Italian adaptation of the Multidimensional Iowa Suggestibility Scale (MISS)

Introduction

Suggestibility is a personality trait that reflects a general tendency of individuals to accept and internalize messages (Kotov, Bellman & Watson, 2004). Suggestion refers to a process of communication in which one or more persons produce a change in one or more individuals’ opinions, attitudes, behaviors, and judgments, without the involvement of rational thinking (Eysenck, Arnold & Meili, 1975), to the extent that the response to suggestion is usually considered unreasoned and uncritical (Coffin, 1941). Research on suggestion and suggestibility originates from the first studies on hypnosis in the nineteenth century. However, it is important to distinguish hypnotic suggestibility—responsiveness to suggestions given after the induction of hypnosis—from nonhypnotic suggestibility—in which the suggestions are given outside hypnosis and are aimed at convincing people that the world is not the same as it actually is (Braffman & Kirsch, 1999). With the passing of time, numerous authors have expanded the field of research and, especially sociologists and social psychologists, used the term suggestion to explain some social implications and clarify how propaganda has some influence on the acquisition or change of attitudes (Gudjonsson, 1987).

To date, a deep comprehension of suggestibility is lacking and there is little agreement about its dimensionality. Numerous authors (Eysenck & Furneaux, 1945; Grimes, 1948; Benton & Bandura, 1953; Hammer, Evans & Bartlett, 1963) have conducted different studies using factor analysis methodologies, but no common structure emerged from them and the question of whether suggestibility is a multifactored construct or not is still unanswered. In a recent study, Polczyk (2016) followed a factor analytic approach and found that suggestibility involves two uncorrelated factors, providing new evidence for the existence of direct suggestibility—when the influence is overt and the subject is aware that his susceptibility is being measured—and indirect suggestibility—when the influence is hidden. This is a result that other studies failed to find, probably due to methodological issues. However, as Oakley et al. (2021) highlighted, it is reasonable to believe that suggestibility covers a broader domain of traits, of which direct verbal suggestibility constitutes only one of them, while placebo suggestibility and interrogative suggestibility others, as increasing evidence suggests.

A crucial topic is that there is a huge variability among all these studies—in terms of design, sample, and types of measures—that does not allow a confident comparison nor a replication, with the result that all conclusions are uncertain (Tasso et al., 2020). There is a shared need to overcome these limitations, understand the real nature of suggestibility, and improve new standardized measures, with a notable benefit in many areas. This is true, especially for clinical practice, where the placebo effect—a phenomenon in which the administration of a pharmacologically inert substance or therapy produces beneficial clinical outcomes—plays a pivotal role, and a deep comprehension of its mechanisms has important implications. It would allow for improving strategies to control placebo responses minimizing their effect in randomized clinical trials, helping in the interpretation of study results, and maximizing it to improve treatment benefits (Enck et al., 2013). We know that placebo responses vary substantially between, and even within, individuals, and this may depend on many factors: cognitive factors, such as expectations and anxiety (Linde et al., 2007; Flaten et al., 2011), conditioning, prior learning strategies (Benedetti et al., 2016) and individual differences. An interesting study by Parsons et al. (2021) investigated whether direct verbal suggestibility is a predictor of individual differences in placebo hypoalgesia responsiveness, and found that it depends on the use of conditioning. From their study emerged that suggestibility was significantly correlated with placebo responsiveness when the manipulation only included suggestion and not conditioning.

A further difficulty is that the suggestive process is more complex than it could seem, since it consists of different stages, and involves diverse elements. As Gheorghiu (1972) remarks, the process starts with a suggestive stimulus, it continues with suggestibility—the predisposition of an individual to be influenced—and, if present, leads to a suggestible reaction. Except for auto-suggestion, the stimulus derives from interpersonal interactions, which has some important implications. The first is the importance of context which can accentuate the nature of interaction; the second concerns a wide range of motives able to facilitate the suggestive influence, such as expectations, wishful thinking, and interpersonal trust (Gheorghiu, 1972; as cited in Gudjonsson, 1987).

