Disjunctive answer options complicate communication – a linguistic analysis of the danish EQ-5D (5 L) version

Introduction: EQ-5D is an internationally acknowledged tool for assessing health-related quality of life. Our aim was to examine how pragmatic dynamics may influence answers to the EQ-5D-5 L in items where the structure of answer options is disjunctive. Methods: We performed a 3-step linguistic analysis building on the seminal work of Grice, including (1) examination of the lexical meanings of the answer options, (2) considerations of how conversational maxims might affect the respondent’s interpretation of compatible answer options under a single item, and (3) analysis of how the questionnaire’s context might counteract the problem of omitted answer options by shifting the meaning of context-sensitive expressions. Results: All items with disjunctive answer options exhibit both compatibilities and omissions. In combination with the disjunctive form of answer options these features of the EQ-5D-5 L complicates the communicative task for respondents relying on conversational norms to identify the most suitable answers to the instrument’s questions. Discussion: In items where answer options have a disjunctive structure, respondents relying on Gricean conversational maxims will have to depend on their individual understanding of fine-grained details concerning the questionnaire’s purpose and may have to weigh how conflicting norms should be balanced. While such dynamics are likely to go undetected in cognitive interviews, linguistic analysis may help to identify them.


Background
Health-related quality of life has increasingly become a core outcome in health delivery as guidance for treatment, care, and rehabilitation.EQ-5D is an internationally acknowledged generic tool for assessing health-related quality of life and, thus, measures, compares, and values health status across disease areas (Devlin and Brooks, 2017).The EQ-5D was developed in 1990, covering five dimensions: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression.It is based on the assumption that health status can be modeled on a unidimensional continuum and represented by a single index score (Gudex, 2005).Furthermore, the instrument allows for calculating quality-adjusted life-years (QALYS) reflecting the quality and quantity of life to be combined into a single index (Prieto and Sacristán, 2003).The index score is based on population norms used as reference values (Sørensen et al., 2009), and QALYS are, among other things, valuable for recommendations of drug effects (Council], 2020).Initially, each dimension had three levels (e.g., No pain or discomfort, Moderate pain or discomfort, or Extreme pain or discomfort).However, due to ceiling effect challenges, a five-level version was introduced in 2009 (Devlin and Brooks, 2017) with an expected increased discriminative capacity and sensitivity to change compared to the EQ-5D-3 L as well as smaller ceiling effects (Herdman et al., 2011).The Danish version of EQ-5D-3 L was developed and ratified in 1996 (Sørensen et al., 2009), and the five-level version (5 L) in 2009 demonstrated valid redistribution, reduced ceiling, and improved discriminatory power (Janssen et al., 2013).
To decide on their answers to questionnaire items, co-operative respondents presumably opt for the answer options that they think best represent the facts about the questions they are asked.When questions and answer options are expressed in ordinary language, such choices, to a large extent, depend on the lexical meanings of the expressions used.Lexical meaning, however, is not the only factor affecting how answer options are interpreted.As linguistic pragmatics explain, a speaker's use and interpretation of language are heavily influenced by conversational norms and the assumption that others rely on his/her conformity with such norms to interpret his/her conversational contributions (Grice, 1989, Huang, 2017, Levinson, 1983).
Our aim was to examine how pragmatic dynamics may influence answers to the EQ-5D-5 L in items where the available answer options have a disjunctive form.Rather than focusing on a single parameter that might affect responses, we considered the possible effects of both scalar implicatures (Horn, 1972, Horn, 2006) and less predictable pragmatic inferences.
1. Do not say what you believe to be false 2. Do not say that for which you lack evidence QUANTITY: 1. Make your contribution as informative as is required (for the current purposes of the exchange) 2. Do not make your contribution more informative than is required RELATION: Be relevant.
Traditional Griceans and more recent Neo-Gricean approaches (Huang, 2017) argue that these norms, or closely adjacent norms, are crucial (Horn, 2004, Levinson, 2000) to linguistic communication.For example, if someone says that he/she is hungry and asks for the way to a diner, and you know that the nearest diner closed for the night, conformity with CP will prohibit you from giving him/her the directions to the nearby diner without informing them that it is closed.If you did, in fact, give your interlocutor the directions without further qualification, the presumption that you are observing CP would lead them to the mistaken conclusion that you have no more relevant information to convey (Hansen et al., 2017).
Your answer should reflect that the nearest diner is closed since this would be more cooperative given the purpose of the conversation.Beliefs about what your interlocutor needs to know affects how you answer his/her question.Pragmatic dynamics also influence answers to questionnaires (Clark andSchober, 1992, Schwarz, 1996).Respondents have been shown to rely on conversational norms for disambiguation of ambiguous questions (Schwarz, 1995) as well as other aspects of questionnaire interpretation (Schwarz and Hippler, 1991).
Pragmatic dynamics, however, are not the only source of contextual effects on linguistic communication.The significance of context also comes from semantic context-sensitivity.Ordinary language is full of context-sensitive expressions that have their meaning on a particular occasion determined by features of the utterance context.On the standard account of gradable adjectives, for example, such terms are context-sensitive (Kennedy, 1999).According to standard account, the meaning of a gradable adjective, like "little", changes because shifting the context may alter the threshold degree of smallness for qualifying as little.That an administered amount of morphine is "little" means one thing when the patient is a large man with extensive third-degree burns but plausibly means something else when the patient is a small child.Hence, semantic context-sensitivity potentially affects the interpretation of answer options in any questionnaire using gradable adjectives.

