A little knowledge is a dangerous thing: excess confidence explains negative attitudes towards science
AA little knowledge is a dangerous thing: excess confidence explains negative attitudestowards science
Frederico Francisco
Instituto Gulbenkian de Ciência, Rua da Quinta Grande 6, 2780-156 Oeiras, Portugal andCentro de Física do Porto, Departamento de Física e Astronomia,Faculdade de Ciências da Universidade do Porto,Rua do Campo Alegre 687, 4169-007 Porto, Portugal ∗ Simone Lackner
NOVA School of Business and Economics, Universidade Nova de Lisboa,Campus de Carcavelos, Rua da Holanda 1, 2775-405 Carcavelos, Portugal
Joana Gonçalves-Sá
Instituto Gulbenkian de Ciência, Rua da Quinta Grande 6, 2780-156 Oeiras, Portugal andNOVA School of Business and Economics, Universidade Nova de Lisboa,Campus de Carcavelos, Rua da Holanda 1, 2775-405 Carcavelos, Portugal (Dated: March 20, 2020)Scientific knowledge has been accepted as the main driver of development, allowing for longer,healthier, and more comfortable lives. Still, public support to scientific research is wavering, withlarge numbers of people being uninterested or even hostile towards science. This is having serioussocial consequences, from the anti-vaccination community to the recent "post-truth" movement.Such lack of trust and appreciation for science was first justified as lack of knowledge, leading tothe “Deficit Model” [1, 2]. As an increase in scientific information did not necessarily lead to agreater appreciation, this model was largely rejected, giving rise to “Public Engagement Models”[3]. These try to offer more nuanced, two-way, communication pipelines between experts and thegeneral public, strongly respecting non-expert knowledge, possibly even leading to an undervaluingof science. Therefore, we still lack an encompassing theory that can explain public understanding ofscience, allowing for more targeted and informed approaches. Here, we use a large dataset from theScience and Technology Eurobarometer surveys, over 25 years in 34 countries [4], and find evidencethat a combination of confidence and knowledge is a good predictor of attitudes towards science.This is contrary to current views, that place knowledge as secondary, and in line with findingsin behavioral psychology, particularly the Dunning-Kruger effect, as negative attitudes peak atintermediate levels of knowledge, where confidence is largest. We propose a new model, based onthe superposition of the Deficit and Dunning-Kruger models and discuss how this can inform sciencecommunication.
Keywords: Attitudes towards science | Deficit model | Dunning-Kruger effect
I. INTRODUCTION
Scientific research has been strongly supported by so-cieties through agencies that channel public funds to-wards research grants and fellowships, under the assump-tion that science drives the “Knowledge-based Society”.This investment is dependent on public support [5] and,from the 1960’s onward, a number of surveys began tobe applied trying to gauge both “hard knowledge” andthe public’s attitude towards science and scientific dis-coveries [2, 6]. The surprising finding that some of thepublic was, not only unknowledgeable, but also disen-gaged or even actively hostile led to the establishment ofa “Deficit Model”. In simple terms, this model claimedthat public skepticism towards science was due to lackof understanding [7] and that the more one knows aboutscience, the more positive one’s attitude towards science ∗ [email protected] is (“to know it is to love it”) [1, 2]. Its corollary was thatexperts and educators should engage with the ignorantpublic to improve their knowledge, directly leading to animprovement in support.In the 1980’s this model, that can be crudely repre-sented by the plot in Fig. 3A, started to face severe crit-icism for several reasons, and by the early 2000’s it hadmostly been discredited [2, 3, 8–12]. First, the concep-tion of a unidirectional communication between scien-tific experts and the community implied a disregard forthe lay public’s views and has been replaced with a two-way stream of dialogue, debate, and discussion, leadingto “Public-engagement" or “Interactive” models [3]. Sec-ond, the definitions of both knowledge and attitude be-came more fluid: knowledge is no longer seen as simpletextbook information that can be uniquely tested and as-signed to a single variable [13], and the notion of a singlepositive or negative “attitude” towards scientific subjectshas been replaced with the possibility of nuanced “atti-tudes”, which can vary widely depending on the subject,question at hand, context, time [7, 14–16], and even po- a r X i v : . [ phy s i c s . e d - ph ] M a r litical identity [17–19]. Third, there is growing evidencethat offering information on controversial issues, or onissues where people hold strong prior beliefs, does notchange people’s minds and can even backfire [20, 21], bypolarizing opinions [22] or even by eroding trust in thescientific method itself [23].Thus, while this relationship between knowledge(s)and attitude(s) has guided most of the discussion aroundscience communication and public understanding of sci-ence in the past decade [24], it is now clear that knowl-edge alone cannot fully predict attitudes [25]. However,when some knowledge and attitude variables can be iden-tified, close re-examinations of survey data have con-firmed that there is a central role to knowledge in thedetermination of attitudes: this role is much more com-plex than the linear relation purported by the “DeficitModel”, but it is real and in general there is a positiveassociation between higher knowledge and an overall pos-itive attitude [2, 12, 16, 24, 26, 27].Interestingly, this correlation disappears when the sub-ject is controversial and the respondent tends to beknowledgeable [24]. Offering “too easy” science textsmight lead to overconfidence and underrate the needfor experts [28], and just searching for information on-line on one subject leads to people to overestimate theirknowledge on an unrelated subject [29]. Dunning andKruger have shown that confidence grows faster thanknowledge [30] and this effect might be relevant in theanti-vaccination movement, with surveyed “anti-vaxxers”overestimating their knowledge on autism, and overcon-fidence being largest for lowest knowledge bins [31]. To-gether, this suggests that confidence might play an im-portant, while overlooked, role in modulating the rela-tionship between knowledge and attitudes towards sci-ence.In this work, we take advantage of 5 rounds of theScience and Technology Eurobarometer questionnaires,a dataset including 34 countries between 1989 and 2005,and ask whether confidence modulates public under-standing of science. By analyzing the relation betweenknowledge ( k ), attitudes ( att ) and a new confidence vari-able ( c ), we find that there is a consistent and strongnon-linear correlation between attitudes and knowledge,and that this relation can be explained by varying levelsof confidence. We propose a new testable model and dis-cuss how it can guide future research and interventions. II. MATERIALS AND METHODS
Computations were performed using R 3.4.4, MicrosoftExcel 16, Wolfram Mathematica 10 and Jupyter Note-book 6.01.
