library(tidyverse)
read_csv('https://wegweisr.haim.it/Daten/breaking_bad_deaths.csv') |>
count(method, sort = TRUE) |>
head(n = 5)Sommersemester 2025
| Sitzung | Datum | Thema |
|---|---|---|
| 1 | 23.04.2025 | Einführung |
| 2 | 30.04.2025 | GLM Grundlagen |
| 3 | 07.05.2025 | Lineare Regression |
| 4 | 21.05.2025 | Mittelwertvergleiche |
| 5 | 28.05.2025 | Multiple Regression |
| 6 | 04.06.2025 | Modellannahmen |
| Sitzung | Datum | Thema |
|---|---|---|
| 7 | 11.06.2025 | Modellvorhersagen |
| 8 | 18.06.2025 | Moderationsanalyse I |
| 9 | 25.06.2025 | Moderationsanalyse II |
| 10 | 02.07.2025 | Logistische Regression |
| 11 | 09.07.2025 | Multilevel-Regression |
| 12 | 16.07.2025 | Abschluss |
Was macht dieser Code?
Was macht dieser Code?
Field, A., Miles, J., & Field, Z. (2012). Discovering statistics using R. London: Sage.
Miles, J., & Shevlin, M. (2001). Applying regression and correlation: A guide for students and researchers. London: Sage.
Darlington, R. B., & Hayes, A. F. (2016). Regression analysis and linear models: Concepts, applications, and implementation. Guilford Publications.
McElreath, R. (2020). Statistical rethinking: A Bayesian course with examples in R and Stan. CRC press. (für Interessierte)
Interpretieren Sie die folgenden Analysen:
“Die Inferenzstatistik (d.h. schließende Statistik) beschäftigt sich mit der Frage, wie man aufgrund von Stichprobendaten auf Sachverhalte in einer zugrundeliegenden Population schließen kann.” (Eid et al., 2010, p. 191)



Die Mittelwerte der einzelnen Stichproben streuen um den wahren Populationsmittelwert von 170 = Standardfehler (SE).
SE = \(SD(x)/\sqrt(n-1)\), den wir anhand einer Stichprobe berechnen können, als Schätzer für die Streuung der Stichprobenmittelwerte.
SE auf Basis unserer ersten Stichprobe: SE = \(11/\sqrt(29)\) = 2.


Rot: Normalverteilungskurve mit Mittelwert und Standardfehler aus der ersten Stichprobe.

95%-Konfidenzintervall auf Basis unserer ersten Stichprobe (M und SE): 167.8 - 175.8
Je größer die Stichprobe (n), desto kleiner der Standardfehler (SE), d.h. desto enger das Konfidenzintervall. Es gilt aber immer, bei 95%-CI enthalten langfristig 5 von 100 Intervallen nicht den Populationswert.
p(Daten|H0)
p(Daten|H1): Die Wahrscheinlichkeit, die empirischen Daten zu beoachten, wenn die Alternativhypothese gilt.
p(H0|Daten): Die Wahrscheinlichkeit für die Richtigkeit der Nullhypothese im Lichte der Daten.
p(H1|Daten): Die Wahrscheinlichkeit für die Richtigkeit der Alternativhypothese im Licht der Daten.
Der p-Wert sagt also nichts über die Wahrscheinlichkeit der Null- oder Alternativhypothese!
außerdem:
Quelle: https://www.statisticssolutions.com



Inferenzstatistik ‚funktioniert’, weil…
Quelle: https://onishlab.colostate.edu/wp-content/uploads/2019/07/which_test_flowchart.png
In der klassischen Statistikausbildung (auch bei uns) als Rezeptesammlung:
Fokus auf Unterschieden und Spezifika statt auf Gemeinsamkeiten
Viele Verfahren sind aber mindestens funktional, oft auch mathematisch äquivalent!
There has been little attempt to understand the influence on children of branded products that appear in television programs and movies. A study exposed children of two different age groups (6–7 and 11–12) in classrooms to a brief film clip. Half of each class was shown a scene from Home Alone that shows Pepsi Cola being spilled during a meal. The other half was shown a similar clip from Home Alone but without branded products. All children were invited to help themselves from a choice of Pepsi or Coke at the outset of the individual interviews.
| id | pepsi_placement | pepsi_chosen |
|---|---|---|
| 49 | 1 | 0 |
| 54 | 1 | 0 |
| 19 | 1 | 1 |
| 6 | 1 | 1 |
| 52 | 1 | 0 |
| pepsi_chosen | no_placement | placement |
|---|---|---|
| 0 | 57 | 37 |
| 1 | 43 | 63 |
| Chi2(1) | p | Cramer’s V (adj.) | Cramers_v_adjusted CI |
|---|---|---|---|
| 4.14 | 0.042 | 0.17 | (0.00, 1.00) |
| Parameter1 | Parameter2 | r | 95% CI | p |
|---|---|---|---|---|
| pepsi_placement | pepsi_chosen | 0.20 | (0.01, 0.38) | 0.042 |
Alternative hypothesis: true correlation is not equal to 0
| Parameter1 | Parameter2 | tau | z | p |
|---|---|---|---|---|
| pepsi_placement | pepsi_chosen | 0.20 | 2.03 | 0.043 |
Alternative hypothesis: true tau is not equal to 0
| Difference | 95% CI | t(103) | p | d |
|---|---|---|---|---|
| -0.20 | (-0.39, -0.01) | -2.06 | 0.042 | -0.41 |
| Parameter | Sum_Squares | df | Mean_Square | F | p | Eta2 |
|---|---|---|---|---|---|---|
| pepsi_placement | 1.03 | 1 | 1.03 | 4.23 | 0.042 | 0.04 |
| Residuals | 25.10 | 103 | 0.24 |
“The only formula you’ll ever need.” Andy Field
\[ outcome_i = Model_i + error_i \]
Frage: Wenn wir nur einen Schätzwert \(a\) für \(Y\) haben, welcher ist der beste Schätzer?