Analysis
To examine the interconnected dynamics of lexical meaning, semantic context-sensitivity, and pragmatics in the Danish EQ-5D, we employed a three-step analysis.
The first step of the analysis examined the lexical meanings of the answer options belonging to a questionnaire item to describe how they relate semantically to each other and to the properties that the items are intended to measure.This part of the analysis served to identify what we refer to as compatibilities and omissions.A compatibility occurs when a questionnaire item only allows a single answer, although two or more answer options under the item may be true simultaneously.Consider, for example, the question "How tall are you?" with the answer options "More than 0 cm and less than or equal to100 cm", "More than 150 cm and less than or equal to 200 cm", "More than a 150 cm and less than or equal to 220 cm", "More than 220 cm".Because the second and third answer options may both be true simultaneously, this is an example of compatibility.The example also illustrates that there is always compatibility when the truth of one answer option entails the truth of another.An omission occurs when the lexical meanings of answer options under a questionnaire item are such that there could be situations where no answer option is true on a literal interpretation.In the example above, there is an omission because none of the available answer options would be true of a person with a height between 100 and 150 cm.Both compatibilities and omissions may have the consequence that respondents cannot rely solely on lexical meaning when instructed to find the most suitable answer under a questionnaire item.Accordingly, respondents confronting compatibilities or omissions are likely to sometimes draw on pragmatic norms to identify the most appropriate answer option available.
The second step of our analysis examined how conversational maxims might affect a respondent's interpretation of answer options incorporating a compatibility.
The third step examined the same question with respect to omissions and also considered how the context of the questionnaire might counteract the problem with omissions by shifting the meaning of context-sensitive expressions.

Ethics statement
As this study presents a theoretical analysis and no human subjects were included, no approval was required.

