>>Um, so yeah I’m going to talk about psychographic targeting in politic- political campaigns.  Um, psychographic meaning using personality and behavior, uh to target uh ad specific groups of people.  Uh, I’m going to speak a little bit more about the title of the talk in a couple of slides.  Um, but let me explain how we got into this uh research and this particular talk. 
So, in 2016 we did some research into the EU Referendum, some of you might know as Brexit.  Uh, we did six studies consisting of about eleven thousand participants and we looked at personality differences.  Differences in numerous skills uh, thinking styles and cognitive biases.  And in three of those studies we added a question around if, can, can you all hear me?  I guess.  [off-mice audience comments] Closer?  Wow.  I’ll probably be eating it but, uh ok.  So, in, um, in three of those studies with about five thousand people we added an additional question that’s, uh, asked people’s opinions to how much they agreed on the nothing uh, nothing to hide nothing to fear argument.  So, if you’ve got nothing to hide you’ve got nothing to fear.  What we saw in that research was that people who voted to leave the European Union were much more likely to support the nothing to hide nothing to fear argument than those that voted to remain.   So, we wanted to study that in greater depth to understand kind of what, what was going on there.  Uh and understand a little bit more about the drivers.  Then in around January of uh this year, some of my friends started pointing out articles like this which appeared in Scout AI and this is where we got the title of the talk from really.  They introduced it as the rise of the weaponized AI propaganda machine.  So, using psychographic information in our political campaigns.
So, my friends weren’t the only people that were beginning to sort of question the efficacy of ya know, how well does this work essentially. Uh, other journalists started looking, ya know looking at that and saying well does this really hold water or, or not?  So, we’ve got a little bit of background here in this topic. In 2013 we presented at Def Con on predicting the susceptibility to social bots on Twitter.  So, seeing which users would interact with a social bot based on their Twitter activity. Uh and we also looked at personality in, in that.  In 2011 and 2012 we looked at determining personality through Facebook activity, uh and through Twitter.  Both of those were pretend at Def Con also.  So, what we decided to do was use the, uh, using Attitudes Online Surveillance with digital rights. Use that as like the central thesis for exploring the efficacy of psychographic marketing in political campaigns. 
So, the rest of the talk is structured into five main sections.  First of all, we look at why people are divided.  Then we move onto what influences their views and nudges them one way or another.  Then we look at, once we, once we understand how people are divided, assuming they are, which spoiler alert, they are, how can you effectively target groups with different attitudes in relation to online communication surveillance?  Uh, once we know how we can target different users, or different people essentially, we look at um, how persuasive are targeted ads.  And finally, we wrap up with looking at, okay once the message is out there, how effective it, how, how able are you to sort of debunk misleading information or misinterpreted information once it gets out of the gate?
So, onto the first question.  Looking at why people are divided.  Well we chose the nothing to fear, nothing to hide argument.  But we could have chosen the, the death penalty.  We could have chose the topic of same sex marriage and we could have chosen the topic of immigration. So, what, what seems to be at the center or at least a major factor in all of these differing world views is uh, something called Authoritarianism.  Which, uh began as a study in the, I guess following the second world war but lost some favor, some credibility in those early years and has regained credibility in more recent years, in the last few decades essentially.  Um, so Authoritarianism, as you look at it you might be forgiven for thinking it’s a bimodal distribution with people either low or high and not a great deal in the middle.  But what we’ve actually, ya know what we actually find is that it, it’s a much more skewed distribution than that.  At least in uh the Facebook samples that we see. 
So, what, what characterizes Authoritarianism is that people high in Authoritarianism tend to see the world in, uh in so, black and white.  Whereas people lower in Authoritarianism see it tend to see in sort of shades, shades of grey really and it’s a sliding scale.  Ya know, so people are, it, it’s very rare to have people at the certain extremes essentially.  People who are higher in Authoritari-, Authoritarianism um, have, are more concerned about the upkeep of societal norms and traditions whereas people lower in Authoritarianism, uh are more embracing of new cultures.  Um, new ideas.  So, uh, the, the footloose image there.  Uh, additionally what we see is that those that are higher in Authoritarianism have lower tolerance for out-groups. Uh, whereas people who are lower in Authoritarianism uh are much more embracing of uh different cultures or ya know difference races and what have you. 
