1. TELLTALE SIGNS OF CONSERVATISM AND LIBERALISM IN
LANGUAGE:
Approach-Avoidance Emotions and Rhetoric
LucasCzarnecki
University of Calgary
Dept. of Political Science
2. “LANGUAGE IS AWINDOW INTOTHE MIND”
- S. PINKER
LucasCzarnecki
University of Calgary
Dept. of Political Science
3. The Canadian Context Left
Right
Center
(Photo Credit: Office of the Prime Minister, via Twitter)
www.canadianelectionsdatabase.ca
(shameless self-promotion, full database launches July 1st 2017 )
4. Understanding
the Left/Right Divide
See: Cochrane, C. (2015). Left and Right: the small world of political ideas. McGill: University Press
And: Jost, J.T. (2006). “The End of the End of Ideology”, American Psychologist, 61(7), 651-670.
(Photo Credit: Koren Shadmi)
The Classical (or
Essentialist)View
TheAlternative
View
Shifting
away…Old
Paradigms
Laponce (1981)
It’s…Religiosity vs Secularism…
Bobbio (1996) & Inglehart (1990)
Nope… extent of egalitarianism
Noel &Therien (2008)
Nope… definition of egalitarianism
Mervis & Rosch (1981)
Rejects “conceptual utopia”
Dif. are gradual & incremental
Alford (2015), Oskarsson et al. (2015) Jost & Amodio (2012)
Inheritance/Predisposition?
Cohen (2003), Haidt (2001), Danziger et al. (2010)
Humans are not so rational!
Where do ideologies & partisan
loyalties come from? … …
5. *One* explanation: Approach-Avoidance Emotions
This is… Anxiety (avoidance-based)
Overestimates risk
Status-quo oriented (risk-avoidance choices)
Risk reduction (concerned with uncertainty)
This is… Anger (approach-based)
Underestimates risk
Change oriented (risk-seeking choices)
Moral anger (addresses injustices “no dessert!”)
See: Marcus, G.E. (2003). The Psychology of Emotion and Politics. In Sears, D.O., Huddy, L., & Jervis, R. (Eds.) The Oxford Handbook of Political Psychology. (pp. 183-207).
See also: Lerner J.S., and Keltner, D. (2001). “Fear, Anger, Risk”. Journal of Personality and Social Psychology 81(1), 146-159.
Liberals
Personality (Big 5):
Openness to New
Experiences
Conservatives
Personality (Big 5):
Contentiousness
Neuroticism
Predisposition != Determinism
6. DO IDEOLOGICAL DIFFERENCES MANIFEST IN
LANGUAGE?
H: Freq. of Approach-Avoidance emotions words will depend on a group’s ideology
H(i): Higher Freq./Volume of AngerWords - left
H(ii): Higher Freq./Volume of AnxietyWords – right
H(iii): Similar patterns for other correlates – e.g. happiness
7. Data Collection
The Canadian 2015 General Election Campaign Strengths & Limitations
Dates (78-day Campaign)
4/Aug/2015 until 19/Oct/2015
Leader Posts
N= 1,712
Harper = 525 |Trudeau = 531 | Mulcair = 656
Comments
N = 297, 830
Con = 150,529 | lib = 97,041 | ndp = 50,260
Pros:
ExternalValidity (not perfect)
Sample Size
Cons:
Controlling variables (e.g. age, sex, education, etc.)
Data Independence
Barbera, P. (2016). Rfacebook (version 0.63)
Data Scraped/Preprocessed
10. Frequency of Anger-related words from Facebook Commentators during the 2015
Canadian General Election Campaign
See: Lerner J.S., and Keltner, D. (2001). “Fear, Anger, Risk”. Journal of Personality andSocial Psychology 81(1),
146-159.
11. Frequency of Anxiety-relatedWords from Facebook Commentators during the 2015
Canadian General Election Campaign
See: Sylwester, K. and Matthew Purver. (2015). “Twitter Language Use Reflects Psychological Differences
between Democrats and Republicans”. PLoS ONE, 10(9), 1-18.
