The Origins of Happiness

Notes to Pages 165–171

In other words, they correct for the fact that some mothers who de-
cided to return to work early did so because of their preferences for
work and/or their children were well developed enough to allow them
to go back to work. With fixed effects and IV estimation, Ermisch and
Francesconi (2013) basically estimated the effect of an exogenous effect
of maternal employment (induced by regional and time variation in
aggregate female unemployment rates) on child’s outcome. In other
words, there will be mothers who returned to work not because they
are ready to, but because the opportunity cost of not going back is sim-
ply just too high for them.

10. Ruhm (2004) building on Waldfogel, Han, and Brooks- Gunn
    (2002). See also Joshi and Verropoulou (2000) for the UK.


11. Becker and Tomes (1986); Duncan and Brooks- Gunn (1999);
    Clark and Oswald (1994); Goldsmith, Veum, and Darity (1996); Powd-
    thavee and Vignoles (2008); Gruber (2004); McLoyd (1989); Christof-
    fersen (1994); and S. Brown and Taylor (2008).


12. The standard errors in columns (1) and (2) exceed those in (3)
    because we have GCSE scores for all children, from the National Pupil
    Database, whereas there are a significant number of missing observa-
    tions on emotional and behavioral outcomes at age 16.

Chapter 12. Parenting and Parents’ Mental Health

1. See Plomin et al. (2013).


2. Lykken (1999) and Caprara et al. (2009).


3. Tellegen et al. (1988).


4. See Plomin et al. (2013).


5. Tienari et al. (1994).


6. Bohman (1996) and Cadoret et al. (1995).


7. See Plomin et al. (2013).


8. Pluess (2015) and Okbay et al. (2016).


9. The same applies to all claims about the role of genes. In par-
   ticular we should be wary of estimates offered of the “heritability” of a
   trait. There are two problems here:

(i) These estimates assume that gene and environment effects are

additive (no gene/environment interaction). The trait (T) is thus

determined by T = G + E and VarT = VarG + VarE + 2Cov(G, E).

(ii) These estimates assume that the covariance term Cov(G, E)

reflects a causal impact of G on E and therefore heritability is

measured as (VarG + 2Cov[G, E])/VarT. However this implies a