Exercise 4.3: Scatterplots

Graph are very useful tools for studying the relationship between attitude change and context variables in greater detail. Scatter plots, for example, are very effective in detecting outlying countries that might have distorted the correlational analysis. Since our aggregate analyses are performed on a very small number of observations only (17 countries), there is a considerable risk that this is the case.

Drawing scatter plots

Click ‘Graphs’ in the main horizontal tool bar, and then ‘Chart Builder’.

The Chart Builder wizard pops up. We start by drawing a scatter plot with FLOWCAP on the X-axis and REJECT_CHANGE on the Y-axis. First, define FLOWCAP as a scale variable (now it is still defined as a nominal variable). To do this, right-click the variable name FLOWCAP and select ‘scale’ instead of ‘nominal’.

We continue by defining the graph type. In the lower pane of the gallery, select ‘scatter/dot’. Drag the first icon into the upper pane of the chart builder. Now you can select the variables to be displayed by dragging the variable names (in the upper left corner) into the graph. Drag REJECT_CHANGE to the box next to the Y-axis, and FLOWCAP to the X-axis.

The scatter plot becomes even more informative if we put the country abbreviations next to the dots. Click the ‘Groups/Point ID’ tab in the lower pane of the Chart Builder. Select the box ‘Point ID Label’. Now, an additional box appears in the upper pane (namely Point ID). Drag the variable ‘Country’ into this box.

Click OK. Repeat these steps in order to obtain plots with GDP_GROWTH and UNEMP_CHANGE on the X-axis.

SPSS Output

Figure 4.3. Scatter 1

Figure 4.3. Scatter 2

Figure 4.3. Scatter 3

Question

What do you conclude on the basis of these scatter plots?

Solution

The graphs indicate that, for the relationship with at least two of the context variables, the correlational analysis is distorted by a clear outlier, namely Poland. In Poland, we witnessed a marked decrease in anti-immigration attitudes. At the same time, Poland has known high levels of immigration and a remarkable decline in the unemployment rate. As a result of this very particular combination, the correlations between REJECT_CHANGE, on the one hand, and FLOWCAP and UNEMP_CHANGE on the other, are misleading. If Poland were omitted from the analysis, we might find near-zero correlations, or even correlations with the opposite sign. The relationship between REJECT_CHANGE and GDP_GROWTH seems somewhat more stable. These findings further weaken the already shaky evidence for the role of group conflict in attitude change. At the same time, they illustrate the risk of conducting analyses on a small number of observations, and show the usefulness of scatter plots to detect outliers.