A. Dataset
The Science and Technology Eurobarometer cam-paigns from 1989 and 2005 surveyed a total of 34 coun-tries, including EU members, candidates at the time,and other European Economic Area (EEA) countries, to-talling 84469 individual interviews [4]. Unlike previousand subsequent campaigns, this set asked questions thattried to gauge bot knowledge and attitudes, in a con-sistent way. However, there were differences both in thequestions asked and in the possible answers, and the maindataset results from an harmonization effort that tookthe November 1992 (EB 38.1) round as a base and iden-tified similar variables in the remaining four rounds (seeTable S1). The harmonization was performed by takingthe variable in the 1992 Eurobarometer and identifyingitems with similar wordings on the other four campaigns.For simplicity, this harmonized dataset is referred to asthe Eurobarometer dataset throughout the text.
B. Attitude variables
In each Eurobarometer round, a number of questionsregarding possible attitudes towards science were asked.For each item, the interviewee is asked to declare agree-ment or disagreement with a given statement. As statedabove, the November 1992 (EB 38.1) round was chosenas a basis and similar variables (with almost identicalwording) were identified in the remaining four rounds.Thus, the Eurobarometer dataset contains an intersec-tion of the questions that were asked in each round andcontains the 10 attitude variables, listed in Table S2, thatare found in all rounds except, in some cases, 1989.The possible answers to the attitude questions are alsonot consistent: 1) the “don’t know” option was alwayspresent but a neutral option such as “neither agree nordisagree” was only offered in 1989, 1992 and 2005; 2) theavailable options on the Likert scale were sometimes fiveand and others two, as shown in Table S3.As these differences may have an impact on the re-spondents’ behaviour [16], we tested its impact in threedifferent ways: 1) by treating all the categories in the Lik-ert scale either separately or fusing them into less options(adding the “strongly agree” with the “agree to some ex-tent” and the “disagree to some extent” with the “stronglydisagree”); 2) by either including or disregarding the “nei-ther agree nor disagree”; and 3) by either aggregating the“neither agree nor disagree” with the “dont’ know” an-swers, or by treating them separately. These alternativesmake up for a total of six different approaches to thedata. We performed many of the calculations that followin all six ways in order to establish that the choice for agiven approach does not significantly affect the results.To obtain a measurement or a smaller set of measure-ments for attitude(s) towards science, we computed their × Spearman correlation matrix and performed aPrincipal Components Analysis (PCA) (see Figure S3A).We found that the answers are mostly uncorrelated andthat there is no single component explaining a large per-centage of the variation. We describe these findings ingreater detail in the main text and thus treat all attitudevariables independently.It is important to note that, regardless of the polarityof the questions, “Agree" and “Strongly Agree" answersare typically more prevalent than disagreement answers,a common effect, known as “acquiescence bias" [15, 32].Therefore, in the results we focus particularly on the“Agree" answers, and these tend to show a stronger effect.
C. Knowledge Variables
The Eurobarometer dataset includes 13 “true or false”questions, listed in Table S4, designed to assess knowl-edge on science related subjects, with a “don’t know”option always available.Similarly to the attitude questions set, we tested in-dependence by calculating Spearman correlation and byperforming a PCA (see Figure S3B). We created a singleknowledge variable, k , computed from the ratio of cor-rect answers to the number of questions each individualwas asked. Thus, a “don’t know” is considered equiva-lent to an incorrect answer as far as the measurement ofknowledge is concerned. D. Confidence Measurement
The neutral and “don’t know” answers can offer a pos-sible measure of confidence. We use the aggregates of the“neither agree nor disagree” and the “don’t know” answersto the attitude questions, to which we call “neutral" an-swers, and the “don’t know” answers to the knowledgequestions as a measure of confidence. As before, thisclassification does not offer a direct measurement of con-fidence, but serves as a general indicator, when comparedto the other variables.
E. Mathematical Model
The Deficit Model (DM) can be represented as a linearrelation between attitudes, att , and knowledge, k , of theform att DM ( k ) = α k + β, α ≤ , β ≥ , (1)with higher knowledge leading to a more positive atti-tude. However, from 1A, we can observe a quadraticrelation between confidence, c , and knowledge. This re-lation (that has been reported for the Dunning-Kruggereffect, D-K), can be derived directly from the curve andbe written as c D-K ( k ) = γ k + δ k + (cid:15), (2) by fitting these curves we find that γ (cid:39) − , δ (cid:39) , (cid:15) (cid:39) . . (3)The proposed model is obtained by multiplying thesetwo relations, with the Deficit Model inverted for nega-tive attitudes, -att ( k ) = c D-K ( k ) × (1 − att DM ( k )) , (4)leading to an inverted-U shaped curve. Taking the confi-dence curve as an experimental result, better fits to thecurves in each attitude item can be obtained by adjust-ing the α and β parameters in our representation of theDeficit Model. III. RESULTSAttitudes towards science
Public attitudes towards science depend on several fac-tors and it is not clear how much of a role knowledgeplays. By using a large-scale database we tested: 1)whether it is possible to define “attitude(s)” towards sci-ence, 2) whether these vary with knowledge, and 3) whatmodulates such variation.We thus started by asking whether it is possible toidentify single, or a small subset of attitudes towards sci-ence. We extended the work of [16] and included all Eu-robarometers and countries, offering not only more dataand statistical power, but also the possibility of compar-ing the results longitudinally.First, we compared all attitude variables and foundthat they are weakly correlated ( < . ), with only twogroups of variables with relatively higher correlations:one that might be associated with an optimistic attitudeand another with overall distrust, as shown in Figure S1.Second, we performed a PCA and found, as [16] beforeus, that this system does not justify the grouping of someattitudinal questions, as can be seen in Figure S3A. In-deed, the first and most significant principal componentaccounts for less than of the variance and even thefirst 5 components only represent around , with thelast and less significant of 10 components still holdingalmost of the variance.Third, a series of attempts at factor analysis did notidentify any set of factors modelling the behaviour of theattitude variables.Thus, we found no mathematical justification for theconstruction of an attitude scale or of a small set of scales.In fact, these attempts indicate that there is a high levelof independence between the variables. Thus, all attitudevariables are treated separately, in the rest of the work. A. Attitudes and Knowledge
In the surveys, respondents were asked to statewhether 13 science-related statements were true or false.We started by testing independence and found that, sim-ilarly to the attitude questions, the knowledge answersare poorly correlated. However, this can be explained ingreat part by the fact that the questions have differentdifficulty levels, with some questions displaying a muchhigher number of correct answers. Also, contrary to theattitudes questions, the PCA reveals that the first com-ponent explains of the variance, with all componentshaving the same sign, indicating that answering one ques-tion correctly, increases the likelihood of giving the rightanswer to other questions, as depicted in Figure S3B. Infact, the distribution of correct answers is approximatelyNormal, as expected (see horizontal axes distributions inFig. 1).Therefore, and as for the purposes of this project wewere not so much interested in measuring individualknowledge as in finding relations between this measureand the identified attitudes, we created a single k vari-able, where k corresponds to the fraction of correct an-swers, from (no correct answers) to (all questionsanswered correctly).When we plotted the different attitudes by knowledge,we found that they also vary differently. Table I showsthe slopes and fit of the linear regressions for the propor-tion of “agreement” answers for all attitude questions. Wefind that while some have strong dependencies on knowl-edge (higher absolute slopes), either positive or negative,others are virtually independent (lower absolute slopes).Fig. 2A and D show examples of the attitude questionsthat fall within each of these two groups (full results inFig. S6).Our analysis does not identify any interesting pattern,with both controversial and less controversial issues (fromthe possibility of harming animals in research to whetherscience makes our liver more interesting), being basicallyindependent from k , and strong dependencies appearingin issues of faith and comfort.These results, seem to support the current views thatnot only there is no single variable that describes a set of“attitudes” towards science, but also there is no simple re-lationship between such attitudes and knowledge. How-ever, both ours and past analysis, have focused only onrespondents that state either an agreement or a disagree-ment with the questions. And it has long been knownthat many people offer answers to survey questions whenthey are unknowleageable of the subject, and even whenthe subjects at hand are fictitious [33].Therefore, we decided to study the impact of the “don’tknow” and “neither agree nor disagree” answers in thiscontext. TABLE I. Slopes of “agreement” linear regressions of attitudevariables plotted against knowledge as measured by k vari-able. “Agree” slope R att_comfort .