Results
Three items in the EQ-5D have disjunctive answer options: Item 2, 4, and 5.We focus on item 4 to report the results of our analysis.Rather than the wording from the English version of the questionnaire, we provide our translation of the Danish version to best capture the features of the Danish EQ-5D.The translations are presented in brackets surrounded by double quotation marks.
If levels of pain and discomfort were always perfectly aligned, the analysis of compatibilities and omissions in item 4 would be relatively simple.On this assumption, a lexical analysis can ignore the disjunctive form of the answer options, and treat them exclusively as questions about pain, because a respondent's true answer about the disjunct concerning pain would correspond to her true answer about the disjunction.Given the assumption that pain, and discomfort are aligned, our analysis concludes that the first answer option is incompatible with all the rest.In the second option, "Lidt smerter" ("Little pain") has a semantic link to a scale ordering pains by their littleness, with the degree of littleness closest to zero as the maximum value.This semantic relation gives "Jeg har lidt smerter" the same meaning as "I have pain (at least as small as) little pains"."Moderat" (English "Moderate") has a similar link to a scale, which gives "Jeg har moderate smerter" the same meaning as "I have pains (at least as small as) moderate pains".Hence, because the threshold degree of littleness for qualifying as moderate is lower than the threshold degree of littleness for qualifying as little, a true answer to the second option entails a true answer to the third.The lexical meaning of "Staerke smerter" ("Strong pains") relates the fourth answer option to a scale ranking pains by order of their degree with the most extreme pain as the maximal value.The expression "Staerke smerter" means the same as "(pains as least as strong as) strong pains".So, the fifth answer option "Jeg har ekstreme smerter" has the meaning of "I have (pain at least as great as) extreme pain" and "Jeg har staerke smerter" means the same as "I have (pain at least as great as) strong pain", the fifth answer option entails the fourth because extreme pain is at least as strong as strong pain.
In addition, because a respondent's level of pain and discomfort could simultaneously be stronger than moderate but less than strong, the item also has an omission, if a respondent's pain and discomfort are at identical levels.
Things are more complicated regarding situations where a respondent's levels of pain and discomfort come apart.Whereas any degree of pain may be assumed to entail the same degree of discomfort, there is no entailment in the opposite direction.Nausea or dizziness, for example, may involve extreme discomfort with little or no related pain.Thus, every answer option that reports a non-zero level of discomfort is compatible with every answer option that reports a lower degree of pain.The truth of "Jeg har ekstreme smerter eller ubehag" is compatible with all other answer options under the fourth item because the truth of its second disjunct "Jeg har ekstremt ubehag" ("I have extreme discomfort") is compatible with the truth of the first disjunct in all answer options.Put differently, the truth of the fifth answer option permits the truth of all the remaining disjunctive answer options because one 1 3 might suffer extreme discomfort together with strong pains that are not extreme, moderate pains, little pain, and absence of pain, and a disjunction is true whenever one of its disjuncts is true (Aloni, 2016).Analogously, the truth of the fourth answer option permits the truth of the third, second, and first answer options because a respondent may experience strong discomfort in combination with any of the following: moderate pain, little pain, or absence of pain.Similarly, the truth of the item's third answer option permits the truth of the second and first options, and the truth of the second option permits the truth of the first.None of these compatibilities are due to entailments because facts about pain levels do not follow from facts about discomfort.

