Get documents about "AS ECONDARY ANALYSIS OF DATA MID CAREER FELLOWSHIP
Document Sample
EMPLOYMENT STATUS AND HEALTH
TRAJECTORIES
Gopalakrishnan Netuveli
Imperial College London
1 Jan 2007 – 31 March 2008
Leeds, 18 March 2008
Employment and health
“… there is a strong theoretical case,
supported by a great deal of background
evidence, that work and paid employment
are generally beneficial for physical and
mental health and well-being.” Waddel and
Burton, 2006.
Debate: selection vs. causation
“…there is a strong case for all health
strategies to consider employment as an
outcome, where appropriate. There is also
a strong case for employment policy to
evaluate the health impact of all its
relevant interventions.” McLean et al. 2005
Leeds, 18 March 2008
Problem with the direction of causation
Employment and health may mutually
influence each other and the direction of
causation might depend on context and
contingency.
This makes the relationship between
employment and health complex.
Data form that might capture context and
contingency is longitudinal trajectories.
Study of trajectories might help to
understand part of this complexity
Leeds, 18 March 2008
Objective
To explore trajectories of health and employment in
a sample of BHPS
Data: 2852 subjects between 16 and 50 years in 1991
who were employed and reported no health
problems and self-rated health as good or better
Employment trajectory: 1 =(self) employed; 0=Else
Health trajectory: 1 = Good or better SRH; 0=Else
W9 & 14 excluded > SRH question different
Leeds, 18 March 2008
Methods
1. Summarising trajectories: are there
classes of trajectories?
Latent Class Growth Analysis
W1 W14
1
1
13
0
S
I
Age
C
Sex
Leeds, 18 March 2008
Methods
2. Comparing employment and health
trajectories within individuals: are
trajectories of health and employment
similar?
Leeds, 18 March 2008
Measuring similarity: requirements
Both trajectories coded similarly
Same number of states in each point of each
trajectory
The states coded similarly have the same
relative position in the vector of states for
each trajectory
Present study:
Codes
Trajectory 0 1
Health Health problems present No health problems
Employment Out of employment In employment
Leeds, 18 March 2008
Methods contd…
Common distance measures of similarity:
Euclidean, Hamming, Levenshtein
Present study: a new approach using
Kolmogorov-Smirnov D – statistic
D-statistics is the maximum distance between the cumulative
fractile/percentile distribution of the two trajectories.
A significant test for the H0:D=0 can be done (if necessary
exact test accounting for small number of points)
The individual P-values can be combined using meta-
analysis, even adjusting for co-variates using meta-
regression
It is also possible to identify which distribution ‘dominates’
Applications used: Mplus, STATA
Leeds, 18 March 2008
Results
Distribution of the sample according to W1 age and sex
Age group Men Women All
16-30 564 456 1020
31-40 425 320 745
41-50 587 500 1087
All 1567 1276 2852
Leeds, 18 March 2008
Results
Distribution according to W1 social class
Leeds, 18 March 2008
Employment trajectories
Latent Class Growth Analysis of employment
trajectory identified 7 classes. Classification
forced to stop at 7 when number of people
in any class fell below 5%
Employment trajectories Freq. Percent AUC*
Immediate drop 446 15.48 0.06
Early rectangle 353 12.25 0.21
Early drop - slow decline 137 4.76 0.37
Middle rectangle 237 8.23 0.46
Early drop- recovery 264 9.16 0.68
Late rectangle 247 8.57 0.72
Persisting 1,197 41.55 0.91
*AUC Average proportion of person-time in employment
Leeds, 18 March 2008
Employment trajectories
Leeds, 18 March 2008
Age and sex distribution of employment
trajectories
Leeds, 18 March 2008
Social class distribution of employment
trajectories by sex
Leeds, 18 March 2008
Propensity to different types of employment
trajectories
Narrative description of a multinomial logistic
regression:
Employment trajectories Characeristics
Immediate drop >30years, manual class
Early rectangle not 31-40 (might be manual class)
Early drop - slow decline Women <31or >40 years manual class
Middle rectangle >40 years
Early drop- recovery Women <31 years manual class
Late rectangle <30 or >40 years
Persisting Reference category
Leeds, 18 March 2008
Health trajectories
LCGA identified 6 classes. Classification
stopped when there was no significant
statistical difference between six and seven
class solutions
Health trajectories N % AUC* Characteristics
Immediate drop 376 15.06 0.06 <31 years manual class
Early rectangle 247 9.89 0.24 <31 years
Early drop - slow decline 173 6.93 0.30 Women >40 years manual class
Late rectangle 248 9.93 0.53 (<31 or >40 years manual class)
Early drop- recovery 402 16.1 0.54 Women <31 or >40 years manual class
Persisting 1,051 42.09 0.75 Reference
*AUC Average proportion of person-time in employment
Leeds, 18 March 2008
Cross-tabulation of health and employment
trajectories
Health
Immediate Early Early drop Early drop- Late
Employment drop rectangle - slow dec recovery rectangle Persisting All
Immediate drop 307 12 4 3 8 3 337
Early rectangle 50 191 9 10 4 8 272
Early drop - slow dec 2 1 37 24 28 19 111
Middle rectangle 6 30 23 81 12 22 174
Early drop- recovery 1 1 24 12 90 121 249
Late rectangle 3 3 23 70 32 73 204
Persisting 7 9 53 48 228 805 1150
All 376 247 173 248 402 1051 2497
Chi-square= 3994; df=30 p-value: <0.0001
Pivotal cells contributing to greatest to chi-square
Correlation between trajectories: 0.8
Leeds, 18 March 2008
Are the health and employment trajectories
within indivuals similar?
