Analysis - Experiment

This file includes code used to clean, analyze data, and draw plots from the study’s experiment. The source data is also stored in this repository.

renv.lock lists all package versions used for the analysis; (https://rstudio.github.io/renv/articles/renv.html)[renv] can be used to install them.

Setup

Data Loading and Cleaning

The dataset has one row per participant-article combination. Each participant saw eight articles. The dataset is joined with the participant’s pre-survey data collected four months previously.

As part of the data cleaning, friendly labels are assigned to some raw qualtrics pre-survey responses.

## Read in the baseline info and the new survey info

path <- c("")

paired_data <-
  read.csv(paste(path, "input/experiment-data-paired-deidentified.csv", sep = ""))


## Drop participants with no set political affiliation -- data collection error

paired_data <- paired_data %>% filter(!is.na(POL_AFFIL_updated))


## Drop the example article from the dataset

paired_data <- paired_data %>% filter(article_true_leaning != "neither")

paired_data$POL_AFFIL_updated[which(paired_data$POL_AFFIL_updated ==
                                                1)] <- 0
paired_data$POL_AFFIL_updated[which(paired_data$POL_AFFIL_updated ==
                                                4)] <- 1
paired_data <-
  paired_data %>%
  mutate(political_affiliation = case_when(
    POL_AFFIL_updated == 0 ~ 'Liberal',
    POL_AFFIL_updated == 1 ~ 'Conservative'))

paired_data <- paired_data %>% mutate(
  political_affiliation = as.factor(political_affiliation),
  label_type = as.factor(label_type),
  article_user_is_opposed_content = as.factor(article_user_is_opposed_content),
  article_user_is_opposed_source = as.factor(article_user_is_opposed_source),
  condition = as.numeric(condition),
  num_decetively_seen = as.numeric(num_decetively_seen),
  actual_article_presentation_index = as.numeric(actual_article_presentation_index),
  DEMOG_Age = as.numeric(DEMOG_Age),
  DEMOG_Gender = as.factor(DEMOG_Gender),
  DEMOG_White = as.factor(DEMOG_White),
  DEMOG_College_Degree = as.factor(DEMOG_College_Degree),
  all_cum_lib_deceptive = as.numeric(all_cum_lib_deceptive),
  all_cum_cons_deceptive = as.numeric(all_cum_cons_deceptive),
  article_is_content_blended = as.factor(article_is_content_blended),
  num_content_blended_seen = as.numeric(num_content_blended_seen),
  article_is_source_blended = as.factor(article_is_source_blended)
)

loadings = function(res.pca){
  # <http://factominer.free.fr/question/FAQ.html>
  #
  # Loadings (i.e. standard coordinates) are not given by FactoMineR's methods. They return principal coordinates.
  # You can calculate them by dividing variables' coordinates on a dimension by this dimension's eigenvalue's square root.
  # Just type:
  # sweep(res.pca$var$coord,2,sqrt(res.pca$eig[1:ncol(res.pca$var$coord),

  return(sweep(res.pca$var$coord,2,sqrt(res.pca$eig[1:ncol(res.pca$var$coord),1]),FUN="/"))
}

Feature Engineering

DV_agreeable

Items:

• agreeable (Kim & Dennis, 2019)
• confirming (Moravec et al., 2019)
• rational Inspired by (Moravec et al., 2019)
• supporting my views Inspired by (Moravec et al., 2019)
• meeting my expectations Inspired by (Moravec et al., 2019)

DV_agreeable.items = c(
    "article_survey_q_find_issue_agreeable",
    "article_survey_q_find_issue_confirming",
    "article_survey_q_find_issue_rational",
    "article_survey_q_find_issue_supporting_my_views",
    "article_survey_q_find_issue_meeting_my_expectations"
  )

pca.tmp = PCA(paired_data[, DV_agreeable.items], graph = FALSE)
as.data.frame(pca.tmp$eig)

	eigenvalue	percentage of variance	cumulative percentage of variance
comp 1	4.11	82.24	82.24
comp 2	0.31	6.15	88.39
comp 3	0.22	4.43	92.83
comp 4	0.21	4.21	97.03
comp 5	0.15	2.97	100.00

as.data.frame(loadings(pca.tmp))