Conversational norms and pragmatic effects
With respect to the effects of pragmatic norms on answer choices in item 4, the respondent's judgments about his/her levels of pain and discomfort are crucial.The key difference is between cases where the respondent judges that his/her level of pain is the same as his/her level of discomfort (relative to the categories of the questionnaire) and where he/she does not.Table 1 shows which answer options may be true when the levels of discomfort and pain are identical.
In the situations represented in Table 1, pragmatic dynamics are likely to affect how a respondent decides between compatible answer options.In these cases, the second option "Lidt smerte eller ubehag" ("Little pain or discomfort") is compatible with the third option "Moderat smerte eller ubehag" ("Moderate pain or discomfort") but also entails that the third option is true.In addition, because confirmation of the third option does not entail that the second option is true, the second option is more informative than the third.Therefore, if both the third and second options are true of a respondent, the first maxim of Quantity enjoins the respondent to opt for the second option to comply with the co-operative principle.Consequently, if the third answer option is true, pragmatic norms imply that selecting this option is only pragmatically permissible when the respondent is not in a position to affirm the second option.Assuming that pragmatic norms remain in place, scalar implicatures related to informativity (Janssen et al., 2013, Grice, 1989, Huang, 2017), therefore, determine what the respective answer choices communicate.
Things are different regarding respondents who judge that their level of discomfort exceeds their level of pain.Table 2 provides a schematic overview of the answer options that may be true simultaneously for such respondents: In several of these situations, there are true answer options that are not entailed by another true option.These situations are represented by combination 2, 3, 4, 5, 6, 7, 9, and 10.The absence of an entailment relation implies that the Quantity maxim does not suffice to resolve which answer a respondent should opt for.Consider, for example, the situation represented by row 3 where a respondent judges his/her discomfort to be extreme but considers his/ her pain to be little.In such a case, option 5 is more informative than option 4 because the former entails the latter, and option 2 is more informative than option 3 because the truth of option 2 entails the truth of option 3.But there is no entailment from option 5 to option 2 or vice versa.Hence, the respondent is confronted with a problem that neither Quality nor Quantity resolves.
The only maxim that may be helpful in this predicament is Relevance.A respondent's decision about the appropriate answer may be aided by his/her assessment as to which information is most relevant to the addressee.Which option Relevance recommends, however, depends on the respondent's construal of the questionnaire item's communicative purpose.Although we can speculate about a respondent's understanding of the item's communicative purpose, conversational norms hence leave it an open question what a respondent with extreme discomfort and little pain would answer to item 4. The same kind of problem arises for respondents in the situations represented by rows 2, 3, 4, 5, 6, 7, 9, and 10 in Table 2.
Depending on how the respondent construes the questionnaire's purpose, Relevance may conflict with Quantity.Consider a respondent who assumes that the questionnaire is primarily intended to capture the maximum extent to which a person is affected by either discomfort or pain.Given this understanding of the purpose of the questionnaire, Relevance should lead him/her to choose option 5 in the situation represented by row 3.But if the same assumption is upheld with respect to a respondent regarding the situations represented by row 8, there is a conflict between the recommendations of the Relevance and Quantity maxims.Relevance would suggest that the respondent should choose option 3 to avoid signaling that both pain and discomfort are minor.In contrast, because the truth of option 2 entails the truth of option 3, Quantity would suggest that the respondent should choose option 2 to avoid indicating that he/she is not in a position to affirm option 2. Accordingly, answering item 2 of the EQ-5D may require a respondent to resolve a tension between conflicting recommendations from different pragmatic norms.With other conceptions of the questionnaire's purpose, conflicts between Relevance and other maxims may arise in situations represented by other rows in Table 2.
There are also situations in which a respondent's level of discomfort is greater than the level of pain while either discomfort or pain is at a level below strong but above moderate.These situations are represented in Table 3.Some of these situations may also involve conflicts between Relevance and other maxims.In the situation represented by row 4, both answer options 2 and 3 will be true, whereas option 2 will be more informative than option 3. Option 3, however, would be closer to a true representation of the respondent's level of discomfort.If the respondent does not experience any pain, this tension will be even higher because there will be an even more significant difference between the only true answer, option 1, and the answer option that best represents the respondent's level of discomfort, either options 3 or 4. Furthermore, if a respondent in either of these situations assumes that it is important for the addressee that reported answers do not underestimate the degree of either discomfort or pain that is highest for the respondent, Relevance might encourage him/her to answer "Staerk smerte eller ubehag" ("Strong pain or discomfort"), although this option is false, and whether available answer options are true.In the situation represented by row 1 in Table 3, the potential conflict confronting the respondent is between selecting option 5 to ensure truthfulness or abandoning truthfulness to somehow take into account that the level of pain is noticeably lower than the degree of discomfort.
It is possible, however, that the complexity of choosing an answer in the situations represented by Table 3 will be mitigated somewhat by contextual adjustments of meaning.If such modulation adjusts the extension of either "Staerk smerte eller ubehag " or "Moderat smerte eller ubehag" to eliminate the gap between the two categories, the subsequent interpretative circumstances will be akin to those represented by row 2 or 3 in Tables 1, or row 1, 2, 5, 6, 7, 8, or 9 in Table 2.But although this dynamic might reduce interpretative complexity considerable interpretative obstacles would remain for some respondents.

Omissions and context-sensitivity
There is also an omission in the fourth item.It is possible for a respondent to simultaneously have pain that is less than strong but worse than moderate, while experiencing discomfort, which is less than strong but worse than moderate.There may hence be cases where none of the item's five answer options is true about the respondent.
None of the Gricean maxims determine how a respondent in such a predicament should respond.Considering the central significance of truthfulness in communication (Dal Negro  , 2016, , Mokkink et al., 2010)), a likely course of action would be for the respondent to decide his/her answer by choosing the option that he/she considers closest to the truth.In that case, he/she would be observing the first maxim of Quality to the highest extent possible, given their circumstances.The other maxim most likely to affect a respondent's decision is relevance because of how close the relevance maxim is to the overarching cooperative principal (CP).How this maxim might affect a respondent's choice of answer cannot be determined based on basic pragmatic principles because its influence depends on how the respondent construes the purpose of the information exchange mediated by the questionnaire.
It is also possible that the semantic context sensitivity of "Staerk" and "Moderat" ("Strong") and ("Moderate") affects how the omission in item 4 affects a respondent's answer.Because the meaning of a context-sensitive term is determined by its context of use, the questionnaire may shift the meanings of "Staerk" and "Moderat" to close the logical gap between answer options 4 and 3. Thus, although the linguistically encoded meanings of the questionnaire's terms do not generally exclude the possibility of a level of impairment between the categories "Staerk" and "Moderat", the contextually determined meanings of the terms in the EQ-5D might preclude this possibility.This situation might occur either by lowering the level of impairment that qualifies as "Staerk" or raising the level of impairment that qualifies as "Moderat".