12. Frequency of Positive Emotion Words from Facebook Commentators during the 2015
Canadian General Election Campaign
See: Schlenker, B.R., Chambers, J.R., and Le, B.M. (2012). “Conservatives are happier than liberals, but why?
Political ideology, personality, and life satisfaction”. Journal of Research in Personality. 46, 127-146.
14. “If I can only hold your attention for a few minutes,
and I can either tell you how I helped an old lady cross the street
or how my opponent kicked a cat,
it makes more sense for me to tell you about the cat.”
- Jon Krosnick
15. “…The wrong decisions on taxes, spending, and deficits
will expose Canada to the very real global economic instability that surrounds us.
They will negatively impact our economy and jobs from coast to coast to coast.
We have to make sure that doesn’t happen…”
- Stephen Harper
16. “The status quo isn't working.
Our plan will make a positive difference in people's lives”
- JustinTrudeau
17. Political Rhetoric During a Campaign
Party Leader Affiliation Achieve Power Reward Risk
Harper
N = 525
6.99 3.12 4.92 2.03 1.57
Trudeau
N = 531
5.84 3.88 4.41 1.88 0.72
Mulcair
N = 656
4.36 3.46 4.38 1.68 0.57
Mean Frequency of Word Categories relating to ‘Psychological Drives’ for Canadian Party Leaders. Cell values are
percentages calculated as (raw count of word category / total words). N are the number of Facebook posts.
18. Future Research
Latent Semantic Analysis (LSA)
Procedures:
1. Create text matrix (m x n)
2. Cell transformations
• Raw count is weighed (local vs. global freq)
3. SingularValue Decomposition
Logic ofWord Association:
IF X does not co-occur with Z,
BUT co-occurs withY, which co-occurs w/ Z,
THEN X is associated with Z
(assoc. depends on strength of XY,YZ pairs, & associations w/ all other potential
word pairs)
Word
/doc
1 … … 6,000
1 x x x x
… x x x x
… x x x x
45,000 x x x x
See: Landauer,T.K. & Dumais, S.T. (1997). “A Solution to Plato’s Problem:The Latent Semantic AnaysisTheory of Acquisition, Induction, and Representation
of Knowledge.” Psychological Review, 104(2), 211-240.
And:Wild, Fridolin. (2015). lsa: Latent Semantic Analysis (version 0.73.1) [Computer Software].
20. THANKYOU
Books & Chapters
Cochrane, C. (2015). Left and Right: the small world of political ideas. McGill: University Press
Marcus, G.E. (2003).The Psychology of Emotion and Politics. In Sears, D.O., Huddy, L., & Jervis, R. (Eds.) The Oxford Handbook of Political Psychology.
(pp. 183-207).
Journal Articles
Grimmer, J. & Stewart, B.M. (2013).“Text as Data:The Promise and Pitfalls of AutomaticContent Analysis Methods for PoliticalTexts.” Political Analysis, 21,
267-297.
Landauer,T.K. & Dumais, S.T. (1997). “A Solution to Plato’s Problem:The Latent Semantic AnaysisTheory of Acquisition, Induction, and Representation of
Knowledge.” Psychological Review, 104(2), 211-240.
Oskarsson, S., et al. (2015). “LinkingGenes and PoliticalOrientations:Testing the Cognitive Ability as Mediator Hypothesis”. Political Psychology, 36(6), 349-
655.
Pennebaker, J.W., Mehl, M.R., & Niederhoffer, K.G. (2003). “Psychological Aspects of Natural Language Use: OurWords, Our Selves.” Annual Review of
Psychology, 54, 547-577.
Piurko,Y., Schwartz, S.H., & Davidov, E. (2011). “Basic PersonalValues and the Meaning of Left-Right PoliticalOrientations in 20 Countries.” Political
Psychology, 32(4), 537-561.
Software
Barbera, Pablo. (2016). Rfacebook: Access to FacebookAPI via R (version 0.63) [Computer Software].
Pennebaker, J.W., Booth, R.J., & Francis, M.E. (2016). Linguistic Inquiry andWord Count: LIWC2015 (version 2.1.0). Austin,TX: Pennebaker Conglomerates.
