Too many zeros: using habitus to estimate long-term time

Too many zeros: using habitus to estimate
long-term time-use from day diaries.
Jonathan Gershuny CTUR May 2011
1. Objective
6. Key insight
9. Three steps to LTTU estimates
Combine diary and questionnaire measures
....to improve estimates of time use.
Need to estimate population time distributions.
Want to use diaries to measure time use, but low
participation in some ―rare‖ activities, in which
majority have zero on random day. Examples:
• Cultural participation
• Sport, purposive exercise
Stylised participation frequency questions imply
daily frequencies in respondents’ own diaries.
If the same respondents are asked participation
rate questions, and on assumption of random
daily sampling, it is then possible to use the
diary to calibrate questionnaire answers.
1.
2. The Problem
Activities vary, day by day.
Need long term time-use (LTTU) estimates of:
• Gender divisions of labour
• Work/leisure inequality measures
• Exercise across populations
BUT long diaries  high respondent burden…
Shortmany non-participants in some activities.
3. Alternative measurement options
Types of “stylised” questionnaire items:
• How often do you… (activity)?
• Who usually does the... (activity)?
• How much time do you spend in (activity)?
Major data quality issues:
• Observation or relevancy period?
• Activity definition, coverage, overlap
• Recall and desirability problems
Diaries avoid problems, but TOO MANY ZEROs
2.
3.
Victor Kipnis,and others “Modeling Data with Excess Zeros and
Measurement Error: Application to Evaluating Relationships between
Episodically Consumed Foods and Health Outcomes” , Biometrics 65,
1003–1010 December 2009
Question: How often do you go to the pub?
Expected
Actual observations day diary
participation
rate
participation
rate
participants’
mean
minutes
subsample
mean
minutes
Most days
> .50
.42
162
68
>=once a week
> .14
.25
141
35
>= once/month
> .03
.14
114
15
several times/year
< .03
.10
104
10
Once a year, less
< .003
.07
109
7
Answers:
8. Meaning of stylised/diary differences
―most days‖:
Actual (diary) lower than expected
―at least once a week‖:
Actual nearly twice as high as expected
―> once/month & several times/year‖:
3-4 times higher than expected
―once a year or less‖:
Actual 20 times higher than expected
...ie some, imprecise, expected/actual association
25
20
%
10
Truncate negative estimates at zero (very few cases,
only in care, paid work and exercise).
Adustment for total time < or > 1440 mins:
• Adj factor = (estimated total mins)/1440
• Mean 1.00; sd .02; min .90; max 1.10
• Adjusted act time= (act time)/(adj factor)
Active sports and exercise (decile percentages): single day estimate
Pierre Bourdieu Distinction London RKP 1985
managers medical &
education
11. Variables
Dependents 33 activity variables =1440 mins
Control variables (C1): age, age squared, sex,
marital status, carer/family status, educ.
attainment, emp. stat, diary day-of-week
Habitus variables: 18 ―how-many-times-lastmonth?‖ variables, normal paid work hours.
Logistic regression: predict daily particip. prob (P)
OLS regression: predict daily participants’ time for
all respondents (t)
Long term time use = P * t
Distribution by population decile:
other professions
20
15
10
5. Habitus (or just habit)
Bourdieu in Distinction: rational use of economic,
social, cultural capitals to maximise
―distinctiveness‖.
Recursive relationship habits and ―performance‖:
the more you do it the better it gets.
―Habitus‖ is the individual’s distinctive style of life
constituted by the long-term mix of activities.
Tn = f(C1…Ci, Ha,Hb,Hc... Hk)
Harmonised European Time Use Study (HETUS)
1 weekday, 1 weekend, whole household age>10
UK: 5K Households, 10K resps., 20K diaries.
Primary & secondary activities, where, who with
40 leisure and exercise ―habit‖ estimate questions:
• ― How often did you (activity) in last 4 weeks?‖
%
15
…….
13. Minor Adjustment
25
sports
hi culture
pop culture
non-tv leisure
tv leisure
Ta is diary estimate of minutes in activity a,
Ha is q’naire participation estimate for activity a,
Note: each habit answer relates to all acts since:
Ta + Tb + Tc +…. Tn =1440 minutes
Hence we may estimate:
Ta = f(C1…Ci, Ha,Hb,Hc…Hk)
Tb = f(C1…Ci, Ha,Hb,Hc... Hk)
10. Data: UK HETUS 2000
Distribution by population decile:
medical and educational professions
4. Intuition
Short diaries give evidence of differential
participation probabilities.
Variation relates to socio-ec . characteristics and...
…also to individual “tastes” or “habits”.
Habits, indicated by participation frequencies
and other similar measures, may be combined
with diary data to produce longer term estimates.
Logistic regression estimates respondents’
predicted participation probabilities
OLS estimations of participants’ time in each
activity from diaries; generate predicted
participants time for all respondents.
Product of predicted daily participation
probabilities and participants’ time gives
individual long-term means
12. Estimation using habitus
5
5
0
0
sports
hi culture
pop culture
non-tv leisure
tv leisure
other
profs
clerical secur, sales
& self emp assembly
farm &
forest
no occ
N
decile 1
2
3
4
5
6
7
8
9
10
11
11
10
9
13
9
1699
decile 10
11
8
13
9
9
17
10
1697
100
100
100
100
100
100
100
1700
1699
1697
1700
No information!
1700
1699
1699
1699
Active sports and exercise (decile percentages): long term estimates
managers medical &
other
clerical
secur, sales
farm &
profs
4
& self emp
12
assembly
11
forest
11
education
20
10
9
16
12
4
6
13
13
11
11
9
8
10
9
7
9
13
11
9
9
7
7
10
11
11
7
12
12
100
decile 1
2
3
4
5
6
7
8
9
decile 10
no occ
3
N
1699
1
6
8
1699
1698
11
10
2
2
10
13
1698
1700
10
8
10
11
6
7
13
14
1698
1699
14
6
10
11
13
1699
7
5
15
20
7
7
8
7
31
41
11
10
1697
1699
100
100
100
100
100
100
14. Information about activity distributions
Sports example shows, everyone does some
purposive exercise, but doctors and teachers do
less than average (also less high culture—both
reflect long work hours)
This information would not emerge from either
diary or stylised alone, LTTU estimates require
both to be collected for the same respondents.
http://www.timeuse.org/