\[ Y_i = a + \epsilon_i \]
Antwort: Mittelwert \(\bar{x}\) als der beste Modellkoeffizient im Nullmodell
Problem: damit erklärt das Modell aber nichts, es fehlt eine Prädiktorvariable \(X\)
\[ Y_i = b_0 + b_1 X_i + \epsilon_i \]
\[ Y_i = b_0 + b_1 X_1 + + b_2 X_2 + b_3 X_3 + ... + \epsilon_i \]
| Parameter | Coefficient | 95% CI | t(103) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 0.43 | (0.29, 0.57) | 6.24 | < .001 | 0.00 | |
| pepsi placement | 0.20 | (0.01, 0.39) | 2.06 | 0.042 | 0.20 | |
| AICc | 153.96 | |||||
| R2 | 0.04 | |||||
| R2 (adj.) | 0.03 | |||||
| Sigma | 0.49 |
Quelle: Scharkow, Festl, Vogelgesang & Quandt, 2013
Quelle: https://www.cjr.org/tow_center_reports/the_curious_journalists_guide_to_data.php
| tv_time | age | games_time | music_time |
|---|---|---|---|
| 0.0 | 22 | 0 | 4.00 |
| 0.0 | 43 | 0 | 2.50 |
| 2.0 | 38 | 0 | 0.17 |
| 5.0 | 30 | 0 | 2.00 |
| 1.5 | 29 | 1 | 0.75 |
| 2.0 | 57 | 0 | 0.00 |
| Variable | Summary |
|---|---|
| Mean tv_time (SD) | 2.73 (3.67) |
| Mean age (SD) | 46.95 (14.67) |
| Mean games_time (SD) | 0.93 (2.41) |
| Mean music_time (SD) | 2.14 (2.75) |
| Parameter1 | Parameter2 | r | 95% CI | p |
|---|---|---|---|---|
| age | music_time | -0.22 | (-0.26, -0.17) | < .001 |
Alternative hypothesis: true correlation is not equal to 0
| Parameter | Coefficient | 95% CI | t(2101) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 4.04 | (3.65, 4.42) | 20.53 | < .001 | 0.00 | |
| age | -0.04 | (-0.05, -0.03) | -10.14 | < .001 | -0.22 | |
| AICc | 10125.19 | |||||
| R2 | 0.05 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 2.68 |
| Parameter | Coefficient | 95% CI | t(2101) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 3.31 | (3.06, 3.57) | 25.49 | < .001 | 0.00 | |
| age18 | -0.04 | (-0.05, -0.03) | -10.14 | < .001 | -0.22 | |
| AICc | 10125.19 | |||||
| R2 | 0.05 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 2.68 |
| Parameter | Coefficient | 95% CI | t(2101) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 2.14 | (2.03, 2.25) | 36.56 | < .001 | 0.00 | |
| age centered | -0.04 | (-0.05, -0.03) | -10.14 | < .001 | -0.22 | |
| AICc | 10125.19 | |||||
| R2 | 0.05 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 2.68 |
| Parameter | Coefficient | 95% CI | t(2101) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 128.40 | (121.51, 135.28) | 36.56 | < .001 | 0.00 | |
| age centered | -2.43 | (-2.90, -1.96) | -10.14 | < .001 | -0.22 | |
| AICc | 27346.01 | |||||
| R2 | 0.05 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 161.06 |
| Parameter | Coefficient | 95% CI | t(2101) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 0.00 | (-0.04, 0.04) | 0.08 | 0.936 | 0.00 | |
| age zstd | -0.22 | (-0.26, -0.17) | -10.14 | < .001 | -0.22 | |
| AICc | 5872.65 | |||||
| R2 | 0.05 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 0.98 |
| Parameter | Coefficient | 95% CI | t(2101) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 3.91 | (3.39, 4.43) | 14.63 | < .001 | 0.00 | |
| age | -0.03 | (-0.04, -0.01) | -4.62 | < .001 | -0.10 | |
| AICc | 11414.93 | |||||
| R2 | 0.01 | |||||
| R2 (adj.) | 0.01 | |||||
| Sigma | 3.65 |
| Zugehörigkeit | Gruppe B | Gruppe C | Gruppe D |
|---|---|---|---|
| Gruppe A | 0 | 0 | 0 |
| Gruppe B | 1 | 0 | 0 |
| Gruppe C | 0 | 1 | 0 |
| Gruppe D | 0 | 0 | 1 |
| Zugehörigkeit | Gruppe A | Gruppe B | Gruppe C |
|---|---|---|---|
| Gruppe D | 0 | 0 | 0 |
| Gruppe A | 1 | 0 | 0 |
| Gruppe B | 0 | 1 | 0 |
| Gruppe C | 0 | 0 | 1 |
Coming across news on social network sites (SNS) largely depends on news-related activities in one’s network. Although there are many different ways to stumble upon news, limited research has been conducted on how distinct news curation practices influence users’ intention to consume encountered content. In this mixed-methods investigation, using Facebook as an example, we first examine the results of an experiment (study 1, n = 524), showing that getting tagged in comments to news posts promotes news consumption the most.
| modus | rw | modus_tag |
|---|---|---|
| Tag | 5 | 1 |
| Chronik | 2 | 0 |
| Post | 3 | 0 |
| DM | 1 | 0 |
| Chronik | 1 | 0 |
| Chronik | 2 | 0 |
| Variable | Summary |
|---|---|
| Mean rw (SD) | 3.04 (1.30) |
| modus | n | M | SD |
|---|---|---|---|
| Chronik | 141 | 2.88 | 1.20 |
| Post | 97 | 2.79 | 1.25 |
| Tag | 152 | 3.51 | 1.33 |
| DM | 134 | 2.84 | 1.28 |
| Difference | 95% CI | t(522) | p | d |
|---|---|---|---|---|
| -0.67 | (-0.91, -0.43) | -5.51 | < .001 | -0.48 |
| Parameter | Coefficient | 95% CI | t(522) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 2.84 | (2.71, 2.97) | 43.26 | < .001 | 0.00 | |
| modus tag | 0.67 | (0.43, 0.91) | 5.51 | < .001 | 0.23 | |
| AICc | 1738.89 | |||||
| R2 | 0.05 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 1.27 |
| Parameter | Sum_Squares | df | Mean_Square | F | p | Eta2 |
|---|---|---|---|---|---|---|
| modus | 49.12 | 3 | 16.37 | 10.17 | < .001 | 0.06 |
| Residuals | 837.19 | 520 | 1.61 |
| Parameter | Coefficient | 95% CI | t(520) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 2.88 | (2.67, 3.09) | 26.95 | < .001 | -0.12 | |
| modus (Post) | -0.09 | (-0.41, 0.24) | -0.51 | 0.609 | -0.07 | |
| modus (Tag) | 0.63 | (0.34, 0.93) | 4.27 | < .001 | 0.49 | |
| modus (DM) | -0.04 | (-0.34, 0.26) | -0.28 | 0.776 | -0.03 | |
| AICc | 1742.69 | |||||
| R2 | 0.06 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 1.27 |
| Parameter | Coefficient | 95% CI | t(520) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 2.84 | (2.62, 3.05) | 25.87 | < .001 | -0.15 | |
| modus dm (Chronik) | 0.04 | (-0.26, 0.34) | 0.28 | 0.776 | 0.03 | |
| modus dm (Post) | -0.04 | (-0.37, 0.29) | -0.25 | 0.804 | -0.03 | |
| modus dm (Tag) | 0.68 | (0.38, 0.97) | 4.50 | < .001 | 0.52 | |
| AICc | 1742.69 | |||||
| R2 | 0.06 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 1.27 |
| term | contrast | estimate | std.error | statistic | p.value | conf.low | conf.high | p_adjusted |
|---|---|---|---|---|---|---|---|---|
| modus | DM - Chronik | -0.04 | 0.15 | -0.28 | 0.78 | -0.34 | 0.26 | 1 |
| modus | DM - Post | 0.04 | 0.17 | 0.25 | 0.80 | -0.29 | 0.37 | 1 |
| modus | DM - Tag | -0.68 | 0.15 | -4.50 | 0.00 | -0.97 | -0.38 | 0 |
| modus | Post - Chronik | -0.09 | 0.17 | -0.51 | 0.61 | -0.41 | 0.24 | 1 |
| modus | Tag - Chronik | 0.63 | 0.15 | 4.27 | 0.00 | 0.34 | 0.92 | 0 |
| modus | Tag - Post | 0.72 | 0.16 | 4.36 | 0.00 | 0.40 | 1.04 | 0 |
| modus | estimate | std.error | statistic | p.value | conf.low | conf.high | df |
|---|---|---|---|---|---|---|---|
| Chronik | 2.88 | 0.11 | 26.95 | 0 | 2.67 | 3.09 | Inf |
| Post | 2.79 | 0.13 | 21.69 | 0 | 2.54 | 3.05 | Inf |
| Tag | 3.51 | 0.10 | 34.14 | 0 | 3.31 | 3.71 | Inf |
| DM | 2.84 | 0.11 | 25.87 | 0 | 2.62 | 3.05 | Inf |
Bender, R., & Lange, S. (2001). Adjusting for multiple testing—when and how?. Journal of clinical epidemiology, 54(4), 343-349.
