# Pakete laden ####
library(report) # Einfaches Erstellen von statistischen Berichten
library(marginaleffects) # Vorhersagen aus Regressionsmodellen
library(modelsummary) # Darstellung von Regressionsmodellen
library(parameters) # Darstellung von Regressionsmodellen
##
## Attaching package: 'parameters'
## The following object is masked from 'package:modelsummary':
##
## supported_models
library(performance) # Prüfen der Voraussetzungen
library(tidyverse) # Datenmanagement und Visualisierung: https://www.tidyverse.org/
# Datensatz lesen und relevante Variablen auswählen und umbenennen ####
d <- haven::read_stata(here::here("data/Vanerkel_Vanaelst_2021.dta")) |>
rename(
Political_knowledge = PK,
Personalized_news = personalized_news,
Radio = News_channels_w4_1,
Television = News_channels_w4_2,
Newspapers = News_channels_w4_3,
Online_news_sites = News_channels_w4_4,
Twitter = News_channels_w4_5,
Facebook = News_channels_w4_6
) |>
mutate(
Gender = as_factor(Gender),
Education = as_factor(Education),
trad = factor(trad, labels = c(
"traditional news diet: no",
"traditional news diet: yes"
))
)
# Modelle in Tabellen 5 & 6 ####
# PNE
# we presented respondents **who use social
# media** for news consumption three statements about the extent to which they encounter
# news on SNSs in the first wave: 1) On social media I often encounter news from media that
# I normally don’t use, 2) On social media I often encounter news that I am not interested in
# and 3) On social media I rarely encounter political news with which I disagree. For each
# statement they could indicate whether they agree or not on a five-point scale, running
# from fully disagree (0) to fully agree (4). [...]
# Based on the three items an additive scale of personalized news environment was created,
# in which all items were coded in the same direction so that a higher score indicates a more
# personalized SNSs environment
# IO
# We selected four
# items from their extensive information overload scale where citizens had to indicate on
# a five-point scale whether they fully disagreed (0) to fully agreed (4) (wave 16). These
# four statements are: 1) I regularly feel overwhelmed by too much information these
# days, 2) There is so much information available on topics of interest to me that
# I sometimes have difficulties to determine what is important and what is not, 3) I am
# being confronted by an avalanche of personal messages (via e-mail, text messages, chat,
# social media, etc.), 4) When I search for information on a topic of interest to me,
# I usually get too much rather than too little information. [...]
# Moreover, the additive information overload scale, which
# runs from 0 (no overload) to 16 (full overload),
# Übersicht
d |>
select(Personalized_news, Information_overload) |>
haven::zap_labels() |>
report_table()
## Variable | n_Obs | Mean | SD | Median | MAD | Min | Max
## ----------------------------------------------------------------------
## Personalized_news | 993 | 5.16 | 1.92 | 5 | 1.48 | 0 | 12
## Information_overload | 993 | 8.46 | 3.16 | 8 | 2.97 | 0 | 16
##
## Variable | Skewness | Kurtosis | percentage_Missing
## ---------------------------------------------------------------
## Personalized_news | -0.02 | 0.35 | 21.55
## Information_overload | -0.22 | 0.09 | 0.00
# Erklärung von Personalized News Environment (Table 5)
# av: Personalized_news
# Alle anderen Variablennamen finden Sie im Modell M4 oben
# Achten Sie auch darauf, welche Variablen NICHT im Modell sind
m_pne <- lm(Personalized_news ~ Radio + Television + Newspapers + Online_news_sites + Twitter +
Facebook + Gender + Age + Education, data = d)
# Erklärung von Information overload (Table 6)
# Alle Variablennamen finden Sie im Modell M4 in der Vorlesung
m_io <- lm(Information_overload ~ Radio + Television + Newspapers + Online_news_sites + Twitter +
Facebook + Gender + Age + Education + Political_interest, data = d)
