Direct and indirect effects in a logit model


 Joan June Snow
 1 years ago
 Views:
Transcription
1 Department of Social Research Methodology Vrije Universiteit Amsterdam
2 Outline The aim
3 The Total Effect X Y
4 The Total Effect parental class attend college
5 The Indirect Effect Z a b X Y
6 The Indirect Effect during high school a b parental class attend college
7 The Direct Effect Z a b X c Y
8 The Direct Effect during high school a b parental class c attend college
9 The aim Z The aim is to find the size of the indirect effect relative to the total effect. a b X c Y
10 Outline The aim
11 Estimation When using regress: 1. college = class + 2. college = class
12 Estimation When using regress: 1. college = class + 2. college = class The direct effect is the effect of class in model 1.
13 Estimation When using regress: 1. college = class + 2. college = class The direct effect is the effect of class in model 1. The total effect is the effect of class in model 2.
14 Estimation When using regress: 1. college = class + 2. college = class The direct effect is the effect of class in model 1. The total effect is the effect of class in model 2. The indirect effect is the total effect  direct effect.
15 Estimation When using regress: 1. college = class + 2. college = class The direct effect is the effect of class in model 1. The total effect is the effect of class in model 2. The indirect effect is the total effect  direct effect. This won t work when using logit
16 Why the naive method doesn t work Easiest explained when there is no indirect effect.
17 Why the naive method doesn t work Easiest explained when there is no indirect effect. The total effect = the direct effect + the indirect effect.
18 Why the naive method doesn t work Easiest explained when there is no indirect effect. The total effect = the direct effect + the indirect effect. So, the total effect should be the same as the direct effect when there is no indirect effect.
19 Why the naive method doesn t work Easiest explained when there is no indirect effect. The total effect = the direct effect + the indirect effect. So, the total effect should be the same as the direct effect when there is no indirect effect. So, the effect of class in a model that controls for (the direct effect ) should be the same as the effect of class in a model that does not control for (the total effect ).
20 Effect while controlling for log odds proportion high medium low 1.5 transformation controlled 3 proportion high status log odds log odds low status proportion effect controlled
21 Averaging the proportions 3 log odds proportion high medium low not controlled 1.5 transformation controlled 3 proportion high status log odds log odds low status proportion effect controlled
22 Effect while not controlling for 3 log odds proportion high status log odds log odds low status proportion proportion high medium low not controlled transformation controlled not constrolled effect controlled not constrolled
23 Outline The aim
24 Indirect effect present log odds prop. high status log odds log odds low status prop proportion high medium low not controlled transformation controlled not constrolled effect controlled not constrolled
25 Indirect effect 3 log odds prop. high status log odds indirect effect log odds low status prop proportion factual high medium low not controlled counterfactual high low not controlled
26 Direct effect 3 log odds prop. high status log odds direct effect log odds low status prop proportion factual high medium low not controlled counterfactual high low not controlled
27 Direct and indirect effects in logit 3 log odds proportion high status log odds indirect effect direct effect total effect log odds low status proportion proportion factual high medium low not controlled counterfactual high low not controlled
28 The logic can be reversed 3 log odds total effect direct effect proportion factual high medium low not controlled prop. high status log odds indirect effect log odds low status prop. counterfactual high low not constrolled
29 Extension Erikson et al. (2005) propose to compute the average proportions given the observed and counterfactual distribution of by assuming that is normally distributed, and then integrate over this normal distribution.
30 Extension Erikson et al. (2005) propose to compute the average proportions given the observed and counterfactual distribution of by assuming that is normally distributed, and then integrate over this normal distribution. Alternatively, these averages can be computed by predicting the observed and counterfactual proportions, add them up and divide by the number of respondents in that group.
31 Extension Erikson et al. (2005) propose to compute the average proportions given the observed and counterfactual distribution of by assuming that is normally distributed, and then integrate over this normal distribution. Alternatively, these averages can be computed by predicting the observed and counterfactual proportions, add them up and divide by the number of respondents in that group. The latter method has the advantage of making less assumptions about the distribution of, as it integrates over the empirical distribution of instead of over a normal distribution.
