Order Online or Call it in? Conceptualization and Measurement of Preferences for Computer versus Human Interaction

Abstract

Technology is increasingly integrated into everyday life and consequently, traditional social exchanges between human agents have evolved to include human-computer interaction, providing scientists new directions for understanding and predicting consumer behavior. Despite progress, there are conceptual and empirical limitations in current measures applied to consumer/user preferences. This paper documents the development and validation of the 10-item preference for computer versus human interaction (PCHI) scale across three distinct samples and incorporating items that (a) include direct comparisons between computers and humans, (b) are independent of specific contexts or technologies, and (c) capture major theoretical domains of social, consumer, and human factors psychology. Results support the hypothesized three-factor structure (efficiency, ease of use, and trust) and demonstrate the utility of this measure to predict everyday consumer decisions beyond extant measures. Additionally, the PCHI offers marketing, user experience, and other practitioners a brief instrument for developing interventions, training protocols, and modeling attitudes.

Keywords

Human-computer interaction preference trust

As automated technology becomes more commonplace in everyday life, consumers face more discretion between interacting with a computer or a human (Brynjolfsson & McAfee, 2014; Meuter et al., 2000; Talwar, 2015). Whether a customer orders a cheese pizza online instead of calling in the order or chooses a self-checkout machine instead of a human cashier, we are surrounded with options in who or what we interact within a variety of situations. Furthermore, these choices are increasingly pervasive in everyday life as businesses work to improve efficiency, optimize cost-effective strategies, and expand the capacity to provide service.

The breadth of human-computer interactions is illustrated by the dramatic rise of automated services in industries as diverse as education (Ivanov, 2016; Timms, 2016), financial trading (Dunis et al., 2017), healthcare (Mirheydar & Parsons, 2013), transportation (Maurer et al., 2016), and hospitality (Ivanov & Webster, 2017). A frequent issue faced in education is predicting the acceptance of technology (e.g., tablets, Smartboards, etc.) as it is introduced into a teaching environment (Dündar & Akçayır, 2014), which concerns both students’ and teachers’ attitudes toward technology. Similarly, a major question facing ride-sharing companies, such as Uber and Lyft, is whether potential passengers will choose to accept a ride from a self-driving car over a human driver. Indeed, a recent American Automobile Association (AAA) found that less than 20% of drivers would trust an autonomous vehicle to drive them, although the majority of drivers with semi-autonomous vehicles trust the autonomous technology (Stepp, 2016).

Given the importance of customer attraction and loyalty to most businesses (Flint et al., 2011; Wood & Neal, 2009) it is critical for consumer psychology research to understand and predict consumer choices to reject or embrace technological interaction (Ivanov & Webster, 2017). In response to the growing industry trends and the explosion of customer loyalty programs that attempt to nudge consumers into repeatedly choosing a product or service, many streams of research have emerged in information management, communication, and psychological domains, all seeking to explain the intersection of technology with human attitudes and behaviors (Coovert & Goldstein, 1980; Fetler, 1984; Morrow et al., 1986). Some of the most promising areas of understanding human attitudes towards computers involve trust in technology and technology acceptance. Empirical research examining these psychological constructs has resulted in many scales meant to measure attitudes toward machines, however, existing scales in this area possess characteristics that constrain their use in predicting consumer choices.

Against this background, we first present several important theoretical vehicles that take human-computer interaction into account. However, the aim of this paper is not to scrutinize the long and highly complex literature in exhaustion. Instead, we focus on the most widely used models for understanding computer attitudes. Then, we consider contemporary measures of these constructs and argue the need for a scale that (a) directly compares computers and humans, (b) is context independent but adaptable, and (c) derives content from appropriate psychological constructs.

Conceptualizing Preferences for Computer/Machine versus Human Interaction

Several approaches to consumer and user interaction with technology involve subjective attitudes and therefore have involved understanding the psychological organization of consumer’s technology preferences. Given a large number of intersecting construct domains from many different psychological fields, our goal is to integrate these approaches in a fashion that advances current understanding based on their commonalities.

One source for approaching PCHI is decision-making, marketing, and consumer research. Recent technology, such as e-commerce, has prompted scholars to consolidate complex theoretical frameworks (see Darley et al., 2010), however, research into dispositional antecedents of consumer intention and behavior appear to overlap substantially with the broader social psychological findings (i.e., Kim et al., 2008). Within consumer psychology, technology readiness has been proposed as a dispositional construct capturing individual differences in propensity to engage and use technology (Parasuraman, 2000). Recent research predicting user intentions to adopt technology found technology readiness to have no direct effect beyond constructs of the technology acceptance model (Lin et al., 2007), while a large field study on self-service technology adoption by airline customers found technology readiness to contribute little in terms of prediction (Liljander et al., 2006). Related concepts like brand attachment and loyalty, technology readiness, trust, and unitarian/hedonic consumer attitudes inform the current study, however, they are rooted in contemporary social literature (Eagly & Chaiken, 1993) and do not particularly address PCHI.

A separate source of conceptualizing the dispositional framework of PCHI stems from human factors, management, and human-computer interaction research. Here, scientists have addressed technology’s emergence in similar terms of user attitudes that also fall in line with other areas of attitude research (Eagly & Chaiken, 1993). Importantly, technology acceptance and trust are prominent construct domains that are conceptually congruent with those of consumer attitudes.

We provide an overview of two domains commonly investigated in applicable research on attitudes toward technology, technology acceptance and trust, and argue that integrating these constructs provides a theory-driven and practical solution to modeling and predicting preferences for computer versus human interaction. These domains have seen strong cross-pollination between consumer-driven research agendas (Kulviwat et al., 2007; Lin et al., 2007; Pavlou, 2003) and user-driven agendas (Carter & Bélanger, 2005), and therefore these domains offer the most empirically supported groundwork for developing the PCHI scale while potentially maximizing generalizability.

Technology Acceptance

User acceptance of technology is considered important in both the development and implementation of technology along with understanding the factors that lead a customer to choose a technology-based option for service.

The technology acceptance model (TAM; Venkatesh & Davis, 2000) derives acceptance of technology from perceptions of usefulness and ease of use. Although there have been numerous iterations and extensions of TAM, the model itself has experienced widespread use throughout information management literature (Venkatesh et al., 2007). Perceived usefulness (PU) refers to judgments regarding the ability of a technology or automated service to reach its functional goal. On the other hand, perceived ease of use (PEU) refers to the effort that one must expend to engage with a technology (Davis, 1993). A similar model, the consumer acceptance model (CAT; Kulviwat, et al., 2007) has recently been proposed in marketing literature, however, it has seen far less investigation and, consequently, less evidence of generalizability––although the core factors of PU and PEU are considered in both models.

Empirical studies have produced support for the PU and PEU constructs, often accounting for around 30–50% of the variance in intentions to use technology and/or actual technology usage (Venkatesh & Bala, 2008; Venkatesh & Davis, 2000). Most of this research involves the implementation of specific technologies for use by employees (e.g., account management or personnel systems), however, recent research has linked technology acceptance by consumers to important marketing outcomes. For instance, Ha and Stoel (2009) found that TAM could help explain customer acceptance for e-shopping, shopping via the internet. Similarly, Suh and Han (2002) applied TAM to predicting user intentions and behavior regarding internet banking. While the technology acceptance model has found broad support in the literature (King & He, 2006), recent research has begun to expand TAM by incorporating trust as another antecedent of behavioral intentions.

