Social and Media Trust in Russia
Project 4
Hypotheses
Research question Which factors can help to predict the time which person spends daily on watching, reading or listening news about politics and current affairs (newspol)?
Hypotheses
It is widely assumed that the more the person is interested in political subjects, the more time he/she will spend on watching or reading news of political nature.
However, it is not that straightforward in what direction the relationship goes taking into account the level of government satisfaction. Intuitively, it seems like the more the person is dissatisfied with the government, the more he/she would be prone to observe closely every action the government undertakes via social media, TV, and newspapers.
In addition, it is also important to consider the age of a person. Some theories claim that the younger generation is extremely politicized nowadays, whereas older ones are becoming indifferent to the state policy as they feel powerless to make a significant impact on the situation. So, it is then logical to expect greater time spent on political news by younger generations.
Another aspect to consider is gender: intuitively, men are supposed to be more politically engaged than women, so they would spend more time trying to catch the info about politics and current affairs.
Finally, we should also take into account the level of urbanization. In is common knowledge that big city dwellers have direct access to the governmental structures. Additionally, in a big city it is much easier to be always up to date with all the new info concerning state policy, and there are also plenty of educated people with whom governmental decisions can be discussed and critically evaluated. Therefore, they are thought to be the most politically engaged groups, compared to people from small cities and rural areas.
Based of the theoretical assumptions stated above, we have come up with the following hypotheses:
formattable(VarTable1,
align =c("l","l"),
list(`Indicator Name` = formatter(
"span", style = ~ style(color = "grey",font.weight = "bold"))))| Hypotheses | Description |
|---|---|
| Hypothesis 1: | The time spent on political news is positively associated with the level of interest in political subjects |
| Hypothesis 2: | The time spent on political news is negatively associated with the level of satisfaction with the government |
| Hypothesis 3: | The time spent on political news is negatively associated with the age of the respondent |
| Hypothesis 4: | The time spent on political news is greater for males than for females |
| Hypothesis 5: | The time spent on political news is greater for people from big cities compared to people from rural areas |
Descriprive statistics of the outcome variable and predictors
For our linear regression models we`ve tried to come up with the best predictors for the outcome variable:
formattable(VarTable,
align =c("l","l","l"),
list(`Indicator Name` = formatter(
"span", style = ~ style(color = "grey",font.weight = "bold"))))| Predictor | Description | Range |
|---|---|---|
| polintr | How much interest a person has towards political subjects | 0 (Not at all interested) - 10 (Very interested) |
| stfgov | How a person is satisfied with the current national government | 0 (Extremely dissatisfied) — 10 (Extremely satisfied) |
| agea | Age of the respondent | [1-78] |
| gndr | Gender of the respondent | Male/Female |
| domicil | Area of living | A big city - Farm or home in countryside |
- Distribution of the outcome variable
nwspol.
- Distributions of predictors; correlations.
2.1 nwspol and polintr
As can be seen, the majority of people which is 39.79% is quite interested in politics. However, the great minority is hardly interested in politics (33.17%). Furthermore, as can be seen from the boxplot, there is a positive correlation between time spent on political news and the level of interest in politics: the higher amount of time spent in political news is associated with a higher level of political interest.
2.2 nwspol and stfgov
So, here it is evident that the most number of people (21.4%) are neutral to the government. Moreover, there is a slight positive association between time spent on political news and level of satisfaction with the government (r = 0.09).
2.3 nwspol and agea
So, the distribution of respondents’ age is bimodal with peaks at 12 and 45 years old. The mean age is about 32. Besides, the association between age and time spend on political news is positive and moderate: the higher age is associated with a higher amount of time spent on watching news about politics and current affairs.
2.4 nwspol and gndr
2.5 nwspol and domicile
The majority of people live in a big city (41.23%). People who live in town or in the country village are 32.99% and 22.43% respectively. Also, there is no association between an area of living and times spent on political news with the exception of those living on a farm who spend much more time on political news compared to others.
Our models
We have a set of hypothetical predictors (stated in the begining of the project) and we are going to add them one by one to reach a better result in R-squared.
mod1 <- lm(nwspol ~ polintr, data = variables)
tab_model(mod1)| nwspol | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 21.18 | 19.99 – 22.36 | <0.001 |
| Quite interested | -5.57 | -6.89 – -4.24 | <0.001 |
| Hardly interested | -10.01 | -11.37 – -8.66 | <0.001 |
| Not at all interested | -13.52 | -15.02 – -12.01 | <0.001 |
| Observations | 2234 | ||
| R2 / adjusted R2 | 0.155 / 0.153 | ||
The basic model explains about 15 % of the variance. But can we do better at explaining the variation in the outcome variable? Let`s try to add other predictors:
mod2 <- lm(nwspol ~ polintr + scale(agea), data = variables)
mod3 <- lm(nwspol ~ polintr + scale(agea) + domicil, data = variables)
mod4 <- lm(nwspol ~ polintr + scale(agea) + domicil + gndr, data = variables)
mod_int <-lm(nwspol ~ polintr * stfgov + scale(agea) + domicil + gndr, data = variables)anova(mod1, mod2, mod3, mod4, mod_int)From the output table of ANOVA, we can see that each model is statistically significantly better than the previous one. Therefore, we end up with the final model with interaction effect which is the best model according to p-value (and R-squared as well).