32 Outline The aim
33 Descriptives. table ocf57 if!missing(hsrankq, college), /// > contents(mean college mean hsrankq freq) /// > format(%9.3g) stubwidth(15) occupation of r father in 1957 mean(college) mean(hsrankq) Freq. lower ,218 middle higher ,837
34 The ldecomp package ldecomp depvar [ if ] [ in ] [ weight ], direct(varname) indirect(varlist) [ obspr predpr predodds or rindirect normal range(##) nip(#) interactions nolegend nodecomp nobootstrap bootstrap_options ]
35 Decomposition of log odds ratios. ldecomp college, direct(ocf57) indirect(hsrankq) rind nolegend (running _ldecomp on estimation sample) Bootstrap replications (50) Bootstrap results Number of obs = 8923 Replications = 50 Observed Bootstrap Normalbased Coef. Std. Err. z P> z [95% Conf. Interval] 2/1 total indirect direct indirect direct /1 total indirect direct indirect direct /2 total indirect direct indirect direct
36 Relative effects 2/1r 3/1r 3/2r method method average method method average method method average
37 Decomposition of odds ratios. ldecomp college, direct(ocf57) indirect(hsrankq) or nolegend (running _ldecomp on estimation sample) Bootstrap replications (50) Bootstrap results Number of obs = 8923 Replications = 50 Observed Bootstrap Normalbased Odds Ratio Std. Err. z P> z [95% Conf. Interval] 2/1 total indirect direct indirect direct /1 total indirect direct indirect direct /2 total indirect direct indirect direct
38 Does it matter? Table: Comparing different estimates of the size of indirect effect relative to the size of the total effect generalization (Erikson et al. 2005) naive middle v. low method method average high v. low method method average high v. middle method method average
39 Discussion This is an area of active research
40 Discussion There are unanswered questions:
41 Discussion There are unanswered questions: The need to take the average indirect effect is less than elegant.
42 Discussion There are unanswered questions: The need to take the average indirect effect is less than elegant. How does it relate to the alternative method proposed by Fairlie (2005) and implemented by Ben Jann as the fairlie package?
43 References Buis, M. L.. Erikson, R., J. H. Goldthorpe, M. Jackson, M. Yaish, and D. R. Cox. On class differentials in educational attainment. Proceedings of the National Academy of Science, 102: , Fairlie, R. W. An extension of the BlinderOaxaca decomposition technique to logit and probit models. Journal of Economic and Social Measurement, 30: , 2005.
Calculating EffectSizes
Calculating EffectSizes David B. Wilson, PhD George Mason University August 2011 The Heart and Soul of Metaanalysis: The Effect Size Metaanalysis shifts focus from statistical significance to the direction
More informationInteraction effects between continuous variables (Optional)
Interaction effects between continuous variables (Optional) Richard Williams, University of Notre Dame, http://www.nd.edu/~rwilliam/ Last revised February 0, 05 This is a very brief overview of this somewhat
More informationSocioeconomic gaps in HE participation: how have they changed over time?
Socioeconomic gaps in HE participation: how have they changed over time? IFS Briefing Note BN133 Claire Crawford Executive summary Socioeconomic gaps in HE participation: how have they changed over time?
More informationMarry Your Like: Assortative Mating and Income Inequality
Marry Your Like: Assortative Mating and Income Inequality By JEREMY GREENWOOD, NEZIH GUNER, GEORGI KOCHARKOV AND CEZAR SANTOS Has there been an increase in positive assortative mating on the marriage market
More informationEffect Sizes. Null Hypothesis Significance Testing (NHST) C8057 (Research Methods 2): Effect Sizes
Effect Sizes Null Hypothesis Significance Testing (NHST) When you read an empirical paper, the first question you should ask is how important is the effect obtained. When carrying out research we collect
More informationIs There a Glass Ceiling in Sweden?