A crucial question arises “about the relationship between suggestibility and private acceptance of the suggestions” (Gudjonsson, 1987). Individuals may accept the suggestions knowing that they are false, implying that they are behaving as others expect them to. In this case, it would be more appropriate to talk about compliance, since there is only a behavioral change without the internalization of a message (Kotov, Bellman & Watson, 2004). On the other hand, individuals may accept the suggestions when they privately believe them to be true and this implies a non-conscious volitional decision. The problem is that it is difficult to understand which is the case since there could not be observable differences in their behavior (Gudjonsson, 1987). For this reason, adequate instruments are required. In fact, over time, numerous attempts to develop instruments to measure suggestibility have been made.

At first, they were limited to reproducing simple motor and sensory reactions, then more complex stimuli to reach changes in opinions and attitudes. In this regard, the few developed scales investigating attitudes mainly focus on specific domains of suggestibility. For example, we can mention the “Gudjonsson Suggestibility Scale” created by G. Gudjonsson which investigates interrogative suggestibility, especially in the forensic field, and the “Barber Suggestibility Scale” developed by T. X. Barber, which focuses on the hypnotic field. It is also worth mentioning the “Harvard Group Scale of Hypnotic Susceptibility, Form A” (HGSHS:A; Shor & Orne, 1962). This scale started as a group version of Weitzenhoffer and Hilgard’s “Stanford Hypnotic Susceptibility Scale: Form A” (SHSS:A; Weitzenhoffer et al., 1959), and nowadays is the most widely used instrument in hypnosis research (Angelini, Kumar & Chandler, 1999), also in Italy (De Pascalis, Russo & Marucci, 2000).

From a historical point of view, these scales were based on the Friedlander-Sarbin Scale (Friedlander & Sarbin, 1938), which was the first standardized scale used to measure hypnotic susceptibility. Over the years numerous changes have occurred. Initially, the SHSS:A required a large amount of time, as it had to be administered individually and was therefore impractical. In contrast, the HGSS:A was used as a preliminary screening device to provide an initial estimate of hypnotic responsiveness (Kallio, 2021). However, both scales have undergone further modification. Despite its popularity, the HGSS:A has received several criticisms. Most of them concern the high presence of false-positive responses in some subscales (e.g., posthypnotic motor movement), or the problematic interpretation of behavioral components. Furthermore, some scales conflate non-hypnotic and hypnotic suggestibility through the inclusion of non-hypnotic suggestions (Acunzo & Terhune, 2021). However, whereas the SHSS:A solely focuses on hypnotic suggestibility, recent studies suggest that hypnotic suggestibility is part of a wider psychological trait of direct verbal suggestibility (DVS), which can be investigated even without hypnotic induction (Oakley et al., 2021). With this premise, Oakley et al. (2021) suggest that by making minor changes, the Harvard Group Scale could be used to measure DVS, but not everyone completely agreed with this solution. In light of this consideration, Kallio (2021) suggested that, rather than revising the currently existing scales, a possible solution could be to completely revise the nature of the concept of “suggestibility” itself.

On the other hand, a wider assessment of suggestibility can be made using the Multidimensional Iowa Suggestibility Scale (MISS). The MISS is a self-report scale about individuals’ perceptions of their suggestibility based on various statements, devised by Kotov, Bellman & Watson (2004) in the United States. The underlying hypothesis is that suggestibility is a personality trait halfway between voluntary and automatic information processing. For the authors, it would also consist of a tendency to accept without particular pressure messages coming from ourselves, from others, or any type of media. After various revisions and analyses, the MISS is presented in its final version consisting of sixty-four items, divided into five subscales: consumer suggestibility, persuadability, sensation contagion, physiological reactivity, and conformism. There are also two companion scales—psychosomatic control, and stubborn opinionatedness—which bring the items to a total of ninety-five. In their validation study, Kotov, Bellman & Watson (2004) showed that the MISS displayed reliabilities of αs >.80 for all the final scales, except persuadability and mental control (α = .79 each). They also provided some evidence for its construct validity, investigating correlations between MISS and self-report measures of Big Five, dependency, absorption, dissociation, obsessive-checking, self-concept clarity, and self-monitoring. Results indicated that the suggestibility scales are distinct from the Big Five traits (r < .32), and showed associations with other constructs, but they were not redundant with any of them (r < .48). However, the scale has not been translated and adapted in Italian.

