ThesisPDF Available

Using Clusters to Predict Vaccine Belief Revision

Authors:

Abstract

This study identified and analyzed the relationships between individual differences in cognitive processes undertaken during reading and belief revision.While other studies have shown that words reflective of analytical thinking are use more in anti-vaccine comments than in pro-vaccine ones and that flexible thinking is a predictor of an increase in inaccurate vaccine beliefs,this study adds value by analyzing cohesion, linguistic categories, and the clustering of verbal protocol measures as predictors of harmful belief change.Forty-four University of Memphis students completed a prescreening, think aloud, screening immediately after the think aloud and then again two days after the initial day of the experiment. Word analysis software LIWC and Coh-Metrix were used to transform linguistic responses to numerical data to examine potential pattern categories. Clustering was used to identify patterns and similarities in the numerical data. A series of single and multiple linear regression analyses were used to predict belief revision within the algorithm defined clusters. Preliminary findings have shown a significant increase in harmful beliefs from prescreening to immediately after reading, and from prescreening to two days later, providing the focus for the current study.The results of this study indicate that, when grouped by clustering, the use of referential cohesion, causal connectives,analytical language, emotional language, and personal pronouns do not significantly predict harmful belief change.Unexpectedly, causal connective use had a tendency to appear as a significant predictor of belief change, appearing as a significant predictor for two of the three clusters and showing to be a predictor of an increase in harmful beliefs in one cluster and a decrease in harmful beliefs in another.
Using Clusters to Predict Belief Change 1
Using Clusters to Predict Vaccine Belief Revision
Paul Anderson
University of Memphis
Using Clusters to Predict Belief Change 2
Abstract
This study identified and analyzed the relationships between individual differences in cognitive
processes undertaken during reading and belief revision. While other studies have shown that
words reflective of analytical thinking are use more in anti-vaccine comments than in pro-
vaccine ones and that flexible thinking is a predictor of an increase in inaccurate vaccine beliefs,
this study adds value by analyzing cohesion, linguistic categories, and the clustering of verbal
protocol measures as predictors of harmful belief change. Forty-four University of Memphis
students completed a prescreening, think aloud, screening immediately after the think aloud and
then again two days after the initial day of the experiment. Word analysis software LIWC and
Coh-Metrix were used to transform linguistic responses to numerical data to examine potential
pattern categories. Clustering was used to identify patterns and similarities in the numerical data.
A series of single and multiple linear regression analyses were used to predict belief revision
within the algorithm defined clusters. Preliminary findings have shown a significant increase in
harmful beliefs from prescreening to immediately after reading, and from prescreening to two
days later, providing the focus for the current study. The results of this study indicate that, when
grouped by clustering, the use of referential cohesion, causal connectives, analytical language,
emotional language, and personal pronouns do not significantly predict harmful belief change.
Unexpectedly, causal connective use had a tendency to appear as a significant predictor of belief
change, appearing as a significant predictor for two of the three clusters and showing to be a
predictor of an increase in harmful beliefs in one cluster and a decrease in harmful beliefs in
another.
Using Clusters to Predict Belief Change 3
Using Clusters to Predict Vaccine Belief Revision
As humanity progresses in an age of information and the 24-hour news cycle, it becomes
increasingly common for people to be exposed to more controversial ideas, topics, and articles.
This is especially true regarding information acquisition using the Internet. Because of the lack
of filters for information posted online and the increasing popularity of social media websites as
a means to transfer information, it is important for people to develop skills necessary to view
information through an analytical lens so that they may discern what information is accurate and
what sources are reliable. The current study reflects that, because they are bombarded with
information that is both accurate and inaccurate, people must make decisions as to whether the
information received is assimilated into an existing belief set or not. As such, it is necessary to
differentiate between the accurate and inaccurate information, so that individuals develop
properly informed beliefs and adapt pre-existing beliefs to accommodate new information.
Because of this abundance of information, especially controversial information, it is
expected that individuals may revise their beliefs on any given topic in light of new information.
This change in beliefs could be towards more accurate information or towards more inaccurate
information. Therefore, it would be beneficial to identify those that may exhibit belief revision,
particularly those that tend to sustain or even increase their inaccurate beliefs. Early
identification of such tendencies may allow for an intervention to teach the individual skills that
allow for in-depth analysis and the identification of accurate, reliable information.
