Explaining question framing effects

This study demonstrates the presence of question wording and order effects in the context of a survey of attitudes towards the legalization of AD. These effects were not only statistically significant, but also in most cases sufficiently large to be of considerable practical importance.

As predicted, the contextual version (#2) of the questionnaire produced greater assent to the legalization of AD than did the concept-focused version (#1). In contrast to the latter version, the contextual version evokes an image of a particular patient portrayed in ways that engage the respondent’s sympathy (e.g., “in great pain”), while at the same time underscoring the rationale for AD (“avoid great suffering”) and reassuring that decisions will not be taken lightly (“thorough evaluation”). The introduction to the contextual version also describes AD in a way that is likely to have positive connotations, namely as “self-determined ending of life”, and avoids the concept-focused version’s terms “intentional killing” and “aid in a person’s suicide” and the negative connotations associated therewith. On our interpretation, these features are most likely responsible for the wording effects observed. Notably, the effect carries over to the subsequent questions; assent to the proposals of AD for mental illness (Q4) and tiredness of life (Q5) is significantly higher in the contextual version.

The variations in question order exhibited a contrast effect, with higher assent to proposals to legalize AD for terminal and chronic disease (Q1-3) when the most controversial proposals (Q4-5) were presented first. The suggestion that AD could be offered even for individuals with mental illness or people who are merely tired of life appearently made the proposal to legalize AD for terminal and chronic disease less controversial and more socially acceptable. Rejecting the former controversial proposals (Q4-5) would mean that the respondent could still accept the latter (Q1-3) and yet perceive their own position as nuanced, avoiding both extremes. Similarly, the respondent who was first exposed to the less controversial questions Q1-3 might have experienced a need to distance him- or herself from the more extreme proposals of Q4-5.

Question framing effects are of practical importance

We find the size of the wording and order effects to be sufficiently large to be important for policy formation and public debate, as well as for attitude research. First, it is well known to policy makers and activists that framing the issue in carefully chosen terms and colouring it with evocative metaphors, is instrumental in shaping the public’s views [17]. The present study demonstrates such framing effects for the topic of AD, and suggests that effect sizes are rather substantial. From this it follows that proponents of AD are likely to win more support for their cause if they portray actual, suffering patients and by invoking the normative language of self-determination. In contrast, opponents of AD appear to benefit from using the Dutch AD definitions which include the terms “killing” and “suicide”, as well as discussing the topic detached from stories of individual patients in extreme circumstances.

A further interpretation is that AD proponents would most likely benefit from sharply demarcating their own proposal for legislation from more extreme proposals that would include, for instance, patients with mental illnesses or those who suffer from (mere) tiredness of life. AD proponents may thus be able to invoke contrast effects deliberately: by explicitly rejecting the more extreme AD proposals, their own position appears more nuanced and responsible. For the camp of AD opponents, an option which presents itself is trying to undermine the contrast effect by, for instance, portraying the assent to AD even in a carefully circumscribed set of cases as merely a first step onto an inevitably slippery slope.

Second, in demonstrating that framing and order effects are of significant size, the present study can be read as an implicit critique of most previous attitude surveys on AD. There is no denying that many surveys are quite naïve and simplistic in their presentation of the issue, and often also in using just one or two questions and/or constraining or forcing respondents into answering either “yes” or “no”, thereby missing out on nuances and ambivalence among respondents.

We believe that more attitude research in bioethics should be done by way of survey experiments. Such study designs enable multiple perspectives on the topic in question within the same study. There are, however, two drawbacks with such designs that should be mentioned: firstly, study design becomes more complex; secondly, large sample sizes, as in the present survey, are needed to achieve statistical power.

What does the apparent malleability of people’s attitudes towards AD disclose about these attitudes? First of all, the presence of significant framing effects is not unique to the issue of AD; such effects have been demonstrated on a wide range of topics [8]. It is debated whether or not people’s susceptibility to framing is a good or a bad thing [8]. On the one hand, the stability, depth and consistency of people’s attitudes may be questioned in light of large framing effects; can such attitudes really be said to be informed and well thought through, and if not, should they have any political significance? On the other hand, susceptibility to framing shows that respondents are sensitive to arguments and context; such sensitivity is an important human capacity as well as key in democratic processes. The significant framing effects may also indicate that large portions of the public have not engaged thoroughly with the issue of AD. As Chong and Druckman state, “Theoretically, we expect that framing effects diminish with active engagement with issues. In particular, biased representations of issues should be less influential as citizens become exposed to the full array of alternative arguments” [8]. In light of this contention, an hypothesis worth exploring in future research is that groups such as health professionals, politicians and professional bioethicists are less susceptible to framing effects than the general public.