To fill this gap, this study aims to adapt the Italian version of the MISS to a large sample of Italian healthy individuals, exploring its reliability, validity, and factor structure. To this purpose, in the first study, we translated and administered the Italian adaptation of the MISS to an Italian sample, tested for its structural validity, and refined the scale. In a second study, we cross-validated the adapted scale to another sample. In the end, the final version of the MISS showed good psychometric properties.

Materials & Methods

Results

Descriptive statistics. First, after recording the two reversed-key items, we examined the descriptive statistics to inspect the normality of the item distribution and found that the distribution of several items was problematic (see Table 1). We identified 16 items with a median that corresponded to the minimum value (1). Moreover, we found that seven items showed an absolute skewness index greater than two, and eleven items presented an absolute kurtosis greater than 2. Furthermore, the results of Mardia’s test confirmed that the assumption of multivariate normality was not supported (Mardia, 1970) (Mardia skewness = 26497.29 (p < .001), Mardia kurtosis = 41.24 (p < .001)). Then, we ascertained the presence of redundancy within the subscales. We identified two items with an r >.7, both belonging to the subscale “Persuadability” (“A logical argument can make me change my mind” and “I can be convinced by a good argument”).

We noted that two items of the Peer Conformity’s subscale showed a weak correlation with the total score (corrected r < .25) and that the subscale “Sensation Contagion” was almost entirely composed of items with a weak discriminative power—since they had a median corresponding to the minimum or maximum value—and/or with absolute skewness and/or kurtosis strongly deviated from the norm and/or did not contribute to the Cronbach’s alpha. For this reason, we decided to remove this subscale from the successive analyses. We also removed the two items that presented a weak correlation with the total score (“I dress very differently from my friends” and “I don’t like most of the movies, my friends, like”). Moreover, we eliminated the two redundant items, the one that slightly seemed more difficult to understand (“A logical argument can make me change my mind”). Finally, we removed three other items that presented excessive skewness and kurtosis and had a median coincident with one of the extremes (“Sometimes I want a product because I like the person endorsing it”, “I get my style from certain celebrities” and “I use advertisements as a guide for shopping”).

We verified the absence of Heywood cases using the analysis of residuals and variances, and then we examined the modification indices to understand how to improve the model through the removal of other items and/or the addition of covariances between the residuals, where theoretically justifiable. Cross-loadings were admitted in three cases (e.g., we allowed the item “Commercials sometimes make me want products that I did not know I needed” to reflect the “Persuadability” construct, as well as “Consumer”). In the end, we arrived at a forty-six-item reduced version of the scale.

Model estimation and evaluation. As reported in Table 2 the values of the fit indices for all the tested models. The one-factor model revealed inadequate, and all the fit indices considered were below the recommended values (CFI = 0.86, TLI = 0.85, SRMR = .13, RMSEA = .14, χ2/df = 8.07). The four factors and the higher-order models yielded acceptable fit and did not show appreciable differences, but we decided to concentrate subsequent analysis on the bifactor model since this one was more theoretically motivated and showed better fit (CFI = 0.98, TLI = 0.97, SRMR = .07, RMSEA = .06, χ2/df = 2.25). The χ2 difference test also revealed that the bifactor model had a significantly better fit than the other models (the output of the model comparison is shown in Table 3). Values of PUCs and ECV for the bifactor model were respectively 0.76 and 0.47 and did not reach the suggested cutoffs (PUCs >0.80 and ECV >0.60) to consider the MISS as essentially unidimensional.

Internal consistency. The reliability in terms of internal consistency showed a ωh of .77 for the suggestibility factor. The group (specific) McDonald’s Omega of the subscales ranged from .37 to .54 (Consumer Suggestibility = .48; Persuadability = .54; Physiological Reactivity = .43; Peer Conformity = .37).

Results

Descriptive statistics. Descriptive statistics of the items are reported in Table 4.