	Dim.1	Dim.2	Dim.3	Dim.4	Dim.5
article_survey_q_find_issue_agreeable	0.44	-0.51	0.66	-0.31	-0.10
article_survey_q_find_issue_confirming	0.45	-0.30	-0.13	0.76	0.32
article_survey_q_find_issue_rational	0.45	-0.03	-0.52	-0.56	0.45
article_survey_q_find_issue_supporting_my_views	0.46	0.07	-0.35	0.04	-0.81
article_survey_q_find_issue_meeting_my_expectations	0.43	0.80	0.38	0.07	0.15

paired_data = paired_data %>%
  mutate(DV_agreeable = rowMeans(select(., DV_agreeable.items)))

DV_credible

Items:

• believable (Kim et al., 2019)
• truthful (Kim & Dennis, 2019)
• credible (Moravec et al., 2019)

DV_credible.items = c(
    "article_survey_q_believable",
    "article_survey_q_truthful" ,
    "article_survey_q_credible"
  )

pca.tmp = PCA(paired_data[, DV_credible.items], graph = FALSE)
as.data.frame(pca.tmp$eig)

	eigenvalue	percentage of variance	cumulative percentage of variance
comp 1	2.87	95.75	95.75
comp 2	0.08	2.53	98.28
comp 3	0.05	1.72	100.00

as.data.frame(loadings(pca.tmp))

	Dim.1	Dim.2	Dim.3
article_survey_q_believable	0.57	0.82	0.06
article_survey_q_truthful	0.58	-0.35	-0.74
article_survey_q_credible	0.58	-0.46	0.67

paired_data = paired_data %>%
  mutate(DV_credible = rowMeans(select(., DV_credible.items)))

DV_sharing_nonnegative

Items:

• “liked” (Kim & Dennis, 2019)
• post support comment (Kim & Dennis, 2019)
• share without comment (Moravec et al., 2019)
• share with quote (Novel to this study)
• share with supporting comment (Novel to this study)

DV_sharing_nonnegative.items = c(
  "article_survey_q_like_article",
  "article_survey_q_post_support_comment",
  "article_survey_q_share_without_comment",
  "article_survey_q_share_with_quote",
  "article_survey_q_share_with_supporting_comment"
)

pca.tmp = PCA(paired_data[, DV_sharing_nonnegative.items], graph = FALSE)
as.data.frame(pca.tmp$eig)

	eigenvalue	percentage of variance	cumulative percentage of variance
comp 1	4.02	80.35	80.35
comp 2	0.43	8.69	89.04
comp 3	0.28	5.58	94.62
comp 4	0.17	3.38	98.00
comp 5	0.10	2.00	100.00

as.data.frame(loadings(pca.tmp))

	Dim.1	Dim.2	Dim.3	Dim.4	Dim.5
article_survey_q_like_article	0.41	0.85	-0.24	0.24	-0.04
article_survey_q_post_support_comment	0.47	-0.15	-0.36	-0.53	0.59
article_survey_q_share_without_comment	0.44	0.09	0.87	-0.18	0.05
article_survey_q_share_with_quote	0.45	-0.43	-0.04	0.75	0.19
article_survey_q_share_with_supporting_comment	0.47	-0.26	-0.21	-0.24	-0.78

paired_data = paired_data %>%
  mutate(DV_sharing_nonnegative = rowMeans(select(., DV_sharing_nonnegative.items)))

DV_sharing_negative

Items:

• share with opposing comment (Kim & Dennis, 2019; Kim et al., 2019)
• post opposing comment (Moravec et al., 2019)

DV_sharing_negative.items = c(
  "article_survey_q_post_opposing_comment",
    "article_survey_q_share_with_opposing_comment"
)

pca.tmp = PCA(paired_data[, DV_sharing_negative.items], graph = FALSE)
as.data.frame(pca.tmp$eig)

	eigenvalue	percentage of variance	cumulative percentage of variance
comp 1	1.87	93.49	93.49
comp 2	0.13	6.51	100.00

as.data.frame(loadings(pca.tmp))

