Chapter 3in Part II of the book. The idea is very simple and explained
more fully in online Annex 3b. Suppose that the full esti-
mated equation is
Life- satisfaction = a 1 Education + a 2 Income + etc.
but we also estimate the simple relationship
Life- satisfaction = b 1 Education
Then b 1 will be larger than a 1 and the difference will equal
a 2 times the “effect” of education on income, plus all the
other similar terms for all the other variables. Some of
these variables will be “mediating” variables, and some will
be “confounders” (where for example the “effect” of ed-
ucation on father’s unemployment is clearly not a causal
relationship).
The variables that have a good claim to be mediating
variables are those listed in Table 3.2. Some of the relation-
ship between education and these variables is not causal,
but we cannot separate out the part that is confounding.
Treating the whole as a mediating effect we find the total
effect of education on happiness as shown in Table 3.2. This
total is less than the simple estimate of 0.19, and the resid-
ual is due to the role of confounding variables, like father’s
unemployment.
Thus, to conclude our BCS analysis, education in itself
has a small positive direct effect on life- satisfaction, but a
bigger overall effect due to its effect via income and other
mediating variables.^7
Like the BCS, the household panel surveys also find that
happiness is moderately affected by education per se. The