# Betrachten Sie die Ergebnisse der Modelle indem Sie aus der Vorlesung entsprechende
# Code-Zeilen kopieren und das Modell austauschen.
# z.B. modelsummary(), report_table(), parameters() |> plot(), predictions()
# Erklärung von Personalized News Environment (Table 5)
m_pne |>
report_table(metrics = "R2_adj")
## Parameter | Coefficient | 95% CI | t(768) | p
## -------------------------------------------------------------------
## (Intercept) | 6.79 | [ 5.98, 7.59] | 16.56 | < .001
## Radio | 0.07 | [-0.01, 0.16] | 1.65 | 0.099
## Television | -0.07 | [-0.19, 0.04] | -1.24 | 0.215
## Newspapers | -0.08 | [-0.17, 0.01] | -1.73 | 0.084
## Online news sites | 0.02 | [-0.07, 0.11] | 0.45 | 0.655
## Twitter | 0.05 | [-0.08, 0.18] | 0.78 | 0.434
## Facebook | -0.22 | [-0.29, -0.14] | -5.48 | < .001
## Gender [female] | -0.01 | [-0.29, 0.26] | -0.09 | 0.926
## Age | -0.01 | [-0.02, 0.00] | -2.14 | 0.033
## Education [Middle] | -0.35 | [-0.76, 0.06] | -1.66 | 0.098
## Education [High] | -0.09 | [-0.51, 0.33] | -0.44 | 0.663
## | | | |
## R2 (adj.) | | | |
##
## Parameter | Std. Coef. | Std. Coef. 95% CI | Fit
## ----------------------------------------------------------
## (Intercept) | 0.10 | [-0.10, 0.29] |
## Radio | 0.07 | [-0.01, 0.14] |
## Television | -0.05 | [-0.13, 0.03] |
## Newspapers | -0.07 | [-0.15, 0.01] |
## Online news sites | 0.02 | [-0.06, 0.10] |
## Twitter | 0.03 | [-0.04, 0.10] |
## Facebook | -0.22 | [-0.30, -0.14] |
## Gender [female] | -6.82e-03 | [-0.15, 0.14] |
## Age | -0.08 | [-0.16, -0.01] |
## Education [Middle] | -0.18 | [-0.40, 0.03] |
## Education [High] | -0.05 | [-0.27, 0.17] |
## | | |
## R2 (adj.) | | | 0.05
m_pne |>
parameters() |>
plot()
m_pne |>
parameters(standardize = "refit") |>
plot()
# Facebook in den folgenden Darstellungen nur ein Beispiel. Gerne andere
# Variablen einfügen.
m_pne |>
plot_predictions(by = "Facebook",
newdata = datagrid(Facebook = 1:6,
grid_type = "mean_or_mode", # Werte der anderen Prädiktoren
FUN_integer = mean)) # Notwendig, da sonst für Integer Prädiktoren (keine Nachkommastellen) der Modus ausgewählt wird.
m_pne |>
predictions(newdata = datagrid(Facebook = 1:6,
grid_type = "mean_or_mode", # Werte der anderen Prädiktoren
FUN_integer = mean)) # Notwendig, da sonst für Integer Prädiktoren (keine Nachkommastellen) der Modus ausgewählt wird.
##
## Facebook Estimate Std. Error z Pr(>|z|) S 2.5 % 97.5 %
## 1 5.71 0.142 40.3 <0.001 Inf 5.43 5.98
## 2 5.49 0.124 44.4 <0.001 Inf 5.25 5.73
## 3 5.27 0.117 45.0 <0.001 Inf 5.04 5.50
## 4 5.06 0.123 41.0 <0.001 Inf 4.82 5.30
## 5 4.84 0.141 34.3 <0.001 855.4 4.56 5.12
## 6 4.62 0.166 27.8 <0.001 562.7 4.30 4.95
##
## Type: response
m_pne |>
avg_comparisons(variables = list(Facebook = "2sd", Newspapers = "2sd",
Online_news_sites = "2sd", Twitter = "2sd",
Gender = "reference",
Education = "pairwise"))
##
## Term Contrast Estimate Std. Error z Pr(>|z|)
## Education High - Lower -0.0931 0.213 -0.4362 0.6627
## Education High - Middle 0.2545 0.149 1.7122 0.0869
## Education Middle - Lower -0.3476 0.210 -1.6555 0.0978
## Facebook (x + sd) - (x - sd) -0.8510 0.155 -5.4768 <0.001
## Gender female - male -0.0131 0.140 -0.0935 0.9255
## Newspapers (x + sd) - (x - sd) -0.2705 0.156 -1.7298 0.0837
## Online_news_sites (x + sd) - (x - sd) 0.0696 0.156 0.4474 0.6546
## Twitter (x + sd) - (x - sd) 0.1130 0.144 0.7832 0.4335
## S 2.5 % 97.5 %
## 0.6 -0.5112 0.3251
## 3.5 -0.0368 0.5459
## 3.4 -0.7590 0.0639
## 24.5 -1.1555 -0.5464
## 0.1 -0.2873 0.2611
## 3.6 -0.5770 0.0360
## 0.6 -0.2353 0.3744
## 1.2 -0.1698 0.3958
##
## Type: response
# Erklärung von Information overload (Table 6)
m_io |>
report_table(metrics = "R2_adj")