This study identifies some of those patterns of cognitive processes that reliably predict
individual belief revision from their verbalized thoughts regarding articles relating to a common
controversial topic seen online: vaccine usage and safety. Three key concepts or methods
motivate the current research: belief revision, word analysis, and clustering.
Using Clusters to Predict Belief Change 4
Belief Revision
Belief revision is the process by which an individual accommodates new information into
an existing belief set. For example, people with pre-existing beliefs that include accurate and/or
inaccurate information, processes may occur in two directions: an increase in accurate beliefs
and an increase in inaccurate beliefs. This means that individuals may acquire new information
to correct inaccurate pre-existing beliefs as well as that individuals may acquire new
misinformation to overtake the accurate belief set. The Knowledge Revision Components
(KReC) framework defines the processes and mechanisms required to revise knowledge and, by
extension, beliefs (Kendeou & O’Brien, 2014). Specifically, the KReC framework identifies five
principles that underlie knowledge revision in terms of reading comprehension: encoding,
passive activation, co-activation, integration, and competing activation. This study focuses
particularly on co-activation and integration, and their role in belief revision. It has been shown
that for belief revision to occur, both the new ideas and pre-existing ideas must be activated
simultaneously in working memory through a process known as co-activation (Kendeou &
O’Brien, 2014). In terms of revising accurate/inaccurate beliefs from those that are pre-existing
to accommodate newly introduced opposing beliefs, two conflicting ideas must be integrated into
one working belief set. In addition to conflicting ideas, belief revision includes the
accommodation of new information that is in support of pre-existing beliefs when that new
information significantly alters the belief set. For this case, two different ideas supporting the
same belief work to increase perceived validity of that belief set.
Naturally, the most appropriate use for belief revision is to correct misinformation.
However, revising inaccurate beliefs to be more accurate proves easier said than done, especially
regarding controversial topics. In fact, it has been shown that people holding beliefs that vaccines
Using Clusters to Predict Belief Change 5
are harmful are resistant to new information debunking the myth. Furthermore, retractions of
articles containing the inaccurate information may even lead to increased belief in inaccurate
information (Lewandowsky, Ecker, Seifert, Schwarz, & Cook, 2012). Additionally,
misinformation has been shown to be present in perception and cognition when processing
information (Rapp, 2016). People do not critically evaluate information received. The persistence
of misinformation coupled with the intense rate at which information is received serve as
evidence to support the importance of identification and refutation of misinformation.
One way to examine what may underlie belief revision is to measure the participants’
cognitive processes while reading. In this study, this is accomplished using a think-aloud study:
having participants read text aloud and verbally respond with their thoughts at the time of
reading. Research conducted using a think-aloud study to model belief revision in terms of the
KReC framework has shown that observing participants’ comments while reading may be used
to indirectly measure knowledge revision and, by extension, belief revision (Kendeou,
Butterfuss, Kim, & Boekel, 2018). Furthermore, it has been shown that measures of participants’
vaccine beliefs taken in this way may be compared to yield a measure of belief revision (Kessler,
Braasch, & Kardash, 2019).
Word Analysis
In order to quantitatively analyze participants’ think aloud comments (as evidence of
their underlying cognitive processes), they must first be translated into numerical data. This
study used Coh-Metrix (Graesser & McNamara, 2017) and Linguistic Inquiry and Word Count
(LIWC) to do so, (Pennebaker, Booth, Boyd, & Francis, 2015). Coh-Metrix is a computational
tool that identifies over one hundred indices of the linguistic and discourse representations of
written or spoken text including indices of descriptive language, ease of text and discourse
Using Clusters to Predict Belief Change 6
comprehension (easability), referential cohesion, latent semantic analysis, lexical diversity,
connectives, situation modeling, syntactic complexity, syntactic pattern density, word
information, and readability (Graesser, McNamara, Louwerse, & Cai, 2004). With the use of
these indices, Coh-Metrix may be used to numerically code and analyze written texts as well as
categorize the ways in which words in those texts are used. When used in conjunction with other
word counting/analysis software, Coh-Metrix can help yield accurate numerical representations
of verbal data, particularly regarding cohesion, defined as the interaction between linguistic and
knowledge representations, may be use as a practical measure of the connections between what
an individual reports and what they know, as is shown in previous research investigating
differences in referential and causal cohesion using, among others, think-aloud responses (Allen,
McNamara, & McCrudden, 2015).
