Why cultural predispositions matter & how to measure them: a fragment …

Here’s a piece of something I’m working on–the long-promised & coming-soon “vaccine risk-perception report.” This section discusses the “cultural predisposition” measurement strategy that I concluded would be most useful for the study. The method is different from the usual one, which involves identifying subjects’ risk predispositions with the two “cultural worldview” scales. I was going to make this scheme the basis of a  “MAPKIA!” contest in which players could make predictions relating to characteristics of the 4 risk-disposition groups featured here and their perceptions of risks other than the ones used to identify their members. But I decided to start by seeing what people thought of this framework in general. Indeed, maybe someone will make observations about it that can be used to test and refine the framework — creating the occassion for the even more exciting CCP game, “WSMD? JA!”

 C.  Cultural Cognition

1.  Why cultural predispositions matter, and how to measure them

Public contestation over societal risks is the exception rather than the norm.  Like the recent controversy over the HPV vaccine and the continuing one over climate change, such disputes can be both spectacular and consequential. But for every risk issue that generates this form of conflict, there are orders of magnitude more—from the safety of medical x-rays to the dangers of consuming raw milk, from the toxicity of exposure to asbestos to the harmlessness of exposure to cell phone radiation—where members of the public, and their democratically accountable representatives, converge on the best available scientific evidence without incident and hence without notice.

By empirical examination of instances in which technologies, public policies, and private behavior do and do not become the focus for conflict over decision-relevant science, it becomes possible to identify the signature attributes of the former. The presence or absence of such attributes can then be used to test whether a putative risk source (say, GM foods or nanotechnology) has become an object of genuine societal conflict or could (Finucane 2005; Kahan, Braman, Slovic, Gastil & Cohen 2009).

Such a test will not be perfect. But it will be more reliable than the casual impressions that observers form when exposed either to deliberately organized demonstrations of concern, which predictably generate disproportionate media coverage, or to spontaneous expressions of anxiety on the part of alarmed individuals, whose frequency in the population will appear inflated by virtue of the silence of the great many more who are untroubled. Because they admit of disciplined and focused testing, moreover, empirically grounded protocols admit of systematic refinement and calibration that impressionistic alternatives defiantly resist.

One of the signature attributes of genuine risk contestation, empirical study suggests, is the correlation of positions on them with membership in identity-defining affinity groups—cultural, political, or religious (Finucane 2005). Individuals tend to form their understandings of what is known to science inside of close-knit networks of individuals with whom they share experience and on whose support they depend. When diverse  groups of this sort disagree about some societal risk, their members will thus be exposed disproportionately to competing sources of information. Even more important, they will experience strong psychic pressure to form and persist in views associated with the particular groups to which they belong as a means of signaling their membership in and loyalty to it. Such entanglements portend deep and persistent divisions—ones likely to be relatively impervious to public education efforts and indeed likely to be magnified by the use of the very critical reasoning dispositions that are essential to genuine comprehension of scientific information (Kahan, Peters et al. 2012; Kahan 2013b; Kahan, Peters, Dawson & Slovic 2013).

These dynamics are the focus of the study of the cultural cognition of risk.  Research informed by this framework uses empirical methods to identify the characteristics of the affinity groups that orient ordinary members of the public with respect to decision-relevant science, the processes through which such orientation takes place, the conditions that can transform these same processes into sources of deep and persistent public conflict over risk, and measures that can be used to avoid or neutralize these conditions (Kahan 2012b).

Such groups are identified by methods that feature latent-variable measurement (Devellis 2012). The idea is that neither the groups nor the risk-perception dispositions they impart can be observed directly, so it is necessary instead to identify observable indicators that correlate with these phenomena and combine them into valid and reliable scales, which then can be used to measure their impact on particular risk perceptions.

One useful latent-variable measurement strategy characterizes individuals’ cultural outlooks with two orthogonal attitudinal scales—“hierarchy-egalitarianism” and “individualism-communitarianism.” Reflecting preferences for how society and other collective endeavors should be structured, the latent dispositions measured by these “cultural worldview” scales, it is posited, can be expected to vary systematically among the sorts of affinity groups in which individuals form their understandings of decision-relevant science. As a result, variance in the outlooks measured by the worldview scales can be used to test hypotheses about the extent and sources of public conflict over various risks, including environmental and public-health ones (Kahan 2012a; Kahan, Braman, Cohen, Gastil & slovic 2010).