## Parameter | Coefficient | 95% CI | t(981) | p | Std. Coef.
## -------------------------------------------------------------------------------
## (Intercept) | 6.05 | [ 4.80, 7.30] | 9.50 | < .001 | -0.03
## Radio | 0.04 | [-0.09, 0.16] | 0.58 | 0.565 | 0.02
## Television | 0.12 | [-0.06, 0.29] | 1.32 | 0.186 | 0.05
## Newspapers | -0.02 | [-0.15, 0.12] | -0.25 | 0.801 | -9.21e-03
## Online news sites | 0.03 | [-0.10, 0.16] | 0.43 | 0.665 | 0.02
## Twitter | 0.22 | [ 0.01, 0.44] | 2.06 | 0.040 | 0.07
## Facebook | 0.16 | [ 0.04, 0.27] | 2.67 | 0.008 | 0.10
## Gender [female] | 0.55 | [ 0.14, 0.96] | 2.62 | 0.009 | 0.17
## Age | 0.02 | [ 0.01, 0.04] | 2.79 | 0.005 | 0.10
## Education [Middle] | -0.19 | [-0.81, 0.43] | -0.61 | 0.540 | -0.06
## Education [High] | -0.19 | [-0.83, 0.44] | -0.60 | 0.547 | -0.06
## Political interest | -0.05 | [-0.14, 0.03] | -1.23 | 0.218 | -0.04
## | | | | |
## R2 (adj.) | | | | |
##
## Parameter | Std. Coef. 95% CI | Fit
## ---------------------------------------------
## (Intercept) | [-0.21, 0.15] |
## Radio | [-0.05, 0.09] |
## Television | [-0.02, 0.12] |
## Newspapers | [-0.08, 0.06] |
## Online news sites | [-0.06, 0.09] |
## Twitter | [ 0.00, 0.14] |
## Facebook | [ 0.03, 0.17] |
## Gender [female] | [ 0.04, 0.30] |
## Age | [ 0.03, 0.17] |
## Education [Middle] | [-0.26, 0.13] |
## Education [High] | [-0.26, 0.14] |
## Political interest | [-0.11, 0.03] |
## | |
## R2 (adj.) | | 0.03
m_io |>
parameters() |>
plot()
m_io |>
parameters(standardize = "refit") |>
plot()
# Facebook in den folgenden Darstellungen nur ein Beispiel. Gerne andere
# Variablen einfügen.
m_io |>
plot_predictions(by = "Facebook",
newdata = datagrid(Facebook = 1:6,
grid_type = "mean_or_mode", # Werte der anderen Prädiktoren
FUN_integer = mean)) # Notwendig, da sonst für Integer Prädiktoren (keine Nachkommastellen) der Modus ausgewählt wird.
m_io |>
predictions(newdata = datagrid(Facebook = 1:6,
grid_type = "mean_or_mode", # Werte der anderen Prädiktoren
FUN_integer = mean)) # Notwendig, da sonst für Integer Prädiktoren (keine Nachkommastellen) der Modus ausgewählt wird.
