In surveys, it is common to ask people to answer questions about health, policies, job satisfaction, life satisfaction, etc. For example, consider the question: "How are you satisfied with your life?". In general, the researcher provides the informant with a scale whose responses can range from "very dissatisfied" to "very satisfied". If the objective of the research is to analyze the real differences between the groups, the direct answers to the self-assessments of the respondents may be biased, since they can interpret the categories of responses in different ways. This heterogeneity in the behavior of informing on survey responses, also called differential item functioning (DIF), reflects an interpersonal incomparability (VOŇKOVÁ; HULLEGIE, 2011).

An approach to dealing with DIF in surveys was proposed by King et al. (2004). The authors introduced vignettes, which are brief descriptions of hypothetical situations about people's lives, and which are important for a certain domain of interest. The concept implies that the respondent evaluates both the situation of the person in the hypothetical description and his/her own position in relation to the scenario (EVANS et al., 2015; GROL-PROKOPCZYK; FREESE; HAUSER, 2011). Given that the same scenario is presented to several people, similarly, the evaluation of vignettes can provide information on respondent styles (VOŇKOVÁ; HULLEGIE, 2011).

In spite of the use of vignettes in surveys, such as in the health area (CHRISTEN, HILTON et al., 2016, OREMUS, XIE, GAEBEL, 2016, SILVA et al., 2015, WINSOR, MCLEAN, 2016), in Sociology (CHRISTEN et al, 2016; GANONG; JAMISON; Chapman, 2015) and empirical research in marketing (COVA; PARANQUE, 2016; ZHANG; SIMON, 2016), the recommendations for its establishment are limited in the literature (WASON; POLONSKY; HYMAN, 2002). Thus, this study aims to review and describe criteria presented in the literature for robust textualization of vignettes.

Psychometrics recognize the need and importance of standardization of the conditions of application of measurement instruments, one of the most important assumptions of the evaluation. In relation to the conditions of application of measurement instruments, an area of the item response theory dedicates to its study through DIF (ANDRADE; LAROS; GOUVEIA, 2010; ANDRIOLA, 2001).

The presence of DIF in measurement instruments leads to lack of equity in evaluation processes (GROL-PROKOPCZYK; FREESE; HAUSER, 2011). For example, consider a situation in which two groups of people are submitted to a five-item assessment, whose base text is about football. Group A, made up of female people, did not match any item, while people in group B, made up of males, matched all items. The hypothesis arises, that the items on soccer would favor the male, to the detriment of the female people. It is inferred, therefore, that the items have DIF. The DIF, in other words, is an empirical evidence that the subgroups do not present the same probability of correctness in the item (ANDRADE; LAROS; GOUVEIA, 2010; ANDRIOLA, 2001; GROL-PROKOPCZYK; FREESE; HAUSER, 2011).

Studies on DIF have been widely used to reduce the effects of incomparability among people in surveys. Researchers try to improve the problems of interpersonal and transcultural incomparability in surveys through careful questioning, translation (retranslation), focus group, prior cognitive information, and other techniques (KING; WAND, 2007).

Among these, anchoring by vignettes is a technique developed to improve the problems that occur when different groups of respondents interpret and use the responses of an ordinal scale (for example, from full approval to complete disapproval) in a varied way (KING et al., 2004). The anchorage with vignettes addresses DIF by introducing supplementary questions so that we are able to construct a common scale of measures through respondents (KING; WAND, 2007).

For King et al. (2004), the objective of constructing a survey through vignettes is not to create DIF-free questions, but to find consistency in responses and equivalence in vignettes. Therefore, a vignette must be written to describe, in a clear and concrete language, the character's status to the point where the respondent positions himself/herself in all situations and is able to compare his/her own status with the character's status (EVANS et al. KING et al., 2004). The next section provides a review of what vignettes are, as well as arguments that justify their superiority to the use of direct questions in surveys.

A scenario is a story that presents a hypothetical situation that demands action or judgment from the respondents (WASON; POLONSKY; HYMAN, 2002). Likewise, a vignette is a brief description of a person or a social situation that contains precise references to what is believed to be the most important factors for the respondent to make a decision or make a judgment (WASON; POLONSKY; HYMAN, 2002).

Because they are stimuli used to collect data in surveys, like any other research tools, vignettes must be designed properly, otherwise they will produce invalid data. The simple adaptation of previously published scenarios can be a problem for the researcher, especially if the original research presents faults in its constructions. Any changes in the context of the scenarios can have different effects on the study variables (WASON; POLONSKY; HYMAN, 2002).