Davis, M. J. (2010). Contrast coding in multiple regression analysis: Strengths, weaknesses, and utility of popular coding structures. Journal of data science, 8(1), 61-73.
Kümpel, A. S. (2019). Getting tagged, getting involved with news? A mixed-methods investigation of the effects and motives of news-related tagging activities on social network sites. Journal of Communication, 69(4), 373-395.
Wir vergleichen die Tanzbarkeit (danceability) und musikalische Stimmung (valence) der Top 10-Hits über 4 Dekaden (1990er bis 2020er) auf Basis von Billboard und Spotify-Daten.
Beide Variablen sind von 0 (niedrig) - 100 (hoch) skaliert. Die Mittelwerte und Fallzahlen pro Dekade sind wie folgt:
| decade | danceability | valence | n |
|---|---|---|---|
| 1990s | 64.72 | 56.09 | 588 |
| 2000s | 67.34 | 57.98 | 558 |
| 2010s | 67.31 | 51.93 | 499 |
| 2020s | 66.10 | 51.38 | 69 |
Interpretieren sie die Ergebnisse der beiden linearen Modelle, in denen die Mittelwertunterschiede getestet werden, Zeile für Zeile.
Welche Dekaden werden nicht miteinander verglichen, d.h. für diese bräuchten wir Post-Hoc Vergleiche?
Lösung bitte bis 04.06.2025, 10 Uhr in Moodle eintragen.
| danceability | |||
| Predictors | Coefficient (B) | SE (B) | p |
| (Intercept) | 64.72 | 0.59 | <0.001 |
| decade [2000s] | 2.62 | 0.85 | 0.002 |
| decade [2010s] | 2.59 | 0.88 | 0.003 |
| decade [2020s] | 1.38 | 1.83 | 0.452 |
| valence | ||
| Predictors | Coefficient (B) | 95% CI (B) |
| (Intercept) | 51.38 | 45.97 – 56.79 |
| decade [1990s] | 4.71 | -1.01 – 10.43 |
| decade [2000s] | 6.60 | 0.87 – 12.34 |
| decade [2010s] | 0.56 | -5.22 – 6.33 |
| Parameter | Coefficient | 95% CI | t(520) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 3.51 | (3.31, 3.72) | 34.14 | < .001 | 0.37 | |
| modus (Post) | -0.72 | (-1.04, -0.40) | -4.36 | < .001 | -0.55 | |
| modus (Chronik) | -0.63 | (-0.93, -0.34) | -4.27 | < .001 | -0.49 | |
| modus (DM) | -0.68 | (-0.97, -0.38) | -4.50 | < .001 | -0.52 | |
| AICc | 1742.69 | |||||
| R2 | 0.06 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 1.27 |
Does exposure to news affect what people know about politics? This old question attracted new scholarly interest as the political informa- tion environment is changing rapidly. In particular, since citizens have new channels at their disposal, such as Twitter and Facebook, which increasingly complement or even replace traditional channels of information. This study investigates to what extent citizens have knowledge about daily politics and to what extent news on social media can provide this knowledge. It does so by means of a large online survey in Belgium (Flanders), in which we measured what people know about current political events, their so-called general surveillance knowledge. Our findings demonstrate that unlike following news via traditional media channels, citizens do not gain more political knowledge from following news on social media. We even find a negative association between following the news on Facebook and political knowledge.
| Age | Gender | Education | TV | Newspaper | Websites | PK | |
|---|---|---|---|---|---|---|---|
| 21 | female | High | 3 | 3 | 5 | 6 | 1 |
| 21 | male | Middle | 4 | 5 | 5 | 5 | 3 |
| 67 | male | High | 5 | 4 | 5 | 4 | 3 |
| 63 | female | Middle | 3 | 1 | 1 | 1 | 2 |
| 61 | male | High | 5 | 5 | 5 | 5 | 5 |
| Variable | Summary |
|---|---|
| Mean Age (SD) | 52.98 (13.96) |
| Gender [female], % | 47.7 |
| Education [Lower], % | 13.7 |
| Education [Middle], % | 40.7 |
| Education [High], % | 45.6 |
| Mean TV (SD) | 4.43 (1.33) |
| Mean Newspaper (SD) | 3.52 (1.69) |
| Mean Websites (SD) | 3.44 (1.72) |
| Mean Facebook (SD) | 2.69 (1.95) |
| Mean PK (SD) | 3.04 (1.36) |
| Parameter | Coefficient | 95% CI | t(988) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 1.35 | (0.96, 1.74) | 6.81 | < .001 | -0.20 | |
| Gender (female) | -0.73 | (-0.89, -0.58) | -9.23 | < .001 | -0.54 | |
| Age | 0.03 | (0.02, 0.03) | 9.46 | < .001 | 0.28 | |
| Education (Middle) | 0.51 | (0.27, 0.75) | 4.23 | < .001 | 0.38 | |
| Education (High) | 0.89 | (0.66, 1.13) | 7.43 | < .001 | 0.66 | |
| AICc | 3217.50 | |||||
| R2 | 0.20 | |||||
| R2 (adj.) | 0.20 | |||||
| Sigma | 1.22 |
| Parameter | Coefficient | 95% CI | t(983) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 0.74 | (0.28, 1.20) | 3.18 | 0.002 | -0.12 | |
| Gender (female) | -0.63 | (-0.78, -0.48) | -8.21 | < .001 | -0.46 | |
| Age | 0.02 | (0.01, 0.02) | 6.23 | < .001 | 0.19 | |
| Education (Middle) | 0.39 | (0.16, 0.61) | 3.33 | < .001 | 0.28 | |
| Education (High) | 0.67 | (0.44, 0.90) | 5.69 | < .001 | 0.49 | |
| TV | 0.14 | (0.08, 0.20) | 4.46 | < .001 | 0.14 | |
| Newspaper | 0.12 | (0.07, 0.17) | 4.91 | < .001 | 0.15 | |
| Websites | 0.12 | (0.07, 0.16) | 4.77 | < .001 | 0.15 | |
| -0.07 | (-0.11, -0.03) | -3.29 | 0.001 | -0.10 | ||
| -0.07 | (-0.15, 0.01) | -1.81 | 0.070 | -0.05 | ||
| AICc | 3114.32 | |||||
| R2 | 0.29 | |||||
| R2 (adj.) | 0.28 | |||||
| Sigma | 1.15 |
| Modell | Res.Df | RSS | Df | Sum of Sq | F | Pr(>F) |
|---|---|---|---|---|---|---|
| 1 | 988 | 1466.83 | NA | NA | NA | NA |
| 2 | 983 | 1308.59 | 5 | 158.24 | 23.77 | 0 |


| Parameter | Political_interest | Age |
|---|---|---|
| PK | 0.49 | 0.3 |
| Age | 0.14 | NA |
$VIF

Quelle: https://catalogofbias.org
Nicht-Berücksichtigung einer relevanten Kovariate, die \(X\) und \(Y\) beeinflusst, verzerrt den geschätzen Zusammenhang zwischen \(X\) und \(Y\).
