMIN-rotated simple structure yielded, with only three items loading on Component 3—i.e., ATT2, PQ6 and HQI8; the remaining items, which tapped on either positive or negative attitudes towards the robot use, unequivocally loaded on either Component 1 or Component 2, respectively. ATT2, PQ6, and HQI8 items were thus inspected for their wording, which in fact appeared not to explicitly convey an inquiry on whether the robot use was either positively or negatively perceived. Therefore, the three-component structure was constrained to a two-component one; this resulted in ATT2, HQI8, and PQ6 items not loading anymore on any factor. After dropping such items, the constraint to the latent structure was removed, and a two-component, OBLIMIN-rotated simple structure yielded. The complete list of the items (described both in Italian and translated into English) is reported in the Appendix A of the paper.

The final PCA proved to meet sphericity (χ² (465) = 3812.46; p < 0.001) and sampling adequacy (KMO = 0.92) assumptions. It yielded a simple, two-component structure accounting for 50.51% of the variance (

Table 3. Items loadings of the Principal Component Analysis for the RAQ.

	Component 1	Component 2
ATT9	0.833
ATT7	0.793
HQF5	0.791
ATT5	0.790
ATT3	0.763
HQF7	0.755
HQI3	0.687
HQF1	0.681
HQF3	0.659
HQF9	0.655
PQ9	0.643
PQ5	0.635
ATT1	0.607
PQ1	0.605
HQI7	0.535
HQI1	0.530
HQI9	0.445
HQF10		0.805
ATT10		0.803
HQI2		0.783
HQI6		0.769
ATT8		0.749
HQI4		0.729
ATT4		0.720
PQ4		0.701
HQI10		0.685
PQ8		0.660
PQ10		0.640
HQF6		0.634
ATT6		0.631
PQ2		0.622

Notes. RAQ = Robot Acceptance Questionnaire; PQ = Pragmatic Quality; HQI = Hedonic Quality—Identity; HQF = Hedonic Quality—Feeling; ATT = Attractiveness.

), with items (N = 17) indexing a positive attitude towards the potential use of the robot (Positive Attitude, PA) loading on the first component (loading range = 0.45–0.83) and those (N = 14) indexing a negative attitude (Negative Attitude, NA) on the second one (loading range = 0.62–0.81). An oblique rotation via the OBLIMIN method proved to be justified (r = −0.26).

Internal consistency was excellent for both PA (Cronbach’s α = 0.93; item–rest correlation range = 0.47–0.75) and NA items (Cronbach’s α = 0.93; item–rest correlation range = 0.51–0.75). PA and NA measures (computed as the sum of respective items) were inversely associated among each other (r_s (208) = −0.38; p < 0.001), while also being intuitively related with EoU, MoU, and WtI, but not FoU, measures (Table 1).

Given that both PA and NA measures were normally distributed (i.e., skewness and kurtosis values < |1| and |3|, respectively) (Kim, 2013) [43], linear models were adopted to test whether they were affected by demographic confounders. At α_adjusted = 0.017, age negatively predicted the PA measure (β = −0.31; t (204) = −4.62; p < 0.001), with neither sex nor education yielding significance (p ≥ 0.021); as to the NA measure, α_adjusted = 0.017, both age (β = 0.41; t (204) = 6.48; p < 0.001) and female sex (β = 0.34; t (204) = 2.7; p = 0.007) were yielded as significant predictors, whilst education did not (p = 0.03). The 5th empirical percentile of the PA measure was 36 and the 95th percentile of the NA one was 57.

5. Discussion and Conclusions

With the rapid advancements in robotic technologies, robot functionalities are continuously growing and diversifying beyond environments requiring structured and repetitive actions, such as factories, to become established in other sectors such as workplaces, education, home services, or healthcare [44]. The most significant aspect of such settings is that they involve interactions with users. Robots are now inserted in social contexts and, consequently, users unconsciously expect that these systems are capable of reproducing behaviors that are intrinsically human [45]. Because of the centrality of these expectations towards automated systems, it becomes vital to consider human user’s perspective since it is the early steps of the robot development by focusing on the evaluation of these systems in terms of preferences, usability, and usefulness [46].