First, we assessed multivariate normality using the Mardia’s test, but results showed that the assumption was not supported (Mardia skewness = 23714.10 (p < .001), Mardia kurtosis =28.98 (p < .001)). Then, we inspected the means and medians of items and found that no one was falling into an extreme value (1 or 5). We also checked items skewness and kurtosis and there were no items with values greater than 2. Moreover, we checked item correlations within the subscales and found there were no correlations higher than .7, confirming the absence of redundancy within the subscales.

Model estimation and evaluation. We performed a CFA through the lavaan library (Rosseel, 2012) to verify the structure validity of the scale. Since the assumption of multivariate normality was not met, we used a diagonally weighted least square estimator (DWLS).

The results of fit indices for all models are reported in Table 5. The overall fit statistics suggest a poor fit for the unidimensional solution (CFI = 0.87, TLI = 0.86, SRMR = .11, RMSEA = .12, χ2/df = 4.96). The four factors and the four factors hierarchical models showed similar results and acceptable fit. The bifactor model performed better than the other models and the overall fit statistics suggested a good fit (CFI = 0.97, TLI = 0.97, SRMR = .07, RMSEA = .06, χ2/df = 1.99). Then, we compared the four models, and from the results of the χ2 difference test (reported in Table 6), the bifactor model outperformed the other three models. Then, we computed PUCs and ECV for the bifactor model and their values (respectively 0.76 and 0.54) were below the indicated cutoffs, suggesting a multidimensional structure of the MISS.

To better understand how variance is partitioned, we explored the loadings of the bifactor solution for the general factor suggestibility and the four subscales (results are reported in Table 7). The average loading on the general factor is .46, while those of the subscales are .46 for consumer .22 for persuadability, .49 for physical reactivity, and .40 for conformity. These data show that the general factor explains about 20% of the variance in the items. The subscales consumer, physical reactivity, and conformity explain, over and above the general factor, on average a similar part of the variance, while the subscale persuadability is only 5%. In Table 7 we put in bold the items loading on the general factor greater than .50. These items provide the best discrimination on the suggestibility factor. On the other hand, the grouping factor items loading greater than the general factor loadings (also in bold) could be considered relatively better measures of the specific subscale construct than of suggestibility. Among the subscales, the physical reactivity one is the only one that appears a bit problematic, since nine of the 13 items have greater loadings on the specific grouping factor than on the general one. This could derive from a measurement problem or from the fact that it may measure something separate from the suggestibility construct.

Internal consistency. The scale demonstrated good internal consistency, with a McDonald’s hierarchical Omega of .81 for the general factor suggestibility and values ranging from .28 to .72 for the subscales (Consumer Suggestibility = .50 ; Persuadability = .72; Physiological Reactivity = .28; Peer Conformity = .57).

Divergent validity. A SEM was used to investigate the relationship between the MISS and the SMS. Results highlighted that the two measures, once accounted for the subscales contribution, are strongly correlated (Estimate = 0.72, se = 0.04, z = 17.60, p < .001). The upper limit of the confidence interval (95% CI [0.641–0.801]) barely exceeded the threshold (.8) suggested by Rönkkö & Cho (2022). On the other hand, the MISS was not significantly correlated with any dimension of the HEXACO-60, nor the TIPI (results are reported in Table 8).

Discussion

The Multidimensional Iowa Suggestibility Scale (Kotov, Bellman & Watson, 2004) is a self-report measure of individual differences in suggestibility, which covers different domains, such as conformity, persuadability, physical reactivity, and consumer. Our study aimed to adapt the scale in a large sample of Italian healthy individuals, investigating its factor structure and psychometric properties.

In the first study, we examined item characteristics of the original five subscales and dropped all those that resulted problematic, redundant, or not discriminative. We also removed the subscale “Sensation Contagion” because it had a weak discriminative power. Reliability analyses highlighted good internal consistency with McDonald’s hierarchical omega, and this result is consistent with that reported in the original validation of the scale by Kotov, Bellman & Watson (2004). The confirmatory factor analysis showed good fit indices for the bifactor model. We also compared the bifactor model with the other three models, and the bifactor model outperformed the others. This is due to the fact that it allows us to better explain how variance is partitioned between the general factor suggestibility and the subscales (Reise, Moore & Haviland, 2010).