So, the scales that we use throughout, our studies stem from some work by uh, researchers called Feldman and Stenner in 1997 where they developed uh a very simple way of looking at Authoritarianism.  They developed a, a four question, uh if you like child rearing questionnaire.  So, if you’re um, if you select answers to these questions which, ya know you have to select one of the other for each of the line items.  If you’re predominately choosing answers on the right-hand side than you’re more, you’re higher in Authoritarianism than people um choosing answers on the left-hand side.  Ya know generally, there’s some exceptions to that. 
So, moving onto our study.  Uh and our results for the first thing and looking at why people, uh why people differ.  Ya know we chose Authoritarianism to uh, to, to, to look at in greater depth.  We conducted uh four different studies, uh over, well from February to, uh to May with differing sample sizes.  You’ll note that there’s studies uh, A and studies B there’s a slight difference in, in, in the uh studies.  All of the questions, all of the studies uh took um, basic demographic information, age and sex and they asked, uh a number of questions about people’s attitudes to online surveillance.  Which we essentially replicated form prior studies or prior surveys, so that we could actually go back and reference and see if our numbers were sort of consistent.  The difference between studies B and A was uh, uh the, the B studies also contained a numerical question which we’ll get to a little bit later in the, in the presentation.  Because the studies differed we treat them kind of individually as well.  We don’t group them, uh with, uh, with one exception. 
So, we had about two thousand four hundred people took part in those, those surveys.  Which we ran through Facebook.  So, the questions we asked, were uh, your, how much do, ya know do you agree? And it was a binary choice, the binary choice in the uh, question stemmed from work from Professor Hetherington and Elizabeth Suhay who had looked at uh, previously looked at uh, Authoritarianism in American politics in 2011.  So, we wanted to try and replicate some of their work, so.  Do you have nothing to fear?  Do you agree or disagree with you have nothing to fear if you have nothing, you have nothing to fear if you have nothing to hide?  So, it was, the questions replicate our EU referendum work.  Then the next question was, uh it is acceptable that law enforcement agencies have the right to access the content of citizen’s online communication without a warrant in order to investigate terrorism.  That question was from the, the Hetherington study.  Then the next question we ask is, it is acceptable that immigrants and visitors from potentially dangerous countries should have to reveal their social media account passwords to border agents.  So, we, we used the UK samples, but you can do this on a US sample set, too.  The next question, which came from a survey on attitudes to privacy in 2016, were simply the darknet should be shut down.  Now quite sure how you do that, but um, but it comes from a survey.  So, we wanted to, you know it gave me a chuckle so we added that in.  And the final question we asked was companies should not be allowed to develop technologies that prevent law enforcement from accessing your online conversations.  And that came from that same uh, privacy survey.
So, going back to our surveys I mentioned we treat them differently because, uh there were different um, B, studies B and A have, were, were slightly different so we treated them different statistically.  Uh the other thing that happened was that in March uh, late March there was a terrorist attack in London.  Um, so what we wanted to do was, well see what the effect was of the terrorist attack attitudes to online surveillance.  And there’s been previous work in that space, uh researchers have looked at the differing attitudes before and after the Charlie Hebdo attacks in, in Paris.  So, what we see from the results of uh surveys A1 and A2, the ones without the numeric question is that uh, for most of the questions you get around a sort of a thirty, thir- ya know, thirty percentage agree with the statement.  Uh only one of those was statistically um, different before and after the attacks.  And that was the banning the anti-surveillance technology question.  Uh the rest while it looks like there is a visual difference, well there is a visual difference, um we didn’t see a statistically significant different between opinions. 
Um the thing that interested me or fascinated me the most about the results is that there was uh ya know more support for shutting down the dark net, people just want to shut that circuit down essentially. 
