Professional Documents
Culture Documents
Descriptive Statistics
Amongst all participants, the data showed that liberal students at Northwestern University
tended to skew more towards the liberal extreme, with an average liberal extremism score of
4.95 out of 7. Within this sample, participants seemed to have a difficult time identifying
whether news articles were real or fake; both real and fake news accuracy scores were
approximately 50% across all participants. However, they were slightly better at identifying real
news articles as real than fake news articles as fake. When asked about confidence in identifying
comparison to the post-test self-reported confidence (M = 4.58). Notably, the mean of post-test
confidence was 4.58, which was interestingly high given most participants had only answered
For additional planned data analysis, participants were split by social media news usage.
The data indicated that liberal extremism increased with social media usage, meaning that
regardless of the social media platform, the more participants relied on social media for news,
the more likely they were to be liberally extreme. Descriptive results showed that participants
with High SMU of their news were 11.1% less accurate in identifying fake and real news based
on their mean overall test scores compared to participants with Low SMU of their news
consumption. Likewise, participants with High SMU were on average 17% less accurate in
identifying real news and 5.1% less accurate in detecting fake news than those with Low SMU.
Interestingly, the mean pre-test confidence levels were highest among the 24 participants who
Low 13 12 13 13 13 13 13
Medium 23 23 21 23 23 23 18
N
High 24 22 23 24 24 24 21
Total 60 57 57 60 60 60 52
Low 4.69 (0.56) 6.67 (5.6) 5.15 (2.1) 0.57 (0.17) 0.64 (0.13) 0.61 (0.08) 4.00 (2.1)
Medium 4.64 (0.95) 6.13 (5.3) 6.38 (1.9) 0.46 (0.19) 0.52 (0.25) 0.49 (0.19) 4.89 (1.9)
M (SD)
High 5.38 (0.97) 5.36 (4.6) 6.52 (2.2) 0.52 (0.22) 0.47 (0.19) 0.49 (0.16) 4.67 (1.8)
Total 4.95 (0.95) 5.95 (5.0) 6.16 (2.1) 0.50 (0.20) 0.53 (0.21) 0.52 (0.17) 4.58 (1.9)
Low [4.00, 5.67] [1, 15] [2, 8] [0.38, 1.00] [0.50, 0.88] [0.50, 0.75] [1, 7]
Medium [2.50, 6.00] [1, 20] [1, 10] [0.00, 0.88] [0.00, 0.88] [0.00, 0.88] [1, 9]
Range
High [3.00, 7.00] [1, 22] [2, 10] [0.00, 0.88] [0.00, 0.75] [0.00, 0.69] [2, 9]
Total [2.50, 7.00] [1, 22] [1, 10] [0.00, 1.00] [0.00 0.88] [0.00, 0.88] [1, 9]
Note. SMU for news stands for “social media usage for news” in percentage groups, where Low
ranges 0-25%, Medium is 50%, and High ranges 75-100%. Time spent on news is in hours per
week. Identification scores range from 0 (fully incorrect) or 1 (fully correct).
Planned Analyses
overall news identification score (r(57) = -0.059, p = 0.655, 95% CI: [-0.308, 0.198]): having
more extreme liberal beliefs did not predict lower overall news identification performance.
Although the correlation value demonstrated there is a weak, inverse relationship between the
two factors, the p value indicated this correlation to be statistically insignificant. Given these
results, we failed to reject the null hypothesis that there is no effect of liberal extremism on news
identification performance.
between liberal extremism and fake news identification score (r(57) = 0.058, p = 0.658, 95% CI:
[-0.199, 0.308]): having more extreme liberal beliefs did not predict lower fake news
identification performance. Although the correlation value demonstrated there is a weak direct
relationship between the two factors, the p value indicated this correlation to be statistically
insignificant. Given these results, we failed to reject the null hypothesis that there is no effect of
The same Pearson’s correlation model lastly revealed a non-significant weak inverse
relationship between liberal extremism and real news identification score (r(57) = -0.147, p =
0.263, 95% CI: [-0.386, 0.111]): having more extreme liberal beliefs did not predict lower real
news identification performance. Although the correlation value demonstrated there is a weak
inverse relationship between the two factors, the p value indicated this correlation to be
statistically insignificant. Given these results, we failed to reject the null hypothesis that there is
We conducted one-way ANOVA tests to compare the effect of the amount of news
ingested by social media on news identification scores. There was not a statistically significant
difference in mean fake identification score across social media usage groups (F(2, 33.4) = 1.57,
p = 0.222, ηp2 = 0.045): although 4.5% of variance in fake news identification performance can
be explained by social media usage, the effect was not statistically significant, so we failed to
reject the null hypothesis. However, there was a statistically significant difference in mean real
news identification score across social media usage groups (F(2, 36) = 5.24, p = 0.010, ηp2 =
0.092): 9.2% of variance in real news identification performance could be attributed to social
media usage, with the difference between Low and High usage providing the greatest statistical
and practical significance (t = 0.170, p = 0.051). In addition, there was a statistically significant
difference in mean overall score across social media usage groups (F(2, 37.8) = 5.42, p = 0.009,
ηp2 = 0.082): 8.2% of variance in overall performance could be attributed to social media usage,
with the difference between Low and Medium usage (t = 0.117, p = 0.102) along with Low and
High usage (t = 0.111, p = 0.122) providing the greatest statistical and practical significance.