Berücksichtigung einer Kovariate, die von \(X\) und \(Y\) beeinflusst wird, verzerrt den geschätzen Zusammenhang zwischen \(X\) und \(Y\).
Quelle: https://www.andrewheiss.com/blog/2020/02/25/closing-backdoors-dags/
| Parameter | Coefficient | 95% CI | t(987) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 0.46 | (0.09, 0.83) | 2.45 | 0.014 | -0.13 | |
| Gender (female) | -0.51 | (-0.65, -0.37) | -6.96 | < .001 | -0.37 | |
| Age | 0.02 | (0.02, 0.03) | 8.48 | < .001 | 0.23 | |
| Education (Middle) | 0.35 | (0.13, 0.56) | 3.15 | 0.002 | 0.26 | |
| Education (High) | 0.60 | (0.38, 0.82) | 5.42 | < .001 | 0.44 | |
| Political interest | 0.20 | (0.18, 0.23) | 14.76 | < .001 | 0.40 | |
| AICc | 3021.49 | |||||
| R2 | 0.35 | |||||
| R2 (adj.) | 0.34 | |||||
| Sigma | 1.10 |
| Gender | Age | Education | Political_interest | PK | Predicted_PK |
|---|---|---|---|---|---|
| female | 24 | High | 5 | 4 | 2.11 |
| male | 22 | High | 6 | 0 | 2.78 |
| female | 61 | High | 7 | 4 | 3.33 |
| male | 61 | Middle | 1 | 3 | 2.36 |
| female | 50 | Middle | 6 | 3 | 2.63 |
| Gender | Age | Education | Political_interest | PK | fit | lwr | upr |
|---|---|---|---|---|---|---|---|
| female | 45 | Middle | 3 | 2 | 1.91 | 1.76 | 2.05 |
| female | 59 | High | 7 | 4 | 3.29 | 3.15 | 3.42 |
| female | 52 | High | 7 | 4 | 3.13 | 3.01 | 3.26 |
| Gender | Age | Education | Political_interest | PK | fit | lwr | upr |
|---|---|---|---|---|---|---|---|
| female | 45 | Middle | 3 | 2 | 1.91 | -0.26 | 4.08 |
| female | 59 | High | 7 | 4 | 3.29 | 1.12 | 5.46 |
| female | 52 | High | 7 | 4 | 3.13 | 0.96 | 5.30 |
| id | Age | Gender | Political_interest | PK | Predicted_PK |
|---|---|---|---|---|---|
| 1 | 45 | female | 3 | 2 | 1.91 |
| 1 | 45 | male | 3 | 2 | 2.42 |
| 2 | 59 | female | 7 | 4 | 3.29 |
| 2 | 59 | male | 7 | 4 | 3.80 |
| 3 | 52 | female | 7 | 4 | 3.13 |
| 3 | 52 | male | 7 | 4 | 3.64 |
| Gender | estimate | std.error | conf.low | conf.high | df |
|---|---|---|---|---|---|
| male | 3.29 | 0.05 | 3.19 | 3.38 | Inf |
| female | 2.78 | 0.05 | 2.68 | 2.88 | Inf |
| id | Age | Gender | Political_interest | PK | Predicted_PK |
|---|---|---|---|---|---|
| 1 | 18 | female | 3 | 2 | 1.31 |
| 1 | 40 | female | 3 | 2 | 1.80 |
| 1 | 65 | female | 3 | 2 | 2.35 |
| 2 | 18 | female | 7 | 4 | 2.38 |
| 2 | 40 | female | 7 | 4 | 2.87 |
| 2 | 65 | female | 7 | 4 | 3.42 |
| id | Age | Gender | Political_interest | PK | Predicted_PK |
|---|---|---|---|---|---|
| 1 | 19 | female | 3 | 2 | 1.33 |
| 1 | 44 | female | 3 | 2 | 1.89 |
| 1 | 56 | female | 3 | 2 | 2.15 |
| 1 | 65 | female | 3 | 2 | 2.35 |
| 1 | 71 | female | 3 | 2 | 2.48 |
| 2 | 19 | female | 7 | 4 | 2.40 |
| 2 | 44 | female | 7 | 4 | 2.96 |
| 2 | 56 | female | 7 | 4 | 3.22 |
| Age | estimate | std.error | conf.low | conf.high | df |
|---|---|---|---|---|---|
| 19 | 2.29 | 0.10 | 2.11 | 2.48 | Inf |
| 44 | 2.85 | 0.04 | 2.76 | 2.93 | Inf |
| 56 | 3.11 | 0.04 | 3.04 | 3.18 | Inf |
| 65 | 3.31 | 0.05 | 3.22 | 3.40 | Inf |
| 71 | 3.44 | 0.06 | 3.33 | 3.56 | Inf |
| Age | Gender | estimate | std.error | conf.low | conf.high | df |
|---|---|---|---|---|---|---|
| 19 | male | 2.54 | 0.11 | 2.33 | 2.75 | Inf |
| 19 | female | 2.03 | 0.10 | 1.84 | 2.22 | Inf |
| 44 | male | 3.09 | 0.06 | 2.98 | 3.20 | Inf |
| 44 | female | 2.58 | 0.05 | 2.47 | 2.69 | Inf |
| 56 | male | 3.35 | 0.05 | 3.26 | 3.45 | Inf |
| 56 | female | 2.84 | 0.05 | 2.74 | 2.95 | Inf |
| 65 | male | 3.55 | 0.06 | 3.44 | 3.66 | Inf |
| 65 | female | 3.04 | 0.06 | 2.92 | 3.17 | Inf |
| 71 | male | 3.69 | 0.06 | 3.56 | 3.81 | Inf |
| 71 | female | 3.18 | 0.07 | 3.03 | 3.32 | Inf |
Replizieren sie eine Regressionsanalyse aus van Erkel & van Aelst (2021) mit R oder SPSS oder anderer Software
Studierende mit gerader Matrikelnummer: Tabelle 5
Studierende mit ungerader Matrikelnummer: Tabelle 6
Der Datensatz ist in data/VanErkel_vanAelst2021.sav und enthält alle nötigen Variablen.