This study used a variant of this “cultural worldview” strategy for measuring the group-based dispositions that generate risk conflicts: the “interpretive community” method (Leiserowitz  2005). Rather than using general attitudinal items, the interpretive community method measures individuals’ perceptions of various contested societal risks and forms latent-dispositions scales from these. The theory of cultural cognition posits—and empirical research corroborates—that conflicts over risk feature entanglement between membership in important affinity groups and competing positions on these issues.  If that is so, then positions on disputed risks can themselves be treated as reliable, observable indicators of membership in these groups—or “interpretive communities”—along with the unobservable, latent risk-perception dispositions that membership in them imparts.

The interpretive community-strategy would obviously be unhelpful for testing hypotheses relating to variation in the very risk perceptions (say, ones toward climate change) that had been used to construct the latent-predisposition scales. In that situation, the interdependence of the disposition measure (“feelings about climate change risks”) and the risk perception under investigation  (“concerns about climate change”) would inject a fatal source of endogeneity into any empirical study that seeks to treat the former as an explanation for or cause of the latter.

But where the risk perception in question is genuinely distinct from those that formed the disposition indicators, there will be no such endogeneity. Moreover, in that situation, interpretive-community scales will offer certain distinct advantages over latent-disposition measured formed by indicators based on general attitude scales (cultural, political, etc.) or other identifying characteristics associated with the relevant affinity groups.

Because they are measures of an unobserved latent variable, any indicator or set of them will reflect measurement error.  In assessing variance in public risk perceptions, then, the relative quality of any alternative latent-variable measurement scheme will thus consists in how faithfully and precisely it captures variance in the group-based dispositions that generate conflict over societal risks. “Political outlooks” might work fairly well, but “cultural worldviews” of the sort typically featured in cultural cognition research will do even better if they in fact capture variance in the motivating risk-perception dispositions in a more discerning manner. Other alternatives might be better still, particularly if they validly and reliably incorporate other characteristics that, in appropriate combinations,[1] indicate the relevant dispositions with even greater precision.

But if the latent disposition one wants to measure is one that has already been identified with signature forms of variance in certain perceived risks, then those risk perceptions themselves will always be more discerning indicators of the latent disposition in question than any independent combination of identifying characteristics.  No latent-variable measure constructed from those identifying characteristics will correlate as strongly with that risk-perception disposition as the pattern of risk perceptions that it in fact causes. Or stated differently, the covariance of the independent identifying characteristics with the latent-variable measure formed by aggregation of the subjects’ risk perceptions will in fact already reflect, with the maximum degree of precision that the data admits, the contribution that other those characteristics could have made to measuring that same disposition.

The utility of the interpretive-community strategy, then, will depend on the study objectives. Again, very little if anything can be learned by using a latent-disposition measure to explain variance in the very attitudes that are the indicators of it.  In addition, even when applied to a risk perception distinct from the ones used to form the latent risk-predisposition measures, an “interpretive community” strategy will likely furnish less explanatory insight than would a latent-variable measure formed with identifying characteristics that reflect a cogent hypothesis about which social influences are generating these dispositions and why.

But there are two research objectives for which the interpretive-community strategy is likely to be especially useful.  The first is to test whether a putative risk source provokes sensibilities associated with any of the familiar dispositions that generate conflict over decision-relevant science—or whether it is instead one of the vastly greater number of technologies, private activities, or public policies that do not. The other is to see whether particular stimuli—such as exposure to information that might be expected to suggest associations between a putative risk source and membership in important affinity groups—provokes varying risk perceptions among individuals who vary in regard to the cultural dispositions that such groups impart in their members.

Those are exactly the objectives of this study of childhood vaccine risks.  Accordingly, the interpretive community strategy was deemed to be the most useful one.

2. Interpretive communities and vaccine risks

Figure 14. Factor loadings of societal risk items. Factor analysis (unweighted least squares) revealed that responses to societal risk items formed two orthogonal factors corresponding to assessments of putative “public-safety” risks and putative “social-deviancy” risks, respectively. The two factors had eigenvalues of 4.1 and 1.9, respectively, and explained 61% of the variance in study subjects’ responses to the individual risk items.