Thus, social media usage revealed to be a strong predictor of real news identification
performance and overall news identification performance, especially for lower usage groups
Figure 3.
Figure 5.
Confidence
We found a moderately strong, positive correlation between pre-test confidence and post-
test confidence (r(52) = 0.701, p < 0.001); while average post-test confidence was lower,
participants who were more confident pre-test were also more confident post-test. Pre-test
confidence was not correlated with fake news identification score at all (r(57) = 0.001, p =
0.996); this result was extremely statistically insignificant, indicating that pre-test confidence is
likely a completely random predictor of fake identification score. Post-test confidence was a
somewhat better predictor of fake news identification score (r(57) = 0.057, p = 0.689) but was
still statistically insignificant and weak (Table 2). These results were insufficient to determine
whether greater average confidence would be associated with lower fake news identification
scores; thus, confidence may not actually be a predictor of ability to detect fake news.
Interestingly, weekly time spent on news had a moderate correlation with post-test
confidence and was statistically significant (r(50) = 0.327, p = 0.020) but pre-test confidence was
not (r(55) = 0.047, p = 0.735). This indicated that students who spent more time ingesting news
were more likely to report higher amounts of confidence after taking the test than students who
spent less time ingesting news; furthermore, students who spent more time ingesting news were
no more likely to report higher levels of confidence prior to taking the test in comparison to
News Literacy
Additionally, we found that news literacy was not a statistically significant predictor of
fake news susceptibility. Independent samples t-tests revealed non-significant differences with
regards to fake news identification score (t(57) = -1.220, p = 0.228, d = -0.39, 95% CI: [-1.031,
0.247]), real news identification score (t(57) = -0.265, p = 0.792, d = -0.086, 95% CI: [-0.719,
0.549]), and overall news identification score (t(57) = -0.944, p = 0.349, d = -0.31, 95% CI: [-
0.941, 0.333]) between those that have taken a news class before and those that have not. Even
though the Cohen’s d values for fake news scores and overall news scores indicate a medium
sized inverse relationship with news literacy, they are not statistically significant; hence, students
who took classes focused on news literacy were no more/less likely to identify real news as real
or fake news as fake in comparison to students who have not taken classes focused on news
literacy.
Lastly, we conducted a one-way ANOVA test to compare the effect of the amount of
news ingested via social media on liberal extremism (Figure 4). We found that social media
usage for news is a statistically significant effector of liberal extremism (F(2, 36.7) = 4.51, p =
0.018, ηp2 = 0.141): 14.1% of variance in average liberal extremism could be attributed to social
media usage for news. Thus, social media usage was revealed to be an extremely important
predictor of liberal extremism, where the difference between Low and Medium usage (t = -0.690,
p = 0.073) along with Low and High usage (t = -0.737, p = 0.017) provided the greatest
Avg. liberal r —
extremism
p —
Overall score r -0.059 —
p 0.66 —
Fake ident. r 0.058 0.79*** —
score
Discussion
Unsurprisingly, our study findings suggest that fake news is a very nuanced subject. A
part of our hypothesis was not supported: liberal extremism was not a significant predictor of
fake news identification performance. We hypothesized that being more extremely liberal would
decrease one’s ability to detect fake news, as they might be “blinded” by the actual facts an
article does or does not provide; however, this was not supported. No results for overall score,
fake news identification score, or real news identification score were statistically significant. Nor
did we find any correlation between extremism and ability to detect fake news. This is in line
with trends across the general United States population; regardless of political affiliation, people
generally overestimate their ability to detect fake news and are ultimately susceptible to fake
news online. Additionally, Casino and Jenkins found that members of both political parties and
ideologies can be susceptible to conspiracy theories and fake news. Given their findings on
susceptibility to conspiracy theories, it is possible that the headlines utilized in our study were
not quite to the level of extremity as studied in the past, and that may be why political extremism
did not seem to be a predictor for fake news identification. Alternatively, it is possible that being
more liberally extreme was not a large factor in fake news detection simply because there could
Despite liberal extremism not being a significant predictor of fake news identification, we
found that social media usage is a predictor of fake news identification. Social media usage and
overall identification scores were found to be very statistically and practically significant,
indicating that social media usage appears to have an effect on fake news detection, which
supports our hypothesis. In fact, social media usage accounted for 8% of all variance in possible
reasons for fake news identification, showing there is likely a strong relationship between the
two. This was seen further when examining the real news identification scores: again, there was