Machen sie einen Screenshot der Regressionstabelle als PNG oder JPG und laden Sie diesen in Moodle hoch.
Deadline: 25.06.2025, 10h
| Parameter | Coefficient | 95% CI | t(983) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 6.93 | (5.89, 7.97) | 13.09 | < .001 | 0.03 | |
| Gender (female) | 0.72 | (0.32, 1.12) | 3.51 | < .001 | 0.23 | |
| Age | 0.02 | (0.00, 0.03) | 2.06 | 0.039 | 0.07 | |
| Education (Middle) | -0.40 | (-1.01, 0.21) | -1.29 | 0.198 | -0.13 | |
| Education (High) | -0.62 | (-1.23, -0.01) | -1.98 | 0.048 | -0.20 | |
| Outlets Used | 0.10 | (0.06, 0.13) | 4.97 | < .001 | 0.16 | |
| AICc | 5062.21 | |||||
| R2 | 0.04 | |||||
| R2 (adj.) | 0.03 | |||||
| Sigma | 3.11 |
iv
| Overload (female) | Overload (male) | |||||
| Predictors | Estimates | CI | p | Estimates | CI | p |
| (Intercept) | 7.79 | 6.47 – 9.11 | <0.001 | 7.97 | 6.53 – 9.40 | <0.001 |
| Age | 0.02 | -0.00 – 0.04 | 0.090 | 0.01 | -0.01 – 0.04 | 0.186 |
| Education: Middle | 0.08 | -0.83 – 1.00 | 0.858 | -0.59 | -1.43 – 0.25 | 0.171 |
| Education: High | 0.12 | -0.79 – 1.02 | 0.799 | -0.72 | -1.56 – 0.11 | 0.089 |
| Observations | 474 | 519 | ||||
| R2 / R2 adjusted | 0.006 / -0.000 | 0.009 / 0.004 | ||||
Bei der Moderationsanalyse gehen wir davon aus, dass der Effekt X auf Y eine Funktion von Z ist
\(Y = b_0 + f(Z)X + b_2Z + \epsilon\)
Die Funktion f(Z) sei definiert als lineare Funktion \(f(Z) = b_1 + b_3Z\)
\(Y = b_0 + (b_1 + b_3Z)X + b_2Z + \epsilon\)
Durch Ausmultiplizieren erhalten wir einen Interaktionsterm \(XZ\), der einfach das Produkt von \(X\) und \(Z\) ist
\(Y = b_0 + b_1X + b_2Z + b_3XZ + \epsilon\)
Wichtig: Bei Regressionsmodellen mit Interaktionseffekten \(XZ\) sind die Koeffizienten von \(X\) und \(Z\) nicht mehr unabhängig voneinander interpretierbar, d.h. die Effekte sind nicht mehr unkonditional für alle Fälle \(n\) gültig!
Wichtig: Damit die konditionalen Effekte überhaupt interpretierbar sind, sollten wir metrische Variablen zentrieren und kategorielle Variablen dummy- oder effektcodieren!
lm() aufnehmen: y ~ x + z + x:z oder einfacher y ~ x * z| schwab | geschlecht_w | atol | gesamt |
|---|---|---|---|
| 1 | 1 | 4.8 | 4 |
| 1 | 1 | 3.4 | 3 |
| 1 | 0 | 2.2 | 4 |
| 1 | 1 | 4.2 | 5 |
| 0 | 1 | 3.6 | 5 |
| Parameter | Coefficient | 95% CI | t(361) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 3.81 | (3.67, 3.94) | 55.54 | < .001 | 0.00 | |
| schwab | -0.20 | (-0.39, -0.02) | -2.22 | 0.027 | -0.12 | |
| AICc | 937.93 | |||||
| R2 | 0.01 | |||||
| R2 (adj.) | 0.01 | |||||
| Sigma | 0.88 |
| Parameter | Coefficient | 95% CI | t(360) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 3.63 | (3.46, 3.79) | 43.52 | < .001 | 0.00 | |
| schwab | -0.21 | (-0.39, -0.03) | -2.31 | 0.021 | -0.12 | |
| geschlecht w | 0.34 | (0.16, 0.52) | 3.75 | < .001 | 0.19 | |
| AICc | 926.08 | |||||
| R2 | 0.05 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 0.86 |
| Parameter | Coefficient | 95% CI | t(359) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 3.71 | (3.51, 3.90) | 37.37 | < .001 | 0.00 | |
| schwab | -0.36 | (-0.62, -0.09) | -2.66 | 0.008 | -0.12 | |
| geschlecht w | 0.19 | (-0.07, 0.46) | 1.42 | 0.158 | 0.19 | |
| schwab × geschlecht w | 0.27 | (-0.09, 0.63) | 1.49 | 0.137 | 0.08 | |
| AICc | 925.89 | |||||
| R2 | 0.06 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 0.86 |
Durch Einsetzen in die Gleichung \(f(Z) = b_1 + b_3Z\) lassen sich die konditionalen Effekte von \(X\) (Dialekt) bei verschiedenen Ausprägungen vom Moderator \(Z\) (Geschlecht) schätzen:
| term | geschlecht_w | estimate | p.value | conf.low | conf.high |
|---|---|---|---|---|---|
| schwab | 0 | -0.36 | 0.01 | -0.62 | -0.09 |
| schwab | 1 | -0.09 | 0.48 | -0.33 | 0.15 |
Die durchschnittlichen Effekte von \(X\) über die gesamte Stichprobe (average marginal effects) lassen sich bestimmen, indem für jeden einzelnen Fall der konditionale Effekt berechnet und dann gemittelt wird.