The best model and interpretation of its regression coefficients
tab_model(mod_int)| nwspol | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 18.94 | 16.69 – 21.19 | <0.001 |
| Quite interested | -4.49 | -7.06 – -1.92 | 0.001 |
| Hardly interested | -8.21 | -10.90 – -5.52 | <0.001 |
| Not at all interested | -9.65 | -12.56 – -6.75 | <0.001 |
| stfgov | 0.51 | 0.11 – 0.90 | 0.013 |
| scale(agea) | 2.14 | 1.76 – 2.53 | <0.001 |
|
Suburbs or outskirts of big city |
-0.48 | -2.91 – 1.95 | 0.699 |
| Town or small city | -0.83 | -1.69 – 0.03 | 0.060 |
| Country village | 0.29 | -0.68 – 1.26 | 0.560 |
|
Farm or home in countryside |
4.90 | 0.96 – 8.84 | 0.015 |
| Female | -1.19 | -1.96 – -0.42 | 0.002 |
| polintrQuite interested:stfgov | -0.17 | -0.63 – 0.30 | 0.485 |
| polintrHardly interested:stfgov | -0.18 | -0.66 – 0.31 | 0.481 |
| polintrNot at all interested:stfgov | -0.57 | -1.12 – -0.01 | 0.046 |
| Observations | 2234 | ||
| R2 / adjusted R2 | 0.210 / 0.205 | ||
Even though not all the coefficients are statistically significant, we can interpret them. The nteraction will be described visually afterward.
Coefficient <- c("(Intercept)", "Quite interested", "Hardly interested",
"Not at all interested", "stfgov", "scale(agea)",
"Suburbs or outskirts of big city", "Town or small city",
"Country village", "Farm or home in countryside", "Female")
Description <- c("For Male of average age, who are very interested in politics, scores 0 for the level of satisfaction with the government and lives in a big city the predicted time to spend on political news is about 19 minutes.",
"If this Male is quite interested in politics, the predicted time would decrease by 4 minutes.",
"If this Male is hardly interested in politics, the predicted time would decrease by 8 minutes.",
"If this Male is not at all interested in politics, the predicted time would decrease by 9 minutes.",
"For Male of average age, who are very interested in politics and lives in a big city the predicted time to spend on political news will be associated with 0.5 minutes increase every time government satisfaction level would grow up by 1.",
"For Male, who is very interested in politics, scores 0 for the level of satisfaction with the government and lives in a big city the predicted time to spend on political news will be associated with 2 minutes increase every time age of the person will increase by 1 SD above the mean.",
"For Male of average age, who are very interested in politics and scores 0 for the level of satisfaction with the government, the predicted time to spend on political news will be associated with 0.5 minutes decrease if a person lives in the suburbs or outskirts of a big city.",
"If this Male lives in a town or a small city, the predicted time to spend on political news will be associated with 0.8 minutes decrease.",
"If this Male lives in a country village, the predicted time to spend on political news will be associated with 0.3 minutes increase.",
"If this Male lives on the farm or in the home in the countryside, the predicted time to spend on political news will be associated with 5 minutes increase.",
"For Female of average age, who are very interested in politics, scores 0 for the level of satisfaction with the government and lives in a big city the predicted time to spend on political news will be 1 minute less, then for Male with the same characteristics.")
