Healing broken time-series data

Tineke DeJonge, Akiko Kamesaka, & Ruut Veenhoven
EHERO working paper 2019/3


Periodical survey studies show that average happiness has risen in most developed nations since the 1970s, similar survey data in Japan do not show such a clear trend. Average happiness in Japan was first assessed in 1958, and has since been assessed in at least six survey programs. The questions used in these surveys are not identical and for this reason it is difficult to identify a time trend in the data. In this paper, we apply four methods for homogenizing these data, each of which allows computation of an average on a numerical scale ranging from 0 to 10: 1) simple ordinal ordering of response options by their ranks, starting with 1 for the least happy response, followed by linear stretch to scale 0-10, 2) assigning fixed weights to particular response options, such as 8 for ‘happy’, 3) assessing weights in the context of the response scale, using native speakers as judges, and 4) deriving weights of verbal response options using a reference distribution obtained in the same year from answers to a similar survey question rated on a numerical response scale. We find that the Japanese have not become any happier over the last 50 years.

Full text