12 0 . att_natural_resources − .
31 0 . att_faith − .
42 0 . att_environ − .
35 0 . att_research_animal ∼ att_res_dangerous − .
08 0 . att_interest .
06 0 . att_daily_life − .
57 0 . att_fast − .
26 0 . att_oppor .
05 0 . B. Knowledge and Confidence
We started by analyzing the impact of the “don’t know”answers, in the knowledge questions, knowing that thefraction of correct answers varies with an approximatelyNormal distribution (Fig. 1). The interesting question iswhether there is variation in the ratio of wrong to “don’tknow” answers as we propose that this variation mightoffer us a measure of confidence.A perfectly rational individual would modulate theirconfidence on a specific subject to their knowledge onthat subject. Therefore, a perfect match between howmuch one knows ( k ) and how much one thinks one knows(confidence) would lead to a complete absence of wronganswers, with respondents either answering correctly orselecting the “don’t know” option. In this case, as the per-centage of correct answers increased from to , thenumber of “don’t know” answers would decrease symmet-rically, creating a perfect diagonal. This line would inter-sect at on both axes (solid black line on Fig. 1A).If the incorrect answers did not depend on eitherknowledge or confidence (for example, if wrong answerswere caused by randomly distributed errors), they shouldvary linearly with k and we would also observe the idealline shifting down by an amount equal to the averagefraction of incorrect answers, intersecting the axes atlower values. However, if the incorrect answers are modu-lated by confidence, with individuals overestimating theirknowledge, we should observe non-linear (non-diagonal)relationships. And if the number of wrong answers growsfaster than the number of ”Correct" and “Don’t know”answers, this will be represented as a deviation from thediagonal towards a concave curve, and can be interpretedas the confidence growing faster than knowledge.To study how confidence varies with k , we analyzedhow the number of “don’t know” answers varies with thedifferent k bins. This can be represented by the linearfit of the fraction of “don’t know” as a function of thefraction of correct answers per bin (dashed black line inFig. 1A). Thus, we may use this deviation as a measure ( A ) ( B ) k ' don ' t k no w ' a n s w e r s - - k correct answers k w r ong a n s w e r s - - k correct answers FIG. 1. Density histogram of the distribution of respondents according to the fraction of correct answers and fraction of “don’tknow” answers (panel A) or incorrect answers (panel B). The dotted and dashed lines are the linear and quadratic regressions,respectively. Bars on the axes show distributions for each variable, all in the same [0 , scale. These charts show how thefraction of “don’t know” answers decreases more rapidly than the increase in knowledge, evidence of overconfidence. If eachrespondent only answered to the questions to which they know the answer, then the curve in Panel A would follow the diagonalthin line and there would be no incorrect answers, a flat line at zero in Panel B. Instead, we see the lowest knowledge bins veryclose to this “ideal confidence” line with the highest levels of overconfidence in the intermediate knowledge bins, coinciding withthe highest proportions of incorrect answers. of overconfidence of the respondents.As can be seen in Fig. 1A, the quadratic fit curve isindeed concave, suggesting that confidence tends to growmuch faster than k .An equivalent way of looking at these results is byplotting the incorrect answers as a function of correctanswers, which we tentatively identify with overconfi-dence and knowledge, respectively. The results in Fig. 1Bclearly show how the probability of wrong answers ismaximum in the intermediate levels of knowledge andnot at the lower, as would be expected if overconfidencewas evenly distributed.As we are looking at the sum of all possibilities withinthe same k bin, over more than 80 000 questionnaires,this curve can appear both when the individuals havevery similar behaviours or when we have different popu-lations, with some populations displaying very low wrongto “don’t know” ratios and some displaying very high.Therefore, we repeated this analysis for each of the 34countries individually, and confirm that confidence growsfaster than knowledge in all surveyed countries (Figs. S4and S5). We also find some small but consistent differ-ences between them as, with few exceptions, respondentsfrom the most developed, and generally more educatedcountries (Norway, Switzerland, Denmark, Netherlands,West Germany) show the highest confidence gap, withthe ratio of wrong to right answers in the low k bins, be- ing over 50% (as gauged by the intersect of the linear fitin the y-axis).This is suggestive of an effect similar to what has beenobserved by Dunning and Kruger in the USA [30], leadingus to look for what effects this observed overconfidencemay have on the attitude items. C. Attitudes and Confidence
As described in the methods, the different Eurobarom-eter surveys followed different policies, with some includ-ing the neutral “neither agree nor disagree”, and othersonly allowing the “don’t know” option. As others be-fore us [16], we found that the sum of these two tendsto be constant (a person that would respond “neitheragree nor disagree” to a given item is likely to choose“don’t know” if the first option is not available). Thus,we used the sum of these two variables, generally callingthem “neutral answers”, and compared their usage acrossall attitude variables. Respondents offer either “agree”or “disagree” answers in the large majority of instances,with neutral choices varying between and of thetotal answers. As it is possible that this variation stemsfrom individual options, we looked at the correlation be-tween people who tend to answer “don’t know” to the k questions and people who tend to offer neutral answers to (A)(D) (att_daily_life) "For me, in my daily life, it is not important to know about science."(att_comfort) "Science & Technology are making our lives healthier, easier and more comfortable." (B) k Neutral answersNeutral (C) k With neutral answersAgreement Disagreement k Without neutral answersAgreement Disagreement k Neutral answers
Neutral (E) k With neutral answers
Agreement Disagreement (F) k Without neutral answers
Agreement Disagreement
FIG. 2. Relative frequencies of agreement, disagreement and neutral stance for each knowledge category towards the statements“For me, in my daily life, it is not important to know about science” (upper row) and “Science & Technology are making our liveshealthier, easier and more comfortable” (lower row), shown here as examples of two distinct behaviours of attitude variables.Upper row shows an example of an asymmetric behaviour of agreement and disagreement, with the distinct “inverted U” curveappearing in the negative attitude. Lower row shows an item with a mostly flat disagreement curve and monotonously crescentagreement curve. Shaded areas highlight the four consecutive knowledge bins with highest agreement in each attitude item. the attitude questions. We controlled these relationshipsbetween attitudes and knowledge for education level andobserved that the behaviour remains substantially thesame.As seen in Fig. 1A, the proportion of “don’t know” an-swers decreases more rapidly than the increase in correctanswers (Fig. 1B) with the highest fractions of incorrectanswers encountered in the mid k range and not in thelower k categories.Similarly, we could expect the individuals in the lower k bins (who answered proportionately more “don’t know”to the k questions), to also offer more neutral answers tothe attitude questions. This is indeed what we observe:the neutral answers have a sharp decline