We conducted a second study to cross-validate the new version of the MISS. We administered the scale to a new independent sample to investigate whether the factor structure and reliability were confirmed. This would strengthen and validate the results we found in the first study, also increasing generalizability. As shown in the results, we found that the MISS demonstrated good psychometric properties. Reliability in terms of internal consistency was good. With regard to factorial structure, we tested four measurement models, as in the first study. The first was unidimensional, and suggestibility represented the general factor. We tested this model for comparison only since we already knew from the original validation study that there was evidence of multidimensionality. The second model was a four-factor model, where each subscale represented a factor. The third model was a hierarchical second-order latent variable model (SOLV), with the subscales as first-order factors, and suggestibility as second-order factor. The fourth was a bifactor model, in which all items loaded on the general factor suggestibility and also on a specific grouping factor. CFAs revealed that the bifactor structure was confirmed, and this is consistent with the results found in our first study. This is an important point to discuss since the use of the bifactor model allowed us to investigate and understand the contribution of general and specific factors more deeply than could be done with other models, and for this reason, it should not be considered only as an alternative model to test against (Dunn & McCray, 2020). A deeper analysis of ωh, PUCs, and ECV permitted us to further explore the dimensionality of the measurement model of the MISS. However, results showed mixed evidence since, on the one hand, the high ωh suggests the presence of a relatively strong general factor, but on the other hand PUCs and ECV values did not reach the cut-offs recommended for essential unidimensionality, thus indicating a multidimensional structure of the scale. At this point, we decided to explore in detail the factor loadings on the general and the specific grouping factors to further understand the contribution of each and, in particular, the adequacy of reporting a total or composite score. We observed that on average the general factor suggestibility explained about the 20% of shared variance. The subscales accounted for a similar or lower level of explained variance, apart from the contribution of suggestibility. The loadings of the subscales were tendentially lower than those of the general factor, except for the physical reactivity subscale.

Given these results, and analyzed the items’ content, we concluded that there was no justification to report only the separate scores for the subscales. Moreover, as argued by Sinharay, Haberman & Puhan (2007), we are convinced that the total score is more reliable compared to the composite score and constitutes a better predictor of the true individual’s score on suggestibility traits.

The Italian adaptation of the MISS demonstrated acceptable construct validity. We tested divergent validity and found that there was a lack of correlation between the MISS score and the TIPI and HEXACO-60 scores and a moderate correlation between the MISS score and the Self-Monitoring scale. However, when we modeled with SEM the relationship between the general factor of the MISS and the factor of the SMS, once accounted for the specific factors’ contribution, the correlation became stronger and the upper limit of the confidence interval fell just above the threshold considered marginally problematic by Rönkkö & Cho (2022). So, it remains unclear whether the MISS is measuring a different construct from self-monitoring.

In the end, the MISS revealed good psychometric properties and can represent a precious tool for a wide evaluation of individual differences in suggestibility in many domains. Future studies are needed to confirm all these findings and maybe test the scale, or a revised version, in clinical samples.

Conclusion

To date, in Italy, the most used scale to measure suggestibility is the Italian version of the Harvard Group Scale of Hypnotic Susceptibility, Form A (De Pascalis, Russo & Marucci, 2000). More instruments are needed. To fill this gap, we adapted the Multidimensional Iowa Suggestibility Scale (MISS) in a large sample of Italian individuals and the scale showed good psychometric properties. Measuring individual differences in suggestibility is crucial, in many different domains, such as clinical psychology, neuromarketing, and neuropharmacology. In particular, the objective assessment of suggestibility could become of interest to pharmacological studies assessing the placebo effect. When an inert substance is delivered as a standard treatment and the clinical outcome is improved, this phenomenon is known as the placebo effect. One of the main clinical conditions for which the rate of placebo response is high is Parkinson’s disease (Quattrone et al., 2018). This phenomenon is drastically driven by verbal suggestions that induce an indirect overactivation of the reward network compensating the dopaminergic deficit in patients with Parkinson’s disease. We believe that the adaptation of this scale could be of interest for clinical practice to assess the probability of neurological patients being prone to placebo effects.

Supplemental Information