How is your benchmark calculated

The benchmark data

The benchmark data comes from a large global survey of representatively sampled respondents that covered 8 countries: USA, UK, Canada, Germany, France, Belgium, Netherlands and Australia.

To be statistically robust, we use this research data rather than our own client data. The research data covers a representative sample, where our clients are often organizations that are already interested in this area and so their results are likely to be skewed higher than the general population.

The benchmark calculations

Benchmarks were calculated for overall Happiness at Work and for each of the Five Ways to Happiness at Work – Connect, Be Fair, Challenge, Empower and Inspire. Five Ways scores were created by calculating the mean of the scores for the three questions under each Way.

A multi-variate regression was conducted with Happiness and each of the Five Ways Happiness as dependent variables, plus all nine benchmarking variables as independents. We used the ‘UNIANOVA’ model in the statistics software package SPSS, which allows categorical variables to be entered without having to create dummy variables. The regression produced a set of parameter estimates, then further work was required to convert this into a formula to use for benchmarking.

Ultimately, we were able to produce a formula that can be used to calculate a benchmark based on the demographics of any given respondent in Friday. The formula looks like:

Benchmark = β1 + β2 + β3 + ... + α

Where β1 may be the beta parameter for the country of the respondent, β2 the beta parameter for their sector, β3 for their organization size, and so on.

Using this formula we calculated three benchmark variables for Friday: country, sector and organizational size.

Learn more:

• Selecting your benchmarks
• The benchmark variables
• How to use your benchmark data

Statistical notes:

• We designed and analysed the survey on behalf of Robert Half – a global recruitment agency that specialises in professional workers. Due to their focus, we over-sampled ‘white-collar’ workers. However we ensured that were sufficient numbers of ‘blue-collar’ workers in the survey so that post-sample weights could be generated, enabling accurate estimates to be calculated for the population as a whole.
• We treated organization size slightly differently in the regression. We had 11 response categories in our survey, which we converted into a continuous variable. To do this, we estimated organization size for any respondent as being the midpoint of each band. So those who said their organization size was between 100 and 249 employees were attributed a value of 175. The natural logarithm of this value was then used in the model.