in the lower k bins in every single attitude item, with only small varia-tions, and remains very close to zero, in the mid to high k bins, as exemplified in Fig. 2B and E.We had observed that attitudes vary inconsistentlywith knowledge, with some having strong and othersshowing very little dependence with k . This was doneby calculating the frequency of "agree" versus "disagree"answers, and disregarding the "don’t know" or "neitheragree nor disagree" (neutral) options. When we now re-analyze this dependence, but including the neutral an-swers, we find not only different behaviours across dif-ferent attitudes, but very asymmetrical effects betweenagreement and disagreement positions (Fig. 2). In fact, all previously linear relationships (Fig. 2A and D), nowbecome quadratic, often displaying either “inverted U”shape curves (Fig. 2C) or asymptotic behaviour (Fig. 2F),especially in the agreement answers, as discussed in themethods.Interestingly, by including the neutral answers in theanalysis, this non-linear behaviour now appears in all at-titude items, with the most negative attitudes appearingat intermediate levels of k , that also correspond to thehighest confidence to k ratios. Shaded areas in Figs. 2 andS6 show where the four consecutive k bins with highestagreement are, allowing for a clear distinction between“inverted-U” and asymptotic curves. Therefore, attitudesare not independent of knowledge, as current theories de-fend, neither do they appear to be more negative in lowerknowledge bins, as the Deficit Model would predict.Many of the attitudes that can be identified as nega-tive seem to be modulated by a combination of knowledgeand confidence, as represented in Fig. 3. Therefore, wedeveloped a simple mathematical model, that combinesthe linear relationship predicted by the Deficit Modeland the quadratic relation observed from the curve inFig. 1A, that confirms the Dunning-Kruger effect. Thisnew model, that simply multiplies both relations (as de-scribed in the Methods), leads to an inverted-U shapedcurve, observed in many of the negative attitude items,as shown in Fig. 3. Importantly, the attitude items havedifferent dependencies on knowledge among them, evenbefore accounting for neutral answers and the effect ofconfidence, and this can be easily modulated by changesin the fitting parameters. IV. CONCLUSIONS
Our work builds on the long-lasting and ongoing dis-cussion of what are the best predictors of public attitudestowards science. By creating a dataset of several roundsof the Science and Technology Eurobarometers and an-alyzing the ratio of correct to incorrect answers to theknowledge questions, we found that this does not vary lin-early, with the majority of incorrect answers appearing atintermediate levels of knowledge. Similarly, the numberof neutral answers to the attitude items drops very fast,approaching its minimum for intermediate knowledge lev-els. Arguing that this variation in the number of neutralanswers, both for the knowledge and the attitudes ques-tions, can be used as a proxy for confidence, we foundthat 1) confidence grows much faster that knowledge,in line with previous works that identify the Dunning-Kruger effect as relevant in the anti-science movements[31, 34]; 2) that the least positive attitudes are found forthese high-confidence / average knowledge groups, creat-ing an inverted U-curve; and 3) that public attitudes to-wards science can be explained by a non-linear combina-tion of both knowledge (following from the Deficit Model)and confidence (following from the Dunning-Kruger ef-fect), proposing a new theoretical model (Fig. 3C).Interestingly, and contrary to the cited works [31, 34],the least positive attitudes are not found at the lowestk bins, and four non mutually exclusive possibilities canexplain this difference. First, the anti-vaccine, GMO andclimate change issues are highly controversial with po-larized populations for or against it, while this is not thecase for most of the attitudes tested in the Eurobarometerdataset. The respondents in [31, 34] have strong opinionsand are likely to believe to be very well informed, whilethe respondents in this dataset are least confident in thelow k bins. This is in line with the predictions of theDunning-Kruger effect, as confidence peaks in the middleand not for low k. Second, these are also issues for whichthere are large amounts of false information circulatingonline. Therefore, strong advocates against GMOs, cli-mate change and vaccines are likely to believe to be right.They might know of the scientific consensus and choosenot offer it as the correct answer. Again, this is unlikelyto be the case with the surveyed for these Eurobarome-ters. Third, there is a significant time gap between thedifferent surveys. The last round of the Eurobarometertook place in 2005 and, although we do not see longi-tudinal differences, this dataset was built mostly beforethe wide expansion of the internet and of online socialnetworks. It is easy to argue that this misinformationand polarization might be made worse by these recenttechnologies, with the creation of echo-chambers and in- formation bubbles. These may limit the quantity, quality,and diversity of information accessible to the non-expertpublic, effectively creating large groups of misinformedcitizens. And the politicizing of science together with anincrease in political polarization [35] might deepen thisdivide even further.It also important to note that, to our knowledge, theDK effect had not been consistently shown outside of theUSA, and the most developed and educated countriesseem to display larger confidence to k gaps. Therefore, itis possible that, if this Eurobarometer was to be repeated,we would observe an even larger gap between confidenceand k, across countries, as the citizens become more con-nected and confident, and possibly an even stronger po-larization in the answers to the attitude items. Thus, weargue that, despite its problems, a new round of this ora very similar survey is in order.Taken together, our results have clear implications tocurrent science communication strategies. Our modelpredicts that receptiveness to science will be stronger atthe lowest and highest knowledge bins, where the C/Kratios are also lowest. Offering information that is in-complete, partial, or over simplified, as science commu-nicators often do, might indeed backfire, as it may offera false sense of knowledge to the public, leading to over-confidence, and less support.In fact, if the lowest support for science comes from theover-confident, these might also be the ones more resis-tant to new information, especially if it contradicts theircertainty, creating a negative reinforcement loop. Thisresistance to change has been shown in several behavioralpsychology studies, and presented as cognitive biases,such as the confirmatory tendencies. Importantly, theseintermediate k and high confidence bins, correspond tothe majority of the individuals surveyed. This effect wasnot important in our analysis, as all bins were normal-ized by frequency, but is fundamental at a populationlevel, as they are likely to correspond to a large group ofEuropean demographics.If indeed negative attitudes can be explained by a com-bination of limited knowledge and excess confidence, de-veloping science communication strategies that offer agood balance between sharing not only accurate and pre-cise information, but also large doses of humility, bothon the scientists and the lay public’s side, is likely tobe a fundamental, while very difficult task. A multidisci-plinary approach, building from cognitive and behavioralpsychology, social media and complex systems analysis,should receive a new focus, so that we move from a post-truth world, by avoiding the dangers of the "little knowl-edge".