Trust in Technology

Trust is an additional important consideration when predicting intentions and behaviors involving technology use (Suh & Han, 2002). In fact, trust in automation has been an increasingly studied construct for understanding a range of daily activities, such as receiving tailored news and advertisements to accepting or engaging with new technology. For example, there have been many concerns about whether people will trust (and subsequently use) automated driverless cars in the future. This type of hesitation from (dis)trust may be related, but distinct from, evaluations of whether the driverless technology is perceived to be easy to use and useful. For example, even though in many cases algorithms can predict outcomes better than humans, people still tend to choose human predictions over computer predictions. This was demonstrated in a series of studies in which participants tended to choose human forecasts over computer forecasts after seeing both make a similar error, even when computer forecasts were superior on average (Dietvorst et al., 2015). Therefore, although the computer algorithm had an advantage in efficiency and was equivalent in ease of use compared with human prediction, the computer algorithm was chosen less often.

In this vein, a variety of the choices that consumers face involving technology or human interaction can be influenced by trust as well as PU and PEU. Scientists have linked PU, PEU, and trust to internet-mediated behaviors, such as online shopping (Gefen et al., 2003; Ha & Stoel, 2009; Pavlou, 2003) or online banking (Suh & Han, 2002). It is important to note, however, that previous models have often neglected actual choices faced by consumers and users. Alternatively, while some marketing research has included measures of trust while testing the consumer acceptance model (e.g., Grabner-Kräuter & Faullant, 2008), there is still little theoretical integration of trust into the model and measures fail to explicitly tap into PU, PEU, and trust domains within the same instrument. Consequently, scientists have not yet produced a theoretical and practical measure to capture either technology acceptance or trust concepts regarding a choice between human versus computer interaction. Provided the accelerating pace of technology’s integration into daily life and this increasing behavioral discretion in choosing between human versus computer interaction, this missing measurement framework may offer important applications that are becoming more relevant with time.

Current Computer Attitude Measures

The relationships and predictions regarding technology use are largely examined using self-report measures that ask about attitudes and perceptions regarding computers or technology. There are several existing scales that measure these constructs, such as the attitudes toward computer usage scale (ATCUS; Popovich et al., 1987), technology acceptance (Venkatesh & Bala, 2008), and trust in automation (Hoff & Bashir, 2015). Despite their wide use in the literature, there are important limitations in their ability to advance our understanding and prediction of choices to interact with a computer versus a human.

First, the popular measures to assess attitudes towards and perceptions of computers only assess the extent to which technologies are perceived or utilized in isolation. That is, these measures do not also assess how humans may instead be utilized in similar roles to computers or technology (e.g., cashier or customer service representative). Omitting a comparison to human alternatives to technology interaction may leave out important information in predicting choices and may be a major confound in research examining technology utilization. For example, an individual may perceive a self-checkout machine as easy to use. However, this individual might still choose to check out with an employee cashier because they perceive dealing with the human employee to be even easier than the self-checkout machine. In other words, without capturing the preference for interaction between humans and technology, these scales will ultimately fall short in predicting intentions and behavior when alternatives are available as they solely capture attitudes toward one target (the technology).

Second, many of these measures contain items that are dependent on a certain context or type of technology. For example, items on the ATCUS 2.0 scale involve specific technologies (e.g., compact discs and Powerpoint; Morris et al., 2009) that are or will likely be obsolete, likely resulting in several updates and a lack of generalizability to future questions regarding consumer technology use. Not only does this issue make existing measures less useful over time, it also adds difficulty to examining differences over time, since the construct being assessed may change based on how applicable the items are to the individual. Given the wide breadth of technologies and their inevitable drive toward antiquation, there is a need to cover as much of the construct domain as possible while avoiding overly narrow sampling, such as references to current and specific technologies (Nunnally & Bernstein, 1994).

Finally, none of the existing measures incorporate all of the major relevant factors (PEU, PU, and trust) into a common instrument. For example, scientists measuring PEU and PU often include a separate trust measure in their research (Gefen et al., 2003). Given that shopping, banking, and a host of other consumer-related behaviors can involve human or automated interactions, lack of a validated measure that addresses these limitations is needed. Such a measure could have great utility for areas as diverse as human factors, marketing, education, or employment where decisions frequently involve choices between interacting with a computer/machine or a human.

Preference for Computer versus Human Interaction Scale Development

The goal of this paper is to establish a measure capturing individual preferences for engaging with humans versus machines in a way that addresses three primary limitations with current measures. Specifically, the current measure should (1) include items that reflect explicit comparisons between humans versus computer/machine interaction such that higher scores reflect a greater preference for interacting with computers/machines over humans, (2) possess items that are independent of a specific context or technology and generalize across consumer settings, and (3) integrate current theoretical elements of technology acceptance and trust in technology. Incorporating these elements, this paper documents the creation, development, and initial validation of the Preference for Computer versus Human Interaction (PCHI) scale across two studies and three independent samples.

Study 1

Study 1 involved the initial construct definition, item generation, content validity analysis, subsequent item trimming, and preliminary psychometric evaluation of the reliability and validity of the PCHI scale. The underlying assumption of creating this instrument is that people differ in respect to their preferences for interacting with humans or computers/machines. Twenty-one graduate psychology students from various PhD programs defined preference for human versus machine interaction (PCHI), producing the general definition of PCHI as an individual’s general disposition toward interacting with technology, machines, or computers, versus a human that involves affect (e.g., trust or liking), judgments (e.g., ease of use, better, slower), and behavioral preferences (e.g., allowing machines to make decisions for oneself).

Method

Participants and Procedure

The initial sample consisted of 401 respondents recruited via various social media platforms by the 21 doctoral students. Missing values were determined to be missing completely at random (MCAR) by examining both item and respondent totals for missing values (<5% missing). To avoid case-wise deletion of data and avoid reducing power, multiple imputation chained equations (MICE) was utilized to estimate values for those missing based on the observed values for other variables (Royston & White, 2011). Using predictive mean matching to constrain the imputed values within the observed score range, five imputed datasets were created and averaged to create final estimates for the missing values.

Preference for Human versus Machine Interaction (PCHI) Scale

Following Hinkin’s (1998) recommendations for scale development, 21 doctoral students independently created items meant to draw from the PCHI domain, employing a heterogeneous domain-sampling approach. This initial effort produced 203 items.

Several weeks following the initial item generation, the sample of doctoral students independently rated items on (a) how essential the item is, and (b) the quality of the item. The content validity ratio (CVR; Lawshe, 1975) was calculated for each item, with items identified as essential by less than half of raters being dropped from the item pool. Further items were removed from the pool based on redundancy and conceptual issues. For example, items that did not specifically include a reference to computers, technology, or machines (e.g., “I like face-to-face interaction”) and were dropped. The remaining 13 items (Supplementary Appendix Table 1) were all statements that reflected either reference to machines or computers or were explicit comparisons between humans or computers (e.g., “machines are less prone to mistakes than humans).

