Simultaneous Quantile Regression

1
Simultaneous Quantile Regression

• William Smith
• EPSSA Methods Workshop
• 4/11/13

2
Introduction to Research Project

• The Non-linear effects of social capital on
  occupational prestige.
• Social capital is important in occupational
  attainment.
• Hints of non-linearity
• Informal channels are more effective early in
  careers (Flap Boxman 2000).
• Ceiling effect of weak ties (Lin 1999).
• Females are limited by overreliance on strong
  ties (Moore 1990).

3
Research Questions

1. How do the effects of social capital differ
   across occupational prestige levels?
2. How do the effects of social capital differ by
   gender?

4
Method Selection Appropriateness

• Needed to test non-linearity
• Simultaneous Quantile Regression
• Allow you to identify quantiles (percentiles)
  along a continuum.
• Provide linear projections for each quantile.
• Different projections at different points along
  the continuum.
• Can test for significant differences between
  projections (coefficients)

5
How is it different from OLS?

• Ordinary Least Square (OLS) and Ordinal Logistic
  Regression both provide a mean projection.
• Constant slope
• Acts like a linear relationship
• Since both are linear projections you can compare
  OLS with Simultaneous Quantile Regression
  coefficients.

6
Data

• 2001 International Social Survey Programme
• Focused on social relationships in 27 countries
• Sample
• Limited to ages 25-64 with recorded occupation
• Used country weights to create large sample that
  included participants in 21 countries
• All analysis done in STATA

7
Variable Preparation

• Occupational Prestige
• Available Social Capital
• Strong Weak Ties
• Interaction Terms

gen siop0 replace siop63 if isco881141
isco881142 isco881143 isco881220
gen scjob. replace scjob1 if v461 v462
v463 replace scjob2 if v464 replace scjob0
if v465 v466 v467 v468 v469 v4610
ASC gen scnumb v4r v8r v23r v24r
v25r gen scstrength v7 v11 v13 v15 v17
v18 v19 v20 v21 v28 gen sctotal
scnumb scstrength
gen femwtiefemalewtiejob
8
OLS Regression Syntax and Output

• Full Regression Model
• OLS Output
• See handout

reg siop i.country female age35_44 age45_54
age55_64 married educyrs sctotal /// primary
secondary higher wtiejob stiejob femwtie femstie
pwweight, cluster (country)
9
Simultaneous Quantile Regression Syntax

• Full Simultaneous Quantile Regression Model
• Simultaneous Quantile Regression Output
• See handout

sqreg siop australia germany greatbritain hungary
norway czechrep poland russia /// newzealand
canada phillipines japan spain latvia cyprus
chile denmark switzerland brazil /// finland
female age35_44 age45_54 age55_64 married educyrs
sctotal primary secondary /// higher wtiejob
stiejob femwtie femstie, q(.1 .3 .5 .7 .9)
10
Comparison Table
11
Available Social Capital
12
Job Search Channel - Female
13
Comparing Sexes
14
Checking for Significant Differences

• Check for non-linearity
• Is the difference between the high and low point
  (coefficient) statistically different than zero?

test q30sctotalq90sctotal test
q50stiejobq90stiejob test q30wtiejobq90w
tiejob test q50femstieq90femstie test
q50femwtieq90femwtie
15
Questions? Collaborations?

• William C. Smith
• Education Theory and Policy
• Comparative International Education
• wcs152_at_psu.edu