Um, so looking at studies B1 and B2, uh which we can, the, the B2 study was conducted forty-one days after the, uh, a- attacks.  We see that there’s almost no difference in opinions uh from before and after.  So, the opinion sort of drop off after the, uh, after the terrorist attacks in those surveys.  So, we see broadly the, the level of support there is hovers, hovers around th- thirty percent and we’ll get a little, we’ll get to that a little more in the presentation. 
So, looking at the role of Authoritarianism in, uh, in these attitudes when we look at the nothing to hide, nothing to fear question, uh in particular.  We see, um you know somewhere between a one and a half to a two and a half uh, times inc- increase between those who agree with the statement being more Authoritarian uh versus those who disagree with the statement.  Uh, we also performed like, um, uh, uh the logistic regression essentially to look at plotting the odds and some other statistical methods and we find the same difference. You know, broadly through all of them.  In terms of, the, the other questions we see sort of broadly similar results as well around wiretapping, shutting down the dark net, banning anti-surveillance technology, and extreme vetting. 
So, we see consistently, not just across the questions but also between all of the different age groups and sexes uh that the, these opinions tend to, tend to be fairly consistent.  So, assuming that Authoritarianism is more a, a constant, what is it that can influence people’s uh, differing views? 
So, it’s said, uh by a gentleman called Bob, uh, Altemeyer who developed the uh, what was called the Right Wing Authoritarian Scale in 1996 that uh Authoritarians are characterized as being like ten steps closer to the panic button.  So, we saw a plot there on, on this graph and show the blue line is essentially the, the baseline of support for, uh for surveillance.  Then there are, uh two streams of research that have looked at this previously.  The first stream of research uh, by Feldman and Stenner, had uh, it essentially argued that a threat inc-, the, the perception of threat increases the support among those who are higher in Authoritarianism.  So, uh the, the folks at the higher end of the scale are going to support it more if they are more concerned about the threats of terrorism. Another stream of research by Hetherington and Suhay in two thousand and- 2011, found that the threat increased su-, support among those lower in uh, Authoritarianism so um, people at the higher end didn’t really change their opinion if something, um ya know if they perceived a threat to be high but those much lower more likely to adopt an Authoritarianism position, uh, if they perceived the threat to be high. 
Uh, the researchers that looked at the difference in uh, before and after the Charlie Hebdo attack which are mentioned a little bit early looked at this and suggest that both are partly right.  And that it’s anxiety that, um is responsible for causing the uplift in those that are lower in Authoritarianism if they are concerned about the threat of terror versus uh, anger in uh, people higher in Authoritarianism, if they perceived the threat of terror to be high. 
So, in the study that we conducted uh what we’ve done in all the studies we had uh, uh a question simply asked, uh, how worried are you that you might personally become a victim uh, of a terrorist attack?  And that was coded. People could answer in like four ways from like I’m not worried at all to, yea I’m very concerned. And that replicates the work from Hetherington and Suhay.  So, what we did, is we used a logistic regression.  Essentially uh, replicating Hetherington’s work.  Um plotting the probability for support for each of those questions against Authoritarianism and then looking at the effect of people’s perceptions on, uh, on threats. 
So, Hetherington and Suhay’s work, I figured out that I- I- I am just so sorry about the, the bad quality photo from uh, from Hetherington’s book.  But here you can see is the effect that he was talking about.  Essentially people lower in Authoritarianism, you see that the support, uh really increases from about twenty-five percent of support on their wiretapping question to about uh seventy-five, eighty percent probability of support.  And you’ll see at the higher end of Authoritarianism there’s really not a great deal of difference. 
So, in uh, our study, in terms of the, uh nothing to hide nothing to fear question, we see essentially a similar sort of effect happening to what Hetherington and Suhay found.  Um, so going from about an eighteen percent probability of support for those who are not concerned um and are lower in Authoritarianism.  That jumped to about a sixty percent um level of support.  But there’s not many in that category.  Uh and at the higher end there’s almost no difference.  Looking at all of the other questions uh, at least for wiretapping and banning anti-surveillance tech we see a similar sort of trend going on.  Um, I’ll talk about extreme vetting and shutting down the dark net shortly.