Analytic Approach

Before conducting the analyses, the data were split into two random halves. In the first sample, we conducted an exploratory factor analysis on the 13-item initial PCHI scale. In the second sample, we performed confirmatory factor analysis (CFA) on competing measurement models based on the findings from the first sample. Criteria for our CFAs include a variety of fit indices and unavoidable rules of thumb. Specifically, we examined the chi-square statistic (χ²), Tucker-Lewis Index (TLI), Comparative Fit Index (CFI), Root Mean Square Error of Approximation (RMSEA), and standardized factor loadings (>.35) to assess our measurement models (Hu & Bentler, 1999). When evaluating incremental fit for model comparisons, we will report all changes to these various indices, as well as Δχ² for statistical significance tests.

Results

Sample 1: Exploratory Factor Analysis

An exploratory factor analysis (Thompson, 2004) using principal axis factoring was conducted to examine the initial factor structure (n = 200). An oblique (direct oblimin) rotation was used since there was no expectation for unrelated factors (Fabrigar et al., 1999). Additionally, a parallel analysis (Hoyle & Duvall, 2004) provided an objective criterion for factor retention in concert with more subjective methods, such as the Kaiser rule or Scree test (Thompson, 2004).

Results from the parallel analysis suggested that a two-factor solution should be interpreted. Factor loadings of the two-factor solution indicated interrelated factors representing human efficiency versus machine efficiency (six items) and trust and ease of use for humans versus machines (five items; Supplementary Appendix Table 1). Two items (3 and 4) were dropped due having factor loadings less than .35.

Sample 2: Confirmatory Factor Analysis and Reliability Analyses

To examine the stability of the scale and test competing models of factor structure, several CFAs were conducted via the Lavaan package for R (Rosseel, 2012) using the second half of the sample (n = 201) The two-factor solution [χ²(43) = 120.63, p < .001; TLI = .86; CFI = .89; RMSEA = .095] demonstrated better fit to the data than the single-factor solution [χ²(54) = 235.06, p < .001; TLI = .71; CFI = .76; RMSEA = .129; Δχ²(11) = 114.43 p < .001], however the solution indicated marginal fit to the data. After examining modification indices, it was found that item 13 (“technology is more reliable than people”) exhibited high loadings onto both factors. Removing this item not only significantly improved fit, but also allowed the model to achieve acceptable levels of fit [χ²(34) = 43.81, p < .001; TLI = .90; CFI = .93; RMSEA = .079; Δχ²(1) = 29.27, p < .001].

The final scale consisted of two five-item factors (Supplementary Appendix Table 2). Due to only a moderate correlation (.43) between the two factors, it appeared more appropriate to treat the two factors as separate subscales. The resulting subscales demonstrated acceptable internal consistency (efficiency preference α = .74; trust/ease of use preference α = .79).

Discussion

Findings from this initial validation attempt for measuring PCHI proved fruitful in establishing preliminary evidence for a factor structure that is consistent with current theoretical models of human/technology interaction and choice (e.g., Venkatesh & Davis, 2000). Although a modified two-factor solution demonstrated only a moderate fit to the data, the two factors were both conceptually in line with previous findings that ease of use and efficiency/usefulness are related constructs involved in constituting individuals’ technology preferences (Davis, 1989; Venkatesh & Davis, 2000).

Despite promising evidence, limitations in study 1 prompted the need for additional data to be collected. First, the sample used in study 1 involved social network contacts of 21 doctoral students that participated in the initial item generation, limiting the generalizability of the psychometric assessment. Second, dropping an item to achieve acceptable levels of factor structure fit warrants confirmation in independent samples. Third, while confirming a factor structure consistent with theoretical models provides some evidence for construct validity, additional evidence is required for a complete psychometric evaluation of the PCHI. This includes examining convergent and discriminant validity (Campbell & Fiske, 1959) along with criterion-related validity (Landy, 1986).

Study 2

The purpose of study 2 was to evaluate PCHI from a theoretical basis in contrast to the empirical approach of study 1. While the exploratory factor analysis in study 1 resulted in testing a two-factor solution, this factor structure demonstrated marginally acceptable fit to the data. The authors revised the PCHI model, considering the previously mentioned literature, to conform with theory on human interactions with technology; consequently, a third factor was hypothesized that reflects interaction preferences based on trust. After discussion between the authors and a comparison of existing trust measures with PCHI item content, a third factor was formed by taking two items from the interaction preferences based on ease factor (“I trust people more than technology” and “I trust computers/machines more than people”) and one item from the interaction preferences based on efficiency factor (“I would allow computers/machines to make some decisions for me”). Three factors assessing interaction preferences based on trust, ease of use, and efficiency better reflect the major theoretical domains of human-technology interaction (Gefen et al., 2003; Venkatesh & Bala, 2008). As such, we predicted the following hypothesis:

H1: A three-factor solution reflecting interaction preferences based on trust, ease, and efficiency will exhibit better fit than a two-factor solution reflecting interaction preferences based on ease and efficiency.

Another aim of study 2 was to evaluate the convergent, discriminant, and criterion-related validity of the PCHI. To establish convergent and discriminant validity of the PCHI subscales, it is important to examine correlations with constructs that would be expected to be related to a preference for human versus machine interaction (Clark & Watson, 1995; Hinkin, 1998). As previously mentioned, attitudes toward computer usage (ATCUS) and technology acceptance (PU and PEU) represent constructs that are theoretically similar to PCHI. Consumers that prefer to interact with machines or computers instead of humans may choose to do so because they use computers or often have positive perceptions of computers overall. Additionally, individuals reporting higher perceived usability and perceived ease of use scores should be more likely to prefer interactions with computers versus humans. Specifically, we predicted that:

H2: All PCHI scales will exhibit convergent validity by correlating positively with TAM and ATCUS measures.

An additional area that would be expected to be associated with a preference for interacting with machines/computers versus humans is social interaction anxiety. The Diagnostic and Statistical Manual of Mental Disorders, 5th edition (DSM-5; American Psychiatric Association, 2013) describes generalized social interaction anxiety (such as talking with strangers) as a core feature of social phobia. Individuals with clinical as well as non-clinical levels of social anxiety may then be motivated to interact with a machine/computer instead of a human simply to avoid feeling anxious. Among the PCHI subscales, the interaction preference based on ease would be expected to have the strongest relationship with social anxiety, because social anxiety is likely an obstacle that makes interaction with humans more difficult than machines. On the other hand, the PCHI subscale for interaction preference based on efficiency should have a weak relationship with social interaction anxiety, due to interaction preferences based on efficiency being based more on perceptions of how mistake-prone a human or machine interaction might be.

Similarly, an additional measure that could be used to explore convergent and discriminant validity of the PCHI scale is the personality trait of extraversion. Extraversion refers to an individual’s tendency to be sociable, gregarious, assertive, talkative, and active (Barrick & Mount, 1991). Given the broad nature of extraversion in influencing a variety of choices and preferences, PCHI subscales may also correlate differently with extraversion scores. For example, an individual with a low extraversion score may still choose to interact with a human instead of a computer/machine if that individual sees a human as more efficient or less mistake prone. On the other hand, given the sociable nature of individuals high in extraversion, human interaction may be preferred regardless of how easy to use, trustworthy, or efficient the computer/machine is perceived to be. Despite the conceptual basis for differential relationships between social interaction anxiety, extraversion, and preference for machine/computer interaction, there is a lack of research examining these relationships. Therefore, we examined the following relationships with a general research question:

RQ1: What pattern of relationships are found between the PCHI subscales, social interaction anxiety, and extraversion scores?