Wild, Fridolin. (2015). lsa: Latent Semantic Analysis (version 0.73.1) [Computer Software].
My research to borrow a phrase from Steven Pinker is based on the idea that “language is a window into the mind”. That by studying how language is used we can infer not only what the mind thinks, but also how it thinks.
Today I’ll be focusing on psychological differences across the political spectrum. Examining how traits such as personality and temperament may shape the way individuals identify with political parties and ideologies. These differences, I argue, manifest subconsciously in language – in the words and types of words that people use. Using methods from computerized text analysis I detect quantitatively how language use varies across groups with different ideologies. I also examine how political candidates tailor language, subverting the subconscious, in order to serve the purposes of election campaigns. Here I hope to show that the constraints of campaigning encourage politicians to use language strategically. The result is, as so many of you might guess, that politicians sound a lot alike. My findings are based on an analysis of data that I scraped from Facebook. The data are primarily text based and were collected during Canada’s last federal election. So, given that, it seems appropriate to start with a brief overview of politics in Canada…As many of you may know Canada recently had an election last October…
…during which Canadians elected this charismatic fellow. The man of course is Justin Trudeau, who ran as the Liberal Party leader. Other notable leaders include Stephen Harper, who was the incumbent Prime Minister at the time, and the leader of the Conservative Party. And Tom Mulcair, who was the leader of the New Democratic Party and the official opposition.
Canada uses FPTP, an electoral system that produces majority governments and pushes political parties to the center of the spectrum. Those familiar with Downsian theory and the median voter hypothesis should not be too awestruck by our political parties. The Liberals have historically been very centrists, but has moved slightly center-left in recent decades. Positioned somewhat further to the left of them is the New Democratic Party (aka the NDP). And finally right of center we have the Conservatives.
you will notice that I’m focusing a lot on the left/right political spectrum...
We hear a lot about the left and right, but what do these words mean? Most of us have an intuitive sense as to their meaning but if forced to give specifics many of us would be hard-pressed to specify what the words entail. And rightfully so. Because despite the seeming simplicity there is no universal reference point or particular set of mutually exclusive characteristics that would definitively identify a particular individual or group as one or the other. As left or as right.
Despite that, the dominant approach to studying this left-right divide comes from…
a classical (or essentialist) view
Classicists are those that believe that left and right differences can be identified according to mutually exclusive categories. Laponce (1981), for example, believes that what distinguishes liberals from conservatives is the distinction between religiosity and secularism. Maintaining that “the right is the domain of the sacred” and “the left, that of the profane”. Noberto Bobbio (1996) disagrees; noting prominent examples of secularism and atheism on the right of the spectrum. He argues instead that the key variable is the extent to which each group supports equality in society. A position not unlike Inglehart (1990) who identifies egalitarianism as one of the defining characteristics of liberals. But alas, others disagree, as academics do! Scholars Noel and Therien (2008), for instance, argue that, hey, both groups support equality, but what distinguishes them is how they define equality. One side may favor equality of opportunity, the other equality of outcome.
Everyone agrees on the existence of defining characteristics, but no one seems to agree on which are most important.
There is An Alternative View, however
Which comes from psychological research and asserts that mutually exclusive categories are “conceptual utopias”. That meaningful political differences do exist, but such differences should be thought of as gradual and incremental variations between individuals and groups. Consequently, there is no single defining characteristic which qualifies or disqualifies an individual from being either to the left or the right.
Research on the left-right divide also seems to be shifting away from rational choice. As the theory fails to explain why people hold certain beliefs and identities over others. Rational choice theorist would argue that decision-making is goal-oriented and predictable. That when left to our own devices individuals will evaluate the options by carefully weighing evidence. Choosing in the end the option that provides them with the greatest benefit at the smallest cost. Studies from evolutionary psychology, however, show quite conclusively that human cognition is a biological adaptation; a product of evolution by natural selection rather than an “engine of pure rationality”.
