# Exercises

In this exercise you will investigate how values and some other background variables can explain party preferences along the political continuum from left to right in Norway. To meet the requirements of such a task, we compute a new variable that includes only the parties that have a relatively clear profile on this dimension. This implies that only respondents who have voted for one of the following parties are included (from left to right): Red Electoral Alliance, Socialist Left Party, Labour Party, Conservative Party, and Progress Party.1

The parties on the right place particular emphasis on the market economy as a means to generate wealth, and give people the resources to protect their security. The intended consequences of such a policy are compatible with power, security, and achievement values. But they may harm the opposing values in the value circle - universalism and, perhaps, benevolence. Both these values call for promoting the welfare of others even at a cost to oneself. And universalism values express concern for the weak, those most likely to suffer from market-driven policies. In contrast, the left advocates the merits of the welfare state and expresses strong concern for social justice. The intended consequences of such a policy that focuses on helping the weak are compatible with universalism values and benevolence values as well. On the other hand, they conflict with the pursuit of individual power and achievement values.

Thus, political choice in Norway along this dimension consists of a trade-off between power, achievement, and security values on the right and universalism and benevolence values on the left. Before you start analysing, you should note that many different factors influence behaviour in natural settings. Hence, value-behaviour correlations are likely to be moderate.

Use the extract Human Values to solve the exercises below. The exercises could be solved using data from both ESS 1 and ESS 4. Subset the data to the Norwegian respondents. Use the design weight. The solutions refer to the results obtained using ESS 1. When you have completed the exercises, either focusing upon 2002 (ESS 1) or 2008 (ESS 4), please investigate if you can say something about change from 2002 to 2008.)

- Perform a correlation analysis with the value variables and PartyN. Is the result in accordance with expectations?
- Try to explain party preference along the left-right dimension on the basis of values. Perform a multiple regression analysis with PartyN as the dependent variable and the values that correlate significantly as independent variables. How will you interpret the results?
- How much of the variation in party preference are these variables able to explain?
- How do you expect the variables age, gender, education and income to be related to party preference? Investigate how much of the variation in party preference these variables are able to explain.
- Perform a multiple regression analysis with the significant value variables and the significant variables from the last analysis. How much of the variation in the dependent variable is explained by the linear relationship with these variables? What is the best predictor variable? Write the regression equation.

SolutionsSubset: Find the country variable in the variable list. Click the icon for subset. Click the variable and select “Add to subset”. Highlight Norway in the list of categories, press “Add” and then “OK”.

Weight: Click the icon for weight. Select the "dweight" variable (>), and press “OK”.

1) Correlation: Click “Analysis” on the menu. Select “Correlation”. Find the PartyN variable and the ten values. Include all these variables in the analysis. The correlations are generally weak and insignificant - only five are significant: universalism (-0,26), tradition (0,09), conformity (0,11) security (0,08) and power (0,08). This means that universalism is, as expected, associated with the political left, and that the political right is weakly related to tradition, conformity, security and power.

2) Regression: Click “Analysis” on the menu. Select “Regression”, and add partyN as dependent and the five values as independent variables. The model is significant, but power and security do not have a significant effect in the model. If power and security are omitted, we get the following regression equation: PartyN = 3,74 + 0,09 conformity + 0,09 tradition – 0,41 universalism. If universalism is increased by one unit, partyN will decrease by 0,4 units, all other variables held constant. This is quite large, and it means that universalism makes people vote for leftist parties. If conformity (or tradition) is increased by one unit, partyN will increase by 0,1 unit, all other variables held constant. Conformity makes people vote for parties on the right.

3) The three variables in this model are able to explain about 8 % of the variation on partyN, R square = 0,08.

4) Perform a regression analysis with age, gender, income and education as independent variables. The analysis gives significant results for gender and education. Women are more leftist, and so are people with higher education. The model is able to explain 6 percent of the variation on partyN.

5) The model is able to explain about 12 percent of the variation in partyN. Universalism is the best predictor variable (compare Beta). If we omit conformity, we get the following equation: PartyN = 4,4 – 0,37*universalism – 0,15*education – 0,25*gender + 0,09*tradition.

#### Footnotes

- [1] Four parties represented in the Parliament are omitted: The Centre Party, the Christian Democratic Party, the Liberal Party and the Coastal Party. These parties are often placed in the middle of the left-right-dimension, but their profiles are not primarily associated with this dimension. To ensure that the data reflect the voters' position on this dimension, those who have voted for the centre parties are omitted. Please note that party choice is a nominal variable (only excluding categories). But when the parties are ordered from left to right, an ordinal variable is created. Although this is not a metric variable, using parametric statistics with it is not a serious problem.