Several attempts have been made to provide quantitative and/or qualitative information about social robots and many tools have been developed to measure robots’ acceptability (for a recent review, see [19]). However, as reported in the introduction section, these questionnaires present different limitations or aspects that should be further examined to provide a more complete evaluation of the system. Therefore, the current research aims to present the RAQ, a new instrument to assess a tele-presented social robot (i.e., Pepper) and the process to test its psychometric properties performed on original data. The proposed questionnaire has been already adopted in previous studies [32,35,36,37,38,39] but it has never been submitted to a validation process. The rationale behind the RAQ is to obtain quantitative information about the users’ perspective on several aspects, such as the willingness to interact with the system, the esthetic pleasantness, the positive and negative feelings that may arouse in a potential interaction, the usefulness and ease of use of the proposed robot, as well as its engaging and communicative skills.

The current study provides Italian researchers with data on the psychometric goodness, in terms of validity and reliability, of the first Italian-validated questionnaire assessing the user’s acceptance of a social robot (i.e., Pepper). The RAQ did not meet the original four-dimensional structure with which it was conceived (PQ, HQI, HQF, ATT). Instead, the final PCA suggests a two-factor structure describing two opposite measures of the user’s acceptance construct: positive and negative attitudes (PA and NA) toward the robot. Both measures yielded high internal consistency and were inversely associated with each other. Moreover, PA and NA intuitively correlated direction with the ease of use, the mastery of use, and the willingness to interact with the robot. Subsequent analyses revealed that participants’ age and PA were negatively associated, whereas neither gender nor education affected this dimension. Concerning NA, age was positively associated with its scores, whereas education did not affect it. Results also suggest that female participants seem to be positively associated with NA scores.

Although the items regarding the participants’ approach to technology and robots address similar, but not overlapping, constructs to the RAQ, no specific a priori hypotheses concerning preliminary items’ data and the two obtained dimensions have been established. Since PA and NA dimensions of the RAQ have been obtained by following a data-drive approach, it would have been imprudent to formulate specific hypothesis regarding their relationship with the ease and mastery of use, and the willingness to interact with the robot. Nonetheless, the outcomes that have emerged in this study are in line with the findings in the literature supporting the positive relationship between users’ acceptance towards robots and the perceived ease and expertise with technology [for a review, see Hornbæk & Hertzum, 2017 [47]]. Despite this, the current findings did not support the role of the educational level nor the frequency of use as influencing factors on robots’ acceptance, as reported in other studies [for a review, see Chatzoglou et al., 2023 [48]].

Concerning the role of the gender and age, the results are in line with the literature reporting that these organismic characteristics play a significant role in the evaluation process [49,50]; indeed, they are considered to be moderating variables on how the robot is perceived [51].

Due to their importance, the recruitment process in the current study involved participants of varying ages, with the aim to ensure that the validation process of the questionnaire was carried out on data pertaining to the assessment of potential end-users spanning different life stages. To this regard, a limitation concerning the sample should be pointed out. We acknowledge that the recruited sample does not encompass all possible participants’ ages are and, due to this, the findings should be interpreted considering this aspect. Nonetheless, the methodological choice of adopting the described sample stratification is supported by previous studies investigating users’ preferences towards social robots through the proposed questionnaire [32,35,36,37,38,39]. Therefore, even though not every individual age was included, the considered ages still manage to represent distinct life stages, as they fall within the age ranges that define them [50].

Another aspect that should be pointed out is that the additional questions of the RAQ that were present in the previous studies [37] concerning the preferred age of the robot and the entrusted occupations were not considered in the validation process since they would not provide quantitative information about the degree of users’ acceptance toward the system. Rather, they intend to qualitatively assess the proposed device by examining the potential application field in which it could be exploited.

A limitation of the current study concerns the brevity of the video stimulus used to present the robot. The 4–7 s clip may not have allowed participants sufficient time to process the robot’s features or reflect on their impressions. Future implementations of the RAQ could benefit from embedding brief pauses between the stimulus and questionnaire items to encourage deeper cognitive engagement and improve data quality.