Another goal of study 2 was to examine the extent the PCHI scale possesses criterion-related validity. Given the intention for the PCHI to assess individual differences in the extent to which a machine or computer is chosen for interaction over a human, it is important to assess the extent to which PCHI scale scores can predict realistic choices. Specifically, we predict:

H3a: The PCHI scale will explain significant variance in individual choices to interact with either a computer/machine instead of a human.

A final intention of study 2 was to examine the usefulness and utility of the PCHI beyond other scales that assess similar constructs, such as attitudes towards computers or computer usage. Indeed, a central tenet of any new measure’s practical value resides in its ability to predict unique variance beyond current measures, thereby substantiating the new measures added value. To examine the PCHI utility over other measures, we also predicted that:

H3b: The PCHI scale will explain incremental variance in individual choices to interact with either a computer/machine instead of a human over and above existing computer attitude measures.

Method

Participants and Procedure

Two independent samples were collected to demonstrate generalizability across a broader range of the population. The first sample consisted of 251 undergraduate students at a large public university in the southeast United States. The second sample consisted of 161 adult workers on Amazon.com’s Mechanical Turk. Respondents on Mechanical Turk choose to complete human intelligence tasks (HITs) that encompass a variety of activities such as taking surveys or proof-reading text. Comparative research investigating differences between Mechanical Turk and traditional student samples have found Mechanical Turk participants to be more diverse and producing similar or superior data quality (Behrend et al., 2011; Buhrmester et al., 2011). Provided the goal to collect evidence for various aspects of validity for the PCHI scale, two disparate samples drawn from separate data sources were deemed an advantage to establishing generalizability.

Prior to conducting analyses, individuals were screened for careless responding by examining responses to an item that read “please select ‘disagree’ for this item.” Individuals that did not respond correctly to this item were considered carelessly responding and were subsequently removed from the data (Meade & Craig, 2012). In addition, individuals that were clear outliers in survey completion times (e.g., more than 1 hour to complete the survey) were also removed. These screenings led to 23 (9.16%) individuals removed from the student sample and 10 respondents (6.2%) removed from the Mechanical Turk sample.

The final student sample was comprised of 228 participants and was 57.5% male with a mean age of 19.5 years, ranging from 18 to 48. The racial breakdown of the student sample was 71.1% White/Caucasian, 7.5% African-American, 8.3% Asian-American, 5.3% Hispanic/Latino, and 7.9% other. Participants in the student sample were compensated with research credits that were required for the introductory psychology course. The final Mechanical Turk sample was comprised of 151 participants and was 60.3% female with a mean age of 40.8 years, ranging from 19 to 85. The racial breakdown of the second sample was 68.9% White/Caucasian, 10.6% African-American, 8.6% Asian-American, 10.6% Hispanic/Latino, and 1.3% other. Students received research credit and Mechanical Turk respondents were compensated $0.25 for completing the survey.

Measures

Preference for Computer versus Human Interaction

The revised PCHI scale from study 1 was used to assess individual differences in computer versus human interaction preferences. The final scale from study 1 contained 10 items across. The three factors proposed included: preference for machine versus human interaction based on efficiency (four items; e.g., “Computers/machines are less prone to mistakes than humans”), preference for machine versus human interaction based on ease (three items; e.g., “It’s easier to work with a computer/machine than it is to work with a human”), and preference for machine versus human interaction based on trust (three items; e.g., “I trust people more than technology”). The scale was scored on a seven-point Likert scale from strongly disagree to strongly agree (composite α = .81 in student sample; .85 in the Mechanical Turk sample).

Attitudes Toward Computer Usage

The second revision of the Attitudes Toward Computer Usage Scale (ATCUS v. 2.0; Morris et al., 2009) was used to assess convergent validity of the PCHI. This scale was originally developed and to assess individual reactions to computers and technology (Popovich et al., 1987) and revised to reflect more recent developments in technology. The ATCUS 2.0 has 22 items across four factors: Computers for tool use (five items; e.g., “I would like to have more computerized features in my car such as GPS, CD player, etc.”), confidence using computers (seven items; e.g., “I feel comfortable hooking up my computer and installing software”), negative reactions to computers (five items; e.g., “I feel that computers limit my creativity”), and positive reactions to computers (five items: e.g., “I would rather shop online than in a physical store”). The scale was scored on a seven-point Likert scale from strongly disagree to strongly agree (composite α = .84 in the student sample, .87 in the Mechanical Turk sample).

Technology Acceptance

Technology acceptance was assessed with the Perceived Usefulness and the Perceived Ease of Use scales from the Technology Acceptance Model 2 (TAM2; Venkatesh & Davis, 2000). The Perceived Usefulness scale has four items (e.g., “Using the system improves my performance in my job”). The Perceived Ease of Use scale also has four items (e.g., “My interaction with the system is clear and understandable”) and is scored in the same format of the Perceived Usefulness scale. To better conform to the context of the current study and other scales within the survey, these items were adapted such that “the system” was changed to “computers/machines” within each item. Both scales were scored on a seven-point Likert scale from strongly disagree to strongly agree (Perceived Usefulness scale overall α = .85 in the student sample, .91 in the Mechanical Turk sample; Perceived Ease of Use scale overall α = .84 in the student sample, .80 in the Mechanical Turk sample.).

Social Interaction Anxiety

Social interaction anxiety was assessed with the Social Interaction Anxiety Scale (SIAS-6, Peters et al., 2012). This scale has demonstrated strong psychometric properties for typical and clinical populations in previous research (Peters et al., 2012). The SIAS-6 has six items (e.g., “I have difficulty making eye contact with others”) assessed on a seven-point Likert scale from strongly disagree to strongly agree (overall α = .84 in the student sample, .92 in the Mechanical Turk sample).

Extraversion

Extraversion was measured with a ten-item scale from the International Personality Item Pool (Goldberg et al., 2006). The items were statements to which participants reported agreement with the extent the statement applied to them (e.g., “I am the life of the party”). The scale was scored on a seven-point Likert scale from strongly disagree to strongly agree (overall α = .92 in the student sample, .87 in the Mechanical Turk sample).

Consumer Choices Between Computers/Machine Interaction and Human Interaction

The authors developed six situations to provide a survey-based criterion to assess the ability of the PCHI to predict realistic choices regarding consumer-based choices in interacting with either a computer/machine or a human. These statements involved common, naturally dichotomous choices such as ordering pizza using an online order form or talking to a person on a phone to make the order (Supplementary Appendix Table 3). Participants responded to the situation by choosing either to utilize, interact with, or rely on a computer/machine or a human. Computer decisions were coded as 1 and human decisions were coded as 0. Individual choices across the six situations were summed, resulting a potential range of scores from 0 to 6. To assess the content validity of these choices, participants were asked after each situation to rate how realistic the situation was on a scale of 1 (“not at all realistic”) to 5 (“extremely realistic”). The average realistic rating for the six situations was 3.72 and 4.06 in the student and Mechanical Turk samples, respectively. These average scores correspond to the situations being rated, on average, as “very realistic” (on a five-point scale with 1 = not at all realistic, 2 = slightly realistic, 3 = moderately realistic, 4 = very realistic, 5 = extremely realistic).