DescTable <- data.frame(Coefficient, Description)formattable(DescTable,
align =c("l","l","l"),
list(`Indicator Name` = formatter(
"span", style = ~ style(color = "grey",font.weight = "bold"))))| Coefficient | Description |
|---|---|
| (Intercept) | For Male of average age, who are very interested in politics, scores 0 for the level of satisfaction with the government and lives in a big city the predicted time to spend on political news is about 19 minutes. |
| Quite interested | If this Male is quite interested in politics, the predicted time would decrease by 4 minutes. |
| Hardly interested | If this Male is hardly interested in politics, the predicted time would decrease by 8 minutes. |
| Not at all interested | If this Male is not at all interested in politics, the predicted time would decrease by 9 minutes. |
| stfgov | For Male of average age, who are very interested in politics and lives in a big city the predicted time to spend on political news will be associated with 0.5 minutes increase every time government satisfaction level would grow up by 1. |
| scale(agea) | For Male, who is very interested in politics, scores 0 for the level of satisfaction with the government and lives in a big city the predicted time to spend on political news will be associated with 2 minutes increase every time age of the person will increase by 1 SD above the mean. |
| Suburbs or outskirts of big city | For Male of average age, who are very interested in politics and scores 0 for the level of satisfaction with the government, the predicted time to spend on political news will be associated with 0.5 minutes decrease if a person lives in the suburbs or outskirts of a big city. |
| Town or small city | If this Male lives in a town or a small city, the predicted time to spend on political news will be associated with 0.8 minutes decrease. |
| Country village | If this Male lives in a country village, the predicted time to spend on political news will be associated with 0.3 minutes increase. |
| Farm or home in countryside | If this Male lives on the farm or in the home in the countryside, the predicted time to spend on political news will be associated with 5 minutes increase. |
| Female | For Female of average age, who are very interested in politics, scores 0 for the level of satisfaction with the government and lives in a big city the predicted time to spend on political news will be 1 minute less, then for Male with the same characteristics. |
Interaction plots with CI and other important stuff to report
Interpretation of interaction plot:
From the interaction plot, we can see that the higher the level of government satisfaction, the lower will be the coefficient for these 3 categories included in the variable “political interest”, where for “Not at all interested” in politics people the coefficient is predicted to be the lowest.
In plain words, the less the person is interested in politics and the more this person is satisfied with the government, the less time he/she is predicted to spend on political news. In even more plain words, people do not care about political news if they are not interested in politics and are okay with the state policies.
Marginal effects:
plot_model(mod_int, type = "int")
For a person who is very or quite interested in politics, we can observe a statistically sighificant difference in time spent on political news on the marginal values of the moderator variable, which is government satisfaction. In other cases, the difference in time is not as vivid as the CIs overlap for the other two categories.
Plainly speaking, if a person is completely dissatisfied with the government but at the same time very or quiet interested in politics, he/she is predicted to spend much less time on political news, compared to being extremely satisfied with the government.
Standartized coefficients:
plot_model(mod_int, type = "std")
By visually observing beta coefficients, we can say that “Hardly interested in politics” category shows the largest effect in our model. In other words, it can predict the time spent on political news better, then other predictors.
Conclusion
After interpreting all the coefficients in our final model, we can check our initial hypotheses:
Hypotheses <- c("Hypothesis 1:", "Hypothesis 2:", "Hypothesis 3:", "Hypothesis 4:", "Hypothesis 5:")
Description <- c("The time spent of political news is positively associated with the level of interest in political subjects", "The time spent on political news is negatively associated with the level of satisfaction with the government", "The time spent on political news is negatively associated with the age of the respondent", "The time spent on political news is greater for males than for females", "The time spent on political news is greater for people from big cities compared to people from rural areas")
Status <- c("Confirmed", "Falsified", "Falsified", "Confirmed", "Falsified")
VarTable_hyp <- data.frame(Hypotheses, Description, Status)formattable(VarTable_hyp,
align =c("l","l"),
list(`Indicator Name` = formatter(
"span", style = ~ style(color = "grey",font.weight = "bold"))))| Hypotheses | Description | Status |
|---|---|---|
| Hypothesis 1: | The time spent of political news is positively associated with the level of interest in political subjects | Confirmed |
| Hypothesis 2: | The time spent on political news is negatively associated with the level of satisfaction with the government | Falsified |
| Hypothesis 3: | The time spent on political news is negatively associated with the age of the respondent | Falsified |
| Hypothesis 4: | The time spent on political news is greater for males than for females | Confirmed |
| Hypothesis 5: | The time spent on political news is greater for people from big cities compared to people from rural areas | Falsified |
Regarding our final model, we also have the following characteristics:
Metric <- c("№ of observations:", "adjusted R-squared:", "F-statistic(13, 2220):", "p-value:")
Value <- c("2234", "0.205", "45.39", "< 2.2e-16")
Stat_tab <-data.frame(Metric, Value)formattable(Stat_tab,
align =c("l","l"),
list(`Indicator Name` = formatter(
"span", style = ~ style(color = "grey",font.weight = "bold"))))| Metric | Value |
|---|---|
| № of observations: | 2234 |
| adjusted R-squared: | 0.205 |
| F-statistic(13, 2220): | 45.39 |
| p-value: | < 2.2e-16 |
Sum up of our findings:
Basically, we have found that the more people are satisfied with the local government, the more time they tend to spend on political news daily. We also found that Males on average spend more time on political news than females, and the effect is also increasing as the persons grow older. Still, the most peculiar finding is that countryside dwellers or people who live on a farm spend much more time on news of politics and current affairs, then people from big cities.