If reason alone does not explain why we identify with certain parties and ideologies, where then do these identities and beliefs come from… Why do some identify as left and others as right?
There are a lot of factors, but today I want to limit my discussion to the role of emotions and personality
This is … Anger. Psychologists find that most feelings of anger have “moral overtones”, which prone individuals towards indignation. Moral anger is a feeling that comes from a sense of injustice- that something has been done wrong – and can motivate individuals towards addressing those injustices. Anger is known to limit an individual’s ability to accurately perceive risk, and encourages risk-taking and action-oriented behavior. Because of this the emotion is often referred to as “approach-based”. Meaning it motivates individuals to take action, and creates a desire or willingness for things to change.
This is…Anxiety, an avoidance-related emotion. Anxiety is in many ways the opposite of anger. Individuals who are prone to higher levels of anxiety are known to overestimate risk; favoring actions that are risk-averse. It is no coincidence then that politicians use messages to inspire fear during election campaigns. Particularly when said politicians are incumbents attempting to stay in power. Fueling feelings of uncertainty, it turns out, has long been a powerful motivator for preserving the status quo.
We know from personality research that liberals and conservatives feel and process emotions differently. Liberals are generally more open to new experiences. A personality trait that reflects the motivational attributes of anger. They also express feelings of anger more frequently than conservatives. Conservatives meanwhile tend to be more conscientious (i.e. mindful of others) and neurotic. And thus typically express feelings of nervousness or anxiety more often than liberals.
Studies have replicated these findings across various journals using methods that range from self-reports, to MRI studies of brain differences, to physiological tests measuring EDC. My method of choice however is computerized text analysis…
Researchers have long studied language for psychological insights. Language is after all the most common -and perhaps the most reliable way- for people to communicate their thoughts and emotions with one another.
Not too long ago the study of language was limited to qualitative methods and controlled lab experiments. Each with merits, but both typically suffering from smaller sample sizes and the latter from a lack of external validity. After all, there is only so much you can learn from experimenting on undergrads. Today with the development of high performance computers, and the proliferation of Internet use, researchers can collect and analyze copious amounts of data. Researchers have thus turned to methods in computerized text analysis as a way to test hypotheses and find meaningful links between everyday language use and psychological correlates.
Using this approach, the question I ask is whether groups with different political ideologies use language in unique ways? Do ideological differences manifest in language? From this question I generated some hypotheses:First, I expect that individuals who support left leaning parties, such as the Liberals or NDP, will collectively, use more words to express anger than groups who support parties that are right of center. While those supporting parties on the right should, on average, use a higher percentage of words expressing anxiety. Relatedly, expressions of both emotions should be dependent on the extent to which a group is left or right of the spectrum. So, for example, we would expect that NDP supporters, who are presumably further left, should be more likely to express feelings of anger. We can expect to see similar results for other emotions. For example, studies show that conservatives tend to report higher levels of life-satisfaction and happiness when compared to liberals. We may therefore expect higher frequencies of positive emotion words in text written by conservatives.
Significant at (F(3,32,)= 13.588, p < .000) with an R2 of .560. Predicted Ideology (con =1, lib=0) increased
Of all the output variables, anger proved to be one of the best predictors of political ideology and party support. Results were statistically significant with higher frequencies of anger related words being correlated with supporters of the two left-of-center parties. NDP supporters contributed to the highest volume of anger-related words throughout the campaign. Averaging .63 percent of total words relating to feelings of anger. That’s .13 percent more than Liberals and .21 more than Conservatives. Expressions of anger were consistently higher for NDP supporters than for any other group throughout the 11 weeks of the campaign - with only one exception in week 2.
With anxiety, however, there appears to be a different story. If you aggregate the data and only look at differences between the left and right than yes, anxiety-related words are typically more prevalent in the language use of those right of center. However, if the unit of analysis is at the level of the political party than one sees a very different observation. While Liberal Party supporters were less likely to use anxiety related words than Conservatives the same could not be said with respect to NDP supporters. The NDP were, in fact, even more likely to use these words. Frankly, I’m hard-pressed to explain these results.