Results

Bivariate correlations, scale alphas, and descriptive statistics for all scales are detailed in Supplementary Appendix Tables 4 and 5 for the student and Mechanical Turk samples, respectively. Supplementary Appendix Tables 8 and 9 detail normative information for the student and Mechanical Turk samples, respectively. Stanine norms are presented with respect to average scale scores. Stanine transformations retain the original raw score distribution and allow scores to be interpreted as below average (1, 2, and 3), average (4, 5, and 6), or above average (7, 8, and 9).

To test the first hypothesis, which predicted creating a third factor for trust would improve factor structure fit, two CFAs were conducted. The first CFA specified the two-factor solution found in study 1 and the second CFA specified the hypothesized three-factor solution. Support was found for the first hypothesis, with the three-factor solution (Supplementary Appendix Table 6) providing strong levels of fit to the data (Hu & Bentler, 1999) and providing significant improvement in fit over the two-factor solution across both samples (student sample: [χ²(32) = 52.51, p < .01; TLI = .96; CFI = .97; RMSEA = .053; Δχ²(2) = 71.34, p < .001]; Mechanical Turk sample: [χ²(32) 39.57, p = .17; TLI = .98; CFI = .98; RMSEA = .04; Δχ ²(2) = 28.23, p < .001]). Additional support for the strong factor fit was substantiated with the 90% confidence intervals for the RMSEA not overlapping .08 in either sample, suggesting a precise estimate of absolute fit with the data.

The second hypothesis predicted PCHI scales would demonstrate convergent validity with constructs that should be theoretically related. Hypothesis 2 predicted the PCHI scales would have positive correlations with other measures assessing computer usage and attitudes. As indicated in Supplementary Appendix Tables 1 and 2, support was found for hypothesis 2, with all three PCHI scale scores showing significant positive correlations with the ATCUS 2.0 and TAM2 subscales across both samples.

A research question was also posed to examine patterns of correlations between the PCHI scale, social interaction anxiety scores, and extraversion scores. The PCHI subscales for interaction preference based on ease of use and trust demonstrated significant positive correlations with SIAS scores (Supplementary Appendix Tables 4 and 5). However, the PCHI scale for interaction preference based on efficiency did not correlate significantly with SIAS scores in either sample. Similar patterns were found between PCHI subscales and extraversion scores in the student sample, with significant negative correlations between the ease of use and trust PCHI subscales and a nonsignificant correlation between extraversion and the efficiency subscale. However, in the Mechanical Turk sample, extraversion did not significantly correlate with any PCHI subscale.

The third hypothesis predicted PCHI scales would demonstrate criterion-related validity by predicting individual choices for machine/computer versus human interaction. To test this hypothesis, a multiple regression analysis was conducted in each sample using the total number of situational machine/computer interaction choices made as the outcome. As indicated in Supplementary Appendix Table 7, support was found for hypothesis 3, with the multiple regression analysis explaining significant variance in interaction choices for the student sample (9% variance explained) as well as the Mechanical Turk sample (18% variance explained).

The fourth and final hypothesis predicted the PCHI scales would demonstrate incremental validity in predicting interaction choices beyond related scales concerning computer usage and attitudes. Four hierarchical regression analyses were conducted to examine the incremental validity of the PCHI scales over the TAM2 subscales and the ATCUS 2.0 subscales across the two samples. In the student sample, PCHI scales explained incremental variance beyond the TAM2 scales (ΔR² = .06, p < .01) and ATCUS 2.0 scales (ΔR² = .03, p < .05). In the Mechanical Turk sample, PCHI scales explained incremental variance beyond the TAM2 scales (ΔR² = .11, p < .001) as well as the ATCUS 2.0 scales (ΔR² = .06, p < .001). Thus, support was found for hypothesis 4, with PCHI scales explaining incremental variance in all four hierarchical regressions.

Discussion

The first goal of study 2 was to integrate the major theoretical areas of technology and computer-based attitudes into our conceptualization of PCHI. Adopting a three-factor structure including efficiency, ease of use, and trust provided a superior fit to the data in both samples compared to a two-factor structure that only included efficiency and ease of use. This finding supports efforts of past research that incorporates trust into research on technology acceptance models (e.g., Gefen et al., 2003). Therefore, trust is an active dimension of PCHI, as opposed to being a distinct construct that should be assessed separately from efficiency and ease of use.

The PCHI scale also demonstrated convergent validity by correlating significantly with similar measures that are used to assess computer attitudes, such as the ATCUS 2.0 and the TAM2 scales. However, results demonstrate how the PCHI’s ability to assess preferences for interaction goes beyond simply assessing attitudes toward computers in several ways. For example, all significant correlation coefficients between ATCUS 2.0 and TAM2 scales with the SIAS were negative, indicating that individuals reporting high levels of social interaction anxiety tended to also report negative perceptions of and attitudes towards computers. However, all significant correlations between PCHI subscales and SIAS scores were positive, indicating that individuals reporting high levels of social interaction anxiety still preferred to interact with a computer/machine despite not always viewing computers/machines as favorable. This key insight suggests that attitudes towards interacting with people (social anxiety) or attitudes towards computers (technology acceptance/trust) alone are not sufficient to understanding and predicting choices between interacting with a person or a computer/automated system. In addition, PCHI scales demonstrated an ability to significantly predict interaction choices beyond the other computer attitude scales, further indicating that interaction choice goes beyond simply having a favorable view of computers/machines. The consistency of this pattern of results across two samples with a mean difference in age of 21 years further substantiates these findings.

The pattern of bivariate correlations between SIAS scores and PCHI scores also provide evidence for both convergent and discriminant validity. In both samples, a significant positive correlation was found between SIAS and PCHI ease of use scores, such that higher social interaction anxiety was associated with a greater preference for computer interaction based on ease of use. The same pattern was found for the PCHI trust subscale in both samples but at a lesser magnitude. However, discriminant validity evidence was found with nonsignificant relationships between the PCHI efficiency subscale and SIAS scores. This suggests that individuals with elevated social interaction anxiety may still choose to interact with humans when humans are perceived to be more efficient (or computers as less efficient).

Findings were less consistent between samples for extraversion scores. In the student sample, the findings between SIAS scores and PCHI subscales were exactly mirrored for extraversion, such that higher extraversion scores were associated with a preference for human interaction over computer interaction based on ease of use and trust, but not efficiency. However, the same could not be said for Mechanical Turk sample, where extraversion scores did not demonstrate any significant relationship with PCHI subscales.

In addition to finding evidence of convergent and discriminant validity, criterion-related validity was also found for the PCHI, with simple three to four-item subscales collectively explaining 9–18% of the variance in realistic choices involving interacting with either a computer/machine or a human. Across both samples, the variance explained in interaction choices by the PCHI was found to be unique from variance explained by currently used measures, further highlighting the added utility of directly assessing preference between computer versus human interaction. In sum, the PCHI measure predicted significant variance in the criterion and did so beyond the TAM2 and ATCUS 2.0 scales.