	Dim.1	Dim.2
article_survey_q_post_opposing_comment	0.71	-0.71
article_survey_q_share_with_opposing_comment	0.71	0.71

paired_data = paired_data %>%
  mutate(DV_sharing_negative = rowMeans(select(., DV_sharing_negative.items)))

Confirmation bias

A measure of confirmation bias is suggested by (Kim et al., 2019); Modified from: (Brown et al., 2014):

Confirmation_Bias = ("Position on Article" x "Importance") / 100

Divided by 100 to keep on the same scale as all other DVs.

paired_data = paired_data %>%
  mutate(DV_confirmation_bias = article_survey_q_find_issue_important * article_survey_q_article_position / 100)

is_unexpected

An article is “unexpected” if a friendly source presents unfriendly content, and vice versa.

paired_data = paired_data %>%
  mutate( unexpected = ( article_user_is_opposed_content != article_user_is_opposed_source ) )

Confirmation Bias Correlations

Correlations between each dependent variable and confirmation bias.

correlation_data <-
  paired_data[, c(
    "DV_confirmation_bias",
    "DV_sharing_nonnegative",
    "DV_sharing_negative",
    "DV_agreeable",
    "DV_credible"
  )]
correlations = rcorr(as.matrix(correlation_data))

correlations.r = as.data.frame(correlations$r)['DV_confirmation_bias']
knitr::kable(correlations.r, caption="Pearson's `r`")

Pearson’s r

	DV_confirmation_bias
DV_confirmation_bias	1.0000000
DV_sharing_nonnegative	0.5549678
DV_sharing_negative	0.0725429
DV_agreeable	0.6675339
DV_credible	0.5448545

correlations.P = as.data.frame(correlations$P)['DV_confirmation_bias']
knitr::kable(correlations.P, caption='p-values')

p-values

	DV_confirmation_bias
DV_confirmation_bias	NA
DV_sharing_nonnegative	0.0000000
DV_sharing_negative	0.0183895
DV_agreeable	0.0000000
DV_credible	0.0000000

Modeling

This section shows the fitting of each model included in the paper.

Model specification

All models used the following independent variables:

• Three-way interaction term:

  article_user_is_opposed_content * article_user_is_opposed_source * political_affiliation

  This is a full three-way interaction of three factors that specifies all possible main effects, two-way, and three-way interactions between the factors.

• Repeated measures:

  Repeated measures were accounted for by specifying a random intercept per participant:

  (1 | uid)

• Control variables:

  All models used the following control variables:

  • Count_TRUST: The number of publication sources the given participant said they “trusted” in the presurvey.
  • condition: The randomly assigned condition.
  • all_cum_lib_deceptive: For liberals, how many “unexpected” articles had been seen so far.
  • all_cum_cons_deceptive: For conservatives, how many “unexpected” articles had been seen so far.
  • actual_article_presentation_index: How many articles the participant had seen so far.
  • DEMOG_Age: A participant’s age.
  • DEMOG_Gender: A participant’s gender.
  • DEMOG_White: Whether a participant self-identified as white.
  • DEMOG_College_Degree: Whether a participant had a college degree.

formula.base =
  " ~ article_user_is_opposed_content * article_user_is_opposed_source * political_affiliation +
      (1 | UID) +
      Count_TRUST + condition +
      all_cum_lib_deceptive + all_cum_cons_deceptive + actual_article_presentation_index +
      DEMOG_Age + DEMOG_Gender + DEMOG_White + DEMOG_College_Degree"

Attitudes and Beliefs

This section describes the two attitudes and beliefs-related dvs: “Agreeable” and “Credible”

Agreeable

Agreeable.formula = formula(paste0('DV_agreeable', formula.base))
Agreeable.model <-
  lmer(
    Agreeable.formula,
    data = paired_data
  )

# Agreeable -- pairwise contrasts ----------------------------------------

Agreeable.model.pairwise = do_pairwise_content_source(Agreeable.model)

