This site uses cookies

Some of these cookies are essential, while others help us to improve your experience by providing insights into how the site is being used.

For more detailed information please check our Cookie notice

Necessary cookies

Necessary cookies enable core functionality. This website cannot function properly without these cookies.

Cookies that measure website use

If you provide permission, we will use Google Analytics to measure how you use the website so we can improve it based on our understanding of user needs. Google Analytics sets cookies that store anonymised information about how you got to the site, the pages you visit, how long you spend on each page and what you click on while you’re visiting the site.

Science knowledge and attitudes across cultures: a meta-analysis

Author: Nick Allum
Institution: University College London
Type of case study: Research

About the research

The correlation between knowledge and attitudes has been the source of controversy in research on the public understanding of science (PUS). Although many studies, both quantitative and qualitative, have examined this issue, the results are at best diverse and at worst contradictory. In this study, we review the evidence on the relationship between public attitudes and public knowledge about science across 40 countries using a meta-analytic approach. We fit multilevel models to data from 193 nationally representative surveys on PUS carried out since 1989. We find a small positive correlation between general attitudes towards science and general knowledge of scientific facts and processes, after controlling for a range of possible confounding variables. This general relationship varies little across cultures but more substantially between different domains of science and technology. Our results suggest that PUS research needs to focus on understanding the mechanisms that underlie the clear association that exists between knowledge and attitudes about science.

One of the key insights from programmes of research on public understanding of science that were set in train in the 1980s is that both European and American publics possess low levels of basic ‘textbook’ knowledge about science. For some, findings of this nature are taken as strong empirical confirmations of the existence of a ‘scientifically illiterate’ public and provide the first pillar in the construction of the pervasive ‘deficit model’ of public understanding of science (Irwin and Wynne, 1996; Sturgis and Allum, 2001). The deficit model sees public resistance to science and technology as underpinned by ignorance, superstition and fear. Public scepticism about technological innovations such as nuclear energy, microwave cooking and genetic science would be markedly reduced if citizens were better able to grasp the science upon which they are based. That is, a judgement when informed by scientific fact would tend to be more favourable and consistent with expert opinion than one expressed without recourse to such ‘objective’ knowledge. Since the late 1980s, much academic debate has focused on examining and understanding the link between knowledge and attitudes about science. Numerous surveys have been carried out to measure public attitudes, knowledge, beliefs and behaviour in relation to science and technology. Up until the present study, no systematic review had been carried out of this burgeoning database.

In the study, we review the evidence on the relationship between public attitudes and public knowledge about science and technology from the multitude of national surveys that have been carried out across the world during the past fifteen years in Europe, North America and beyond. To do this we use a formal, meta-analytic model. We go beyond a simple summary to present a model that helps systematise the apparent diversity of findings in the PUS literature. More specifically, we firstly evaluate the associations between particular forms of scientific knowledge and particular attitude domains; secondly, we control for a range of possible confounding variables across all our analysis; thirdly, we estimate disparities between countries in the knowledge-attitude relationship and employ macro-level predictors that might explain these disparities.


Our data collection and analytic methods took the following form. First we made a comprehensive search for data sources and assembled a large number of raw survey datasets. Second, we derived a number of comparable knowledge and attitude scales in each survey (hereafter we refer to each national survey as a ‘sample’). Third, we ran ordinary least square regressions of knowledge on attitudes, with additional control variables, for each sample. This yielded effect size estimates in the form of standardised partial regression coefficients of knowledge on attitude. Fourth, we compiled a new dataset that included these effect sizes, their standard errors and a range of other higher level variables relating to the year of the study, country, length of attitude scale and some aggregate country-level variables such as GDP per capita and proportion of 18-24 yr olds in tertiary education. Finally, this composite dataset was analysed using MLwiN 2.0 (Rasbash et al 2004), running a number of multilevel regression models with ‘effect size’ as the dependent variable.

Note on the meta-analysis:

Meta-analytic techniques see effect sizes estimated from single studies as units drawn from a hypothetical population of possible studies. As such, relying on single studies for effect size estimates relies on the unlikely event that the single study is representative of all possible studies that could have been sampled from this population (Rosenthal, 1991). Thus the basic objective of meta-analysis is to provide pooled estimates of effect sizes through a weighted average of the effect sizes of the individual studies, with sampling variances calculated as a function of the sample size of each individual study. The meta-analysis that we present here can be viewed as a special case of a multilevel model, as we have a hierarchical dataset with effect sizes nested within samples, nested within countries. By using a multilevel model, one can incorporate multiple effect sizes from the same sample into the same analysis (Hox, 2002). The multilevel model is also a more flexible meta-analytic model, allowing characteristics at each level to be included in the model to explain heterogeneity in effect sizes across and within samples. In the standard approach, effect sizes from each study are assumed to be fixed effects – estimates of an overall unknown population effect. However, when there is substantial heterogeneity between these estimates, a random effects model may be more appropriate. This model assumes that all studies estimate their own unique, unknown ‘study effects’ which are themselves distributed around an unknown population effect (Lambert and Abrams, 1995). In the present investigation, by using a multilevel model, we estimate parameters for a random effects model but with the addition of fixed effects to explain between sample and between country heterogeneity.


Our findings suggest that, if one examines all measured knowledge and attitude domains, there is a small but positive relationship.

Of equal, or perhaps greater, interest is the discovery of systematic differences in this relationship according to the degree of consonance between knowledge and attitude domains. The correlation between general ‘textbook’ knowledge and attitudes towards science as a whole is almost twice as high as the overall estimate whereas, for example, the correlation between general knowledge and attitudes to GM food is practically zero. However, when knowledge relates to biology and genetics, it becomes a considerably stronger predictor of a person’s attitudes towards GM food.

Previous work has suggested that there is a great deal of variation in the association between science literacy and attitudes to science between countries and that some of this variation is related to the degree of local economic development. Our results suggest that much of what has been interpreted as cultural variation can be accounted for mainly by variation in the relative proportion of individuals with particular attributes within countries rather than ‘culture’ per se. Our analysis shows that only 10 percent of the variation in the knowledge-attitude correlation can be explained by country level processes or mechanisms – much less than would be indicated by considering aggregate measures, as has previously been the rule. Neither GDP per capita nor Internet diffusion were significantly associated with cultural variation. However, the percentage of young people enrolled in tertiary education in a country is linked to stronger knowledge-attitude correlations and can account for all of the cross-country variation in our initial variance components analysis.


Allum, N., Sturgis, P., Tabourazi, D. and Brunton-Smith, I. (2008) ‘Science knowledge and attitudes across cultures: A meta-analysis’, Public Understanding of Science, 17(1), pp. 35-54. doi: 10.1177/0963662506070159 Retrieved 11 September 2013 from