So, extreme vetting is uh, is, is an anomaly but Hetherington had seen in uh, or, had seen a similar sort of uh interaction in one of his other questions.  So, that needs more, more analysis.  But I didn’t want to shirk away from calling that one out. 
Uh the other thing, which interested me was that you, here you see reflected the high level of support for shutting down the dark net and you see how that jumps up even from like a medium high, very high level.  You see the interactions so that if people are concerned of the threat of terror, you know, there’s broad support for uh, tackling the, the dark net. 
So, okay, now we’ve looked at how people differ around the, the uh disposition of Authoritarianism.  Uh we can see the role, uh perception of threat has uh on people’s attitudes.  Now we look at, okay how can we target those different groups of people? How can we target people who are low in Authoritarianism versus people who are high in Authoritarianism?   So essentially a lot of research is looked at, um the relationship between age, sex, geography, Facebook interest, Facebook activity and such like and personality.  What we want to be able to do here is almost get uh, personality, in our case we use Authoritarianism, put it through a function and get the age, sex, geography and Facebook interests that are going to be most closely associated with people that are either low or high in Authoritarianism.  So, if you like, we create two book-it, one book-it for low Authoritarian and one book-it for uh, high Authoritarian.
So, onto the study on our, our findings.  Uh first of all, tackling age and sex.  What we saw in the EU Referendum studies was that the difference between female voters who voted to leave or remain in the EU had much closer level of Authoritarianism.  There wasn’t a great deal to separate them and indeed looking at the results there are very low percentages of females in eighteen to thirty-five age group who wanted to uh leave, the, the EU.  We also saw that reflected in uh the studies that we, we’d look at there.  It, with regards to uh, the, um the nothing to fear nothing to hide argument, online um, communication and surveillance.  You’ll see that as females get older the difference extends so there is a starker difference.  And that gets like more consistent with males.  Of males, there was, uh there was differences at every age.  So, in terms of the book-its we created and we created these book-its so we could add them into the Facebook advertising audiences.  
Uh, we have the low Authoritarian groups, of males of any age but females only if they’re under thirty-five and for the high Authoritarian category we used males of any age and females who are thirty-five and over.  So, that’s a simple uh, like age and sex classification.  And, uh it, it is, you know rather typically simple but as we see it becomes quite effective. 
So, for geography, uh Rentfrow and uh, and other researchers have looked at regional differences in personality.  These are maps of the United Kingdom and the letters at the bottom reflect what’s called the Big Five personalities.  So, you’ve got openness to experience, uh which is you know strongly correlated with creativity.  Conscientiousness, which is more concerned about some of the, the traditions and what have you or relate, or correlates with that.  You’ve got extroversion, agreeableness and neuroticism.  So, he’s looked at different areas with a, a very large uh samples that across the UK and found different regional, uh regional differences.  He’s also done that with the United States as well though if you’re interested in those regional differences you can check that out. 
So, he’s presented uh, a lot of results and findings for each of about three hundred different local area districts in the United Kingdom.  So, what we did is we mapped Authoritarianism to the big five using uh some matter analysis from uh researchers Sibley and Duckett.  And in terms of Authoritarianism it’s classified uh or high Authoritarianism is classified by high degrees of conscientious and lower degrees of openness mostly.  There’s a slight difference in neuroticism but we can you know, pretty safely ignore that.  So, for our research what we did, we plotted openness on the uh, openness here, but we reverse it. So, we get low openness essentially.  Um, against conscientiousness.  And then we take the results from Rentfrow’s study of uh over three hundred areas in the UK and we plot them on a graph that’s a little bit like this.  So, you get the high Authoritarians at the top right hand corner and you get the low Authoritarians towards the uh bottom left.  And um, since it’s a hacker conference it’s quite interesting that the least Authoritarian place in the United Kingdom is called Hackney in London.  And I didn’t make that up for the, for the talk. 