The vignettes emerge as an analog research model. This type of research constitutes a reasoning through an example. As opposed to in vivo research, whose factors the researcher cannot manipulate or control, in analogical research we are able to control the variables of interest (LANGER, 2016).

Analog research can be done through live models, audio or video recordings, or written descriptions of a customer encounter. Written scenarios are also termed as vignettes. A set of vignettes can be evaluated over a relatively short period of time by the respondents (HEVERLY; FITT; NEWMAN, 1984). In a set of five or more vignettes, for example, it is possible to incorporate several factors into each vignette and study them simultaneously. An additional advantage in using vignettes relates to the customer's first impression bias, such as physical appearance, dressing, speech, etc., being free from such bias (HEVERLY; FITT; NEWMAN, 1984).

Wason, Polonsky and Hyman (2002) identified six arguments in the literature that justify the superiority of vignettes in relation to research with direct questions. For these authors, a vignette: (a) provides greater realism because it provides a variety of situational or contextual factors; (b) provides a standardized stimulus for all respondents, which increases internal validity, measure reliability, and ease of replication; (c) improves the validity of the construct because it directs the attention of the respondent to specific aspects of the research problem question; (d) it transposes difficulties, such as time and expenses, to study real business decisions; (e) reduces the social convenience bias, especially if the questions about the intention to behave are written in the third or first person; and (f) increases the involvement of respondents and dramatizes the issues.

As an example, Figure 1 represents two self-assessments (respondents 1 and 2) and three vignettes (characters 1, 2 and 3) regarding the level of difficulty in practicing physical activities. The degree of self-assessment of the level of difficulty in practicing physical activity is lower for respondent 1 than respondent 2 (and both agree on the ordering of the difficulty level of the characters).

Figure 1 Source: based on King et al. (2004).

Following is a review of the recommendations on the construction of vignettes.

The greatest challenge in the construction of vignettes is to strike a balance between the rigor of the experimental method and the reality of the environment they describe. An ideal set of vignettes should present a descriptive summary that resembles the real case stories and whose variation exists only in the factors studied (EVANS et al., 2015, LANGER, 2016). A vignette should be short and informative, and reflect an average level of operation, so that there is enough variance to reveal the influence of each factor studied (LANGER, 2016). The set of factors should be small and avoid sensitive topics such as rape, suicide, euthanasia, etc., as they distort the assessment and position people's responses at the extremes of the response scale (HEVERLY; FITT; NEWMAN, 1984).

According to King et al. (2004), the respondent does a self-assessment and evaluation of several hypothetical people described in vignettes. Vignettes are arranged on an ordinal scale, from the most effective to the least effective. The authors recommend that the self-assessment of the respondent be positioned at the end of the scenarios. The authors' approach assumes two key assumptions: consistency of response and vignette equivalence.

The consistency of the response assumes that each person uses the category of responses to a particular question of the survey the same way, whether in self-assessment or evaluation of hypothetical situations. The type of DIF can vary between respondents and even for a single respondent among the questions of a survey, but it cannot vary between the self-assessment and the set of vignettes associated with a single question for any respondent. This property can be violated if the respondent feels inferior in relation to the hypothetical situation (KING et al., 2004).

The equivalence of vignette assumes that all respondents perceive the level of the variable represented in any vignette in the same way, on the same one-dimensional scale, independent of the random error. Respondents may differ in how they perceive the level of variation established in vignettes, but any difference is random and independent of the characteristic being measured. This premise can be violated if the respondent interprets the writing of some vignette as directed to some particular interest of the researcher (KING et al., 2004).

In this section, we present the development of an application of the method for creating vignettes to a survey of value co-creation in health services. It should be noted, that this study does not aim to demonstrate how vignettes minimize the effects of DIF in surveys. For this, we recommend a review of the study by King et al. (2004), where such authors demonstrate these properties. The research proposes to identify elements of the literature that contribute to the robust construction of vignettes, whose nature approaches a more prescriptive rather than descriptive approach to the method.

Initially, we construct a blueprint in the model of structural equations modeling to propose a hypothetical model considering the constructs flexibility, responsiveness and co-innovation of the front-line employees of a health service who are positively associated with the value creation construct (SILVA et al., 2015).