Is There a Glass Ceiling in Sweden? James Albrecht, Georgetown University Anders Björklund, Swedish Institute for Social Research (SOFI), Stockholm University Susan Vroman, Georgetown University Using
More informationChapter 18. Effect modification and interactions. 18.1 Modeling effect modification
Chapter 18 Effect modification and interactions 18.1 Modeling effect modification weight 40 50 60 70 80 90 100 male female 40 50 60 70 80 90 100 male female 30 40 50 70 dose 30 40 50 70 dose Figure 18.1:
More informationXPost: Excel Workbooks for the Postestimation Interpretation of Regression Models for Categorical Dependent Variables
XPost: Excel Workbooks for the Postestimation Interpretation of Regression Models for Categorical Dependent Variables Contents Simon Cheng hscheng@indiana.edu php.indiana.edu/~hscheng/ J. Scott Long jslong@indiana.edu
More informationNational Evaluation of Student Support Services: Examination of Student Outcomes After Six Years
U.S. DEPARTMENT OF EDUCATION National Evaluation of Student Support Services: Examination of Student Outcomes After Six Years Final Report National Evaluation of Student Support Services: Examination of
More informationIntroduction to. Hypothesis Testing CHAPTER LEARNING OBJECTIVES. 1 Identify the four steps of hypothesis testing.
Introduction to Hypothesis Testing CHAPTER 8 LEARNING OBJECTIVES After reading this chapter, you should be able to: 1 Identify the four steps of hypothesis testing. 2 Define null hypothesis, alternative
More informationESTIMATING THE PAYOFF TO ATTENDING A MORE SELECTIVE COLLEGE: AN APPLICATION OF SELECTION ON OBSERVABLES AND UNOBSERVABLES*
ESTIMATING THE PAYOFF TO ATTENDING A MORE SELECTIVE COLLEGE: AN APPLICATION OF SELECTION ON OBSERVABLES AND UNOBSERVABLES* STACY BERG DALE AND ALAN B. KRUEGER Estimates of the effect of college selectivity
More informationNumber boards for mini mental sessions
Number boards for mini mental sessions Feel free to edit the document as you wish and customise boards and questions to suit your learners levels Print and laminate for extra sturdiness. Ideal for working
More informationFindings from the Michigan School Readiness Program 6 to 8 Follow Up Study
Findings from the Michigan School Readiness Program 6 to 8 Follow Up Study October, 2007 Elena Malofeeva, Marijata DanielEchols, and Zongping Xiang High/Scope Educational Research Foundation 600 North
More informationHOW SOCIAL TIES AFFECT PEERGROUP EFFECTS: A CASE OF UNIVERSITY STUDENTS
Oleg Poldin, Dilyara Valeeva, Maria Yudkevich HOW SOCIAL TIES AFFECT PEERGROUP EFFECTS: A CASE OF UNIVERSITY STUDENTS BASIC RESEARCH PROGRAM WORKING PAPERS SERIES: SOCIOLOGY WP BRP 15/SOC/2013 This Working
More informationTHE DETERMINANTS OF THE GLOBAL DIGITAL DIVIDE: A CROSSCOUNTRY ANALYSIS OF COMPUTER AND INTERNET PENETRATION
ECONOMIC GROWTH CENTER YALE UNIVERSITY P.O. Box 208269 New Haven, CT 065208269 http://www.econ.yale.edu/~egcenter/ CENTER DISCUSSION PAPER NO. 881 THE DETERMINANTS OF THE GLOBAL DIGITAL DIVIDE: A CROSSCOUNTRY
More informationTeacher Attrition and Mobility:
NCES 2014077 U.S. DEPARTMENT OF EDUCATION Teacher Attrition and Mobility: Results From the 2012 13 Teacher Followup Survey First Look Teacher Attrition and Mobility: Results From the 2012 13 Teacher
More informationDesign of Experiments (DOE)
MINITAB ASSISTANT WHITE PAPER This paper explains the research conducted by Minitab statisticians to develop the methods and data checks used in the Assistant in Minitab 17 Statistical Software. Design
More informationThe changing social patterning of obesity: an analysis to inform practice and policy development
The changing social patterning of obesity: an analysis to inform practice and policy development Final report to the Policy Research Programme, Department of Health Version 2, 5 th November 2007 Martin
