This project sought to identify ways of measuring lay beliefs about the relationship of genes and behavior that did not face the limitations identified in previous measurement attempts (presumed to be primarily ceiling effects). As we began the project, we assumed that it would be possible to use standard scale-building methods to construct a nonquantitatively explicit, reliable and unidimensional scale that would capture the desired content dimensions of the gene-behavior interaction concept (interaction, amplificative relationship and transitive relationship of healthy/unhealthy). We presumed that such a scale would be roughly isomorphic with direct measures that included quantitative estimates. We pursued measurement strategies in both quantitatively explicit and nonquantitatively explicit modes and produced independent measurement tools (the ‘Gene-Environment Interaction Scale’ and the additive/amplificative test question pair). However, although the scale has good psychometric properties, it is only weakly correlated with the latter, more direct quantitative measurement tool.
This research thereby has identified several challenges for the development of optimal means of communicating with the public about gene-behavior interaction. This discussion will first address the issue of the beliefs as measured and then address issues related to the need for future scale development (including limitations of this reported research project).
This research suggests that substantial segments of the public harbor different views of the relationship between genes and behavior. As measured via the gain framed additive/amplificative test question (R.J.), about a third of the population endorses an amplificative model, but the largest segment appears to hold a mathematical model that is additive rather than interactive, and there is also a small minority that endorses a sub-additive model. As measured by the scale, on the other hand, the mean answer leans toward acceptance of an amplificative account. Both of these results are different from those of the prior research garnered using the contrasts between numeric scales, which identified the sub-additive model as the dominant understanding. Our results, thus, support French et al.'s [26
] suggestion that the earlier results were produced by a ceiling effect inherent to the measurement approach used there. The present results also are consonant with the qualitative research on lay people's discourse, which showed a predominant separate tracks model but some interactive metaphors [25
It is also interesting that our research showed that higher education is not associated with a more interactive model, given that many studies have shown other facets of genetic knowledge to be associated with higher education [42
]. Because prior research has shown that even highly educated people tend to have low numeracy skills [43
], future research might explicitly measure numeracy to more precisely assess whether differences exist along these dimensions.
Second, the results suggest that the mathematical model and underlying causal accounting employed by many lay people may differ depending on whether the issue is understood as ‘gaining’ good health from good genes and good behaviors or ‘losing’ good health from weak genes and unhealthy behaviors. This may result from a general difference between gain and loss framing, but it also may be more specifically related to a tendency to understand genes primarily as causal factors for bad health and healthy behaviors as causal factors for good health outcomes. This difference may arise because good health is taken as normative and as a product of normal or typical genes, but this may also be related to a model in which genes are more powerful at causing bad health and behavior is more powerful in causing good health. The former explanation is consistent with a well-documented health optimism bias [44
], but the latter may also be true and perhaps may be a product of news coverage, which typically links genes with disease [45
Whether the former or latter (or both) explanations are correct may have implications for presenting information from personalized genetic testing in order to motivate behavior and for modes of teaching of gene-environment interaction. If this is simply a gain/loss frame issue, then the general recommendations for use of gain and loss framing would apply [46
]. If, on the other hand, it results from differential assignment of outcomes to genes and to behaviors for disease as opposed to for normal health, this issue may have to be explicitly addressed in teaching the gene-behavior interaction concept and in messages about the implications and utilities of personalized genetic tests.
Such implications, however, presume that the measured difference between gain and loss messages is real, rather than an artifact of inadequate measurement approaches. We suggest that the consistency of this finding with previous qualitative findings on the variation of health accounting, resulting from motivational effects [25
] and with the larger research tradition on gain/loss framing [46
], weighs in favor of the conclusion that the questions are detecting a substantive difference in response. The fact that both the recommended additive/amplificative test questions (R.J./C.W.) detect it and that the less parsimonious ‘Emmeline/Sarah’ questions using a different format (ranking) detect it as well, also weighs in favor of such an interpretation. However, one limitation may be that the gain framed question (which always followed the loss framed question, although other questions were randomized in order) does not explicitly use the 20%, 20 numbers in the stem. A version of that question using those numbers and separated from the loss framed (R.J.) question should be tested to ensure that the difference in that question pair is not a result of the exclusion of the numbers in that case.
Third, the Gene-Behavior Interaction Beliefs Scale reported here meets the psychometric tests for good scale qualities. However, the level of public agreement with the gene-environment interaction concept as measured by the scale is greater than the level of agreement as measured by the more concrete additive/amplificative test questions, and the correlation between the two is low (approximate r < 0.10). It may be that this weak correlation reflects an insufficiency in one or the other measurement approaches. Further research to validate the 2 scales may reveal ways in which one or the other is superior. Validation efforts might proceed via a close comparison of the scales with lay qualitative responses through cognitive interviewing. It may also prove that one or the other scales is more predictive of health behavior change, which would make it more useful, even if both detect substantive features of belief structures. Thus, using effectiveness rather than referential validity might provide the most desirable approach. Comparing the 2 approaches in studies assessing impact of genetic information on behavioral intentions would, in that case, provide the appropriate approach to validation.
It may also be that the inclusion of a specific disease (heart disease) in the additive/amplificative (R.J./C.W.) question pair and the reference to general diseases in the general scale is the cause of the difference. This possibility seems particularly worth exploring because existing surveys show sizeable differences in the role assigned to behavior and genes (independently) across different human outcome characteristics [13
]. It may be that the public generally affirms an amplificative model for gene-behavior relationships but in the case of heart disease specifically has an additive approach. Testing of the additive/amplificative (R.J./C.W.) question pair using different diseases would illuminate this possibility.
In spite of the need to test these alternate possibilities, there is some reason to believe that perhaps the 2 different measurement approaches reflect a real schism between globalized beliefs that affirm the notion of gene-environment interaction in the abstract, but which do not apply the concept in concrete situations. In seeking to account for that possibility, we have discovered that such schisms between precise and concrete concepts and general diffuse beliefs seem likely to be undergirded by differences in what have been identified as ‘qualitatively different representational frameworks’ associated with abstract and concrete knowledge [47
]. If this is the case, it has substantial consequences for the way in which teaching of the gene-behavior concept should go forward. Rather than concentrating on teaching the general idea (which the majority of people already appear to be able to affirm when it is presented to them abstractly), teaching may need to focus on applying the general idea to specific instances, including specific mathematical instances.
This line of explanation may also account for other message content choices. For example, it provides a plausible explanation for the superior performance of the ‘elevator’ metaphor over the ‘bridge’ metaphor in the research by Kaphingst et al. [29
], as the elevator metaphor draws on pre-existing linkages between mathematical concepts (numbered floors and concepts of acceleration and deceleration related to numbers through driving), whereas the appearance of holes and patches in the bridge metaphor does not draw on a neural semantic network that contains previously established links among mathematical and concrete applications. On the other hand, this discrepancy may also or merely reflect the well-documented challenges of relating risk estimates to personal health issues [48
]. Immediate research is needed to disentangle these potential causal factors.
Researchers are just making a beginning on understanding the complexities of human cognition necessary to translate the wealth of genetic information into forms that will allow people to maximize the utility of that information in their lives. This report provides some preliminary tools and findings toward that effort.