# Agreeable.model.pairwise$emm
as.data.frame(Agreeable.model.pairwise$contrasts) %>%
  mutate(across(where(is.numeric), ~as.numeric(number(., accuracy=0.01))))

contrast	estimate	SE	df	t.ratio	p.value
Fully friendly vs Friendly-source/unfriendly-content	18.02	2.77	937.43	6.51	0.00
Fully friendly vs Unfriendly-source/friendly-content	8.75	2.76	937.19	3.17	0.00
Fully friendly vs Fully unfriendly	25.40	2.06	908.32	12.34	0.00
Friendly-source/unfriendly-content vs Unfriendly-source/friendly-content	-9.27	2.79	908.18	-3.32	0.00
Friendly-source/unfriendly-content vs Fully unfriendly	7.38	2.76	937.43	2.68	0.01
Unfriendly source-friendly content vs Fully unfriendly	16.65	2.75	937.23	6.05	0.00
Friendly source vs Unfriendly source	8.07	1.73	908.19	4.65	0.00
Friendly content vs Unfriendly content	17.33	1.73	908.28	10.00	0.00

# Agreeable -- plots ----

plot_agreeable(Agreeable.model,custom.ylim = c(40,85))

plot_opposed_vs_blended_content(Agreeable.model, 'Agreeable', custom.ylim = c(40,80))

Credible

Credible.formula = formula(paste0('DV_credible', formula.base))
Credible.model <-
  lmer(
    Credible.formula,
    data = paired_data
  )


# Credible -- contrasts ----------------------------------------------

Credible.model.pairwise = do_pairwise_content_source(Credible.model)
# Credible.model.pairwise$emm
as.data.frame(Credible.model.pairwise$contrasts) %>%
  mutate(across(where(is.numeric), ~as.numeric(number(., accuracy=0.01))))

contrast	estimate	SE	df	t.ratio	p.value
Fully friendly vs Friendly-source/unfriendly-content	13.62	2.84	942.09	4.81	0.00
Fully friendly vs Unfriendly-source/friendly-content	6.93	2.83	941.77	2.45	0.01
Fully friendly vs Fully unfriendly	25.71	2.11	908.40	12.17	0.00
Friendly-source/unfriendly-content vs Unfriendly-source/friendly-content	-6.70	2.86	908.24	-2.34	0.02
Friendly-source/unfriendly-content vs Fully unfriendly	12.09	2.83	942.00	4.28	0.00
Unfriendly source-friendly content vs Fully unfriendly	18.79	2.82	941.73	6.66	0.00
Friendly source vs Unfriendly source	9.51	1.78	908.24	5.34	0.00
Friendly content vs Unfriendly content	16.21	1.78	908.35	9.10	0.00

# Credible -- plots ----------------------------------------------

plot_credible(Credible.model, custom.ylim = c(40,85))

plot_opposed_vs_blended_content(Credible.model, 'Credible', custom.ylim = c(40,80))

Model Output

Model output for Credible and Agreeable Attitudes and Beliefs models.

tab_model(Credible.model, Agreeable.model)