More informationMICROLEVEL ESTIMATION OF POVERTY AND INEQUALITY. By Chris Elbers, Jean O. Lanjouw, and Peter Lanjouw 1. 1 introduction
Econometrica, Vol. 71, No. 1 (January, 2003), 355 364 MICROLEVEL ESTIMATION OF POVERTY AND INEQUALITY By Chris Elbers, Jean O. Lanjouw, and Peter Lanjouw 1 1 introduction Recent theoretical advances have
More informationMissing Data & How to Deal: An overview of missing data. Melissa Humphries Population Research Center
Missing Data & How to Deal: An overview of missing data Melissa Humphries Population Research Center Goals Discuss ways to evaluate and understand missing data Discuss common missing data methods Know
More informationWe show that social scientists often do not take full advantage of
Making the Most of Statistical Analyses: Improving Interpretation and Presentation Gary King Michael Tomz Jason Wittenberg Harvard University Harvard University Harvard University Social scientists rarely
More informationWhich Immigrants Are Most Innovative and Entrepreneurial? Distinctions by Entry Visa
DISCUSSION PAPER SERIES IZA DP No. 4745 Which Immigrants Are Most Innovative and Entrepreneurial? Distinctions by Entry Visa Jennifer Hunt February 2010 Forschungsinstitut zur Zukunft der Arbeit Institute
More informationRecall this chart that showed how most of our course would be organized:
Chapter 4 OneWay ANOVA Recall this chart that showed how most of our course would be organized: Explanatory Variable(s) Response Variable Methods Categorical Categorical Contingency Tables Categorical
More informationInteraction effects and group comparisons Richard Williams, University of Notre Dame, http://www3.nd.edu/~rwilliam/ Last revised February 20, 2015
Interaction effects and group comparisons Richard Williams, University of Notre Dame, http://www3.nd.edu/~rwilliam/ Last revised February 20, 2015 Note: This handout assumes you understand factor variables,
More informationSupplementary PROCESS Documentation
Supplementary PROCESS Documentation This document is an addendum to Appendix A of Introduction to Mediation, Moderation, and Conditional Process Analysis that describes options and output added to PROCESS
More informationWere Minority Students Discouraged From Applying to University of California Campuses After the Affirmative Action Ban?
Were Minority Students Discouraged From Applying to University of California Campuses After the Affirmative Action Ban? Kate Antonovics Ben Backes UCSD June 2012 Abstract This paper uses studentlevel
More informationSF2.2: Ideal and actual number of children
Definitions and methodology SF.: Ideal and actual number of children Childbearing preferences are difficult to measure since they depend on different factors, including social norms, personal circumstances,
More informationBIS RESEARCH PAPER NO. 112 THE IMPACT OF UNIVERSITY DEGREES ON THE LIFECYCLE OF EARNINGS: SOME FURTHER ANALYSIS
BIS RESEARCH PAPER NO. 112 THE IMPACT OF UNIVERSITY DEGREES ON THE LIFECYCLE OF EARNINGS: SOME FURTHER ANALYSIS AUGUST 2013 1 THE IMPACT OF UNIVERSITY DEGREES ON THE LIFECYCLE OF EARNINGS: SOME FURTHER
More informationMarginal Effects for Continuous Variables Richard Williams, University of Notre Dame, http://www3.nd.edu/~rwilliam/ Last revised February 21, 2015
Marginal Effects for Continuous Variables Richard Williams, University of Notre Dame, http://www3.nd.edu/~rwilliam/ Last revised February 21, 2015 References: Long 1997, Long and Freese 2003 & 2006 & 2014,
More informationGrowth Is Good for the Poor
Forthcoming: Journal of Economic Growth Growth Is Good for the Poor David Dollar Aart Kraay Development Research Group The World Bank First Draft: March 2000 This Draft: March 2002 Abstract: Average incomes
More information