General Discussion

The aim of these studies was to develop and evaluate the PCHI scale. After an initial item creation, the first study proceeded to evaluate the initial scale in an exploratory fashion with a convenience sample. Although a two-factor (ease and efficiency) solution demonstrated adequate fit with the data, the authors’ review of current human factors literature prompted several hypotheses warranting further testing with a revised model. Thus, study 2 was undertaken with the purpose of evaluating (a) whether a three-factor solution including trust offered better fit with the data, (b) whether the PCHI scale demonstrated acceptable construct validity, and (c) whether the PCHI scale predicts choices that consumers commonly face.

There is no current construct found in the literature that cleanly captures preferences for interacting with a computer or machine instead of a human. The validity of the PCHI to assess preferences for interaction based on trust, ease, and efficiency using comparative items without anchoring to a specific context or technology was confirmed in two independent samples. Our use of different sourcing methods (university and Mechanical Turk) produced diverse samples that map quite closely to latest United States Census data (2016). That these two samples, with a mean age difference of 20 years, produced results in line with our three-factor model suggests that our findings may be quite generalizable across age, gender, and race.

The second contribution of these studies is evidence for the construct validity of the PCHI scale. Specifically, support was found for hypothesized relationships between conceptually related and unrelated measures of constructs, providing strong evidence for the construct validity of the PCHI scale. Although further documentation of the theoretical space surrounding PCHI is warranted, this preliminary evidence is promising in substantiating a distinct construct measure.

The most straightforward and practically useful outcome of this research is the development of an instrument that can be used to predict individual choices. Outside of research, practitioners are faced with decisions in how to create, implement, and forecast the use of technologically-mediated tools and services. Existing measures for computer attitudes possess serious limitations for the choice between a computer of a human by only examining attitudes towards computers, being technology or context-specific, and not incorporating the major psychological constructs involved in preference attitudes. The PCHI addresses these limitations and thus may be useful in understanding and predicting the choices that potential users make. To this end, the PCHI scale demonstrated criterion-related validity for the choice outcome in this study––computer/machine or human interaction. Thus, the primary evidence presented in study 2 suggests that the PCHI scale can be employed to predict choice across a variety of contexts and in very different samples (students and working adults). Perhaps more importantly, the PCHI scale demonstrated incremental validity beyond alternative measures currently used in human factors and computer utilization research.

Conclusions

Limitations

Despite the encouraging psychometric properties and validity evidence presented in this paper, there are several limitations that bear mention. First, the PCHI subscale for interaction preference based on trust demonstrated low internal consistency across both samples. This may be due to constructing this subscale based on a theoretical rationale instead of generating trust-based PCHI items empirically. However, this limitation is mitigated by the strong factor structure fit and consistent pattern of significant correlations in the expected direction with related variables across two independent samples. Nevertheless, future research should aim to develop this subscale with further item generation or integration with existing trust measures. Although we used theory to define and flesh out the theoretical domains of our subscales, our paramount concern was offering a useful tool for practice. Future research should investigate further whether the subscales themselves offer scientific uses, such as mapping particular constructs to closely related phenomena (e.g., does the trust subscale relate more closely to decision making for risky interactions?).

Second, data were collected from a single, common source using the same method (self-report) which can result in common source bias––potentially limiting conclusions drawn from data analysis. Although the influence and pervasiveness of common method bias are hotly debated (see Spector, 2006), we took steps to mitigate the influence of CMV. For example, we offered model tests of our instrument across three separate samples, all solicited through different means. Indeed, recent empirical investigations recommend researchers consider how student samples and crowd sourced samples are reflective of the populations of interest (Cheung et al., 2017), which prompted this study to examine multiple distinct samples to bolster generalizability. Further, there were several nonsignificant correlations between self-report measures, suggesting that there was not likely a source of common variance across all the measures used in these studies. If there were some shared method bias that inflated observed correlations, then it would be expected that all self-reported measures would be statistically significant (Spector, 2006). Finally, different anchors and response formats (e.g., Likert vs. dichotomous choice) were used to reduce the likelihood that a response bias would occur.

Another limitation of the study was the outcome variable used in the criterion-related validity evidence. We took care to ensure that we asked appropriate and generalizable scenarios where our population would make decisions regarding human or computer interaction. Unfortunately, we could not capture all the domains these choices could occur, nor did we examine actual behavioral choices. Nevertheless, participants rated each of the scenarios used in our analyses as very realistic, suggesting that they were good representations of situations many respondents have faced or are likely to face.

Our data was collected prior to the onset of the COVID-19 pandemic that brought a quick and visceral shift in the manner in which many people exchanged information and goods globally. While the United States and large portions of the world are moving toward a return to pre-pandemic life or a “new normal,” we believe that future research should consider the ways in which acute (quick onset of a nascent pandemic) or chronic (technological advancements) may affect individuals’ preferences for interaction.

Future Research

Technology, and automated systems in particular, is experiencing an incredible pace of development, resulting in further technological integration into our everyday lives. Given this increasing presence, we believe it is equally pressing that consumer psychology, human factors, marketing, and various other related disciplines begin to investigate the individual differences that can account for technology use and acceptance alongside human alternatives. Specifically, we believe a crucial foundation to these investigations is an understanding of individual preferences.

Future research should seek to conceptualize and substantiate the nomological network surrounding PCHI. Although understanding the consequences of PCHI provides strong practical utility for those wishing to predict intentions and choice, scholarly understanding would benefit from documenting the antecedents and related constructs that are already studied relevant to technology-related behaviors and attitudes. While the studies described in this paper included comparisons of the PCHI scale with measures of similar and unrelated constructs, future research should continue to map relations between PCHI and other extant variables.

We chose to adopt a multidisciplinary approach to conceptualizing and measuring PCHI based on frameworks developed in human factors research, with an emphasis on implications for consumer psychology. We believe the PCHI scale’s three-factor structure is optimal for many of the reasons outlined previously, however, there are other theoretical frameworks which are congruent with the assumptions and theory we have outlined in this paper. For instance, the tenets of social exchange theory (Blau, 1964) purport that social interactions involve the weighing of various characteristics by the social actor(s). Characteristics such as perceived benefit, costs, and uncertainty can be likened to efficiency, ease of use, and trust, respectively. Joining disparate streams of research from human factors, social psychology, and consumer behavior will require rigor, but may ultimately shed light on important generalizations that may transfer across these disciplines.

Implications

Interactions between humans and computers are increasing in frequency and complexity. This integration is occurring across nations and functions (e.g., leisure and work) and is especially salient in industrialized nations where artificial intelligence and automated services (e.g., self-driving automobiles; self-service) are becoming a reality. The importance of these increasing interactions may be determined by several factors, but the individualistic and choice-driven culture of capitalist societies, like the United States, present especially unique circumstances to understand how individual preferences influence decision-making. That is, one may expect that preferences bear more weight on decisions when there are competing choices to make. The outcome of our studies is the development of a psychometrically-sound and practical instrument that can be used by practitioners who need a direct measure of personal preference that is brief but also captures the most pertinent domains––ease of use, usefulness, and trust. From setting up a ride to the airport to purchasing the plane ticket, the PCHI scale is a useful tool that may help to explain the reactions people have toward technology versus human options, as well as their decision-making processes leading to a choice.