LIWC, while similar to Coh-Metrix in that it may be used to analyze entered words, the
difference is that its focus is primarily on word counting rather than cohesion. To that end, it
scans a given text and counts the percentages of words that reflect different emotions, thinking
styles, social concerns, and parts of speech using linguistic dictionaries designed to capture
individuals’ social and psychological states, that may be compared, contrasted, and analyzed
(Tausczik & Pennebaker, 2009). Specifically, LIWC has been used to show that anti-vaccination
comments showed evidence of higher usage of analytical thinking and lower usage of references
to family and social processes when compared to pro-vaccination comments (Faasse, Chatman,
& Martin, 2016).
Clustering
One way to organize data into groups of similar features is to use cluster algorithms,
resulting in aggregated data based on algorithm defined similarity. This can be used to clearly
Using Clusters to Predict Belief Change 7
determine the types of words and phrases used by participants exhibiting belief revision that may
differ, by comparison, with those that do not. Because the criteria different data must meet to be
assigned into a certain group is determined by the algorithm itself, the goal of the current
research is to identify if the patterns produced can be used to predict belief change. Cluster
algorithms have been previously shown to reliably organize data into similar groups for further
research (Borgen & Barnett, 1987). Thus, the current research is exploratory in nature, to first
identify patterns of cognitive processes that may underlie belief revision, which may prove
useful for additional more experimental and intervention research.
Current Study
The current study is motivated by the above three concepts and methods in identifying
patterns of language use found in participants’ think aloud responses that may be applied to both
single and multiple linear regression analyses to predict belief revision. While previous research
has been done using this data to identify belief revision, none before having attempted to predict
levels of belief revision using participant response tendencies as evidence of the cognitive
processes in which they may engage. In this study, there were three defined time periods on
which vaccine beliefs were measured: prior to reading, immediately after reading, and two days
following reading (i.e. at a delay). Think aloud comments produced during reading were coded
and analyzed using Coh-Metrix and LIWC to allow for the use of a cluster algorithm to sort the
data into the optimal number of related groups. Finally, single and multiple linear regression
analyses were used to predict belief revision. In order to check for prediction accuracy, initial
analyses have been conducted so as to identify significant changes in belief. Ultimately, this
study attempts to predict levels and direct of belief change based on patterns found in the
Using Clusters to Predict Belief Change 8
individual differences of participants’ think aloud responses to a vaccine belief task by utilizing
word analysis software in conjunction with clustering.
Method
Participants
There was a total of 44 participants (65.9% female, 31.8% male, 2.3% other). Participants
were all college students at the University of Memphis ranging in age 18 years old to 48 years
old (M = 22.55, SD = 6.56). Participants were recruited as volunteers from the University of
Memphis. Participants were compensated with course credit.
Materials
The study was conducted in a lab room in the Psychology Building on the University of
Memphis campus. It took place on a computer utilizing a standard keyboard and mouse. The
study used the Childhood Vaccination Belief Inventory (CVBI) as well as many additional
surveys outside the scope of the current study including, the Intention to Vaccinate Scale, the
Certainty/Simplicity of Knowledge Scale, the Need for Cognition Scale, the Flexible Thinking
Scale, the Actively Open-Minded Thinking Scale, the Belief Identification Scale, and the Need
for Closure Subscale. Additionally, the study used three prior knowledge questions, and an essay
prompt in relation to a reading for understanding focal task. The data for the current study
incorporates the CVBI and the focal reading task.
CVBI. The CVBI is intended to measure accurate and inaccurate beliefs about vaccines.
Statements claiming that vaccines are helpful reflect accurate beliefs and statements claiming
that vaccines are harmful or unnecessary reflect inaccurate beliefs. Helpful statements include
ones such as “The side effects of vaccines are minimal compared to the benefits,” whereas the
Using Clusters to Predict Belief Change 9
harmful and unnecessary statements include “Vaccines carry harmful ingredients,” and “Most
illnesses that vaccines target are not that serious, so it is unnecessary to vaccinate children,”
respectively. We quantified belief revision in terms of changes between three time periods
comparing each type of belief measured before the reading task, and assessments of the same
beliefs after the reading task is completed, both immediately and at a two-day delay.