But um, so we chose some towns, um that were most likely to be in our opinion high Authoritarian and some towns and cities that were most likely to be low in Authoritarianism.  We also looked at like how many people were there for reach and stuff like that.  And so, we have some other characteristics that we, we looked at.  The towns that we chose for low Authoritarian were towns that most of you have probably heard of, Cambridge, Liverpool, Edinburgh and now you have all heard of Hackney.  Um, and in the high Authoritarian town we’ve got high Authoritarian districts we’ve got places like Basildon, Thurrock and Mansfield.  Uh and Swindon, you might have heard of Swindon if you like the game, like my son does, The Amazon Frog and Pungence’s YouTube channel.  Other than that, I’m not sure that any of those towns are that famous outside of the United Kingdom.  So now we’ve got a book-it for age and sex and we’ve got the regions that we’re going to use in our Facebook advertising. And incidentally the results that we found using just a simple logistic regression were actually really good at correctly classifying the EU Referendum vote.  And we did the same thing when we looked at the uh regional differences in the US, um predicting the, uh recent election that you guys had.
So, onto Facebook interests.  There’s a tool here called the PreferenceTool, uh that’s available to researchers where you use like slider bars if you like to put in uh difference effects of personality and that shows you some of the Facebook interests that those groups use.  Now, we did not use the preference tool because it’s available only to uh, to researchers.  Um, and we did, we didn’t need to do that really because we looked at some of the Facebooks interests that we already knew were highly correlated with Authoritarianism and low Authoritarians.  So, the low Authoritarian category we chose just simply the interest of liberalism, uh which is a Facebook interests.  And um, the Guardian newspaper which I guess is like your Washington Post, I guess.  Uh, for the high Authoritarian category we chose conservatism and, and the Mail Online.  I’m not sure what the equivalent US newspaper is.  [laughter]
Somebody shouted something out but I didn’t hear it.  But um, so the, the baseline we’ve got to, to, to beat if you like is uh the nothing to hide, nothing to fear argument we had uh sort of a baseline of thirty-eight percent support for that particular argument in the studies that we use in this sample set.  For the targeted audience just based on uh, age, sex and geography we saw a drop from thirty-eight percent to twenty-two, uh thirty-two percent but it wasn’t statistically significant.  We saw an uplift in um, support for the high Authoritarian towns from thirty-eight to forty-six.  Um when we added the Facebook interests in that we looked at for the correlation to uh to, to Authoritarianism we saw that we started to get significant results although for the first group it was relatively low.  Possibly due to the overall skew to look towards low Authoritarianism anyway.  So, from thirty-eight agreement with the argument to twenty-five is not bad.  Uh, for a very simple experiment.
And in the high Authoritarian category adding in personality we see the jump go from thirty-eight percent to sixty-one percent, um which was highly statically significant.  A quick note on the significance, um if, if I’m doing this in the wild I might not be so interested or worried about P levels and stuff like that.  If I consistently see the results like that maybe I just take a gamble on it anyway and it might give me sort of a higher return on my advertising investment. 
So, okay, we’ve seen the differences in personality and we’ve seen how we can target different groups based on just some um, simple factors that uh, uh enable us to reach audiences with uh significantly different views on online communication surveillance.  But, now we want to turn to, well what about persuasive ads and uh, targeting of ads?  So, this is uh Alexander Nix from Cambridge Analytica and he states and this video is, is online, if you knew the personality of the people you’re targeting you can nuance your messaging to resonate more effectively with those key audience groups. 
So, there’s been some research in this space.  The study by Hirsh, Kang and Bodenhausen. They looked at, they created five separate telephone uh ads, that would design to appeal to different levels of the Big Five of personality.  So, this is the app that they used or at least this is the text that they used and uh, in uh app that was aimed at uh, extroversion.  The XPhone, with Xphone you’ll always be where the excitement is.  And then for neuroticism, they have Stay safe and secure with the Xphone.  And they always saw statistically significant differences.  Uh, Matz, Popov, Kosinski and Stillwell did a similar thing looking at aiming at introverts and extroverts, uh for a beauty product and they saw that indeed uh, targeting those ads based on, um, on the personality had uh, significant uh effect on their click through rates so their return on their advertising investment.  