Leeds, 18 March 2008
Meta-analysis of p-values: full and subgroups
Groups D-statistics (95%CI) z-value p
All 0.28 (0.27 to 0.30) -9.040 1.000
Men, non-manual, 16-30 years 0.19 (0.15 to 0.23) -7.083 1.000
Men, non-manual, 31-40 years 0.26 (0.22 to 0.30) -4.307 1.000
Men, non-manual, 41-50 years 0.31 (0.27 to 0.34) -0.698 0.757
Men, manual, 16-30 years 0.21 (0.17 to 0.25) -7.327 1.000
Men, manual, 31-40 years 0.30 (0.24 to 0.35) -0.870 0.808
Men, manual, 41-50 years 0.36 (0.31 to 0.40) 1.823 0.034
Women, non-manual, 16-30 years 0.26 (0.22 to 0.29) -5.364 1.000
Women, non-manual, 31-40 years 0.26 (0.21 to 0.31) -4.912 1.000
Women, non-manual, 41-50 years 0.33 (0.29 to 0.36) 0.358 0.360
Women, manual, 16-30 years 0.31 (0.24 to 0.37) -0.403 0.657
Women, manual, 31-40 years 0.31 (0.25 to 0.38) -0.669 0.748
Women, manual, 41-50 years 0.33 (0.28 to 0.39) 0.215 0.415
Leeds, 18 March 2008
Distribution of Employment and health
trajectories in men, non-manual, 41-50 years
Health
Early
drop Early
Early slow Late drop Persistin
Employment rectangle decline rectangle recovery g Total
Middle rectangle 0 0 1 1 2 4
Early drop- recovery 0 1 0 1 7 9
Late rectangle 0 3 1 0 0 4
Persisting 1 3 5 19 41 69
Total 1 7 7 21 50 86
Pearson chi2(12) = 32.3433 Pr = 0.001
Leeds, 18 March 2008
Average D according to employment and
health latent classes
Health
Early Early
Immediat Early drop - drop- Late
Employment e drop rectangle slow dec recovery rectangle Persisting All
Immediate drop 0.02 . 0.07 . 0.65 0.36 0.31
Early rectangle 0.14 0.14 0.11 0.11 0.07 0.48 0.32
Early drop - slow dec . . 0.11 . 0.34 0.38 0.30
Middle rectangle 0.21 0.29 0.26 0.23 0.25 0.36 0.28
Early drop- recovery . 0.79 0.31 0.38 0.30 0.20 0.25
Late rectangle 0.64 0.64 0.39 0.26 0.32 0.18 0.26
Persisting 0.64 0.63 0.56 0.47 0.39 0.22 0.29
All 0.41 0.57 0.44 0.40 0.37 0.23 0.28
Emboldened: significant p-value after synthesis
Leeds, 18 March 2008
Conclusions
Are there classes of trajectories? YES
Are trajectories of health and employment
similar? YES for the majority (80%)
Selection or causation?
Weak evidence (if any) for causation
Leeds, 18 March 2008
Acknowledgements
ESRC and UPTAP programme
Professor David Blane, ICL
Professor Mel Bartley, UCL
Professor Richard Wiggins, IOE
Members of Q3 seminar group, Imperial
College London
Leeds, 18 March 2008