Now it may be that LIWC’s dictionary did not target the appropriate words. But I doubt that given that Pearson Correlation analyses find its dictionary to be an externally valid method for identifying emotions. Alternatively, one might argue that the approach-avoidance framework is incorrect. Again, I doubt this given the extent of evidence that observes higher levels of anxiety among conservatives.
I am also not the first person to find these results. A previous study examining Twitter users in the US found that anxiety was closely correlated with Democrat rather than with Republican twitter users. The study also happened to use LIWC. So if anyone in the audience has any thoughts or hypotheses, I would be happy to hear them!
One final example. This one tracking expressions of positive affect. Again, studies and self-reports have shown that conservatives report higher levels of life-satisfaction and happiness. Such findings, however, have been dismissed by critics who argue that self-reports are unreliable because of the social desirability bias. That conservatives will exaggerate their happiness because they know that it is a socially desired trait. The argument has never been a convincing criticism for me, as I see no reason why liberals could not do the same. One of the strengths of text analysis is that you can study individuals that do not know they are being studied. Thus reducing the chances of an observer effect or a social desirability bias. To be clear I suspect that these results have little to do with any inherent differences. Instead they may be indicative of the benefits that come from the environments and social support structures that are more common to conservative households. Specifically, with respect to community and religiosity, which studies have found to be positively correlated with happiness.
Alright, those are the results for everyday Facebook users. But what happens when we look at party leaders. Do these patterns hold up? In short, No... When people say that all politicians sound the same, there is some truth to that. And it might be especially true in countries like Canada where the FPTP electoral system creates incentives for parties to appeal to voters in the middle of the spectrum. Since the election is winner take all, a party only needs a plurality of the votes to win an election, so it makes strategic sense to tailor your message to the median voter. Since the use of emotion words emphasizes consensus values, party leaders may want to restrict their use of such wording in order to appeal to as broad a voting base as possible. The language of rhetoric also has little to do with political ideology. Regardless of the political party all candidates are incentivized to use negative emotion words. This is because Psychologically, people are more likely to remember negative information. The same is true with respect to campaign messaging. Negative messages are far easier to remember than positive ones. As Jon Krosnick so eloquently puts it, imagine I was a political candidate…
“If I can only hold your attention for a few minutes, and I can either tell you how I helped an old lady cross the streetor how my opponent kicked a cat, it makes more sense for me to tell you about the cat.”
Indeed we see this strategy throughout the campaign.
[INSERT HARPER QUOTE].
Here is Harper discussing the possible risks associated with electing a Liberal or NDP government. The language is simple, concise, a repetitive. It is designed to reinforce the prospects of uncertainty by uses words to induce a fear response.
The onus is then [INSERT TRUDEAU QUOTE] on whoever is challenging the incumbent to convince voters that they are worth the risk. That changing government is necessary.
All in all there where very few linguistic and psychological differences in how language was used between the three candidates. One of the main exceptions was the rate at which candidates communicated risk with their voters. As an incumbent Stephen Harper used more than twice as many risk-related words. The strategy being to reinforce that Justin Trudeau was a liability. That he was young. Unexperienced. And as per the mantra of advertisements…that he was Just Not Ready. Turns out Justin Trudeau won the election anyway, but hey, you can’t blame a guy for trying.
The research I presented today comes from an approach within text analysis known as word count strategy. This approach however is only interested in the presence and frequency with which words appear. It assumes that the same dictionary can be used to measure and detect psychological correlates across different groups. When we know from research that different groups understand the same words in different ways. An alternative approach is word pattern analysis, which originates from work in information retrieval and artificial intelligence. Rather than exploring text ‘top down’ within the context of previously defined word categories,” it instead approaches word patterns mathematically by detecting “‘bottom-up’ how words co-vary across large samples of text.
One such method is known as Latent Semantic Analysis, which is used for extracting and representing the meaning (or association) between words. Its creators also argue that it serves as an accurate model for how the human mind learns and represents knowledge. It can therefore be used to infer how different groups think about the same words or topics. Based on some preliminary findings, I’ve noticed this to be the case. Conservatives for instance when talking about economics seem to think about things like institutions, and risk. While liberals seem to associate economics with social responsibility.
