Top Banner
Analysis of propensity-score matched data HHS Methods Meeting 2012 Felix Thoemmes 1
21

Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

May 22, 2020

Download

Documents

dariahiddleston
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Analysis of propensity-score matched data

HHS Methods Meeting 2012 Felix Thoemmes

1

Page 2: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Propensity scores and regression

1. Propensity scores are a principled approach to causal inference, regression adjustment is not

2

Page 3: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

(𝑌0,𝑌1 ) ⊥ 𝑇|𝑋 Ignorability

Propensity scores and regression

1. Propensity scores are a principled approach to causal inference, regression adjustment is not

3

Both rely on same assumption of no unmeasured confounders

Page 4: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Propensity scores and regression

2. Propensity scores are outcome-free and are not victim to (unintentional) biases

4

Page 5: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Propensity scores and regression

2. Propensity scores are outcome-free and are not victim to (unintentional) biases

5

In ideal study in which outcomes and covariates are clearly separated avoidance of bias is possible

Page 6: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Propensity scores and regression

3. Functional form assumptions of outcome relationship are more restrictive

6

Page 7: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Propensity scores and regression

3. Functional form assumptions of outcome relationship are more restrictive

7

Regression models do not have to be modeled linearly. Linearity violations can to some degree be assessed. Progress with data mining techniques.

Page 8: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Propensity scores and regression

4. Regression extrapolates, whereas propensity scores put focus on common support

8

Page 9: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Propensity scores and regression

4. Regression extrapolates, whereas propensity scores put focus on common support

9

Common support in multivariate space can be assessed without propensity scores (convex hull).

Page 10: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Propensity scores and regression

Regression adjustment

≈ Propensity score matching

10

Page 11: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Effects of military training

• Big Five personality traits are conducive to change

• Important life events can impact personality development

• Jackson, J., Thoemmes, F., Jonkmann, K., Lüdtke, O., & Trautwein, U. (2012). Military training and personality trait development: Does the military make the man or does the man make the military? Psychological Science.

11

Page 12: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Effects of military training

12 Photo Credit: Stock Photograph by Department of Defense Public Domain. Copyright: United States Government Works

Page 13: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Effects of military training

• Until very recently, 18 year old men in Germany had to do either civil service or military training

• Any impact of military training on personality that is observed is likely confounded by selection effects

13

Page 14: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Effects of military training

• Large sample (~2,000) of college-bound youth were equated using PS matching on 38 covariates

• Pre-tests • Socio-economic status & cultural capital • Academic achievement • Vocational interests • …

14

Page 15: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

15

Selection effects (Differences t1)

Page 16: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

16

Effects after matching (Differences t2)

Page 17: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

17

Effects after adjustment (Differences t2)

Page 18: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Longitudinal effects of military training

18 Latent growth model after propensity score matching

Page 19: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Summary

• Effects of military training on agreeableness are present after 1 year and even after 4 years, as assessed in latent growth model

• Potential unobserved confounders ?

19

Page 20: Analysis of propensity-score matched dataopremethodsmeeting.org/wp-content/uploads/2017/07/... · Analysis of propensity-score matched data HHS Methods Meeting 2012 ... were equated

Summary

• In this particular example regression adjustment ≈ propensity score matching

20