##
## Facebook Estimate Std. Error z Pr(>|z|) S 2.5 % 97.5 %
## 1 7.91 0.198 39.9 <0.001 Inf 7.52 8.29
## 2 8.06 0.185 43.5 <0.001 Inf 7.70 8.43
## 3 8.22 0.191 43.1 <0.001 Inf 7.85 8.60
## 4 8.38 0.213 39.3 <0.001 Inf 7.96 8.80
## 5 8.54 0.248 34.4 <0.001 861.4 8.05 9.02
## 6 8.70 0.291 29.9 <0.001 650.4 8.13 9.27
##
## Type: response
m_io |>
avg_comparisons(variables = list(Facebook = "2sd", Newspapers = "2sd",
Online_news_sites = "2sd", Twitter = "2sd",
Gender = "reference",
Education = "pairwise"))
##
## Term Contrast Estimate Std. Error z Pr(>|z|)
## Education High - Lower -0.19442 0.323 -0.60202 0.54716
## Education High - Middle -0.00147 0.219 -0.00671 0.99465
## Education Middle - Lower -0.19295 0.315 -0.61242 0.54026
## Facebook (x + sd) - (x - sd) 0.61588 0.231 2.66691 0.00766
## Gender female - male 0.54953 0.210 2.61601 0.00890
## Newspapers (x + sd) - (x - sd) -0.05820 0.231 -0.25201 0.80103
## Online_news_sites (x + sd) - (x - sd) 0.10041 0.232 0.43310 0.66494
## Twitter (x + sd) - (x - sd) 0.44362 0.216 2.05817 0.03957
## S 2.5 % 97.5 %
## 0.9 -0.8274 0.439
## 0.0 -0.4317 0.429
## 0.9 -0.8104 0.425
## 7.0 0.1633 1.069
## 6.8 0.1378 0.961
## 0.3 -0.5108 0.394
## 0.6 -0.3540 0.555
## 4.7 0.0212 0.866
##
## Type: response
# Bonus: Prüfen Sie die statistischen Annahmen beider Modelle
m_pne |>
check_model(check = c("linearity", "normality", "homogeneity", "outliers", "vif"))
m_pne |>
check_residuals() |>
plot()
m_pne |>
check_outliers() |>
plot()
correlation::correlation(m_pne$model) |>
summary(redundant = TRUE)
## # Correlation Matrix (pearson-method)
##
## Parameter | Personalized_news | Radio | Television | Newspapers
## -------------------------------------------------------------------------
## Personalized_news | | -0.03 | -0.12** | -0.10*
## Radio | -0.03 | | 0.40*** | 0.31***
## Television | -0.12** | 0.40*** | | 0.34***
## Newspapers | -0.10* | 0.31*** | 0.34*** |
## Online_news_sites | -0.06 | 0.23*** | 0.30*** | 0.35***
## Twitter | -0.02 | 0.10* | 0.05 | 0.12*
## Facebook | -0.20*** | 0.24*** | 0.25*** | 0.11*
## Age | -0.08 | 0.04 | 0.24*** | 0.22***
##
## Parameter | Online_news_sites | Twitter | Facebook | Age
## ---------------------------------------------------------------------
## Personalized_news | -0.06 | -0.02 | -0.20*** | -0.08
## Radio | 0.23*** | 0.10* | 0.24*** | 0.04
## Television | 0.30*** | 0.05 | 0.25*** | 0.24***
## Newspapers | 0.35*** | 0.12* | 0.11* | 0.22***
## Online_news_sites | | 0.26*** | 0.32*** | -0.06
## Twitter | 0.26*** | | 0.29*** | -0.12**
## Facebook | 0.32*** | 0.29*** | | -0.17***
## Age | -0.06 | -0.12** | -0.17*** |
##
## p-value adjustment method: Holm (1979)
m_io |>
check_model(check = c("linearity", "normality", "homogeneity", "outliers", "vif"))
m_io |>
check_residuals() |>
plot()
m_io |>
check_outliers() |>
plot()
correlation::correlation(m_io$model) |>
summary(redundant = TRUE)
## # Correlation Matrix (pearson-method)
##
## Parameter | Information_overload | Radio | Television | Newspapers
## -------------------------------------------------------------------------------
## Information_overload | | 0.05 | 0.08 | 0.02
## Radio | 0.05 | | 0.39*** | 0.28***
## Television | 0.08 | 0.39*** | | 0.31***
## Newspapers | 0.02 | 0.28*** | 0.31*** |
## Online_news_sites | 0.04 | 0.22*** | 0.28*** | 0.33***
## Twitter | 0.09 | 0.09 | 0.04 | 0.09
## Facebook | 0.12** | 0.18*** | 0.19*** | 0.06
## Age | 0.05 | 0.05 | 0.24*** | 0.21***
## Political_interest | -0.02 | 0.20*** | 0.30*** | 0.33***
##
## Parameter | Online_news_sites | Twitter | Facebook | Age | Political_interest
## ----------------------------------------------------------------------------------------------
## Information_overload | 0.04 | 0.09 | 0.12** | 0.05 | -0.02
## Radio | 0.22*** | 0.09 | 0.18*** | 0.05 | 0.20***
## Television | 0.28*** | 0.04 | 0.19*** | 0.24*** | 0.30***
## Newspapers | 0.33*** | 0.09 | 0.06 | 0.21*** | 0.33***
## Online_news_sites | | 0.23*** | 0.27*** | -0.08 | 0.32***
## Twitter | 0.23*** | | 0.33*** | -0.15*** | 0.05
## Facebook | 0.27*** | 0.33*** | | -0.25*** | 0.06
## Age | -0.08 | -0.15*** | -0.25*** | | 0.14***
## Political_interest | 0.32*** | 0.05 | 0.06 | 0.14*** |
##
## p-value adjustment method: Holm (1979)