ACKNOWLEDGMENTS
The authors would like to thank Caetano Souto-Mayor,Michael West and João Nolasco for initial analysis of thedataset, members of the Data Science and Policy group a tt ( A ) Deficit Model c on f . ( B ) Dunning - Kruger Effect - a tt . ( C ) Combined Model: DM and D - K FIG. 3. Proposed model of the observed behaviour of negative attitudes towards science. The Deficit Model is shown on theleft as a simple linear relationship between knowledge and (positive) attitudes, whereas the Dunning-Kruger effect model inthe center is derived directly from the curve in Fig. 1A. The resulting inverted-U curve model on the right is the product ofnegative attitudes Deficit Model with the Dunning-Kruger confidence curve. for valuable discussions, and Tiago Paixão, Marta En-tradas and Joana Lobo Antunes for critical reading of the manuscript. JGS was partially supported by Wel-come DFRH WIIA 60 2011, co-funded by the FCT andthe Marie Curie Actions. [1] J. R. Durant, G. A. Evans, and G. P. Thomas, Nature , 11 (1989).[2] M. W. Bauer, N. Allum, and S. Miller, Public Under-stand. Sci. , 79 (2007).[3] S. Miller, Public Understand. Sci. , 115 (2001).[4] M. W. Bauer, S. R, and K. P, Public understandingof science in Europe 1989-2005. A Eurobarometer trendfile. , Tech. Rep. (2012).[5] J. D. Miller, Public Understand. Sci. , 273 (2004).[6] M. W. Bauer, in Handbook of public communication ofscience and technology (Routledge, 2008) pp. 111–130.[7] B. Wynne, Science, Technology, & Human Values , 111(1991).[8] House of Lords, Science and Society , Tech. Rep.(2000).[9] B. Wynne, Science as Culture , 445 (2001).[10] M. C. Nisbet, D. A. Scheufele, J. Shanahan, P. Moy,D. Brossard, and B. V. Lewenstein, Communication Re-search , 584 (2002).[11] S. Jasanoff, Soc Stud Sci , 389 (2003).[12] P. Sturgis and N. Allum, Public Understand. Sci. , 55(2004).[13] B. Wynne, Public Understand. Sci. , 37 (1992).[14] S. Martin and J. Tait, in Biotechnology in Public , editedby J. R. Durant (1992).[15] G. Evans and J. Durant, Public Understand. Sci. , 57(1995).[16] R. Pardo and F. Calvo, Public Understand. Sci. , 155(2002).[17] L. C. Hamilton, Climatic Change , 231 (2010).[18] A. M. McCright, Climatic Change , 243 (2010). [19] C. Drummond and B. Fischhoff, Proc Natl Acad Sci USA , 9587 (2017).[20] T. v. Gelder, College Teaching , 41 (2005).[21] T. Gilovich, D. Griffin, and D. Kahneman, eds., Heuris-tics and Biases: The Psychology of Intuitive Judgment ,1st ed. (Cambridge University Press, 2012).[22] P. S. Hart and E. C. Nisbet, Communication Research , 701 (2011).[23] G. D. Munro, Journal of Applied Social Psychology ,579 (2010).[24] N. Allum, P. Sturgis, D. Tabourazi, and I. Brunton-Smith, Public Understand. Sci. , 35 (2008).[25] B. Fischhoff and D. A. Scheufele, Proc Natl Acad SciUSA , 13583 (2014).[26] B. C. Hayes and V. N. Tariq, Public Understand. Sci. ,433 (2000).[27] M. Entradas, portuguese journal of social science , 71(2015).[28] L. Scharrer, Y. Rupieper, M. Stadtler, and R. Bromme,Public Understand. Sci. , 1003 (2017).[29] M. Fisher, M. K. Goddu, and F. C. Keil, Journal ofExperimental Psychology: General , 674 (2015).[30] J. Kruger and D. Dunning, Journal of Personality andSocial Psychology , 1121 (1999).[31] M. Motta, T. Callaghan, and S. Sylvester, Social Science& Medicine , 274 (2018).[32] G. Meisenberg and A. Williams, Personality and Individ-ual Differences , 1539 (2008).[33] G. F. Bishop, A. J. Tuchfarber, and R. W. Oldendick,Public Opinion Quarterly , 240 (1986).[34] P. M. Fernbach, N. Light, S. E. Scott, Y. Inbar, and P. Rozin, Nat Hum Behav , 193 (2019). [35] S. Iyengar and D. S. Massey, Proc Natl Acad Sci USA , 201805868 (2018). Appendix A: Supplementary Data
TABLE S1. List of Eurobarometer rounds used to compile the harmonized dataset from Ref. [4], used in this paper. EB 38.1was used as a reference for the identification of similar variables to construct the harmonized dataset, with countries surveyedin each Science and Technology Eurobarometer round. CandidateRound Eurobarometer Eurobarometer Eurobarometer Country EB Eurobarometer31 38.1 55.2 2002.3 63.1Dates Mar-Apr 1989 Nov 1992 May-Jun 2001 Oct-Nov 2002 Jan-Feb 20051 France • • • - • • • • - • • • • - • • • • - • • • • - • • • • - • • • • - • • • • - • • • • - •
10 Northern Ireland • • • - •
11 Greece • • • - •
12 Spain • • • - •
13 Portugal • • • - •
14 East Germany - • • - •
15 Finland - • - •
16 Sweden - - • - •
17 Austria - - • - •
18 Cyprus - - - • •
19 Czech Republic - - - • •
20 Estonia - - - • •
21 Hungary - - - • •
22 Latvia - - - • •
23 Lithuania - - - • •
24 Malta - - - • •
25 Poland - - - • •
26 Slovakia - - - • •
27 Slovenia - - - • •
28 Bulgaria - - - • •
29 Romania - - - • •
30 Turkey - - - • •
31 Iceland - - - - •
32 Croatia - - - - •
33 Switzerland - - - - •
