Multilevel Modeling: A Real Life Example. True or False statement??Before I had children, I didn't have any gray hairs. NONE. Now I have a lot."MK (speaking to his children) . Multilevel Modeling ? Real Life Example. TRUE!But what's wrong with the statement?. Linear and Logistic Regression ?
1. Extreme Makeover -- Data Edition: Outside the Box
2. Multilevel Modeling: A Real Life Example True or False statement?
“Before I had children, I didn’t have any gray hairs. NONE. Now I have a lot.”
MK (speaking to his children) I would substitute “age increment” for “the age of the children” in the first bullet.I would substitute “age increment” for “the age of the children” in the first bullet.
3. Multilevel Modeling – Real Life Example TRUE!
But what’s wrong with the statement? I would substitute “age increment” for “the age of the children” in the first bullet.I would substitute “age increment” for “the age of the children” in the first bullet.
4. Linear and Logistic Regression – Review linear model review:
logistic model review:
5. First Model Y = ß0 + ß1X1 + e
Where Y = # of gray hairs on MK’s head
ß1X1 = The presence of children (the exposure of interest) I would substitute “age increment” for “the age of the children” in the first bullet.I would substitute “age increment” for “the age of the children” in the first bullet.
6. Second Model Y = ß0 + ß1X1 + ß2X2 + ß3X3 + ß4X4 + e
Where Y = # of gray hairs on MK’s head
ß1X1 = The presence of children (the exposure of interest)
ß2X2 = The number of children (the covariate or a variable that needs to be controlled for)
ß3X3 = His age (a covariate)
ß4X4 = Marital status (a covariate)
I would substitute “age increment” for “the age of the children” in the first bullet.I would substitute “age increment” for “the age of the children” in the first bullet.
7. BUT… … notice that all those factors are individual-level factors (age, number of children, marital status).
What if there are broader factors influencing the number of gray hairs between 1989 and 2007?
8. Multilevel Modeling:Definition and Synonyms Multilevel modeling: a method that allows researchers to investigate the effect of group or place characteristics on individual outcomes while accounting for non-independence of observations
synonyms: different models:
multilevel models - fixed effects
contextual models - random effects
hierarchical analysis - generalized estimating equations
9. Why Use Multilevel Models? outcomes may be clustered by some unit of aggregation (contextual unit)
individuals within contexts may be similar in ways that are unmeasured
to take into account clustering / non-independence of observations
to partition the observed variability into within-context and between- context variables
10. Why Context Matters empirically, individual outcomes can’t be explained exclusively by individual-level exposures
persistent contextual effects are observed in all (?) outcomes across populations
exposures are structured; distributions are differential
11. Beyond Individual Determinants of Health
12. Beyond Individual Determinants of Health: Multilevel Analyses
13. Different from Ecological Analyses
14. Area and Health Area—states, counties, cities, neighborhoods--acts as a source of adverse or protective exposures and factors impacting health, such as:
Economic well-being—jobs, unemployment, economic development
Stressors—physical, economic, segregation, “social disorganization”
social capital/social cohesion
proximity to (competition for) resources (goods, services, transportation, employment opportunities, recreation)
15. Other Considerations Unit of analysis: states, counties, zip codes, census tracts, census block groups, etc.
Should “neighborhood” be defined using to geographic boundaries? If so, which one?
Data sources—going beyond census
Characteristics to examine, need rationale for each indicator
16. Sources of Data
17. Direct Mechanisms Community Social Environment
Social relationships? transmit information
Neighborhood cohesion?social control
Shared cultural norms and values
Civic participation? demand services
Access to education and employment
Health care facilities
18. Indirect Mechanisms Chronic Stress
Selye, 1956 “General Adaptive Response”
Alarm, resistance, exhaustion.
Repeated cycles lead to cumulative damage to organism.
McEwen & Stellar, 1993—Allostatic Load
The cost of maintaining stability through change
19. Neighborhood Effects: Evidence Community context consistently has a
“modest association” with numerous health
25 studies reviewed
Individual-level attributes controlled for
23/25 had significant neighborhood effects
Reviewed in Pickett & Pearl, J. Epidemiol Community Health, 2001; 55
20. What We Know About Neighborhoods and Child Well-Being Protective
Incivilities (physical & social
21. Are We Studying the Right Factors for Policy & Program Purposes? Protective
Safe public spaces
Services & resources
Political activity & support
Poor availability of services (health & social)
Norms (health and other)
Areas for play/social interaction
Poor housing quality
22. Neighborhoods & Child Well-Being: Modeling
23. Neighborhoods & Child Well-Being: Modeling
24. Adding Group-Level Variables problem: making cross-level inferences [drawing inferences regarding factors associated with variability in outcome at one level based on data collected at another level]
e.g., making individual inferences based on group-level associations
25. When are Observations Not Independent? when data are collected by cluster / aggregating unit
children within schools
patients within hospitals
drug users within neighborhoods
cholesterol levels within a patient
why care about clustered data?
two children / observations within one school are probably more alike than two children / observations drawn from different schools
does knowing one outcome inform your understanding about another outcome?
26. Back to the Example… Potential aggregate factors to consider:
In 2000, MK moved to a neighborhood with a very high percentage of lawyers;
In 2003, the Red Sox lost in the 7th game of a playoff series to the Yankees, blowing a three run lead in the 8th inning. (MK threw pillow at the tv, and was properly chastised for that).
27. Final Outcome of Real-Life Example (Assume Two-Level Outcome)
28. Acknowledgments Lynne Messer, PhD, University of North Carolina
Pat O’Campo, PhD, University of Toronto
Jennifer Culhane, PhD, Drexel University
29. Contact Information Michael Kogan, Ph.D.
Director, Office of Data and Program Development
5600 Fishers Lane, Room 18-41
Rockville, MD 20857