Supplemental Material

Supplemental Material - Order Online or Call it in? Conceptualization and Measurement of Preferences for Computer versus Human Interaction

Supplemental Material for Order Online or Call it in? Conceptualization and Measurement of Preferences for Computer versus Human Interaction by Justin Travis and Samuel Wilgus in Psychological Reports

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Data Availability

All data collection, storage, and analyses were approved by the IRB and maintained strict adherence to ethical guidelines and protocols.

ORCID iD

Justin Travis

Supplemental Material

Supplemental material for this article is available online.

Author Biographies

Justin A. Travis received his PhD in Industrial/Organizational Psychology from North Carolina State University and teaches in the psychology and business analytics departments at the University of South Carolina Upstate. His research interests involve individual differences and work behavior, and his work has appeared in outlets like the Journal of Educational Psychology and Microbiology Spectrum.

Samuel Wilgus received a BS in Psychology from Towson University in 2015 and his MS in Industrial and Organizational Psychology from North Carolina State University in 2017. He currently works as a Data Scientist at Clarkston Consulting.

References

American Psychiatric Association . (2013). Diagnostic and statistical manual of mental disorders (5th ed.). American Psychiatric Publishing.

Barrick

M. R.

Mount

M. K.

(1991). The big five personality dimensions and job performance: a meta‐analysis. Personnel Psychology, 44(1), 1–26. https://doi.org/10.1111/j.1744-6570.1991.tb00688.x

Behrend

T. S.

Sharek

D. J.

Meade

A. W.

Wiebe

E. N.

(2011). The viability of crowdsourcing for survey research. Behavior Research Methods, 43(3), 800. https://doi.org/10.3758/s13428-011-0081-0

Blau

P. M.

1964. Exchange and power in social life. John Wiley.

Brynjolfsson

McAfee

(2014). The second machine age: Work, progress, and prosperity in a time of brilliant technologies. WW Norton & Company.

Buhrmester

Kwang

Gosling

S. D.

(2011). Amazon’s mechanical turk: A new source of inexpensive, yet high-quality, data? Perspectives on Psychological Science, 6(1), 3–5. https://doi.org/10.1177/1745691610393980

Campbell

D. T.

Fiske

D. W.

(1959). Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin, 56(2), 81–105. https://doi.org/10.1037/h0046016

Carter

Bélanger

(2005). The utilization of e‐government services: Citizen trust, innovation and acceptance factors. Information Systems Journal, 15(1), 5–25. https://doi.org/10.1111/j.1365-2575.2005.00183.x

Cheung

J. H.

Burns

D. K.

Sinclair

R. R.

Sliter

(2017). Amazon mechanical turk in organizational psychology: An evaluation and practical recommendations. Journal of Business and Psychology, 32(4), 347–361. https://doi.org/10.1007/s10869-016-9458-5

10.

Clark

L. A.

Watson

(1995). Constructing validity: Basic issues in objective scale development. Psychological Assessment, 7(3), 309–319. https://doi.org/10.1037/1040-3590.7.3.309

11.

Coovert

M. D.

Goldstein

M. A.

(1980). Locus of control as a predictor of users’ attitude toward computers. Psychological Reports, 47(1), 1167–1173. https://doi.org/10.2466/pr0.1980.47.3f.1167

12.

Darley

W. K.

Blankson

Luethge

D. J.

(2010). Toward an integrated framework for online consumer behavior and decision making process: A review. Psychology and Marketing, 27(2), 94–116. https://doi.org/10.1002/mar.20322

13.

Davis

F. D.

(1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319–340. https://doi.org/10.2307/249008

14.

Davis

F. D.

(1993). User acceptance of information technology: System characteristics, user perceptions and behavioral impacts. International Journal of Man-Machine Studies, 38(3), 475–487. https://doi.org/10.1006/imms.1993.1022

15.

Dietvorst

B. J.

Simmons

J. P.

Massey

(2015). Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology General, 144(1), 114–126. https://doi.org/10.1037/xge0000033

16.

Dündar

Akçayır

(2014). Implementing tablet PCs in schools: Students’ attitudes and opinions. Computers in Human Behavior, 32, 40–46. https://doi.org/10.1016/j.chb.2013.11.020

17.

Dunis

C. L.

Middleton

P. W.

Karathanasopolous

Theofilatos

K. A.

(Eds.). (2017). Artificial intelligence in financial markets: Cutting edge applications for risk management, portfolio optimization and economics. Palgrave Macmillan.

18.

Eagly

A. H.

Chaiken

(1993). The psychology of attitudes. Harcourt Brace Jovanovich

19.

Fabrigar

L. R.

Wegener

D. T.

MacCallum

R. C.

Strahan

E. J.

(1999). Evaluating the use of exploratory factor analysis in psychological research. Psychological Methods, 4(3), 272–299. https://doi.org/10.1037/1082-989x.4.3.272

20.

Fetler

(1984). Television viewing and school achievement. Journal of Communication, 34(2), 104–118. https://doi.org/10.1111/j.1460-2466.1984.tb02163.x

21.

Flint

D. J.

Blocker

C. P.

Boutin

P. J.

(2011). Customer value anticipation, customer satisfaction and loyalty: An empirical examination. Industrial Marketing Management, 40(2), 219–230. https://doi.org/10.1016/j.indmarman.2010.06.034

22.

Gefen

Karahanna

Straub

D. W.

(2003). Trust and TAM in online shopping: an integrated model. MIS Quarterly, 27(1), 51–90. https://doi.org/10.2307/30036519

23.

Goldberg

L. R.

Johnson

J. A.

Eber

H. W.

Hogan

Ashton

M. C.

Cloninger

C. R.

Gough

H. G.

(2006). The international personality item pool and the future of public-domain personality measures. Journal of Research in Personality, 40, 1, 84–96. https://doi.org/10.1016/j.jrp.2005.08.007

24.

Grabner-Kräuter

Faullant

(2008). Consumer acceptance of internet banking: The influence of internet trust. International Journal of Bank Marketing, 26(7), 483–504.

25.

Stoel

(2009). Consumer e-shopping acceptance: Antecedents in a technology acceptance model. Journal of Business Research, 62(5), 565–571. https://doi.org/10.1016/j.jbusres.2008.06.016

26.

Hinkin

T. R.

(1998). A brief tutorial on the development of measures for use in survey questionnaires. Organizational Research Methods, 1(1), 104–121. https://doi.org/10.1177/109442819800100106

27.

Hoff

K. A.

Bashir

(2015). Trust in automation: Integrating empirical evidence on factors that influence trust. Human Factors, 57(3), 407–434. https://doi.org/10.1177/0018720814547570

28.

Hoyle

R. H.

Duvall

J. L.

(2004). Determining the number of factors in exploratory and confirmatory factor analysis. In Handbook of quantitative methodology for the social sciences (pp. 301-315). Thousand Oaks, CA: Sage.

29.

L. T.

Bentler

P. M.

(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. https://doi.org/10.1080/10705519909540118

30.