34 Norway - - - - • Total 13 14 17 13 34 TABLE S2. Set of 9 attitude variables in the Eurobarometer dataset. For each statement respondents were asked to state theiragreement or disagreement. Starred items ( ∗ ) do not have data for 1989.Long Code Statement att_comfort “Science & Technology are making our lives healthier, easier and more com-fortable.” ∗ att_natural_resources “Thanks to scientific and technological advances, the earth’s natural resourceswill be inexhaustible.” att_faith “We depend too much on science and not enough on faith” ∗ att_environ “Scientific and technological research cannot play an important role in protect-ing the environment and repairing it.” ∗ att_research_animal “Scientists should be allowed to do research that causes pain and injury toanimals like dogs and chimpanzees if it can produce information about humanhealth problems.” ∗ att_res_dangerous “Because of their knowledge, scientific researchers have a power that makesthem dangerous.” ∗ att_interest “The application of science and new technology will make work more interest-ing.” ∗ att_daily_life “For me, in my daily life, it is not important to know about science.” att_fast “Science makes our way of life change too fast.” ∗ att_oppor “Thanks to science and technology, there will be more opportunities for thefuture generations.”TABLE S3. Available answers for attitude items in each Eurobarometer campaign contained in the dataset.EB 31 EB 38.1 EB 55.2 Candidate EB 2002.3 EB 63.1Mar-Apr 1989 Nov 1992 May-Jun 2001 Oct-Nov 2002 Jan-Feb 2005Strongly agree • • - - • Agree to some extent • • • • •
Neither agree nor disagree • • - - • Disagree to some extent • • • • •
Strongly disagree • • - - • Don’t know • • • • • a tt _ r e s e a r c h _ a n i m a l a tt _ c o m f o r t a tt _ i n t e r e s t a tt _ o pp o r a tt _ r e s _ d a n g e r o u s a tt _ f a i t h a tt _ f a s t a tt _ n a t u r a l _ r e s o u r c e s a tt _e n v i r o n a tt _ d a il y _ li f e att_research_animal 1 0,094 0,118 0,140 -0,004 0,024 0,047 0,081 0,023 -0,008att_comfort 0,094 1 0,251 0,277 0,029 -0,032 -0,013 0,091 -0,096 -0,097att_interest 0,118 0,251 1 0,364 0,036 0,000 0,046 0,065 -0,047 -0,097att_oppor 0,140 0,277 0,364 1 0,029 -0,011 0,038 0,047 -0,079 -0,090att_res_dangerous -0,004 0,029 0,036 0,029 1 0,135 0,207 -0,014 0,028 0,035att_faith 0,024 -0,032 0,000 -0,011 0,135 1 0,219 0,044 0,103 0,099att_fast 0,047 -0,013 0,046 0,038 0,207 0,219 1 0,034 0,081 0,078att_natural_resources 0,081 0,091 0,065 0,047 -0,014 0,044 0,034 1 0,146 0,106att_environ 0,023 -0,096 -0,047 -0,079 0,028 0,103 0,081 0,146 1 0,167att_daily_life -0,008 -0,097 -0,097 -0,090 0,035 0,099 0,078 0,106 0,167 1 FIG. S1. Spearman correlation matrix of attitude variables, showing their weak correlations and ordered to show the also weakclusters. k _ o x y g e n k _ hu m a n k _ g e n e k _ a n t i b i o t i c s k _ d i n o s a u r s k _ m il k k _ l a s e r s k _ r a d i o a c t i v i t y k _ s un k _ t i m e k _e l e c t r o n k _e a r t h k _ c o n t i n e n t s k_oxygen 1 0,086 0,046 0,020 0,065 0,051 0,047 0,056 0,074 0,076 0,081 0,157 0,139k_human 0,086 1 0,084 0,081 0,102 0,094 0,111 0,105 0,061 0,076 0,120 0,131 0,188k_gene 0,046 0,084 1 0,159 0,120 0,137 0,104 0,121 0,057 0,101 0,103 0,090 0,151k_antibiotics 0,020 0,081 0,159 1 0,256 0,252 0,263 0,292 0,114 0,160 0,117 0,124 0,168k_dinosaurs 0,065 0,102 0,120 0,256 1 0,266 0,263 0,284 0,155 0,221 0,160 0,167 0,212k_milk 0,051 0,094 0,137 0,252 0,266 1 0,280 0,310 0,199 0,211 0,146 0,178 0,211k_lasers 0,047 0,111 0,104 0,263 0,263 0,280 1 0,350 0,188 0,239 0,215 0,160 0,181k_radioactivity 0,056 0,105 0,121 0,292 0,284 0,310 0,350 1 0,231 0,251 0,211 0,201 0,237k_sun 0,074 0,061 0,057 0,114 0,155 0,199 0,188 0,231 1 0,282 0,140 0,175 0,123k_time 0,076 0,076 0,101 0,160 0,221 0,211 0,239 0,251 0,282 1 0,196 0,166 0,177k_electron 0,081 0,120 0,103 0,117 0,160 0,146 0,215 0,211 0,140 0,196 1 0,162 0,214k_earth 0,157 0,131 0,090 0,124 0,167 0,178 0,160 0,201 0,175 0,166 0,162 1 0,261k_continents 0,139 0,188 0,151 0,168 0,212 0,211 0,181 0,237 0,123 0,177 0,214 0,261 1 FIG. S2. Spearman correlation matrix of knowledge variables, showing their fairly weak correlations and ordered to show thealso weak clusters. TABLE S4. Set of 13 knowledge variables in the Eurobarometer dataset, with question statement and possible answers; A“don’t know” option was also available in each question. The correct answer is starred ( ∗ ).Code Question Answers k_earth “The centre of the Earth is very hot.” ∗ “True” or “False” k_oxygen “The oxygen we breathe comes from plants.” ∗ “True” or “False” k_milk “Radioactive milk can be made safe by boiling it.” “True” or ∗ “False” k_electron “Electrons are smaller than atoms.” ∗ “True” or “False” k_continents “The continents on which we live have been moving theirlocation for million of years and will continue to move inthe future.” ∗ “True” or “False” k_gene “It is the father’s gene which decides whether the baby isa boy or a girl.” ∗ “True” or “False” k_dinosaurs “The earliest humans lived at the same time as the di-nosaurs.” “True” or ∗ “False” k_antibiotics “Antibiotics kill viruses as well as bacteria.” “True” or ∗ “False” k_lasers “Lasers work by focusing sound waves.” “True” or ∗ “False” k_radioactivity “All radioactivity is man-made.” “True” or ∗ “False” k_human “Human beings, as we know them today, developed fromearlier species of animals.” ∗ “True” or “False” k_sun “Does the earth go around the sun or does the sun goaround the earth?” “The sun goes around theearth” or ∗ “The earth goesaround the sun” k_time “How long does it take for the earth to go around thesun?” ∗ “Year” or “Month” Knowledge
Attitudes (A) (B)