What research like mine shows is that differences in beliefs and values may come from sources that that are inherent to individuals. Traits such as personality and temperament may predispose people to be more inclined to favor certain parties, ideologies and styles of thinking. Individuals who are prone to experiencing anger may underestimate risk and be more inclined to demand change; perhaps out of a desire to address injustices. Such individuals may be more likely to identify as liberals. Others who are more prone to feeling anxiety will typically overestimate risk and be driven to maintain the status quo. They may be more likely to identify with a conservative party.
None of these psychological traits are, as the classicists might argue, mutually exclusive. Personalities and temperaments will crisscross and overlap; existing across the political spectrum to various degrees in relation to one another. So too will party loyalties - people will change their party identity and vote for reasons that have little to do with their temperament or personality. But if these differences are inherent in groups then it does well to remind ourselves that some political disputes may be irreconcilable. That battles of reason may not change how people feel about certain issues. If we want to make progress on contentious challenges such as climate change or gun control we may need to appeal to more than just reason. At the end of the day it would do good to remind ourselves that politics is about compromise. Compromise between groups of people who may never view the world in the same way.
DELETED FOR TIME
(graph from Paul Fairie’s work on the Globe and Mail)
Finally, I’d like to end with an anecdote from the 2015 election that, I think, summarizes these findings… You may recall that that there was an outlier when tracking expressions of anger. Specifically, in the second week of the election when the highest volume of anger-related words came from conservatives. During this moment in the campaign something else happened. The Syrian Refugee Crisis had become a hot button issue in Canadian politics; as parties began debating on what Canada’s role would be internationally. The cause of this uptick in anger, based on the timing and content of the comments, was a response to Justin Trudeau announcing he would pledge asylum to 25,000 Syrian Refugees; a number 15,000 more than either of the other two parties had promised earlier. Many conservatives felt that this was fiscally irresponsible given that the Canadian economy was struggling at the time. Notice also, that the response from Liberal and NDP supporters is the complete opposite – with the frequency of anger words declining for both groups.It is important to mention that federally the majority of Canadians vote left of center. In the past the Conservative Party’s strategy has therefore been to divide and conquer by splitting the left-of-center vote between the Liberal and the NDP. We can see this strategy unfolding throughout the 2015 campaign. During the early months Harper did well to keep Trudeau and Mulcair at a distance from one another. Presenting neither candidate with the opportunity to distinguish themselves – keeping left-of-center voters split in who they support. As August came to a close Mulcair started to improve in the polls. And by early September he even started to outperform the other candidates. Now with the prospect of Mulcair winning the election the Harper team shifted strategies and began a bizarre campaign that focused on identity politics. Issues regarding the economy which had taken front stage were now sidelined in favor of debates regarding Syrian refugees, what the Conservative Party called “culturally barbaric practices”, and finally a debate as to whether muslim women could be banned from wearing niqabs during swearing in ceremonies.
Likely the reason Harper shifted the election towards identity politics was in order to pander to voters in Quebec. A province that was highly in favor of banning religious symbols from public service, and a province that has historically shaped many election results. Harper may also have been appealing to the same group that had collectively expressed anger at the prospect of admitting additional Syrian refugees. Now at first Harper’s strategy succeeded in that it lowered Mulcair’s approval rating. Unlike Harper, Mulcair took the position that neither federal nor provincial government should impose on peoples’ religious liberties. But this position did little to win over voters. As a result, Mulcair’s approval rating began to decline. Interestingly, however, Harper’s strategy quickly backfired. As Mulcair’s poll numbers declined this provided a signal to left-of-center voters that the candidate to back was Justin Trudeau.
Like Russian babushka dolls political identities can be compartmentalized. Individually Canadians left of the spectrum may have identified with different parties – either the NDP, the Liberals, or the Greens. And while party loyalist from each of these parties may have at one time vehemently disagree with one another, collectively individuals found commonality under the Liberal Party banner.