	DV_credible			DV_agreeable
Predictors	Estimates	CI	p	Estimates	CI	p
(Intercept)	77.81	65.07 – 90.54	<0.001	69.91	56.32 – 83.50	<0.001
article_user_is_opposed_content [TRUE]	-14.38	-23.43 – -5.32	0.002	-22.28	-31.17 – -13.40	<0.001
article_user_is_opposed_source [TRUE]	-5.76	-14.88 – 3.36	0.216	-8.87	-17.83 – 0.09	0.052
political_affiliation [Liberal]	0.79	-6.28 – 7.86	0.827	-6.91	-14.23 – 0.42	0.065
Count_TRUST	0.56	0.28 – 0.85	<0.001	0.66	0.35 – 0.97	<0.001
condition	1.23	-1.83 – 4.30	0.430	-0.07	-3.22 – 3.08	0.965
all_cum_lib_deceptive	2.23	-1.23 – 5.69	0.206	2.16	-1.32 – 5.64	0.223
all_cum_cons_deceptive	0.58	-2.83 – 3.99	0.738	-2.50	-5.91 – 0.91	0.151
actual_article_presentation_index	-0.01	-1.02 – 1.00	0.986	0.38	-0.61 – 1.37	0.449
DEMOG_Age	-0.15	-0.34 – 0.03	0.099	-0.06	-0.26 – 0.14	0.566
DEMOG_Gender [Male]	-4.11	-8.57 – 0.36	0.071	0.27	-4.59 – 5.13	0.913
DEMOG_White [1]	-2.17	-7.90 – 3.55	0.457	-5.85	-12.07 – 0.38	0.066
DEMOG_College_Degree [Yes]	1.12	-3.76 – 6.00	0.654	-2.99	-8.30 – 2.32	0.270
article_user_is_opposed_content [TRUE] * article_user_is_opposed_source [TRUE]	-5.43	-18.95 – 8.09	0.431	7.36	-5.99 – 20.70	0.280
article_user_is_opposed_content [TRUE] * political_affiliation [Liberal]	1.51	-9.13 – 12.14	0.781	8.54	-1.95 – 19.02	0.110
article_user_is_opposed_source [TRUE] * political_affiliation [Liberal]	-2.34	-13.11 – 8.44	0.671	0.24	-10.41 – 10.89	0.964
(article_user_is_opposed_content [TRUE] article_user_is_opposed_source [TRUE]) political_affiliation [Liberal]	0.53	-14.76 – 15.83	0.946	-11.98	-27.18 – 3.21	0.122
Random Effects
σ2	623.21			590.85
τ00	77.16 UID			110.03 UID
ICC	0.11			0.16
N	131 UID			131 UID
Observations	1048			1048
Marginal R2 / Conditional R2	0.186 / 0.275			0.185 / 0.313

Social Media Engagement Intentions

First, plots for positive social media engagement intentions are shown; then, plots negative ones; then, full model output is shown.

Positive Social Media Engagement Intentions

eng_pos.formula = paste0('DV_sharing_nonnegative', formula.base)
eng_pos.model <-
  lmer(
    eng_pos.formula,
    data = paired_data
  )


# contrasts ----

eng_pos.model.emm.pairwise = do_pairwise_content_source_by_political_affiliation(eng_pos.model)
# eng_pos.model.emm.pairwise$emm
as.data.frame(eng_pos.model.emm.pairwise$contrasts) %>%
  mutate(across(where(is.numeric), ~as.numeric(number(., accuracy=0.01))))

contrast	estimate	SE	df	t.ratio	p.value
Fully friendly vs friendly-source/unfriendly-content (Conservative)	14.87	3.19	937.14	4.66	0.00
Fully friendly vs Unfriendly-source/friendly content (Conservative)	6.25	3.24	944.67	1.93	0.05
Fully friendly vs Fully unfriendly (Conservative)	10.75	2.35	908.02	4.58	0.00
friendly-source/unfriendly-content vs Unfriendly-source/friendly-content (Conservative)	-8.62	3.45	927.94	-2.50	0.01
friendly-source/unfriendly-content vs Fully unfriendly (Conservative)	-4.12	3.19	937.26	-1.29	0.20
Unfriendly-source/friendly-content vs Fully unfriendly (Conservative)	4.50	3.23	944.47	1.39	0.16
source-aligned vs source-opposed (Conservative)	1.06	2.09	921.99	0.51	0.61
content-aligned vs content-opposed (Conservative)	9.69	2.09	921.53	4.64	0.00
Fully friendly vs Friendly-source/unfriendly-content (Liberal)	7.87	2.10	929.06	3.76	0.00
Fully friendly vs Unfriendly-source/friendly-content (Liberal)	5.62	2.11	936.38	2.66	0.01
Fully friendly vs Fully unfriendly (Liberal)	14.34	1.55	908.28	9.26	0.00
friendly-source/unfriendly-content vs Unfriendly-source/friendly-content (Liberal)	-2.25	2.29	951.34	-0.98	0.33
friendly-source/unfriendly-content vs Fully unfriendly (Liberal)	6.47	2.10	930.89	3.08	0.00
Unfriendly-source/friendly-content vs Fully unfriendly (Liberal)	8.72	2.09	934.08	4.18	0.00
Friendly source vs unfriendly source (Liberal)	6.04	1.39	940.71	4.34	0.00
Friendly content vs unfriendly content (Liberal)	8.29	1.37	936.16	6.06	0.00