Focal reading task. The reading task instructed readers to think aloud while interacting
with eight articles regarding vaccine usage which required verbal responses after reading each
sentence of an article. The articles, sources, and authors involved modification to actual
documents, but are all indicative of the types of information that are freely available on the
internet. Four of the articles (average of 251.8 words, range: 215 296 words) used reliable
information and were from reliable sources, two stating that vaccines are necessary and two
stating that vaccines are not linked to autism. The average Flesch Reading Ease scores for these
articles was 43.7, with scores ranging from 37.2 55.2. The average Flesch-Kincaid Grade Level
was 12.3 with levels ranging from 10.3 13.4. The other four articles (average of 263.3 words,
range: 201 307 words) used unreliable information and were from unreliable sources, two
claiming that vaccines are unnecessary and two claiming that vaccines cause autism. The average
Flesch Reading Ease scores for these articles was 38.3, with scores ranging from 24.1 46.3.
The average Flesch-Kincaid Grade Level was 12.5 with levels ranging from 10.9 15.7. There
was no indication as to which articles were accurate or inaccurate.
Procedure
Participants were given a brief description of the tasks involved then asked to read and
sign an informed consent after which any questions were answered. As part of pre-screening,
Using Clusters to Predict Belief Change 10
participants completed the CVBI. Participants then completed the following in order: three prior
knowledge questions, the Intention to Vaccinate Scale, a practice for the focal reading task with
a novel text unrelated to vaccine belief, the focal reading task, the Intention to Vaccinate Scale, a
short essay task, then the CVBI after reading. Two days later, participants were asked to
complete the following independently outside of the lab: the Intention to Vaccinate Scale, the
CVBI, and the other previously-mentioned surveys. All participants were debriefed after the first
post-reading measures were taken and were directed to the CDC should they have any follow-up
questions regarding vaccine usage safety or necessity.
Reading task. Participants were asked to, but not required to, read the eight articles.
Participants selected the articles’ order as they progressed through the task. They were allowed
to revisit an article they had already selected. Participants read each article aloud and after every
sentence, were prompted to verbally respond with their thoughts of what they were reading.
There was no mandatory requirement for the number of or order in which the articles were read.
Participants’ completion times varied widely, as did the likelihood that they visited and revisited
various texts.
Using Clusters to Predict Belief Change 11
Results
Regarding belief revision, initial analyses reveal that readers, overall, gained
misinformation as a function of reading. A 3x3 between subjects Analysis of Variance
(ANOVA) was conducted to compare changes across the different 3 types of beliefs (helpful,
harmful, and unnecessary) at 3 time points (before, immediately after, 2 days later) on the CVBI.
There was a significant difference in variance found F (2, 129) = 3.841, p = 0.024. Afterwards, 3
one-way repeated measures ANOVAs were conducted to identify the effects of time (measured
at pre-screen, immediately after reading, and after a two-day delay) on change for helpful,
unnecessary, and harmful beliefs, respectively. There was not a significant effect for time for
changes helpful, Wilk’s Lambda = 0.98, F (2, 42) = 0.37, p = 0.695, partial eta squared = 0.17, or
unnecessary, Wilk’s Lambda = 0.99, F (2, 42) = 0.30, p = 0.740, partial eta squared = 0.014,
beliefs. There was a significant effect for time for harmful beliefs, Wilk’s Lambda = 0.78, F (2,
42) = 6.186, p = 0.004, partial eta squared = 0.23. As there was a significant effect for time was
for harmful beliefs, simple effects analyses were conducted for harmful beliefs.
The simple effects for harmful beliefs have revealed that there was a significant increase
in beliefs reflecting an acceptance that vaccines are harmful held from pre-screen (M = 1.71, SD
= 0.61) relative to those held immediately after reading (M = 2.28, SD = 1.64), t (43) = -2.543, p
= 0.015). There was also a significant increase in harmful beliefs at a delay two days after
reading (M = 2.42, SD = 1.63), t (43) = -2.865, p = 0.006) compared to beliefs from pre-screen.