a relationship with social media usage, this time accounting for 9% of variance. Lastly,
participants who reported Low social media usage for news scored better across all metrics,
compared to those who had Medium and High social media usage, once again supporting our
hypothesis. This is supported by Allcott and Gentzkow, who found that those who use more
social media are more likely to believe ideology-aligned articles—since we were examining
liberal participants, those who had High social media usage likely could have immediately
believed any headlines from liberal news sources, while those with Low social media usage
could have been more skeptical. Alternatively, this could be explained by confirmation bias,
where liberal users with higher social media usage are more likely to be surrounded by liberal
Given the statistically insignificant relationship between liberal extremism and fake news
detection capabilities, as well as the significant relationship between social media usage and fake
news detection capabilities, we looked to identify any relationship between liberal extremism
and social media usage. Social media usage is a significant predictor of liberal extremism,
accounting for 14% of variance. Confirmation bias could explain this large effect. The more time
a person spends on social media, the more likely their social network is to be aligned with their
political ideologies.
samples, and the lack of consideration for misinformation vs disinformation of fake news.
Northwestern University is an institution known for its liberal student body, and our sample
consisted of only Northwestern students. Thus, our sample was heavily concentrated with liberal
extremes, as reflected in the average liberal extremism score of 4.95 out of 7. If our sample
included more participants across the liberal extremism spectrum, we could have looked at
differences between those with a more moderate liberal ideology and more extreme liberals.
Further, had our study included an equal number of liberal, moderate, and conservative
participants, we would be able to examine fake news susceptibility predictors across all political
affiliations. This could provide insight as to whether the same fake news susceptibility predictors
participants was required to maintain a medium effect size. While there were initially 87
respondents to the survey, 25 participants were removed from our analysis after leaving the
survey incomplete or having a conservative or moderate political affiliation. This means that
despite significant findings regarding social media usage as a predictor of fake news
identification, there is a chance of a false positive given that our sample size was not large
enough.
information, whereas disinformation is information that was falsified with the intention of
misleading individuals. The wording in our survey was intentionally ambiguous around the word
“fake” as we were interested in how various participants would respond. However, because it
was not specified whether participants in our study were identifying fake news based on if they
thought the information was false or false with the intentions of misleading, it is unclear whether
participants identified fake news sources as fake due to misinformation, disinformation, or both.
It is possible participants would have scored differently on the overall score and fake news
identification score had there been more specification between the two.
Due to the limited number of studies available regarding liberal extremism and mode of
news consumption as potential predictors of fake news susceptibility, as well as the limitations of
our own study, more research is necessary to provide evidence for a significant relationship
between fake news susceptibility and liberal extremism or mode of news consumption. Future
researchers need to take into consideration the political affiliation of those sampled in order to
provide an inclusive sample of individuals across the liberal extremism spectrum as well as an
ample amount of individuals across the political spectrum, the power of the sample in order to
avoid false positives, and detailed definitions of misinformation and disinformation along with
specific instructions on which one is to be considered when answering questions about fake
news.
Conclusion
In conclusion, we found that liberal extremism is not a significant predictor of fake news
identification. However, social media usage is a large, significant predictor of liberal extremism.
This suggests that low social media usage predicts better performance, especially at identifying
real news as real. Additionally, people are often far more confident at their abilities to detect fake
news than they actually are. It could be easy for malicious actors to prey on this disjunction,
creating disinformation that is intentionally difficult to detect. Possible warning signs or pre-
viewing programs could be implemented on social media based news sites themselves to combat
these findings, as investigated by Pennycook et al., who found that priming individuals to think
about accuracy resulted in a tripled level of discernment between sharing real and fake news in
relation to the control group who were not primed. With the continuous rise of both social media
and fake news on social media, it is recommended that this study be used for educational
purposes for not only politicians and voters, regardless of age, but the general public as well.