# plots ----

plot_opposed_vs_blended_content.by_political_party(eng_pos.model, 'Positive Social Media Engagement', custom.ylim =c(10,55))

## plot 3-way interaction
plot_two_by_two(model = eng_pos.model,
                dv_label_long = 'Estimated marginal means',
                dv_label_short = 'Eng_pos',custom.ylim = c(10,55))

Negative Intentions

eng_neg.formula = paste0('DV_sharing_negative', formula.base)
eng_neg.model <-
  lmer(
    eng_neg.formula,
    data = paired_data
  )

tab_model(eng_neg.model)

	DV_sharing_negative
Predictors	Estimates	CI	p
(Intercept)	35.39	15.21 – 55.56	0.001
article_user_is_opposed_content [TRUE]	8.10	1.35 – 14.85	0.019
article_user_is_opposed_source [TRUE]	5.35	-1.51 – 12.20	0.126
political_affiliation [Liberal]	-15.53	-24.99 – -6.08	0.001
Count_TRUST	0.57	0.07 – 1.07	0.024
condition	-2.03	-5.87 – 1.82	0.301
all_cum_lib_deceptive	1.60	-1.32 – 4.52	0.283
all_cum_cons_deceptive	-1.83	-4.65 – 0.99	0.203
actual_article_presentation_index	-0.42	-1.17 – 0.32	0.263
DEMOG_Age	-0.20	-0.51 – 0.12	0.220
DEMOG_Gender [Male]	6.09	-1.64 – 13.83	0.122
DEMOG_White [1]	-12.56	-22.42 – -2.69	0.013
DEMOG_College_Degree [Yes]	0.39	-8.07 – 8.85	0.928
article_user_is_opposed_content [TRUE] * article_user_is_opposed_source [TRUE]	-4.53	-14.87 – 5.81	0.390
article_user_is_opposed_content [TRUE] * political_affiliation [Liberal]	0.41	-7.68 – 8.51	0.920
article_user_is_opposed_source [TRUE] * political_affiliation [Liberal]	-1.91	-10.23 – 6.40	0.652
(article_user_is_opposed_content [TRUE] article_user_is_opposed_source [TRUE]) political_affiliation [Liberal]	2.14	-9.90 – 14.18	0.728
Random Effects
σ2	322.41
τ00 UID	426.70
ICC	0.57
N UID	131
Observations	1048
Marginal R2 / Conditional R2	0.159 / 0.638

# contrasts ----
eng_neg.pairwise = do_pairwise_content_source(eng_neg.model)
# eng_neg.pairwise$emm
as.data.frame(eng_neg.pairwise$contrasts) %>%
  mutate(across(where(is.numeric), ~as.numeric(number(., accuracy=0.01))))

contrast	estimate	SE	df	t.ratio	p.value
Fully friendly vs Friendly-source/unfriendly-content	-8.31	2.06	916.30	-4.04	0.00
Fully friendly vs Unfriendly-source/friendly-content	-4.39	2.05	916.26	-2.14	0.03
Fully friendly vs Fully unfriendly	-9.24	1.52	908.07	-6.08	0.00
Friendly-source/unfriendly-content vs Unfriendly-source/friendly-content	3.92	2.06	908.02	1.90	0.06
Friendly-source/unfriendly-content vs Fully unfriendly	-0.93	2.05	916.40	-0.45	0.65
Unfriendly source-friendly content vs Fully unfriendly	-4.85	2.05	916.38	-2.37	0.02
Friendly source vs Unfriendly source	-2.66	1.28	908.03	-2.08	0.04
Friendly content vs Unfriendly content	-6.58	1.28	908.05	-5.14	0.00

# plots ----
plot_negative_social_media(eng_neg.model, custom.ylim=c(10,55))

plot_opposed_vs_blended_content(eng_neg.model, 'Negative Social Media Engagement', custom.ylim=c(10,55))

Negative Intentions - expected vs unexpected

A follow-up test prompted by unexpected results for DV_sharing_negative explored whether differences existed between expected and unepected articles.

model.emm = emmeans(eng_neg.model,
                    ~ article_user_is_opposed_content * article_user_is_opposed_source)
model.emm.contrast = contrast(model.emm,
                              list('expected - unexpected' = c(1, -1, -1, 1)))
print(model.emm.contrast)

