Document

Intentions questions in the GGS:
Where did we get it wrong and
how to correct it?
Éva Beaujouan
Second GGP-users conference, Milano, 24-25 October 2013
Counting children intended in
the GGS, outline
• Questions on fertility intentions in the GGS
– In theory / ideas behind the conception of the questionnaire
– In practice / problems with comparability between countries:
filters and pre-codes of the “preceding-question”
• Examples on how the pre-codes of the “precedingquestion” can change the aggregate “intentions”
– General Household Survey (GB)
– Evidence from France and the Netherlands
• Discussion on monitoring of intended fertility
– Suggestion: French example
Ideas behind GGS and the questions
on fertility intentions
Intention questions in the GGS
• “[The GGS questionnaire] includes the core that each
participating country needs to implement in full” (Vikat
et al. 2007)  comparability
• “Prospective focus”  questions on intentions
• Fertility intentions: the concept
– Three-year interval up to next wave
• Theory of Planned Behaviour (behaviour reflects individuals’ informed
decisions)
• Miller and Pasta framework (reference time window)
– Evaluate degree of certainty (Thomson and Brandreth, 1995)
– Parity-specific measure (additional number of children
intended)
Focus: short term v.s. long term
Short term intentions
• “Do you intend to have a/another child during the next three
years?”
– Definitely yes, Probably yes, Probably not, Definitely not, (don’t know)
• Questions give reasonable results
– Fertility intentions are associated with behaviour
– “Certain” answers remain a good predictor of future fertility
Long term intentions
• “Supposing you do not have a/another child during the next three
years, do you intend to have any (more) children at all?”
– Conditional question: people certainly adjust their answer (Schaeffer and
Presser, 2003)
– Does the sum of questions equal one question on long-term intentions?
Questions and filters in GGS:
how do we arrive to the number
intended?
Do you yourself want
to have a/another
baby now?
no
yes
Definitely yes
Probably yes
Probably not
don’t know
As far as you know, is it
physically possible for
you, yourself, to have
a/another baby?
(+ sterilization)
yes
Definitely not
Do you intend to have a/another child during
the next three years?
Definitely yes
Probably yes
Definitely not
Probably not
don’t know
Supposing you do not have a/another child
during the next three years, do you intend to
have any (more) children at all?
Definitely not
Definitely yes
Probably yes
Probably not
don’t know
How many (more) children in total do you intend
to have?
Total number of children intended =
n >= 0
n=0
+ number of own children
Consequences of this complexity?
Long set of questions on fertility intention
• Many filters
• Used inconsistently depending on the country
 We can expect, when calculating mean intended
family size:
• Incomparability between countries
• Many missing / don’t know answers
Intended family size: % missing values
and “don’t know” answers
Age
% DK
coding
U = 18-50 missing
5 = N,PN,PY,Y, DK
Exceptions on number
Australia
18-45
23.3 miss: high by older people
Yes, No, DK
Austria
18-45
1.6
5
Belgium
U
44.3
5
Bulgaria
U
2.9
5
Estonia
21-45
28.7 Women only
5
min / max
France
U
4.3
5
additional number not asked of PN
Georgia
U
0.2
5
Germany
U
21.1 miss: high by younger men
5
problem with fertility histories
Hungary
21-45
7.2
Yes, No, DK
Italy
U
3.3
5
ask total and not add. no. intended
Lithuania
U
3.3
5
Netherlands
18-45
19.1
Yes, No, DK
Norway
U
7.2
Yes, No, DK
Poland
U
3.0
5
additional number not asked of PN
Romania
U
47.3 miss: high by younger people
5
Russia
U
1.6
5
14 GGS countries
In grey: countries dropped
Can imputations deal with these
inconsistencies?