Scope of analysis
Whether this analysis of item 4 can be extended to the other items with disjunctive questions (i.e., items 2 and 5) depends on the relations between the disjuncts in their answer options.Item 2 asks about the degree to which the respondent has difficulty washing themselves or getting dressed, whereas item 5 enquires about the degree to which a respondent experiences anxiety or depression.If a respondent's difficulty with getting washed and getting dressed are always on the same level, and depression and anxiety are always of similar severity, then the situations a respondent might be in when answering either item correspond to those in Table 1 with a line added to represent situations where both properties are present to a degree less than strong but more than moderate (for this possibility, please see supplementary file, Appendix 1).In contrast, if any degree of depression is compatible with any degree of anxiety that does not exceed the degree of depression, or vice versa, then the possible communicative circumstances of a respondent will correspond to a combination of Tables 1 and 2, and 3 with a line added to represent situations where both properties are present to a degree less than strong but more than moderate (for this possibility, please see supplementary file, Appendix 2).Finally, if a person's degree of depression and anxiety do not restrict each other at all, the analog of Table 1, as well as two analogs of Table 2 and two analogs of Table 3, and a further line added to represent situations where both properties are present to a degree less than strong but more than moderate, will be required to represent the situations that a respondent might be in when deciding how to answer item 5 of the EQ-5D (for this possibility, please see supplementary file, Appendix 3).

Discussion
A questionnaire is a widely used, simple, and cheap instrument for research concerning both specific and generic populations' behaviors, attitudes, and beliefs, and it is crucial that the specific questionnaire is constructed to produce valid and trustworthy results.The validity of questionnaires in terms of the comprehension of their questions, and the repeatability of the responses, might still be challenged (Dal Negro and Povero, 2016).Rigorous (statistical) validation of questionnaires has been recommended for decades, and the inclusion of content validity and cross-cultural adaptation has been internationally recommended and acknowledged at least since the description of the COSMIN guidelines in 2010 (Mokkink et al., 2010).Such validation includes cognitive interviews to investigate how responders interpret questions and how they choose an answer.This process of understanding the question and choosing the answer is described in four elements: the responder has to comprehend the question, retrieve the necessary information to answer, decide which information is required to answer, and, finally, choose the adequate answer (de Vet et al., 2011).Cognitive interviews are considered means to revealing these processes and, thus, contribute to the meaningfulness of a questionnaire as judged by the responders.There is, however, a risk that this process ignores the fact that responders (unconsciously) tend to answer in a socially desirable manner (fake good) (Vésteinsdóttir et al., 2019, Brenner andDeLamater, 2016), and sometimes seek to give answers conforming to what they think is expected from them (Bergen and Labonté, 2020).
Linguistic analysis can serve to identify further pitfalls in questionnaires that might otherwise remain unrecognized despite a thorough and rigorous validation process.Empirical results from experimental pragmatics (Noveck, 2018, Geurts andvan tiel, 2013) support the analytical framework from Grice, and other aspects of the framework's significance to survey methodology are already well documented (Schwarz, 1996, Schwarz, 1995, Schwarz and Hippler, 1991).Despite thorough semantic and linguistic testing during the development and cultural adaptation of the EQ-5D (Rabin et al., 2014, Herdman et al., 1998), we found several places where the communication of the EQ-5D is complicated by compatibilities or omissions.Because of the disjunctive form of answer options respondents relying on conversational norms to decide between compatible answer options in item 2, 4, and 5 may be required to find a balance between conflicting maxims based on their personal understanding of a specific item's purpose.A respondent answering the EQ-5D might hence be influenced by well-attested pragmatic aspects of linguistic communication (Rabin et al., 2014, Herdman et al., 1998) that are not directly related to the properties that the questionnaire is intended to measure.
With respect to how such concerns about a questionnaire might be mitigated or avoided by adjusting its language we propose two main suggestions to consider.First, answer options with disjunctive structure may be modified to avoid a disjunction, either by removing a disjunct or by replacing each disjunction with a non-disjunctive construction.Second, problems related to omissions might be avoided by using answer options without gaps between them.In the SF-36, for example, some items appear to avoid omissions by employing answer options without such gaps (Petersen and Nørgaard, 2022).Whether and how the EQ-5D could be amended by integrating these suggestions, however, is far from obvious, and pursuing the issue goes beyond our aim here.