##  contrast              estimate   SE  df t.ratio p.value
##  expected - unexpected    -3.46 3.21 922 -1.078  0.2812 
## 
## Results are averaged over the levels of: political_affiliation, DEMOG_Gender, DEMOG_White, DEMOG_College_Degree 
## Degrees-of-freedom method: kenward-roger

Model Output

tab_model(eng_pos.model, eng_neg.model)

	DV_sharing_nonnegative			DV_sharing_negative
Predictors	Estimates	CI	p	Estimates	CI	p
(Intercept)	48.95	29.82 – 68.08	<0.001	35.39	15.21 – 55.56	0.001
article_user_is_opposed_content [TRUE]	-14.87	-21.12 – -8.62	<0.001	8.10	1.35 – 14.85	0.019
article_user_is_opposed_source [TRUE]	-6.25	-12.59 – 0.09	0.053	5.35	-1.51 – 12.20	0.126
political_affiliation [Liberal]	-15.01	-23.95 – -6.07	0.001	-15.53	-24.99 – -6.08	0.001
Count_TRUST	0.84	0.37 – 1.32	<0.001	0.57	0.07 – 1.07	0.024
condition	0.93	-2.70 – 4.56	0.617	-2.03	-5.87 – 1.82	0.301
all_cum_lib_deceptive	-0.67	-3.38 – 2.04	0.628	1.60	-1.32 – 4.52	0.283
all_cum_cons_deceptive	-1.82	-4.43 – 0.79	0.172	-1.83	-4.65 – 0.99	0.203
actual_article_presentation_index	0.36	-0.33 – 1.04	0.308	-0.42	-1.17 – 0.32	0.263
DEMOG_Age	-0.17	-0.47 – 0.13	0.279	-0.20	-0.51 – 0.12	0.220
DEMOG_Gender [Male]	2.54	-4.80 – 9.88	0.498	6.09	-1.64 – 13.83	0.122
DEMOG_White [1]	-16.67	-26.03 – -7.30	<0.001	-12.56	-22.42 – -2.69	0.013
DEMOG_College_Degree [Yes]	-1.47	-9.51 – 6.56	0.719	0.39	-8.07 – 8.85	0.928
article_user_is_opposed_content [TRUE] * article_user_is_opposed_source [TRUE]	10.37	0.80 – 19.93	0.034	-4.53	-14.87 – 5.81	0.390
article_user_is_opposed_content [TRUE] * political_affiliation [Liberal]	7.00	-0.49 – 14.49	0.067	0.41	-7.68 – 8.51	0.920
article_user_is_opposed_source [TRUE] * political_affiliation [Liberal]	0.63	-7.07 – 8.32	0.873	-1.91	-10.23 – 6.40	0.652
(article_user_is_opposed_content [TRUE] article_user_is_opposed_source [TRUE]) political_affiliation [Liberal]	-11.21	-22.36 – -0.06	0.049	2.14	-9.90 – 14.18	0.728
Random Effects
σ2	275.71			322.41
τ00	386.37 UID			426.70 UID
ICC	0.58			0.57
N	131 UID			131 UID
Observations	1048			1048
Marginal R2 / Conditional R2	0.214 / 0.673			0.159 / 0.638

References

Brown, S. A., Venkatesh, V., & Goyal, S. (2014). Expectation Confirmation in Information Systems Research. MIS Quarterly, 38(3), 729-A729.

Kim, A., & Dennis, A. R. (2019). SAYS WHO? THE EFFECTS OF PRESENTATION FORMAT AND SOURCE RATING ON FAKE NEWS IN SOCIAL MEDIA. MIS Quarterly, 43(3).

Kim, A., Moravec, P. L., & Dennis, A. R. (2019). Combating Fake News on Social Media with Source Ratings: The Effects of User and Expert Reputation Ratings. Journal of Management Information Systems, 36(3), 931-968.

Moravec, P., Minas, R., & Dennis, A. (2019). Fake News on Social Media: People Believe what They want to Believe when it Makes no Sense at All. MIS Quarterly, 43(4).