However, there was not a significant change from harmful beliefs held immediately after reading
(M = 2.28, SD = 1.64) to harmful beliefs held at a two-day delay (M = 2.42, SD = 1.63) t (43) = -
0.456, p = .65). It was shown that participants’ beliefs that vaccinations are harmful significantly
increased after reading (immediately after and at a two-day delay) compared to prior to reading
Using Clusters to Predict Belief Change 12
(pre-screen), but the two measures taken after reading were not shown to be significantly
different from each other. These initial findings, reflective of belief revision, will be predicted by
the outcomes of the cluster algorithm analyses. Because the only significant change in beliefs
was found for harmful beliefs, those data serve as the focus of analyses for this study.
It was hypothesized that participants’ changes in harmful beliefs could be predicted by
first identifying similarities in rates of use of referential cohesion, causal connectives, analytical
language, emotional language, and personal pronouns through clustering, then by performing a
series of regression analyses on those variables.
Clusters. K-means clustering was conducted using the identified optimal number of
clusters to organize the data into groups based on algorithm-defined similarity. The grouping was
conducted while minimizing the differences between grouped data and maximizing the
differences between distinct groups. The differences between individual data points and the
centroids of the clusters were measured based on proximity through the use of their Euclidean
distances. The optimal number of clusters for this data was found to be 3. The measures of
linguistic use for 19 participants were assigned into cluster 1, 5 into cluster 2, and 20 into cluster
3.
Single linear regressions. A series of single linear regressions were calculated to predict
changes in harmful beliefs based on referential cohesion, causal connectives, analytic language,
affective language, and personal pronoun use. Analyses involving each variable were further
separated by cluster identity. The regression equations for referential cohesion within clusters 1
(F(1,17) = 0.003, p = 0.957), 2 (F(1,3) = 0.244, p = 0.655), and 3 (F(1,18) = 1.767) p = 0.200)
were all found to be not significant with an R2 < 0.001, an R2 = 0.075, and an R2 = 0.089
respectively. The regression equation for causal connectives within cluster 1 (F(1,17) = 0.018, p
Using Clusters to Predict Belief Change 13
= 0.896) was found to be not significant with an R2 = 0.001. The regression equations for causal
connectives within clusters 2 (F(1,3) = 9.892, p = 0.051) and 3 (F(1,18) = 4.488, p = 0.048),
however, were found to be significant with an R2 = 0.767 and an R2 = 0.200 respectively. The
regression equations for analytic language within clusters 1 (F(1,17) = 0.195, p = 0.664), 2
(F(1,3) = 0.081, p = 0.795), and 3 (F(1,18) = 1.057), p = 0.317) were all found to be not
significant with an R2 = 0.011, an R2 = 0.026, and an R2 = 0.055 respectively. The regression
equations for affective language within clusters 1 (F(1,17) = 0.384, p = 0.543), 2 (F(1,3) = 2.087,
p = 0.244), and 3 (F(1,18) = 0.057), p = 0.814) were all found to be not significant with an R2 =
0.022, an R2 = 0.410, and an R2 = 0.003 respectively. The regression equations for personal
pronoun use within clusters 1 (F(1,17) = 1.311, p = 0.268), 2 (F(1,3) = 0.887, p = 0.416), and 3
(F(1,18) = 0.093), p = 0.764) were all found to be not significant with an R2 = 0.072, an R2 =
0.228, and an R2 = 0.005 respectively. Thus, of all 5 variables in all 3 clusters, only causal
connectives in the second and third cluster were found to be predictors of harmful belief change.
Therefore, within the first of these two groups, participants’ predicted harmful belief change
measure is equal to 5.191 0.163 (causal connectives) change when causal connectives use is
measured by Coh-Metrix’s word count and analysis. Within the second of the two predicting
clusters, participants’ predicted harmful belief change measure is equal to -1.628 + 0.092 (causal
connectives) change when causal connectives use is measured by Coh-Metrix’s word count and
analysis. Participants’ harmful beliefs increased 0.092 for each use of a causal connective.