| term | estimate | p.value | conf.low | conf.high |
|---|---|---|---|---|
| schwab | -0.21 | 0.02 | -0.39 | -0.03 |
Average Marginal Effects (AME) entsprechen im linearen Modell den unmoderierten Koeffizienten
| term | estimate | std.error | statistic | p.value |
|---|---|---|---|---|
| schwab | -0.21 | 0.09 | -2.31 | 0.02 |
| geschlecht_w | 0.34 | 0.09 | 3.75 | 0.00 |
| schwab | geschlecht_w | estimate | std.error | conf.low | conf.high | df |
|---|---|---|---|---|---|---|
| 0 | 0 | 3.71 | 0.10 | 3.51 | 3.90 | Inf |
| 0 | 1 | 3.90 | 0.09 | 3.72 | 4.08 | Inf |
| 1 | 0 | 3.35 | 0.09 | 3.17 | 3.53 | Inf |
| 1 | 1 | 3.81 | 0.08 | 3.65 | 3.97 | Inf |
| Parameter | Coefficient | 95% CI | t(987) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 2.82 | (2.55, 3.09) | 20.37 | < .001 | -0.16 | |
| Gender (female) | -0.63 | (-1.07, -0.19) | -2.81 | 0.005 | -0.46 | |
| Education (Middle) | 0.50 | (0.18, 0.82) | 3.04 | 0.002 | 0.37 | |
| Education (High) | 0.98 | (0.66, 1.30) | 6.00 | < .001 | 0.72 | |
| Gender (female) × Education (Middle) | -0.11 | (-0.61, 0.40) | -0.42 | 0.672 | -0.08 | |
| Gender (female) × Education (High) | -0.45 | (-0.94, 0.05) | -1.75 | 0.080 | -0.33 | |
| AICc | 3300.42 | |||||
| R2 | 0.14 | |||||
| R2 (adj.) | 0.13 | |||||
| Sigma | 1.27 |
Quelle: Rains et al. (2023)
| Parameter | Coefficient | 95% CI | t(987) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 8.81 | (8.13, 9.49) | 25.59 | < .001 | 0.11 | |
| Gender (female) | -0.12 | (-1.21, 0.98) | -0.21 | 0.833 | -0.04 | |
| Education (Middle) | -0.60 | (-1.40, 0.20) | -1.47 | 0.141 | -0.19 | |
| Education (High) | -0.75 | (-1.54, 0.05) | -1.85 | 0.065 | -0.24 | |
| Gender (female) × Education (Middle) | 0.63 | (-0.63, 1.88) | 0.98 | 0.326 | 0.20 | |
| Gender (female) × Education (High) | 0.74 | (-0.50, 1.98) | 1.17 | 0.242 | 0.23 | |
| AICc | 5108.02 | |||||
| R2 | 0.01 | |||||
| R2 (adj.) | 0.00 | |||||
| Sigma | 3.16 |
| Parameter | Coefficient | 95% CI | t(360) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 3.44 | (3.02, 3.85) | 16.38 | < .001 | 0.00 | |
| schwab | -0.20 | (-0.38, -0.02) | -2.15 | 0.032 | -0.11 | |
| atol | 0.11 | (0.00, 0.23) | 1.88 | 0.060 | 0.10 | |
| AICc | 936.42 | |||||
| R2 | 0.02 | |||||
| R2 (adj.) | 0.02 | |||||
| Sigma | 0.87 |
| Parameter | Coefficient | 95% CI | t(359) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 4.19 | (3.60, 4.79) | 13.79 | < .001 | 0.01 | |
| schwab | -1.53 | (-2.32, -0.74) | -3.80 | < .001 | -0.11 | |
| atol | -0.12 | (-0.29, 0.06) | -1.30 | 0.196 | 0.09 | |
| schwab × atol | 0.40 | (0.17, 0.64) | 3.40 | < .001 | 0.18 | |
| AICc | 926.98 | |||||
| R2 | 0.05 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 0.86 |
| Parameter | Coefficient | 95% CI | t(359) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 4.08 | (3.65, 4.51) | 18.69 | < .001 | 0.10 | |
| schwab | -1.13 | (-1.69, -0.56) | -3.91 | < .001 | 0.06 | |
| atol - 1 | -0.12 | (-0.29, 0.06) | -1.30 | 0.196 | 0.09 | |
| schwab × atol - 1 | 0.40 | (0.17, 0.64) | 3.40 | < .001 | 0.18 | |
| AICc | 926.98 | |||||
| R2 | 0.05 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 0.86 |
| Parameter | Coefficient | 95% CI | t(359) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 3.81 | (3.68, 3.95) | 56.55 | < .001 | 0.01 | |
| schwab | -0.20 | (-0.38, -0.02) | -2.21 | 0.028 | -0.11 | |
| atol - mean(atol, na rm = T) | -0.12 | (-0.29, 0.06) | -1.30 | 0.196 | 0.09 | |
| schwab × atol - mean(atol, na rm = T) | 0.40 | (0.17, 0.64) | 3.40 | < .001 | 0.18 | |
| AICc | 926.98 | |||||
| R2 | 0.05 | |||||
| R2 (adj.) | 0.05 | |||||
| Sigma | 0.86 |
| term | atol | estimate | p.value | conf.low | conf.high |
|---|---|---|---|---|---|
| schwab | 1 | -1.13 | 0.00 | -1.69 | -0.56 |
| schwab | 2 | -0.72 | 0.00 | -1.07 | -0.37 |
| schwab | 3 | -0.32 | 0.00 | -0.51 | -0.13 |
| schwab | 4 | 0.08 | 0.50 | -0.16 | 0.32 |
| schwab | 5 | 0.48 | 0.03 | 0.05 | 0.92 |
JOHNSON-NEYMAN INTERVAL
When atol is OUTSIDE the interval [3.35, 4.74], the slope of schwab is p < .05.
Note: The range of observed values of atol is [1.00, 5.00]
| schwab | atol | estimate | std.error | conf.low | conf.high | df |
|---|---|---|---|---|---|---|
| 0 | 1 | 4.08 | 0.22 | 3.65 | 4.51 | Inf |
| 0 | 3 | 3.85 | 0.07 | 3.70 | 3.99 | Inf |
| 0 | 5 | 3.62 | 0.16 | 3.30 | 3.94 | Inf |
| 1 | 1 | 2.95 | 0.19 | 2.59 | 3.32 | Inf |
| 1 | 3 | 3.53 | 0.06 | 3.40 | 3.65 | Inf |
| 1 | 5 | 4.10 | 0.15 | 3.81 | 4.39 | Inf |
Altay et al. (2022) untersuchen experimentell, ob falsche Nachrichten weniger als wahre geteilt werden, und wie dies von der Interessantheit der Nachricht abhängt.