So, moving onto our study and results, what we did was we had ads that were aimed at um, people in the high Authoritarian book-it and ads that were targeted at people who were in the low Authoritarian book-its.  And we split that into, uh to ads that are designed to be pro-surveillance and ads that are uh, designed to be anti-surveillance.  So, the first ad for pro was really appealing to uh, to the high Authoritarian categories. Terrorists, don’t let them hide online.  Um, so hopefully invoking some anger there.  For the low Authoritarian group, we tried to highlight that uh there was more than just terrorism uh, online.  We had the tagline, Crime doesn’t stop where the internet starts.  And I know it’s cheesy but um, the, the thinking was that people would look at uh, uh broader crimes like human traffic-, trafficking and child exploitation.  
So, for the uh low authoritarian group with the anti-surveillance message we used the image of uh, Anne Frank and said you know, do you really have nothing to fear if you’ve got nothing to hide?  And for the high Authoritarian group we used an image of the D-Day landings um appealing to the Authoritarian characteristics of um, affiliation with military and respect for elders.  So, those were the ads we categorized and then in looking at the results of those ads we, uh references them here and these are the self-reported levels of Authoritarianism and people that took part of our surveys or uh, samples to uh, to, to rate the ads.  So, they were asked a bunch of questions, like how much do you like that ad?  How much does the ad resonate with my beliefs, uh how persuasive is the ad, how likely am I to click like, or share on the app?  That, that kind of thing.  And we split that then in people’s self-reported um, self-reported groups. 
So, this was, this was actually where the, where the uh persuasive of the ad is rated.  And that is, that is what people’s level of Authoritarianism were classed as.  So, high, medium and low.  Okay, so the performance of the high Authoritarian ad.  We see that people who are high in Authoritarianism like the ad much more than people who were like medium and low.  When we moved to the ad for the low Authoritarian group we see people who were high still like it.  It’s kind of bought in because they’re, they’re, they’re more pre, more pro uh surveillance anyway.  But we see then an uplift in uh, uh well an expected uplift in the low and medium.  We see an uplift in the low and medium Authoritarian categories.  So, the ad seems to be having the sort of desired effect if you’d like.  For the low Authoritarian category in anti-surveillance we see kind of a reversal of the higher auth-, uh pro add where people who are low in Authoritarianism like it and people who are high don’t like it so much.
Um, when we target the ad towards or when we created the ads for um, the high Authoritarian group we see a similar sort of uplift to the one we’ve got in the top right hand corner.  Um, and there’s a broad level of uh increase of support from those medium and high.  So, so that’s people’s responses to how much they like the ad.  Then we took it a step further and tested it in the wild with, with um Facebook ads designed or actually, there were Facebook posts and we used the boosting in the Facebook advertising, um to appeal to those different uh, those different audiences.  So, here we see the Facebook target audiences of high and low Authoritarian.  There’s only tow, tow group from our target audience.  And then um, the, the ads in those, um boxes uh re- relate to the ads we just saw.  So, what we do for this, using the, the Facebook advertising essentially is then look at the click through rates or actually, this is actually the ratios of likes and shares by, by the level of reach for each ad. 
Um so the numbers look pretty low but I actually um you know those are fairly consistent with click through rates that you would expect.  So, for the ad aimed at the high auth group we see as expected people who are in high in Authoritarian like the ad, people who are low think it sucks.  Uh, in the low Authoritarian ad uh there was a, a lot less traction altogether and fakebook gives you forewarning. It basically says that you know, ads that got a lot of text uh, do less well.  You might want to consider changing your ad. 
But what I draw your attention to, at least is that there was a, an expected difference in the, the correlation coefficient if you like.  So, the, the gap uh the gap reduced between the, the low and high.  So, if we move onto the low Authoritarian ads, uh for anti-surveillance we see the expected results that low, people in the low Authoritarian target like the ad or interact with the ad much more than those in the high Authoritarian or more than those in the high.  But then what we see for the ad that’s targeted at people in the high Authoritarian, uh, uh, designed to appeal to people who are high in Authoritarianism and anti.  We still get a lot of people who are low in Authoritarian doesn’t like it.  Because they like the message of I don’t, you know, they don’t, um, on, online communication surveillance or the idea of it. But people higher end support it a lot more. That was kind of surprising to us.  We, we thought we’d actually get more of a backlash from those, uh those groups.  Uh rather than, rather than the, the interaction so again we see the correlation coefficient reduces as we would have expected from the ads. 