Multiple linear regressions. A series of multiple linear regressions were calculated to
predict harmful belief change based on referential cohesion, causal connectives, analytic
language, affective language, and personal pronoun use. The regression equations for clusters 1
(F(5,13) = 0.263, p = 0.925) and 3 (F(5,14) = 1.518, p = 0.247) were both found to not be
Using Clusters to Predict Belief Change 14
significant with an R2 = 0.092 and an R2 = 0.351 respectively. As a result of a small population in
the second cluster, a regression equation for this cluster could not be calculated. Thus, none of
the combinations of the 5 variables within each cluster were found to be predictors of harmful
belief change. Therefore, cluster identity was not found to be a predictor of harmful belief
change.
Discussion
It was hypothesized that participants’ harmful belief changes could be predicted
by analyzing the rates of referential cohesion, causal connectives, analytic language, affective
language, and personal pronoun when organized through the use of clustering. Specifically, it
was hypothesized that clusters of participants exhibiting lower rates of referential cohesion,
causal connectives, and analytic language would predict changes in harmful beliefs resulting in
more harmful beliefs when compared to clusters of participants exhibiting higher use rates of
those same linguistic categories. Additionally, it was hypothesized that clusters of participants
exhibiting high use rates of affective language and personal pronouns would also predict changes
in harmful beliefs resulting in more harmful beliefs than clusters of participants exhibiting lower
use rates of those same linguistic categories. Broadly, the results did not support these
hypotheses. Only the regression equations for the second and third cluster of causal connective
use were found to significantly predict changes in harmful beliefs. For participants in the second
cluster, increased incidence of causal connectives resulted in decreased harmful beliefs.
Conversely, for participants in the third cluster, increased incidence of causal connectives
resulted in increased harmful beliefs. These two analyses support the hypotheses by providing
evidence of causal connectives being a predictor of change in harmful beliefs.
Using Clusters to Predict Belief Change 15
Because there has been no previous research on predicting vaccine belief revision
through the use of clustering, this study is uniquely situated as a pioneer of this area within the
existing literature. However, it may be compared to related areas of research on vaccine belief
language use. This study provides no significant evidence to support the findings that
participants in support of vaccination use comments typified by greater analytical thinking
(Faasse, Chatman, & Martin, 2016). Due in part to the low sample of the current study, the
validity should be retested in future research. Furthermore, the study’s participation was limited
to students at the University of Memphis, lowering the generalizability of the results.
Future research in this area should revisit the current study’s linguistic categories using a
larger sample of data as well as explore the predictive power of other linguistic categories of
interest such as positive and negative emotion words, authenticity, health references, or other
measures analyzed by LIWC or Coh-Metrix. Additionally, future research should further explore
the predictive power of multiple linguistic categories in conjunction potentially through the use
of further multiple linear regressions or similar analyses.
The results of this study generally do not support the hypothesis that the rates of
referential cohesion, causal connectives, analytic language, affective language, and personal
pronoun use can be used to predict harmful belief change when organized through the use of
clustering. The two significant analyses show that use rates of causal connectives, when
organized through automatic clustering, can be used to predict belief revision. Since the causal
connective linguistic category includes phrases such as “because of” and “as a result of”, these
findings show that believing that something directly causes something else, in the scope of
vaccine beliefs, indicates the co-activation of new and existing ideas as well as the integration of
new information that either opposes or reinforces the existing ideas. Additional research on
Using Clusters to Predict Belief Change 16
vaccine belief change analysis through clusters should focus on validating the current research
and exploring other linguistic categories as predictors of belief change. Ultimately, this study
provides the groundwork for further research in the analysis of vaccine belief revision through
the use of clustering.
Using Clusters to Predict Belief Change 17
References
Allen, L. K., McNamara, D. S., & McCrudden, M. T. (2015). Change your mind: Investigating
the effects of self-explanation in the resolution of misconceptions. In D. C. Noelle, R.
Dale, A. S. Warlaumont, J. Yoshimi, T. Matlock, C. D. Jennings, & P. Maglio (Eds.),
Proceedings of the 37th Annual Meeting of the Cognitive Science Society (Cog Sci
2015). Pasadena, CA.