Beantworten Sie folgende Forschungsfragen mit Hilfe der Studien-Daten:
| Share | Type | Interesting |
|---|---|---|
| 2 | TN | 3 |
| 1 | TN | 6 |
| 1 | TN | 4 |
Share = Wahrscheinlichkeit, die Nachricht mit anderen zu teilen (1-6, höher ist wahrscheinlicher)Type = Nachrichtentyp (experimentell variiert, TN=wahr, FN=falsch)Interesting = Wahrgenommene Interessantheit der Nachricht (1-7, höher ist interessanter)

\(Logit(Y) = b_0 + b_1 X_i + \epsilon_i\)
plogis()) in eine Baseline-Wahrscheinlichkeit umrechnen| cbbully_gen | cbvictim_gen | age | gender | internetuse | class |
|---|---|---|---|---|---|
| 0 | 0 | 14 | male | 2.5 | G8b |
| 0 | 0 | 13 | male | 1.0 | G7b |
| 0 | 0 | 18 | male | 1.0 | G12 |
| 0 | 0 | 14 | female | 0.5 | G8b |
| 0 | 0 | 13 | male | 1.5 | G7a |
| Variable | n_Obs | Mean | SD | Median | MAD | Min | Max |
|---|---|---|---|---|---|---|---|
| cbvictim_gen | 275 | 0.11 | 0.31 | 0 | 0 | 0 | 1 |
| Parameter | Coefficient | 95% CI | z | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | -2.14 | (-2.52, -1.75) | -10.89 | < .001 | -2.14 | |
| AICc | 187.31 | |||||
| Tjur’s R2 | 0.00 | |||||
| Sigma | 1.00 | |||||
| Log_loss | 0.34 |
| Parameter | Coefficient | 95% CI | z | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | -2.48 | (-3.13, -1.94) | -8.27 | < .001 | -2.48 | |
| gender (female) | 0.69 | (-0.08, 1.50) | 1.74 | 0.082 | 0.69 | |
| AICc | 186.26 | |||||
| Tjur’s R2 | 0.01 | |||||
| Sigma | 1.00 | |||||
| Log_loss | 0.33 |
| Parameter | Coefficient | CI | CI_low | CI_high | z | p | Fit |
|---|---|---|---|---|---|---|---|
| (Intercept) | 0.08 | 0.95 | 0.04 | 0.14 | -8.27 | 0.00 | NA |
| gender [female] | 2.00 | 0.95 | 0.92 | 4.46 | 1.74 | 0.08 | NA |
| term | estimate | std.error | statistic | p.value | conf.low | conf.high |
|---|---|---|---|---|---|---|
| gender | 0.07 | 0.04 | 1.71 | 0.09 | -0.01 | 0.14 |
| gender | estimate | std.error | statistic | p.value | s.value | conf.low | conf.high | df |
|---|---|---|---|---|---|---|---|---|
| male | 0.08 | 0.02 | 3.61 | 0 | 11.65 | 0.04 | 0.12 | Inf |
| female | 0.14 | 0.03 | 4.45 | 0 | 16.85 | 0.08 | 0.21 | Inf |
| Parameter | Coefficient | 95% CI | z | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | -2.98 | (-3.82, -2.23) | -7.41 | < .001 | -2.56 | |
| gender (female) | 0.76 | (-0.02, 1.58) | 1.88 | 0.060 | 0.76 | |
| internetuse | 0.19 | (0.00, 0.37) | 2.10 | 0.036 | 0.34 | |
| AICc | 184.36 | |||||
| Tjur’s R2 | 0.03 | |||||
| Sigma | 1.00 | |||||
| Log_loss | 0.32 |
| term | estimate | std.error | statistic | p.value | conf.low | conf.high |
|---|---|---|---|---|---|---|
| gender | 0.07 | 0.04 | 1.85 | 0.06 | 0 | 0.15 |
| internetuse | 0.02 | 0.01 | 1.98 | 0.05 | 0 | 0.04 |
| Parameter | Coefficient | 95% CI | t(2987) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 0.36 | (0.21, 0.52) | 4.57 | < .001 | -0.11 | |
| Type (TN) | 0.35 | (0.25, 0.46) | 6.89 | < .001 | 0.23 | |
| Interesting | 0.45 | (0.41, 0.48) | 26.98 | < .001 | 0.44 | |
| AICc | 10512.35 | |||||
| R2 | 0.20 | |||||
| R2 (adj.) | 0.20 | |||||
| Sigma | 1.40 |
| Parameter | Coefficient | 95% CI | t(2986) | p | Std. Coef. | Fit |
|---|---|---|---|---|---|---|
| (Intercept) | 0.64 | (0.44, 0.83) | 6.31 | < .001 | -0.11 | |
| Type (TN) | -0.25 | (-0.54, 0.04) | -1.69 | 0.091 | 0.23 | |
| Interesting | 0.38 | (0.34, 0.43) | 17.37 | < .001 | 0.38 | |
| Type (TN) × Interesting | 0.15 | (0.08, 0.21) | 4.38 | < .001 | 0.14 | |
| AICc | 10495.21 | |||||
| R2 | 0.21 | |||||
| R2 (adj.) | 0.21 | |||||
| Sigma | 1.40 |
Level-1-Einheiten gehören zu/sind geschachtelt in Level-2-Einheiten.
Schüler geschachtelt in Klassen oder Schulen
Beiträge geschachtelt in Nachrichtenoutlets, Accounts, Ausgaben oder Sendungen
Messungen geschachtelt in Personen (Personen sind Level-2!)
Daten können auch auf mehr als zwei hierarchischen Ebenen (Schüler, Klasse, Schule) geschachtelt oder kreuzklassifiziert sein
bei kreuzklassifizierten Daten gehören Level-1-Einheiten zu min. zwei verschiedenen Kontexten (z.B. Bewertungen von Werbespots gehören zu bewertenden Personen und bewerteten Spots)
statistische Gründe: Die Schachtelung der Level-1-Einheiten muss auch berücksichtigt werden, wenn nur Level-1-Zusammenhänge von Interesse sind (dependence as a nuisance).
substanziellen Gründe: Multilevel-Modelle erlauben eine Zerlegung in Between- und Within-Group-Varianz und damit die adäquate Modellierung von Zusammenhängen auf verschiedenen Ebenen (dependence as an interesting phenomenon).
Das Multilevel-Nullmodell für geschachtelte Daten mit \(i\) Individuen in \(j\) Gruppen besteht aus je einem Nullmodell pro Ebene:
Level 1: \(Y_{ij} = b_{0j}+ \epsilon_{ij}\) und Level 2: \(b_{0j} = \gamma_{00} + u_{0j}\)
oder zusammengenommen: \(Y_{ij} = \gamma_{00} + u_{0j} + \epsilon_{ij}\)
\(\gamma_{00}\) ist der globale Mittelwert von \(Y\), \(u_{0j}\) ist die gruppenspezifische Abweichung vom globalen Mittelwert und \(\epsilon_{ij}\) ist das individuelle Residuum.