Limitations
We acknowledge that our analysis is disrupted from actual practice and that theoretical analyses only have an indirect relation to practice.On the other hand, such analyses may help to gain insight into subconscious mental processing crucial to the function of the questionnaire, that cannot be revealed by cognitive interviewing.They also avoid the concern that the procedure and setting of a cognitive interview might influence how a respondents reason about a questionnaire and answers its questions (Wilson et al., 1996).
The analysis is restricted to the Danish questionnaire, which could be a limitation of the study considering the small linguistic area.There is, however, no reason to believe that such an analysis would have significantly different results if applied to other languages.
Another limitation concerns the analysis of how a respondent might deal with a situation where no answer option is true of a correspondent because of an omission.While Gricean pragmatics suggest factors that might shape a respondent's answer in such a situation, Gricean norms do not suffice to explain how a respondent in this predicament decides on an answer.
Our purpose, moreover, is not to suggest substantial revisions of an internationally acknowledged and widely used tool but to contribute to the understanding of processes behind questionnaire answers-and to encourage further attention to the linguistic aspects of new instruments.Ideally, the type of analysis employed may aid the prequalification of questionnaires prior to more labor-extensive and costly empirical pre-testing of surface validity.We readily admit, however, that our analysis does not fully specify how a questionnaire's answer options should be formulated to avoid problems regarding the wording of questionnaire items.The aim here is limited to drawing attention to the fact that even data from rigorously validated questionnaires should be interpreted cautiously.

Conclusion
Our analysis showed that the questionnaire has several compatibilities and at least one omission.Because of how answer options are structured respondents cannot rely solely on lexical meanings to choose between them.In addition, respondents who try to choose their answer by relying on Gricean maxims will have to depend on their personal understanding of details concerning the questionnaire's purpose and may have to weigh how conflicting norms should be balanced.These issues complicate the questionnaire's communication considerably and renders its precise interpretation unpredictable.
The purpose of this paper, however, is not to argue that the EQ-5D in its present form cannot be used for its intended purpose.Our purpose is rather to demonstrate that certain linguistic factors that tend to be underappreciated in survey research may have a significant influence on how respondents answer the questionnaire.Our hope is also to call attention to the methodological potential of a type of linguistic analysis that may be highly valuable to survey methodology in health-and quality-of-life research, particularly in the prequalification phase prior to empirical pilot testing.Ideally, our analysis would help questionnaire developers and other researchers in the field see how their work might benefit from drawing on linguistic theorizing about lexical semantics and pragmatic dynamics.

Lessons for practice
• A questionnaire is a widely used, simple, and cheap instrument for research concerning both specific and generic populations' behaviors, attitudes, and beliefs, and it is crucial that the specific questionnaire is constructed to produce valid and trustworthy results.
• When questionnaire answer options are given in ordinary language terms, a respondent's interpretation and selection of answer options depends on both lexical meaning and pragmatic norms governing linguistic communication.
• Certain linguistic factors that tend to be underappreciated in survey research may have a significant influence on how respondents answer the questionnaire and may be important to how their answers should be interpreted.

Table 1
True answer options in item 4 when the level of discomfort equals the level of pain

Table 2
True answer options in item 4 when the level of discomfort exceeds the level of pain

Table 3
True answers in item 4 when the level of discomfort exceeds the level of pain, and discomfort or pain is at a level below strong but above moderate