FIG. S3. Proportion of variance for each principal component resulting from the PCA ran on the knowledge and attitudevariables. (A) Attitude variables PCA, with full line for binning of answers into positive, negative and neutral, other binningmethods as superimposed dotted lines. There is a slow and steady decline in the proportion of variance throughout, with thefirst few principal components failing to provide a large enough proportion of the total variance to be useful. (B) Knowledgevariables PCA, with full line considering the aggregation of incorrect and “don’t know” answers and dotted line keeping themdistinct. The first principal component accounts for a significantly larger part of the total variance and its coefficients all havethe same sign. France
Belgium
Netherlands
West Germany
Italy
Luxembourg
Denmark
Ireland
Great Britain
Northern Ireland
Greece
Spain
Portugal
East Germany
Finland
Sweden
Austria
Cyprus
Czech Republic
Estonia
Hungary
Latvia
Lithuania
Malta
Poland
Slovakia
Slovenia
Bulgaria
Romania
Turkeyk
Iceland
Croatia
Switzerland
Norway
FIG. S4. Fits of the distribution of respondents according to the fraction of correct answers and fraction of “don’t know”answers by country. The dotted and dashed lines are the linear and quadratic regressions, respectively. Compare with Fig. 1A. France
Belgium
Netherlands
West Germany
Italy
Luxembourg
Denmark
Ireland
Great Britain
Northern Ireland
Greece
Spain
Portugal
East Germany
Finland
Sweden
Austria
Cyprus
Czech Republic
Estonia
Hungary
Latvia
Lithuania
Malta
Poland
Slovakia
Slovenia
Bulgaria
Romania
Turkey
Iceland
Croatia
Switzerland
Norway
FIG. S5. Fits of the distribution of respondents according to the fraction of correct answers and fraction of wrong answers bycountry. The dotted and dashed lines are the linear and quadratic regressions, respectively. Compare with Fig. 1B. (att_research_animal)"Scientists should be allowed to research that causes pain and injury to animals like dogs and chimpanzees if it can produce information about human health problems."(att_natural_resources)"Thanks to scientific and technological advances, the earth's natural resources will be inexhaustible." (att_faith)"We depend too much on science and not enough on faith." k With neutral answersAgreement Disagreement Neutral k Without neutral answersAgreement Disagreement k Without neutral answersAgreement Disagreement (att_environ)"Scientific and technological research cannot play an important role in protecting the environment and repairing it." k Without neutral answersAgreement Disagreement k With neutral answersAgreement Disagreement Neutral k With neutral answersAgreement Disagreement Neutral (att_res_dangerous)"Because of their knowledge, scientific researchers have a power that makes them dangerous." (att_interest)"The application of science and new technology will make work more interesting." k With neutral answersAgreement Disagreement Neutral k Without neutral answersAgreement Disagreement k Without neutral answersAgreement Disagreement (att_fast)"Science makes our way of life change too fast." k Without neutral answersAgreement Disagreement k With neutral answersAgreement Disagreement Neutral k With neutral answersAgreement Disagreement Neutral (att_oppor)"Thanks to science and technology, there will be more opportunities for the future generation." k Without neutral answersAgreement Disagreement k With neutral answersAgreement Disagreement Neutral k Without neutral answersAgreement Disagreement k With neutral answersAgreement Disagreement Neutral
FIG. S6. Relative frequencies of agreement, disagreement and neutral stance for each knowledge category towards the remainingattitude items analyzed, with and without the inclusion of neutral answers. Shaded areas highlight the four consecutiveknowledge bins with highest agreement in each attitude item. Curve fit equations on Tables S5 and S6. TABLE S5. Linear and quadratic fit equations for agreement and disagreement curves as a function of knowledge for eachattitude item when neutral answers are not considered.
Linear Projection R^2 Quadratic Projection R^2Agreement 0,109x + 0,795 0,944 0,079x^2 + 0,03x + 0,808 0,976Disagreement -0,109x + 0,205 0,944 -0,079x^2 + -0,03x + 0,192 0,976Agreement -0,327x + 0,486 0,910 -0,316x^2 + -0,011x + 0,433 0,964Disagreement 0,327x + 0,514 0,910 0,316x^2 + 0,011x + 0,567 0,964Agreement -0,393x + 0,823 0,943 -0,37x^2 + -0,023x + 0,761 0,997Disagreement 0,393x + 0,177 0,943 0,37x^2 + 0,023x + 0,239 0,997Agreement -0,353x + 0,554 0,958 -0,262x^2 + -0,091x + 0,511 0,991Disagreement 0,353x + 0,446 0,958 0,262x^2 + 0,091x + 0,489 0,991Agreement -0,024x + 0,589 0,130 0,044x^2 + -0,067x + 0,596 0,159Disagreement 0,024x + 0,411 0,130 -0,044x^2 + 0,067x + 0,404 0,159Agreement -0,088x + 0,75 0,377 -0,437x^2 + 0,349x + 0,677 0,969Disagreement 0,088x + 0,25 0,377 0,437x^2 + -0,349x + 0,323 0,969Agreement 0,037x + 0,794 0,524 -0,101x^2 + 0,138x + 0,777 0,769Disagreement -0,037x + 0,206 0,524 0,101x^2 + -0,138x + 0,223 0,769Agreement -0,568x + 0,789 0,991 -0,207x^2 + -0,361x + 0,754 0,999Disagreement 0,568x + 0,211 0,991 0,207x^2 + 0,361x + 0,246 0,999Agreement -0,271x + 0,883 0,896 -0,357x^2 + 0,086x + 0,823 0,995Disagreement 0,271x + 0,117 0,896 0,357x^2 + -0,086x + 0,177 0,995Agreement 0,029x + 0,857 0,532 0,071x^2 + -0,042x + 0,869 0,729Disagreement -0,029x + 0,143 0,532 -0,071x^2 + 0,042x + 0,131 0,729att_interestatt_daily_lifeatt_fastatt_opporWithout neutral answersatt_comfortatt_natural_resourcesatt_faithatt_environatt_research_animalatt_res_dangerous
TABLE S6. Linear and quadratic fit equations for agreement and disagreement curves as a function of knowledge for eachattitude item when neutral answers are considered.