-Intercepts dürfen gruppenspezifisch variieren, Regressionskoeffizienten \(B\) sind für alle Level-2-Einheiten gleich
\(Y_{ij} = \gamma_{00} + u_{0j} + B_1 x_1 + B_2 x_2 + ... + \epsilon_{ij}\)

| uni | uni_fans | created_time | type | likes_count | comments_count |
|---|---|---|---|---|---|
| U British Columbia | 121.04 | 2013-06-20 18:22:58 | photo | 66 | 5 |
| U Colorado Boulder | 154.47 | 2012-10-30 20:03:57 | link | 253 | 56 |
| U Michigan | 710.50 | 2013-08-01 17:26:54 | photo | 7968 | 261 |
| U British Columbia | 121.04 | 2015-09-08 19:53:00 | photo | 11 | 2 |
| Johns Hopkins U | 113.61 | 2014-04-10 19:59:07 | video | 20 | 0 |
| Variable | n_Obs | Mean | SD | Median | MAD | Min | Max |
|---|---|---|---|---|---|---|---|
| comments_count | 10807 | 12.15 | 36.46 | 3 | 4.45 | 0 | 1556 |
| Parameter | Coefficient | 95% CI | t(10804) | p | Effects | Group | Std. Coef. | Fit |
|---|---|---|---|---|---|---|---|---|
| (Intercept) | 14.34 | (9.23, 19.44) | 5.50 | < .001 | fixed | 0.06 | ||
| 16.71 | random | uni | ||||||
| 33.32 | random | Residual | ||||||
| AICc | 106618.77 | |||||||
| R2 (conditional) | 0.20 | |||||||
| R2 (marginal) | 0.00 | |||||||
| Sigma | 33.32 |
| uni | estimate | std.error | statistic | p.value | conf.low | conf.high | df |
|---|---|---|---|---|---|---|---|
| Columbia U | 2.33 | 2.61 | 0.89 | 0.37 | -2.78 | 7.44 | Inf |
| Cornell U | 9.40 | 2.61 | 3.61 | 0.00 | 4.29 | 14.51 | Inf |
| Duke U | 8.69 | 2.61 | 3.34 | 0.00 | 3.58 | 13.80 | Inf |
| Harvard U | 66.38 | 2.61 | 25.48 | 0.00 | 61.27 | 71.48 | Inf |
| Imperial College London | 3.40 | 2.61 | 1.30 | 0.19 | -1.71 | 8.50 | Inf |
| Johns Hopkins U | 3.18 | 2.61 | 1.22 | 0.22 | -1.92 | 8.29 | Inf |
| MIT | 15.76 | 2.61 | 6.05 | 0.00 | 10.66 | 20.87 | Inf |
| New York U | 10.71 | 2.61 | 4.11 | 0.00 | 5.61 | 15.82 | Inf |
| Northwestern U | 7.88 | 2.61 | 3.03 | 0.00 | 2.78 | 12.99 | Inf |
| Princeton U | 5.63 | 2.61 | 2.16 | 0.03 | 0.53 | 10.74 | Inf |
| Parameter | Coefficient | 95% CI | t(10387) | p | Effects | Group | Std. Coef. | Fit |
|---|---|---|---|---|---|---|---|---|
| (Intercept) | 15.57 | (10.48, 20.67) | 5.99 | < .001 | fixed | 0.06 | ||
| topic research | -4.06 | (-5.53, -2.59) | -5.42 | < .001 | fixed | -0.05 | ||
| AICc | 102577.73 | |||||||
| R2 (conditional) | 0.20 | |||||||
| R2 (marginal) | 0.00 | |||||||
| Sigma | 33.41 |
| npar | AIC | BIC | logLik | -2*log(L) | Chisq | Df | Pr(>Chisq) | |
|---|---|---|---|---|---|---|---|---|
| m1_research | 4 | 102582.7 | 102611.7 | -51287.36 | 102574.7 | |||
| m2_research_vs | 6 | 102532.8 | 102576.3 | -51260.39 | 102520.8 | 53.93 | 2 | 0 |
| term | uni | estimate | std.error | statistic | p.value | conf.low | conf.high |
|---|---|---|---|---|---|---|---|
| topic_research | Columbia U | -0.43 | 1.22 | -0.35 | 0.72 | -2.82 | 1.96 |
| topic_research | Cornell U | -3.15 | 1.22 | -2.58 | 0.01 | -5.54 | -0.76 |
| topic_research | Duke U | -2.72 | 1.22 | -2.23 | 0.03 | -5.11 | -0.33 |
| topic_research | Harvard U | -24.59 | 1.22 | -20.16 | 0.00 | -26.98 | -22.20 |
| topic_research | Imperial College London | -0.91 | 1.22 | -0.75 | 0.45 | -3.30 | 1.48 |
| topic_research | Johns Hopkins U | -0.73 | 1.22 | -0.60 | 0.55 | -3.13 | 1.66 |
| topic_research | MIT | -5.51 | 1.22 | -4.52 | 0.00 | -7.90 | -3.12 |
| topic_research | New York U | -3.48 | 1.22 | -2.85 | 0.00 | -5.87 | -1.09 |
| topic_research | Northwestern U | -2.43 | 1.22 | -1.99 | 0.05 | -4.82 | -0.04 |
| topic_research | Princeton U | -1.67 | 1.22 | -1.37 | 0.17 | -4.07 | 0.72 |
| Parameter | Coefficient | 95% CI | t(10384) | p | Effects | Group | Std. Coef. | Fit |
|---|---|---|---|---|---|---|---|---|
| (Intercept) | 10.88 | (6.56, 15.19) | 4.94 | < .001 | fixed | 0.05 | ||
| topic research | -5.69 | (-8.35, -3.04) | -4.21 | < .001 | fixed | -0.07 | ||
| uni fans | 0.01 | (0.01, 0.01) | 8.18 | < .001 | fixed | 0.21 | ||
| AICc | 102509.01 | |||||||
| R2 (conditional) | 0.16 | |||||||
| R2 (marginal) | 0.05 | |||||||
| Sigma | 33.32 |
in der Kommunikationswissenschaft sind geschachtelte Daten nahezu überall anzutreffen, sei es bei Inhaltsanalysen, Panel- oder Mehrländer-Befragungen
oft sind auch Personen die Level-2-Einheiten, z.B. bei Experience-Sampling-Daten oder Within-Subject-Experimenten
die Grundlogik der Multilevel-Analyse entspricht dem linearen Modell, lediglich die Art und Weise, die Schachtelung zu berücksichtigen, variiert
wenn Verdacht auf Verletzung der Unabhängigkeitsannahme besteht, kann man immer zumindest ein Varying Intercept Modell schätzen, das nie “schlechter” als ein lineares Regressionsmodell ist
technisch muss man in R (fast) nur lm() durch lmer() ersetzen und die Modellformel leicht anpassen (siehe praktische Übung)
| ICC_adjusted | ICC_unadjusted | optional |
|---|---|---|
| 0.45 | 0.45 | FALSE |
| Share | Share | |||||
|---|---|---|---|---|---|---|
| Predictors | Estimates | CI | p | Estimates | CI | p |
| (Intercept) | 2.23 | 2.15 – 2.30 | <0.001 | 2.23 | 2.11 – 2.35 | <0.001 |
| Type [TN] | 0.36 | 0.26 – 0.46 | <0.001 | 0.34 | 0.27 – 0.42 | <0.001 |
| Interesting c | 0.38 | 0.34 – 0.43 | <0.001 | 0.34 | 0.30 – 0.37 | <0.001 |
| Type [TN] × Interesting c | 0.15 | 0.08 – 0.21 | <0.001 | 0.14 | 0.09 – 0.19 | <0.001 |
| Random Effects | ||||||
| σ2 | 1.04 | |||||
| τ00 | 0.92 Participants | |||||
| ICC | 0.47 | |||||
| N | 299 Participants | |||||
| Observations | 2990 | 2990 | ||||
| R2 / R2 adjusted | 0.206 / 0.205 | 0.173 / 0.562 | ||||