Each construct has a cluster of evaluation items (measurement model) that were the objects of the survey. The flexibility construct, for example, has six items. We apply the final questionnaire with the 4 vignettes and the 27 items to the model that will be presented in Figure 3 (SILVA, 2014). However, details regarding structural equation modeling fall outside the scope of this article, whose main objective directed the focus to describe the vignettes method. We will describe the items of each of the constructs in the following sections.

After the definition of the model, with the specification of the evaluation items for each construct, the second step consisted of creating the components of the scenarios that represented each of the items of the constructs, which were validated by an independent panel of ten experts. All have more than 5 years' experience in supervising and/or coordinating clinical services. These qualifications and experiences with health services position them as experts with adequate levels of knowledge about clinical situations to assess the quality of vignettes.

The vignettes were reviewed until there was agreement among the experts that the scenarios described clinical contexts of interactions between clients (patients) and front-line staff (physician, nurse, nurse practitioner, physiotherapist, etc.). In addition, the experts also reviewed the semantic correspondence between the components of the vignettes and the constructs reviewed by the literature. McCrow et al. (2013) used a cutoff of 90% agreement among experts to consider that there should be no changes in the textualization vignettes.

Regarding the content of the four constructs, there was 100% agreement among the experts about the appropriate context for the population of interest, indication of the behaviors of the client and the front-line staff, as well as indication of the communication between the front-line staff. After the adjustments indicated by the experts' observations, there was unanimity of agreement for the clarity, the simplicity and the correspondence between the domains of the four constructs. The four vignettes, from the perspective of all the experts, corresponded to the semantic meaning of the respective constructs. The following are the vignettes of the four researched constructs.

This research aimed to review and describe criteria presented in the literature for a robust vignettes texturing. One of the main goals in the study of judgments and decision making directs the focus to identify the variables that influence the attitude of the service providers on the clients. For example, in the professional meeting of the health-client, it is important to understand how health professionals judge or make decisions about their clients (patients). Many of these studies are based on analogical research, in which the researcher controls the stimuli for the respondents.

Vignettes emerge as an analog research model in which the researcher controls the variables of interest so that we can omit or keep external factors constant. Nevertheless, the first draft of a vignette should be based on a review of the literature or on the judgment of experts, we recommend that a blind review be carried out to improve its validity and reliability.

The presence of DIF in surveys affects the effectiveness of the instruments, so the anchorage with vignettes is presented as a method to repair the measurement problems arising from the interpretation of the scales. Through the vignettes are added anchors in the questions seeking to achieve consistency in the answers and equivalence in the vignettes, allowing the respondent to position himself/herself comparatively to the status of the character.

This study presented some guidelines for the construction of vignettes with a view aimed at methodological rigor in relation to the validity and reliability of the data. A minimum set of criteria (plausibility, clarity, simplicity, content validity and analogy between the vignette and the construct) conceives the essential elements for the judicious development of this methodology. Nevertheless, we developed the application of the process in this study in scenarios involving health services, and since vignettes represent scenarios of real-life situations, it is useful to explore the application of the method in other areas of research, including in administration.

We developed the four hypothetical vignettes following the recommendations presented. A meticulous review of the literature on value creation attempted to make the content of the vignettes describe a real context of health services and thus built up a blueprint for the model. Experts evaluated textualized vignettes regarding the validation of the scenarios and the semantic equivalence of the vignettes, obtaining agreement superior to the recommended cutoff of 90%.

As a theoretical contribution, this research looked at the recommendations on the design of research using the vignettes method. Thus, researchers will rely on a survey of criteria and recommendations for a robust textualization and application of a method to deal with DIF in surveys. In addition, the research explores a model on the background of the experience of value creation for a health service in order to illustrate the application of the method according to the criteria and recommendations presented. In practice, the results of this research contribute to increasing the knowledge about the creation of values in health services by proposing a pattern of development of operative resources that facilitates the creation of strategies to operationalize the concept of value creation with clients.

As the vignettes were applied to three different categories of professionals (physicians, physiotherapists and nurses) to express their opinion on the scenarios, one limitation of this research was the absence of hypothesis tests to verify if the perception of the scenarios diverges or does not due demographic differences and professional backgrounds. In this sense, surveys on attitudes are highly sensitive to textualization; the same questions may have different meanings for respondents when they are in different contexts. In addition, the theoretical and operational platform of this research does not explore the totality of constructs that potentially associate with the experience of value creation in health services.

Although this is a limitation, it also provides an opportunity to verify if there is in fact no difference of perception between the groups of respondents. It is also of interest to future research, the application and development of these criteria to improve the mechanisms for reducing the incomparability of data in interpersonal contexts.