Study subjects indicated their perceptions of a variety of risks in addition to ones relating to childhood vaccines—from climate change to exposure to second-hand cigarette smoke, from legalization of marijuana to private gun possession. These and other risks were selected because they are ones that are well-known to generate societal conflict—indeed, conflict among groups of individuals who subscribe to loosely defined cultural styles and whose positions on these putative hazards tend to come in recognizable packages.

Factor analysis confirmed that the measured risk perceptions—eleven in all—loaded on two orthogonal dimensions.  One of these consisted of perceptions of environmental risks, including climate change, nuclear power, toxic waste disposal, and fracking, as well as risks from hand-gun possession and second-hand cigarette smoke.  The second consisted of the perceived risks of legalizing marijuana, legalizing prostitution, and teaching high school students about birth control.

The factor scores associated with these two dimensions were labeled “PUBLIC SAFETY” and “SOCIAL DEVIANCY,” each of which was conceived of as a latent risk-disposition measure.  Support for the validity of treating them as such was their appropriate relationships, respectively, with the Hierarchy-egalitarianism and Individualism-communitarianism worldview scales, which in previous studies have been used to predict and test hypotheses relating to risk perceptions of the type featured in each factor.

Figure 15. Risk-perception disposition groups.  Scatter plot arrays study subjects with respect to the two latent risk-perception dispositions. Axes reflect subject scores on the indicated scales.

Because they are orthogonal, the two dimensions can be conceptualized as dividing the population into four interpretive communities (“ICs”): IC-α (“high public-safety” concern, “low social-deviancy”);  IC-β (“high public-safety,” “high social-deviancy); IC-γ (“low public-safety,” “low public-safety”); and IC-δ (“low public-safety,” “high social-deviancy”).  The intensity of the study subjects’ commitment to one or the other of these groups can be measured by their scores on the public-safety and societal-deviancy risk-perception scales.

Members of these groups vary in respect to individual characteristics such as cultural worldviews, political outlooks, religiosity, race, and gender.  IC-αs tend to be more “liberal” and identify more strongly with the Democratic Party,” and are uniformly “egalitarian” in their cultural outlooks. IC‑βs, who share the basic orientation of the IC-αs on risks associated with climate change and gun possession but not on ones associated with legalizing drugs and prostitution, are more religious and more African-American, and more likely to have a “communitarian” cultural outlook than IC-αs. IC-γs include many of the “white hierarchical and individualistic males” who drive the “white male effect” observed in the study of public risk perceptions (Finucane et al. 2000; Flynn et al. 1994; Kahan, Braman, Gastil, Slovic & Mertz 2007).  Like IC-βs, with whom they share concern over deviancy risks, IC-δs are more religious and communitarian; they are less male and less individualistic than IC- γs, too, but like members of that group, IC- δs are whiter, more conservative and Republican in their political outlooks, and more hierarchical in their cultural ones than are IC-βs.

These characteristics cohere with recognizable cultural styles known to disagree over issues like these (Leiserowitz 2005). Appropriate combinations of those characteristics, combined into alternative latent measures, could have predicted similar patterns of variance with respect to these risk perceptions, although not as strongly as the scales derived through a factor analysis of the covariance matrixes of the risk perception items themselves.

Vaccine-risk perceptions  . . .

References

Berry, W.D. & Feldman, S. Multiple Regression in Practice. (Sage Publications, Beverly Hills; 1985).

Cohen, J., Cohen, P., West, S.G. & Aiken, L.S. Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences, Edn. 3rd. (L. Erlbaum Associates, Mahwah, N.J.; 2003).

DeVellis, R.F. Scale Development : Theory and Applications, Edn. 3rd. (SAGE, Thousand Oaks, Calif.; 2012).

Finucane, M., Slovic, P., Mertz, C.K., Flynn, J. & Satterfield, T.A. Gender, Race, and Perceived Risk: The “White Male” Effect. Health, Risk, & Soc’y 3, 159-172 (2000).

Finucane, M.L. & Holup, J.L. Psychosocial and Cultural Factors Affecting the Perceived Risk of Genetically Modified Food: An Overview of the Literature. Social Science & Medicine 60, 1603-1612 (2005).