Attitudes towards AD among the Norwegian public

Answers to the four questionnaire versions cannot be straightforwardly pooled and taken to represent Norwegians’ views on AD. If pooled, it must be kept in mind that the resulting average scores stem from questionnaires with significant differences. Notwithstanding this caveat, however, we would argue that exposing respondents to different ways of posing the key questions constitutes a kind of method triangulation in which the question at hand – what are the public’s views on AD? – is perceived from several angles; arguably, a more detailed picture of those views then emerges than if a single method was employed. In a similar vein, quantitative research on opinions on AD can and should be complemented by qualitative research, which has the potential for enriching the account of such attitudes with depth and complexity [18, 19].

If we proceed to combine the results on the four questionnaire versions, keeping in mind the way in which the results have arisen, it appears that majorities support the legalization of AD for terminal and chronic disease, ranging from a preponderant majority of 75.8 % for PAS for terminal disease to a slight majority of 51.4 % for AD for chronic disease. Few support AD in other situations. Because previous Norwegian studies differ radically from ours and typically are flawed in important respects (e.g., only one or a few questions, key terms undefined, biased question wording) [4], our results are not directly comparable. Our finding that the legalization of AD is supported by a majority is however in line with previous surveys.

Most attitude surveys carried out in other Western countries show that majorities support AD, including in most countries where AD is presently illegal [16]. Our findings reveal that the level of support among Norwegians varies across demographic subgroups in a way partly different from what has been shown to be the case in research in other countries. We find that the respondents with the highest level of education (>3 years of higher education) were less accepting of PAS than groups with lower levels of education. For questions on euthanasia and AD for chronic disease, the higher educated were also less accepting than respondents with a medium level of education, but not significantly different from the group with the least education. The finding that those most educated tend to be less accepting contradicts the tendency in a recent survey on 15 other countries [16].

We also observe that the younger groups (16–24, 25–34, and 35–44) were more likely to support AD than the oldest group (55+). This may indicate a cohort effect in which support for AD will increase as time goes by. The finding that Christian and Muslim respondents were less positive towards AD than the non-religious, was expected. However, we note that there is still considerable support for AD among Christians (Table 5; e.g., mean score 3.62 on Q1).

Potential limitations

The study design does not enable separate assessment of the effect of the different questionnaire introductions on the one hand, and of the different question wording on the other: the effects measured are for introductions and questions combined.

There is a difference between the concept- and the context-focused questionnaire versions in that only the latter mentions pain and suffering (cf. Q1-3, Tables 2 and 3). This difference, some might argue, is substantial and does not only involve wording effects. However, we do not agree with this objection. The context-focused version’s portrayal of PAS/E (Q1-2) involves a dying patient in great pain, yet this is precisely the kind of patient for whom a majority of the public would want PAS/E to be available, as research has repeatedly demonstrated. In that sense it is a paradigmatic example, and thus constitutes the background against which many respondents will view the concept-focused questionnaire version too. We accordingly find the propositions to be sufficently equivalent across the two questionnaire versions: They portray the legalization of PAS or E for terminal illness, or AD for chronic disease, respectively.

The response rate is very low at 13.5 %, raising the issue of a possible non-response bias. However, the phenomenon of low and declining response rates is a problem that affects population surveys all over the Western world [20]. Much effort has been put into studying consequences of nonresponse for study validity. Detailed analyses indicate that surveys may well be representative in spite of very low response rates [21, 22]. Furthermore, surveys of respondent attitudes, such as the present, are less at risk for nonresponse bias than surveys of respondent activities [21]. Our survey has attempted to mitigate nonresponse bias through selective invitations to balance respondents on demographic paramenters. Still, we cannot rule out that our results are influenced by a nonresponse bias, for instance in that respondents who are less interested in bioethical issues might have been less likely to complete the survey.