Borgen, F. H., & Barnett, D. C. (1987). Applying cluster analysis in counseling psychology
research. Journal of Counseling Psychology, 34(4), 456468. doi: 10.1037/0022-
0167.34.4.456
Faasse, K., Chatman, C. J., & Martin, L. R. (2016). A comparison of language use in pro- and
anti-vaccination comments in response to a high profile Facebook post. Vaccine, 34(47),
58085814. doi: 10.1016/j.vaccine.2016.09.029
Graesser, A.C. & McNamara, D.S. (2017). Coh-Metrix. Memphis, TN: The University of
Memphis (www.tool.cohmetrix.com).
Graesser, A. C., Mcnamara, D. S., Louwerse, M. M., & Cai, Z. (2004). Coh-Metrix: Analysis of
text on cohesion and language. Behavior Research Methods, Instruments, &
Computers, 36(2), 193202. doi: 10.3758/bf03195564
Kendeou, P., Butterfuss, R., Kim, J., & Boekel, M. V. (2018). Knowledge revision through the
lenses of the three-pronged approach. Memory & Cognition, 47(1), 3346. doi:
10.3758/s13421-018-0848-y
Using Clusters to Predict Belief Change 18
Kendeou, P., & O’Brien, E. J. (2014). The Knowledge Revision Components (KReC)
Framework: Processes and Mechanisms. In D. N. Rapp & J. L.G. Braasch
(Eds.), Processing Inaccurate Information (353 - 369). Cambridge, Massachusetts: The
MIT Press.
Kessler, E. D., Braasch, J. L., & Kardash, C. M. (2019). Individual Differences in Revising (and
Maintaining) Accurate and Inaccurate Beliefs About Childhood Vaccinations. Discourse
Processes, 56(5-6), 415428. doi: 10.1080/0163853x.2019.1596709
Lewandowsky, S., Ecker, U. K. H., Seifert, C. M., Schwarz, N., & Cook, J. (2012).
Misinformation and Its Correction. Psychological Science in the Public Interest, 13(3),
106131. doi: 10.1177/1529100612451018
Pennebaker, J.W., Booth, R.J., Boyd, R.L., & Francis, M.E. (2015). Linguistic Inquiry and Word
Count: LIWC2015. Austin, TX: Pennebaker Conglomerates (www.LIWC.net).
Ramírez-Esparza, Nairán & Chung, Cindy & Kacewicz, Ewa & Pennebaker, James. (2008). The
Psychology of Word Use in Depression Forums in English and in Spanish: Testing Two
Text Analytic Approaches. ICWSM 2008 - Proceedings of the 2nd International
Conference on Weblogs and Social Media.
Rapp, D. N. (2016). The Consequences of Reading Inaccurate Information. Current Directions
in Psychological Science, 25(4), 281285. doi: 10.1177/0963721416649347
Tausczik, Y. R., & Pennebaker, J. W. (2009). The Psychological Meaning of Words: LIWC and
Computerized Text Analysis Methods. Journal of Language and Social
Psychology, 29(1), 2454. doi: 10.1177/0261927x09351676
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
As a research technique that has grown rapidly in applications in many scientific disciplines, cluster analysis has potential for wider use in counseling psychology research. We begin with a simple example illustrating the clustering approach. Topics covered include the variety of approaches in clustering, the times when cluster analysis may be a choice for analysis, the steps in cluster analysis, the data features, such as level, shape, and scatter, that affect cluster results, alternate clustering methods and evidence indicating which are most effective, and examples of clustering applications in counseling research. Although we make an attempt to provide a comprehensive overview of major issues, the reader is encouraged to consult several good recent publications on the topic that are especially relevant for psychologists. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Article
Full-text available
We are in the midst of a technological revolution whereby, for the first time, researchers can link daily word use to a broad array of real-world behaviors. This article reviews several computerized text analysis methods and describes how Linguistic Inquiry and Word Count (LIWC) was created and validated. LIWC is a transparent text analysis program that counts words in psychologically meaningful categories. Empirical results using LIWC demonstrate its ability to detect meaning in a wide variety of experimental settings, including to show attentional focus, emotionality, social relationships, thinking styles, and individual differences.