Flynn, J., Slovic, P. & Mertz, C.K. Gender, Race, and Perception of Environmental Health Risk. Risk Analysis 14, 1101-1108 (1994).

Gelman, A. & Hill, J. Data Analysis Using Regression and Multilevel/Hierarchical Models. (Cambridge University Press, Cambridge ; New York; 2007).

Kahan, D. Why We Are Poles Apart on Climate Change. Nature 488, 255 (2012).

Kahan, D.M. A Risky Science Communication Environment for Vaccines. Science 342, 53-54 (2013).

Kahan, D.M. Ideology, Motivated Reasoning, and Cognitive Reflection. Judgment and Decision Making 8, 407-424 (2013).

Kahan, D.M. in Handbook of Risk Theory: Epistemology, Decision Theory, Ethics and Social Implications of Risk. (eds. R. Hillerbrand, P. Sandin, S. Roeser & M. Peterson) 725-760 (Springer London, Limited, 2012).

Kahan, D.M., Braman, D., Gastil, J., Slovic, P. & Mertz, C.K. Culture and Identity-Protective Cognition: Explaining the White-Male Effect in Risk Perception. Journal of Empirical Legal Studies 4, 465-505 (2007).

Kahan, D., Braman, D., Cohen, G., Gastil, J. & Slovic, P. Who Fears the HPV Vaccine, Who Doesn’t, and Why? An Experimental Study of the Mechanisms of Cultural Cognition. Law and Human Behavior 34, 501-516 (2010).

Kahan, D.M., Braman, D., Slovic, P., Gastil, J. & Cohen, G. Cultural Cognition of the Risks and Benefits of Nanotechnology. Nature Nanotechnology 4, 87-91 (2009).

Kahan, D.M., Peters, E., Dawson, E. & Slovic, P. Motivated Numeracy and Englightened Self Government. Cultural Cognition Project Working Paper No. 116 (2013).

Kahan, D.M., Peters, E., Wittlin, M., Slovic, P., Ouellette, L.L., Braman, D. & Mandel, G. The Polarizing Impact of Science Literacy and Numeracy on Perceived Climate Change Risks. Nature Climate Change 2, 732-735 (2012).

Leiserowitz, A.A. American Risk Perceptions: Is Climate Change Dangerous? Risk Analysis 25, 1433-1442 (2005).

Lieberson, S. Making It Count : The Improvement of Social Research and Theory. (University of California Press, Berkeley; 1985).

[1] A multivariate-modeling strategy that treats all such indicators or all potential ones as “independent” right-hand side variables will not be valid. The group affiliations that impart risk-perception dispositions are indicated by combinations of characteristics—political orientations, cultural outlooks, gender, race, religious affiliations and practices, residence in particular regions, and so forth. But these characteristics do not cause the disposition, much less cause it by making linear contributions independent of the ones made by others.  Indeed, they validly and reliably indicate particular latent dispositions only when they co-occur in signature combinations. By partialing out the covariance of the indicators in estimating the influence of each on the outcome variable, a multivariate regression model that treats the indicators as “independent variables” is thus necessarily removing from its analysis of each predictor’s impact the portion of it that it owes to being a valid measure of the latent variable and estimating that influence instead based entirely on the portion that is noise in relation to the latent variable.  The variance explained (R²) for such a model will be accurate. But the parameter estimates will not be meaningful, much less valid, representations of the contribution that such characteristics make to variance in the risk perceptions of real-world people who vary with respect to those characteristics (Berry & Feldman 1985, p. 48; Gelman & Hill 2006, p. 187). To model how the latent disposition these characteristics indicate influence variance in the outcome variable, the characteristics must be combined into valid and reliable scales. If particular ones resist scaling with others—as is likely to be the case with mixed variable types—then excluding them from the analysis is preferable to treating them as independent variables: because they will co-vary with the latent measure formed by the remaining indicators, their omission, while making estimates less precise than they would be if they were included in formation of the composite latent-variable measure, will not bias regression estimates of the impact of the composite measure (Lieberson 1985, pp. 14-43; Cohen, Cohen, West &  Aiken 2003, p. 419).  Misunderstanding of (or more likely, lack of familiarity with) the psychometric invalidity of treating latent-variable indicators as independent variables in a multivariate regression is a significant, recurring mistake in the study of public risk perceptions.