Ivanov

(2016). Will robots substitute teachers? Paper Presented at the 12th International Conference “Modern Science, Business and Education”, Varna University of Management (pp. 42–47).

31.

Ivanov

Webster

(2017). The robot as a consumer: A research agenda. Paper Presented at the “Marketing: Experience and Perspectives” Conference, University of Economics-Varna.

32.

Ivanov

Webster

(2017). The robot as a consumer: a research agenda. Paper presented at the “Marketing: Experience and Perspectives” Conference, 29-30 June 2017, University of Economics-Varna, Bulgaria.

33.

Kim

D. J.

Ferrin

D. L.

Rao

H. R.

(2008). A trust-based consumer decision-making model in electronic commerce: The role of trust, perceived risk, and their antecedents. Decision Support Systems, 44(2), 544–564. https://doi.org/10.1016/j.dss.2007.07.001

34.

King

W. R.

(2006). A meta-analysis of the technology acceptance model. Information and Management, 43(6), 740–755. https://doi.org/10.1016/j.im.2006.05.003

35.

Kulviwat

Bruner

G. C.

Kumar

Nasco

S. A.

Clark

(2007). Toward a unified theory of consumer acceptance technology. Psychology & Marketing, 24(12), 1059–1084. https://doi.org/10.1002/mar.20196

36.

Landy

F. J.

(1986). Stamp collecting versus science: Validation as hypothesis testing. American Psychologist, 41(11), 1183–1192. https://doi.org/10.1037/0003-066x.41.11.1183

37.

Lawshe

C. H.

(1975). A quantitative approach to content validity. Personnel Psychology, 28(4), 563–575. https://doi.org/10.1111/j.1744-6570.1975.tb01393.x

38.

Liljander

Gillberg

Gummerus

Van Riel

(2006). Technology readiness and the evaluation and adoption of self-service technologies. Journal of Retailing and Consumer Services, 13(3), 177–191. https://doi.org/10.1016/j.jretconser.2005.08.004

39.

Lin

C.-H.

Shih

H.-Y.

Sher

P. J.

(2007). Integrating technology readiness into technology acceptance: The TRAM model. Psychology and Marketing, 24(7), 641–657. https://doi.org/10.1002/mar.20177

40.

Maurer

Gerdes

J. C.

Lenz

Winner

(Eds.). (2016) Autonomous driving: technical, legal and social aspects. Springer Open.

41.

Meade

A. W.

Craig

S. B.

(2012). Identifying careless responses in survey data. Psychological Methods, 17(3), 437–455. https://doi.org/10.1037/a0028085

42.

Meuter

M. L.

Ostrom

A. L.

Roundtree

R. I.

Bitner

M. J.

(2000). Self-service technologies: Understanding customer satisfaction with technology-based service encounters. Journal of Marketing, 64(3), 50–64. https://doi.org/10.1509/jmkg.64.3.50.18024

43.

Mirheydar

H. S.

Parsons

J. K.

(2013). Diffusion of robotics into clinical practice in the United States: Process, patient safety, learning curves, and the public health. World Journal of Urology, 31(3), 455–461. https://doi.org/10.1007/s00345-012-1015-x

44.

Morris

S. A.

Gullekson

N. L.

Morse

B. J.

Popovich

P. M.

(2009). Updating the attitudes toward computer usage scale using American undergraduate students. Computers in Human Behavior, 25(2), 535–543. https://doi.org/10.1016/j.chb.2008.11.008

45.

Morrow

P. C.

Prell

E. R.

McElroy

J. C.

(1986). Attitudinal and behavioral correlates of computer anxiety. Psychological Reports, 59(3), 1199–1204. https://doi.org/10.2466/pr0.1986.59.3.1199

46.

Nunnally

J. C.

Bernstein

I. H.

(1994). Psychometric theory (3rd ed.). McGraw-Hill.

47.

Parasuraman

(2000). Technology Readiness Index (TRI) a multiple-item scale to measure readiness to embrace new technologies. Journal of Service Research, 2(4), 307–320. https://doi.org/10.1177/109467050024001

48.

Pavlou

P. A.

(2003). Consumer acceptance of electronic commerce: Integrating trust and risk with the technology acceptance model. International Journal of Electronic Commerce, 7(3), 101–134. https://doi.org/10.1080/10864415.2003.11044275

49.

Peters

Sunderland

Andrews

Rapee

R. M.

Mattick

R. P.

(2012). Development of a short form Social Interaction Anxiety (SIAS) and Social Phobia Scale (SPS) using nonparametric item response theory: the SIAS-6 and the SPS-6. Psychological Assessment, 24(1), 66–76. https://doi.org/10.1037/a0024544

50.

Popovich

P. M.

Hyde

K. R.

Zakrajsek

Blumer

(1987). The development of the attitudes toward computer usage scale. Educational and Psychological Measurement, 47(1), 261–269. https://doi.org/10.1177/0013164487471035

51.

Rosseel

(2012). lavaan: An R package for structural equation modeling. Journal of Statistical Software, 48(2), 1–36. https://doi.org/10.18637/jss.v048.i02

52.

Royston

White

I. R.

(2011). Multiple imputation by chained equations (MICE): implementation in Stata. Journal of Statistical Software, 45(4), 1–20. https://doi.org/10.18637/jss.v045.i04

53.

Spector

P. E.

(2006). Method variance in organizational research truth or urban legend? Organizational Research Methods, 9(2), 221–232. https://doi.org/10.1177/1094428105284955

54.

Stepp

(2016). Three-quarters of Americans “afraid” to ride in a self-driving vehicle. http://newsroom.aaa.com/2016/03/three-quarters-of-americans-afraid-to-ride-in-a-self-driving-vehicle/

55.

Suh

Han

(2002). Effect of trust on customer acceptance of Internet banking. Electronic Commerce Research and Applications, 1(3-4), 247–263. https://doi.org/10.1016/s1567-4223(02)00017-0

56.

Talwar

(Ed.). (2015). The future of business. Fast Future Publishing.

57.

Thompson

(2004). Exploratory and confirmatory factor analysis: Understanding concepts and applications. American Psychological Association.

58.

Timms

M. J.

(2016). Letting artificial intelligence in education out of the box: Educational cobots and smart classrooms. International Journal of Artificial Intelligence in Education, 26(2), 701–712. https://doi.org/10.1007/s40593-016-0095-y

59.

Venkatesh

Bala

(2008). Technology acceptance model 3 and a research agenda on interventions. Decision Sciences, 39(2), 273–315. https://doi.org/10.1111/j.1540-5915.2008.00192.x

60.

Venkatesh

Davis

F. D.

(2000). A theoretical extension of the technology acceptance model: Four longitudinal field studies. Management Science, 46(2), 186–204. https://doi.org/10.1287/mnsc.46.2.186.11926

61.

Venkatesh

Davis

F. D.

Davis

Morris

2007). Dead or alive? The development, trajectory and future of technology adoption research. Journal of the Association for Information Systems, 8(4), 267–286. https://doi.org/10.17705/1jais.00120

62.

Wood

Neal

D. T.

(2009). The habitual consumer. Journal of Consumer Psychology, 19(4), 579–592. https://doi.org/10.1016/j.jcps.2009.08.003

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.28 MB