Additionally, the sample consisted exclusively of Italian participants, which may limit the generalizability of the RAQ’s psychometric properties to other cultural contexts. Cultural norms, media exposure, and societal attitudes toward automation are known to influence robot acceptance. Therefore, cross-cultural validation studies are recommended to assess the RAQ’s applicability in diverse populations.

Another methodological consideration concerns the timing of the questionnaire administration. In the current study, participants completed the RAQ immediately after viewing the video stimulus. Introducing a brief pause between the stimulus and the questionnaire could encourage deeper reflection and enhance the quality of responses. Future studies may consider implementing such inter-stimulus buffering to foster more deliberate evaluations of the robot.

A further limitation of the present study is the absence of benchmarking the RAQ against an external, standardized acceptance scale. This choice was primarily due to the lack of Italian-validated instruments that assess robot acceptance without requiring physical interaction with the system. While tools such as the UTAUT and the System Usability Scale (SUS) have been adapted for Italian populations, their items typically assume direct experience with the robot, which was not feasible in our remote, video-based design. Future research should consider comparative studies using adapted versions of these scales or newly validated instruments suitable for tele-presented scenarios.

The cultural homogeneity of the sample represents an additional limitation of the present study. All participants were recruited in Italy, and the RAQ was validated exclusively within this cultural context. As robot acceptance is influenced by cultural norms, media exposure, and societal attitudes toward automation, the psychometric properties of the RAQ may not generalize to other populations. Future research should aim to replicate the validation process in different cultural settings to assess the cross-cultural robustness of the instrument.

A further limitation concerns the convergent validity that was established by using the participants’ responses to the eight preliminary questions presented to the participant before administering the proposed questionnaire, rather than considering a completely different one. However, to our knowledge, no other scale was psychometrically tested to be validated for the Italian population. Some authors have adapted the original version of the UTAUT [27,28] and the System Usability Scale [29,30], but these questionnaires entail many items which imply the actual use of the system. This could not have been applied to our experimental design since the RAQ does not evaluate the in-presence experience with a social robot.

Although the sample included participants aged 14 to 90 years, the distribution across life stages was not sufficiently balanced to support meaningful stratified analyses. Despite efforts to ensure generational representation, the underrepresentation of certain age groups limited the reliability and interpretability of subgroup comparisons. To avoid potentially misleading conclusions, the study focused on continuous age effects rather than categorical distinctions. In light of this limitation, future research is strongly encouraged to adopt targeted sampling strategies that enable robust stratification by life stage (e.g., adolescence, young adulthood, middle age, and older adulthood), thereby facilitating a more nuanced understanding of robot acceptance across developmental phases.

Rather, the RAQ is conceived as an instrument able to provide information about the criteria the robot should meet to be positively evaluated by the users through the assessment of several aspects related to the system such as its appearance and its capability to establish a long-lasting interaction. Moreover, the RAQ is structured in a way that there is no need for a co-present robot to investigate the evaluation process. These features could be useful in the early stages of the systems’ development and design, as well as when choosing whether to introduce an automated system in a private or public environment before committing to an expensive purchase.

Future research could extend the validation process of the RAQ to other types of social robots (e.g., androids or gynoids), or to other classes of automated systems such as conversational systems or virtual agents. Indeed, an advantage of this tool is that its formulation easily adapts to the evaluation of other technologies.

From a managerial perspective, the present work provides a tool capable of assessing the features that interactive systems should possess in order to meet users’ needs and expectations. The proposed questionnaire may guide the design of intuitive interfaces with improved supportive functions, allowing users/consumers to efficiently interact with them, diminishing the resistance toward the adoption of robotic technologies in everyday life. Indeed, these systems could be exploited in several fields, such as retail services [52], tourism, and hospitality [53], workplace interventions [54], and pedagogic environments [55]. Moreover, the assessment of robotic technologies could be helpful to customize the human–machine interactions for users with specific needs (e.g., elders, people dealing with mental health issues, and children). Moreover, these systems can be exploited as supportive tools for managing and monitoring mental health [56,57,58].

Within these contexts, social robots and artificial intelligence systems may be helpful to increase productivity and reduce costs, encouraging significant growth in the diffusion of service robots and motivating the research dedicated to understanding their implications.