Article
Full-text available
Advances in computational linguistics and discourse processing have made it possible to automate many language- and text-processing mechanisms. We have developed a computer tool called Coh-Metrix, which analyzes texts on over 200 measures of cohesion, language, and readability. Its modules use lexicons, part-of-speech classifiers, syntactic parsers, templates, corpora, latent semantic analysis, and other components that are widely used in computational linguistics. After the user enters an English text, CohMetrix returns measures requested by the user. In addition, a facility allows the user to store the results of these analyses in data files (such as Text, Excel, and SPSS). Standard text readability formulas scale texts on difficulty by relying on word length and sentence length, whereas Coh-Metrix is sensitive to cohesion relations, world knowledge, and language and discourse characteristics.
Article
This study examined relationships between several individual differences measures and belief revision after reading a text refuting common miscon- ceptions about childhood vaccinations. Individual differences included pre- existing accurate and inaccurate beliefs on the topic, prior knowledge about how vaccinations work, need for cognition, and flexible thinking. Dependent measures included an essay assessing vaccination support and a readministration of the preexisting topic beliefs inventory immediately after reading and after a 2-day delay. Multiple regression analyses demon- strated belief consistency effects in that preexisting accurate and inaccurate beliefs stably predict their postreading counterparts. Above and beyond these effects, readers higher in need for cognition were more likely to gain accurate beliefs about childhood vaccination at immediate and delayed time points. However, a backfire effect was also observed such that readers higher in flexible thinking were more likely to gain inaccurate beliefs, with concomitant lower values on accurate beliefs; this effect was present immediately and at a delay. The findings confirm that people can display a strong “myside bias” even when reading refutations, but that higher need for cognition affords opportunities for appropriate knowledge revision. At the same time, our findings suggest an alarming, potentially detrimental aspect of refutational texts: They may introduce opportunities for some learners to acquire novel misconceptions, especially if they are more flexible thinkers. Future directions of the current work are discussed.
Article
In the present study, we employed the three-pronged approach to determine the actual cognitive processes theorized in knowledge revision. First, the Knowledge Revision Components (KReC) framework was identified as the guiding theory. Second, think-aloud analysis highlighted at which points in refutation texts readers detected discrepancies between their incorrect, commonsense beliefs and the correct beliefs, and the exact processes with which they dealt with these discrepancies—successfully or unsuccessfully, as indicated by posttest scores. Third, corroborating reading-time data and posttest data demonstrated that the structure of the refutation texts facilitated the coactivation and integration of the explanation with the commonsense belief, resulting in knowledge revision. Finally, an analysis directly connected the processes identified during think-aloud to sentence reading times. These findings systematically identify the cognitive processes theorized during knowledge revision and, in doing so, provide evidence for the conditions for revision outlined in the KReC framework.
Article
We are regularly confronted with statements that are inaccurate, sometimes obviously so. Unfortunately, people can be influenced by and rely upon inaccurate information, engaging in less critical evaluation than might be hoped. Empirical studies have consistently demonstrated that even when people should know better, reading inaccurate information can affect their performance on subsequent tasks. What encourages people’s encoding and use of false statements? The current article outlines how reliance on inaccurate information is a predictable consequence of the routine cognitive processes associated with memory, problem solving, and comprehension. This view helps identify conditions under which inaccurate information is more or less likely to influence subsequent decisions. These conditions are informative in the consideration of information-design approaches and instructional methods intended to support critical thinking.
Change your mind: Investigating the effects of self-explanation in the resolution of misconceptions
  • L K Allen
  • D S Mcnamara
  • M T Mccrudden
Allen, L. K., McNamara, D. S., & McCrudden, M. T. (2015). Change your mind: Investigating the effects of self-explanation in the resolution of misconceptions. In D. C. Noelle, R. Dale, A. S. Warlaumont, J. Yoshimi, T. Matlock, C. D. Jennings, & P. Maglio (Eds.), Proceedings of the 37th Annual Meeting of the Cognitive Science Society (Cog Sci 2015). Pasadena, CA.
The Knowledge Revision Components (KReC) Framework: Processes and Mechanisms
  • P Kendeou
  • E J Brien
Kendeou, P., & O'Brien, E. J. (2014). The Knowledge Revision Components (KReC) Framework: Processes and Mechanisms. In D. N. Rapp & J. L.G. Braasch (Eds.), Processing Inaccurate Information (353 -369). Cambridge, Massachusetts: The MIT Press.