But uh, some of you might be thinking, ok well, that’s well and good, we’re seeing how much people are liking click on ads, but does persuasion actually work.  You haven’t actually told us that, fair, fair comment, I haven’t.  Uh and there is very little research in, into that.  But this research  by, um, researchers at the University of Mannheim in, in Germany, Jungherr, Wuttke, and Co.  What they did is say, uh took people’s attitudes to transatlantic trade and the sample set was a large, it was about 8,000 people. Then they sent some of them some ads, um explaining or some literature about the benefits of transatlantic trade, uh and they found that over, the, the baseline which is the zero line, the, the control of people who didn’t receive the literature, they found that people became more, much more supportive after they, of transatlantic trade after they received the uh, the literature.  And they look at um, the difference in opinions shortly after receiving the literature and then some time after and they did not- they did notice a drop off in uh, the time. But uh you know it suggests that if you’re wanting to be, get persuasive material out there, the time is of the essence. 
So, okay, now we’ve seen, you know you can get the message out there but what if the message is erroneous?  SO how would you go about sort of debunking propaganda.  Well to, to this, well it’s not really how we care about debunking it.  TI’s what are the challenges of debunking propaganda. So, for this we turn to some research from Professor uh, Dan Kahan at Yale University.  He divided participants in his experiment into 4 categories, A, B, C, and D.  For categories A and B, he gave them a numerical question about how well uh, skin cream does at treating a rash.  Um, and he gave them a number grid which like this.  So, group A received this number grid and were asked to say, okay did skin cream essentially make the rash better or did it make it worse?  And people had to go figure that out.
Whereas group B, the text was uh, slightly changed.  Um but the numbers remained the same, so the rash gets better or the rash gets worse.  Well they found that people did roughly the, uh had roughly the same um, ability to answer the, the, the question.  But when he changed the question to one about uh, does gun control reduce or increase crime, he found that people’s ideological beliefs really hampered their ability to answer the question. And I’ll demonstrate that because we replicated uh Kahan’s study in our EU Referendum work.  But we changed the question from one about gun control to one about immigration increasing or decreasing crime.  So here we see how um, how people do at the skin cream ad if you’d like so uh, broadly consistent.  Uh just under 60% getting it right in the referendum result, leave vs. remain. 
But when we ask about does immigration uh, decrease crime people who voted to, to leave would just not have any of that or were unable to answer the question or their performance tanked essentially. Whereas people who voted to remain their performance at answering the question actually increased.  Then when we switch it to immigration increases crime then the leave voters start doing better again. But the remain voters performance tanks.  Uh, and uh, the, the, the research that he’s conducted is, is termed ‘Ideologically Motivated Reasoning’ and it was dubbed, uh in one article the most depressing brain finding ever, which I liked.  His study was from a little while ago but it’s pretty exciting.  He’s got uh, a web page you should go check out called the cultural cognition uh project and it’s well worth a read.
So, in our study, um we sort of replicated that research again but used the, the, you know the attitudes to online surveillance.  So in the skin cream ads we see similar sorts of trends between those that agree and disagree with the nothing to fear nothing to hide argument.  But then when we suggest that surveillance actually increases the threat of a terrorist attack than those who agree with the statement, their performance drops quite significantly.  Whereas those who disagree, their performance drops a bit.  Um and I would suggest that the performance drops a bit because the numerical question is actually, the, the text in it is actually harder to read than it was for the immigration and gun control.
When we switched to a surveillance actually does decrease the threat then we see the interaction just not as starkly.  Uh however the interaction does seem to increase as people get older.  Um, so you see a more significant result as people get over 35 and here we just show 44 and over.  So, in terms of uh, wrapping the, wrapping the talk up.  What we’ve seen is that Authoritarianism plays a part in why, why people differ or why their opinions differ.  We’ve found that the perception of, of threat plays a part in actually influencing or nudging people from a baseline position to a different position.  Uh we found that we are able to actually target people with different attitudes to online uh, surveillance. Uh just by using uh some simple knowledge about the, the psychological traits associated with Authoritarianism.  Uh we find that the persuasiveness of, of ads does indeed um seem to work and that challenging or tackling misleading information is going to be a, a very challenging problem because people's ability to, to uh interpret uh evidence essentially becomes a lot harder if it goes against their existing beliefs. 
