Why is the chart interesting?
The first principal component was the most important response pattern overall. It explained 67% of variance on its own. Countries with high negative scores had a low opinion of their child's school overall (ie. lower school ability to prepare the child for the future, lower school teaching quality) and tended to have low engagement as well. They were less optimistic about the child’s future in particular. These were South Korea, Japan, Russia, Germany and France. Countries with high positive scores had the opposite pattern. These were India, Indonesia, Kenya and the United States.
Countries with high negative scores on the second principal component thought that the school teaching quality was low but had high engagement. In particular, they thought that it was very important for their child to attend university in order to achieve the most in life. Countries with high positive scores had the opposite pattern. Russia scored relatively low while Finland, the United Kingdom and France scored relatively high.
Countries with high negative scores on the third principal component thought that the school had low teaching quality but was able to prepare the child well for the future nonetheless. Countries with high positive scores had the opposite pattern. Italy, Poland, Brazil and France scored relatively low while Germany, India, Indonesia and Finland scored relatively high.
Countries scoring the lowest on the first principal component (ie.South Korea, Japan, Russia, Germany and France) formed one cluster. All remaining countries belonged to the other cluster. However, France, Poland and Italy, on the edge of the clusters and scoring low on the third principal component, did not fit as well in their respective groups as the other countries.
In 2018, the Global Education Census was carried out by Cambridge Assessment International Education to explore the experiences of teachers and students around the world. Results can be found here.
PAM cluster analysis was used to determine groups of countries with similar answers. The optimum number of clusters was determined using the silhouette method. The optimum number of principal components was determined using the scree test method. The first three principal components explained 91% of variance of the original variables, so little information was lost.