Annex C. Quantitative analysis

The quantitative analysis presented in this Report was conducted in collaboration with Senior Expert Professor Achim Goerres (University of Duisburg-Essen). Following internal and collaborative discussions on the overarching objective of the analysis, research questions, hypotheses and measurement, the OECD Secretariat provided the Senior Expert with all the original and additional data files. The data files were then cleaned individually, restructured to follow a pattern of one row per country, and finally merged into one major data file. The data file used for the final quantitative analysis was updated in a second step to account for changes from the data cleaning and validation process conducted by the OECD Secretariat in co-operation with the respondent government entities as well as to account for changes in OECD membership (i.e. addition of data for Colombia).

The methods used for the analysis are bivariate and multivariate analyses including:

1. 1. Pearson’s for continuous variables that range from -1 (perfect negative relationship), through 0 (no relationship) to +1 (perfect positive relationship);

2. 2. Pearson’s Chi²-test for tables of categorical variables with a p-value that gives the likelihood of that pattern to occur by chance alone;

3. 3. Fisher’s Exact test for tables in which one cell has a count of zero with a p-value that gives the likelihood of that pattern to occur by chance alone;

4. 4. Wilcoxon test for a context where one continuous variable is tested for its distribution across two groups with a p-value that indicates the likelihood of the two sub-distributions deriving from the same underlying distribution;

5. 5. Kruskal-Wallis test for analyses in which one continuous variables is tested for distribution across more than two groups with a p-value that indicates the likelihood of the n sub-distributions deriving from the same underlying distribution; and

6. 6. Exploratory OLS regression analyses with two predicting variables and one continuous dependent variable.

The sample of countries is not a random sample from a defined population, but a census of OECD countries for which there were data available. Inferential statistics such as p-values of correlation coefficients should hence be interpreted as a measure of precision of observed patterns of co-variation.

When looking at the impact of involving youth organisations in the policy cycle on their satisfaction with government’s performance across a range of public services, country-level observations were calculated as a mean of means of individual youth organisations’ responses on the basis of their country of operation. Whenever a country-level observation had missing information, the mean of the other country-observations was imputed.