Linear Projection R^2 Quadratic Projection R^2Agreement 0,411x + 0,495 0,845 -0,641x^2 + 1,052x + 0,387 0,976Neutral -0,384x + 0,371 0,719 0,887x^2 + -1,271x + 0,52 0,966Disagreement -0,027x + 0,134 0,147 -0,247x^2 + 0,219x + 0,093 0,910Agreement -0,028x + 0,232 0,020 -0,702x^2 + 0,675x + 0,114 0,864Neutral -0,643x + 0,596 0,883 0,907x^2 + -1,55x + 0,748 0,996Disagreement 0,67x + 0,172 0,991 -0,205x^2 + 0,875x + 0,138 0,997Agreement -0,055x + 0,474 0,052 -0,902x^2 + 0,847x + 0,323 0,927Neutral -0,354x + 0,448 0,739 0,768x^2 + -1,122x + 0,577 0,963Disagreement 0,409x + 0,077 0,990 0,134x^2 + 0,275x + 0,099 0,997Agreement -0,002x + 0,264 0,000 -0,807x^2 + 0,805x + 0,129 0,952Neutral -0,664x + 0,588 0,861 1,035x^2 + -1,7x + 0,761 0,995Disagreement 0,666x + 0,148 0,991 -0,228x^2 + 0,894x + 0,11 0,998Agreement 0,214x + 0,346 0,646 -0,528x^2 + 0,742x + 0,258 0,898Neutral -0,403x + 0,414 0,708 0,98x^2 + -1,383x + 0,578 0,976Disagreement 0,189x + 0,239 0,713 -0,452x^2 + 0,641x + 0,164 0,975Agreement 0,306x + 0,38 0,515 -1,163x^2 + 1,468x + 0,186 0,992Neutral -0,541x + 0,513 0,772 1,121x^2 + -1,661x + 0,701 0,984Disagreement 0,235x + 0,106 0,959 0,042x^2 + 0,193x + 0,114 0,961Agreement 0,486x + 0,356 0,786 -0,975x^2 + 1,461x + 0,193 0,989Neutral -0,574x + 0,553 0,787 1,154x^2 + -1,728x + 0,747 0,991Disagreement 0,088x + 0,091 0,780 -0,179x^2 + 0,267x + 0,061 0,988Agreement -0,284x + 0,551 0,641 -0,767x^2 + 0,483x + 0,422 0,939Neutral -0,341x + 0,331 0,726 0,709x^2 + -1,051x + 0,45 0,926Disagreement 0,626x + 0,118 0,998 0,058x^2 + 0,568x + 0,128 0,998Agreement 0,1x + 0,53 0,101 -1,135x^2 + 1,234x + 0,339 0,943Neutral -0,41x + 0,42 0,704 0,984x^2 + -1,394x + 0,585 0,964Disagreement 0,31x + 0,051 0,979 0,15x^2 + 0,16x + 0,076 0,994Agreement 0,447x + 0,461 0,807 -0,8x^2 + 1,247x + 0,327 0,972Neutral -0,487x + 0,459 0,768 1,009x^2 + -1,496x + 0,628 0,979Disagreement 0,04x + 0,08 0,355 -0,209x^2 + 0,249x + 0,045 0,973att_interestatt_daily_lifeatt_fastatt_opporatt_comfortatt_natural_resourcesatt_faithatt_environatt_research_animalatt_res_dangerousWith neutral answers 0 D O H ) H P D O H 1 H Y H U U H V S R Q G ' R Q W . Q R Z 3 U R E D E L O L W \ $ 0 D O H ) H P D O H 1 H Y H U U H V S R Q G ' R Q W . Q R Z 3 U R E D E L O L W \ % FIG. S7. Male respondents answer less often "don’t know" than female respondents (A) Each dot represents a country andits number of respondents that never respond with “don’t know” to knowledge questions, color-coded according to gender(purple for men, green for women). Solid black lines connect male and female respondents of each country. (B) Box-plots showinterquartile range and median of the distribution of male (purple) and female (green) respondents that never respond with“don’t know” to knowledge questions (paired t-test, ***p < 0.001).FIG. S8. Within countries, male respondents systematically show higher C/K ratios than female respondents. Figure showslatitude of each country as a function of the intercept of the linear regressions in Fig. S4. Each dot (purple is male, green isfemale, black is male and female respondents together) represents one of the 31 countries. Green (female), purple (male) andblack (both) lines are the linear regressions respectively. Variable betas**Gender (female = 1, men = 0) 0.04Latitude -0.17"Don't Know" Attitudes 0.29Informed in Science topics -0.12Age 0.18Marital status (married = 1) -0.02Community (large city = 1) -0.02Religion (dominant religion = 1) -0.01Constant 0.18Adj. R2 = .262; F = 3706; N = 83368; p <.00 ** All Variables are significant in the regression equation (p <.000)
TABLE S7. Multiple Ordinary Least Square Regression Model on "don’t know" answers to knowledge questions
Attitudes: att_comfort att_nat_resour att_faith att_environ att_res_animal att_res_danger att_interest att_daily_life att_fast att_oppor
Variable Gender (female = 1, men = 0) -0.07 * 0.07 * -0.13 ** 0.09 ** -0.09 ** 0 0.06 * -0.17 ** -0.12 ** 0.02Latitude -2.07 ** -2.08 ** -1.16 ** -2.35 ** -1.91 ** -1.90 ** -2.02 ** -2.32 ** -1.40 ** -2.37 **Informed in Science topics -1.41 ** -1.25 ** -0.87 ** -1.52 ** -1.00 ** -1.33 ** -1.27 ** -1.36 ** -1.30 ** -1.28 **Age -0.20 ** 0.28 ** -0.42 ** 0.40 ** -0.30 ** -0.12 0.48 ** -0.4 ** -0.45 ** 0.20 **Marital status (married = 1) -0.48 ** -0.36 ** -0.26 ** -0.45 ** -0.38 ** -0.40 ** -0.36 ** -0.39 ** -0.40 ** -0.40 **Community (large city = 1) -0.37 ** -0.29 ** -0.21 ** -0.36 ** -0.31 ** -0.32 ** -0.35 ** -0.38 ** -0.31 ** -0.39 **Religion (dominant religion = 1) -0.13 ** 0.05 0.07 ** 0.08 ** -0.09 ** 0.04 -0.18 ** -0.03 -0.07 ** -0.17 **Pseudo R2 -0.007 0.029 -0.002 0.043 -0.008 0.014 0.026 -0.007 -0.006 0.023Log-Likelihood -26221 -27564 -36442 -24465 -24389 -24966 -26252 -20397 -29921 -22981LL-Null -26038 -27574 -36334 -25573 -24193 -25329 -26960 -20258 -29738 -23529LLR p-value 1 0 1 0 1 0 0 1 1 0No. Observations 67903 56242 67903 56242 56242 56318 56318 56318 67979 56318Coefficients with * p < .01, ** p < .000