Impact of imputations on Mean Intended Family size in GGS
Imputation 1:
Yes & Pyes without
value are given a
number
Imputation 2a:
Pnot in FR & PL are
given values based
on other countries
Imputation 2b:
DKs in HU, NL, NO
are given values
based on
Probablies in other
countries
Imputation 3:
Remaining DKs are
imputed on values
of the Probablies
Comparability in question
Imputations change MIFS only marginally
• But are they (and should they be) a solution for crosscountry comparability?
And what if for instance…
– the proportion planning to remain childless/not to have
further children depends on the pre-codes of the
“preceding-question” and on the filters?
– those who answer Probably Yes finally do not give the
same number as if they had answered Yes (priming effect)?
• The pre-codes and filters would then have an impact
on researchers’ perception of societal phenomenon
such as number of children intended, voluntary
childlessness, etc.
Insights from other surveys: do
numbers depend on the
preceding-question?
Trends disruption in Great Britain
Effect on % wishing
more children?
If No + P No == 0
(like in PL & FR)
Yes + P Yes
increases more here
=> possible increase
in % wishing 1+ child
If only No == 0
(P No are probed,
like in oth. GGS surv)
=> certainly stronger
increase in % wishing
1+ child
Source: General Household Survey ESRC CPC series
See also: Ní Bhrolcháin, M. & Beaujouan, É. 2011. “Uncertainty in fertility intentions
in Britain, 1979-2007”, Vienna Yearbook of Population Research 9, p. 99-129.
Additional number of children intended
just before and just after the change
All women
yes
probably yes
probably no
no
don't know
all
childless
yes
probably yes
probably no
no
don't know
all
before
(1989-90)
2.04
0.00
1.37
0.88
before
(1989-90)
2.29
0.00
1.81
1.75
after
(1991-92)
2.10
1.78
0.00
0.00
0.82
after
(1991-92)
2.34
2.03
0.00
0.00
=> additional number
intended decreases
slightly
+ slight
change in
balance
yes/no
1.64
Source: General Household Survey ESRC CPC series
=> Decrease in additional
number intended
Having the option to say
“probably” changes numbers
declared by childless women,
not only proportionally.
The Netherlands and France: no
consistent series
Total number of children intended
the Netherlands
Problem with
weights
France
Change in
questionnaire
To few births
Change in
questionnaire
Source: CBS, Onderzoekgezinsvorming,
1988, 1993, 1998, 2003, 2008
Not asked of
people with
difficulties
Sources: INED, Enquête fécondité 1988;
INED, French FFS 1994;
INED, Enquête intentions de fécondité 1998;
Ined-Insee French GGS 2005;
INED-INSERM, Enquête Fecond 2011.
Conclusions & suggestions
Comparison across countries
and over time
In GGS, how far can we go in cross-country
comparisons?
• For short term studies seems OK (apart from NO, NL and
HU)
• But for long term studies, rather chaotic
Need to be more systematic in monitoring intentions
data
e.g. the change in formulation between FFS and GGS makes it
impossible to look at the change over time
• Use simple questions
• Avoid too many filters
Other remarks and discussion
About the filters
• Each question costs money
• Fecundity : can we ask a woman whether she wants children
when she said she thinks that she can’t have more?
For aggregate family size
• Dependent on preceding-questions => the more
decomposition, the more potential issues.
• Better to
– either ask at first about the numbers, and then decompose the
timing
– or to ask as a summary question that can be understood more
independently from the preceding-questions, and without filter.
– (This question could be on the total number of children intended)
– or/and to ask other questions!!!
And thinking especially to persons of your
background, and with equivalent resources,
what is the ideal number of children in a family?
Surveys: INED, Enquêtes conjoncture 1955, 1967, 1976, 1978, 1982, 1987; INED, Enquête intentions de
fécondité 1998; INED, Enquête fécondité 1988; INED, French FFS 1994 ; Ined-Insee French GGS 2005;
INED-INSERM, Enquête Fecond: 2011
Acknowledgments
This research was supported by the European
Research Council under the European Union’s
Seventh Framework Programme (FP7/2007-2013) /
ERC Grant agreement n° 284238 (EURREP).
Web: www.eurrep.org