So question we received uh recently was ok, well if both sides are using this do the effects equal each other out?  Uh and I thought about that and though oh yeah in theory they sort of do, uh or could do.  But this, and, and another potential theory is that it’s possible that one side has a home team advantage essentially.  Um, so one of the ways they’d have a home team advantage is, well if you’d got momentum on your side um then you can go with that and just like double down on it.  So here we see the shuttin down the dark net question um and a level of support for it.  Well if you just go with that and just increase the, the already existing fears that people have.  Um, you, it’s going to be a lot harder to be in the defense, to, to defend that ar- that argument, than it is to attack the argument if you like. 
Uh there’s a lot uh the role of uh, of fear.  So, uh if you’ve not see it there’s a great video on line by Dee Madigan on propaganda. And Dee Madigan is a Australian uh, campaign advertiser.  And does a, a really fantastic talk.  And she talks about the uh effect of fear based advertising. And she, she says yeah, people have said that, have said to her that negative campaigns really such and they don’t’ work on my.  And her comment was essentially that, yes negative campaigns do work and that’s why we do them.  Um, something like that.  Anyway, paraphrasing. But uh, the fear actually, or increasing fear works.  We’ve seen some studies on that as well from Brader in 2005.  That shows how political ads motivate and persuade voters uh based on emotion.  People, uh, people have higher attention for uh ads that are, uh create more, more fear if you like.  And you saw grainy black and white images versus ads that have a, a lower um, versus ads that you use pictures of smiling children if you like in front of flags and you know all that kind of good stuff.  So people’s attention is drawn to the more negative ads in the first place. 
So ok, some thoughts on how, um how folks might or how society might begin tackling the problem. One of the things we saw that the effective cognitive biases in the EU referendum uh research that we’ve, we’ve done.  And that research we’re going to be talking about at the international conference on political psychology in October.  And we’ve teamed up with some researchers at Missouri State, to, to help us tackle the challenge of uh, uh extremely non-normal data.  Uh, but one of the ways we can tackle this is, uh or tackle effective cognitive biases is perhaps adopt a method that, uh, this lady here Jane Elliott had used.  And if you’re not familiar with Jane Elliott’s work, she basically splits um, participants, it was with children in a classroom but she’s done this with older group, into people with blue or brown eyes, uh and uses that as the basis for helping educate about racism and the effects of racism and kind of how that works.  So, it’s possible that you could do something similar with, in relation to, to cognitive biases to make people aware of that.  Um, which would have benefits outside the political campaigns to prevent people getting hustled at uh, car dealerships and stuff like that. 
The other thing we noticed is that levels of numeracy were right, really shockingly poor in the UK.  And I don’t think the UK, uh the US is a great deal better.  Um, so increasing people’s numerical literacy skills is being able to determine that 1,000 % APR is a bad deal on a credit card.  Um, but it also has effects quite, um you, uh, it also has effects in people choosing medical options and what not as, as well.  There’s a lot of research actually in that area liter-, risk literacy and like making medical based uh choices.  So, that needs increasing I think uh all around as well in addition to the biases. 
And then the next thing which I think is going to be a lot harder to achieve because literally anyone can do this, so if you’ve got big pockets you’re going to get ya know, a lot more attraction.  Is really start enforcing or going after people who are uh abusing this sort of, uh this sort of political advertising.  So, that wraps up the uh, the presentation. I’m sure folks will have uh, questions.  You can reach me at uh, chris at onlineprivacyfoundation dot org.  Um, I’m happy to address questions afterwards but we’ll have to go outside somewhere I think.  And I’ll leave some uh, some cards on the table if folks have questions.  Aright?  Thanks very much. [applause]