Introduction to Catalog of Happiness in Nations
Abstract chapter 7
HOW THE DATA ARE HOMOGENIZED
This study collects the result of investigations that used acceptable measures of
happiness. These acceptable measures are not quite identical. They differ in five aspects:
1) Variant of happiness measured, 2) Assessment mode, 3) Phrasing of the lead question, 4)
Labeling of response categories and 5) Number of response categories. This chapter
explains how the divergent data were classified into equivalent categories. It further
considers three techniques for transforming responses to dissimilar questions into
comparable scores.
Classification
This database presents the data by kind of happiness
indicator. This breaks the data collection into four main parts: one big part on 'overall
happiness', a smaller one on 'hedonic level' and two minor ones referring to 'contentment'
and 'mixed indicators'. Within these parts the collection is further differentiated in
tables of near-identical indicators. This results in tables by 'question-type. Most of the
tables with identical items concern overall happiness. Among these, three groups of
questions some can be discerned which ask essentially the same thing, but that differ only
in the rating of response. Though not 'identical', the items in these clusters are
'equivalent'. As such they qualify for conversion to a common scale. The possibilities for
converting average scores on divergent indicators of happiness are however limited.
Transformation
Scores on indicators of different happiness variants
can not be converted to the same standard. They measure essentially different
things that do not necessarily coincide.
Scores on different indicators of the same happiness variant can be
converted in principle. However, in practice it is quite difficult to estimate the method
effects involved. If sufficient data are available, we can inspect whether there is a
linear relationship between responses yielded by different indicators in the same
populations. Such data are only available for some single questions on overall happiness.
We found a reliable relation in the nation scores on the two pairs of items: 1) 10-step
life-satisfaction by 4-step satisfaction with way-of-life, and 2) 11-step
life-satisfaction by 11-step best-worst possible life. In these cases missing values on
one variable can be reliably estimated by linear regression on the basis of observed
scores on the other; interpolation is less risky than extrapolation. In three pairs we
found no reliable relation however: 1) happiness-in-life by satisfaction-with-life, 2)
happiness-in-life by best-worst possible life, and 3) happiness-in-life by
delighted-terrible life. In these latter cases we deem transformation inadvisable.
Conversion is better possible when indicators (questions) are
substantially equivalent, and differ only in number and labeling of response categories.
In that case standardization by expert-weighting is justified. The expert-transformation
applied here successfully passed a test for congruent validity.
If differences between equivalent items concern only the length of a
graphic or numerical rating scale, simple linear transformation (stretching) is most
appropriate.
Only the latter two standardization methods (expert-weighting and
stretching) are applied in this data collection. In the tables transformed scores are
mentioned for equivalent items. Transformed means are presented next to the original
means.