Technical Appendix: Childhood Family Structure and Schooling Outcomes: Evidence for Germany

Similar documents
Discrete Mathematics and Probability Theory Spring 2013 Anant Sahai Lecture 17

Discrete Mathematics and Probability Theory Summer 2014 James Cook Note 17

Satellite Retrieval Data Assimilation

1B40 Practical Skills

Genetic Programming. Outline. Evolutionary Strategies. Evolutionary strategies Genetic programming Summary

p-adic Egyptian Fractions

Derivations for maximum likelihood estimation of particle size distribution using in situ video imaging

Convert the NFA into DFA

Quantitative Genetics and Twin Studies

1 Online Learning and Regret Minimization

Review of Gaussian Quadrature method

Lecture 3 Gaussian Probability Distribution

u( t) + K 2 ( ) = 1 t > 0 Analyzing Damped Oscillations Problem (Meador, example 2-18, pp 44-48): Determine the equation of the following graph.

Chapter 4: Techniques of Circuit Analysis. Chapter 4: Techniques of Circuit Analysis

Parse trees, ambiguity, and Chomsky normal form

Non-Linear & Logistic Regression

Designing Information Devices and Systems I Discussion 8B

I1 = I2 I1 = I2 + I3 I1 + I2 = I3 + I4 I 3

Continuous Random Variables Class 5, Jeremy Orloff and Jonathan Bloom

Mathematics Number: Logarithms

4 7x =250; 5 3x =500; Read section 3.3, 3.4 Announcements: Bell Ringer: Use your calculator to solve

A signalling model of school grades: centralized versus decentralized examinations

Chapter 3 Single Random Variables and Probability Distributions (Part 2)

Student Activity 3: Single Factor ANOVA

Bases for Vector Spaces

How do we solve these things, especially when they get complicated? How do we know when a system has a solution, and when is it unique?

7.1 Integral as Net Change and 7.2 Areas in the Plane Calculus

Industrial Electrical Engineering and Automation

New Expansion and Infinite Series

Read section 3.3, 3.4 Announcements:

2.4 Linear Inequalities and Interval Notation

Tests for the Ratio of Two Poisson Rates

1 Probability Density Functions

Chapter 9: Inferences based on Two samples: Confidence intervals and tests of hypotheses

Bayesian Networks: Approximate Inference

Heavy tail and stable distributions

M344 - ADVANCED ENGINEERING MATHEMATICS

Physics 202H - Introductory Quantum Physics I Homework #08 - Solutions Fall 2004 Due 5:01 PM, Monday 2004/11/15

Riemann is the Mann! (But Lebesgue may besgue to differ.)

Conservation Law. Chapter Goal. 5.2 Theory

Section 3.1: Exponent Properties

Data Assimilation. Alan O Neill Data Assimilation Research Centre University of Reading

Quadratic Forms. Quadratic Forms

Acceptance Sampling by Attributes

Quantum Nonlocality Pt. 2: No-Signaling and Local Hidden Variables May 1, / 16

Continuous Random Variables

Reinforcement learning II

List all of the possible rational roots of each equation. Then find all solutions (both real and imaginary) of the equation. 1.

Introduction to Electronic Circuits. DC Circuit Analysis: Transient Response of RC Circuits

Lecture 2: January 27

CS 275 Automata and Formal Language Theory

Fig. 1. Open-Loop and Closed-Loop Systems with Plant Variations

APPROXIMATE INTEGRATION

Designing Information Devices and Systems I Spring 2018 Homework 7

SUMMER KNOWHOW STUDY AND LEARNING CENTRE

Section 4: Integration ECO4112F 2011

LAMEPS Limited area ensemble forecasting in Norway, using targeted EPS

Consequently, the temperature must be the same at each point in the cross section at x. Let:

UNIT 5 QUADRATIC FUNCTIONS Lesson 3: Creating Quadratic Equations in Two or More Variables Instruction

Chapter 4 Contravariance, Covariance, and Spacetime Diagrams

A BRIEF INTRODUCTION TO UNIFORM CONVERGENCE. In the study of Fourier series, several questions arise naturally, such as: c n e int

5 Probability densities

10 Vector Integral Calculus

Designing Information Devices and Systems I Anant Sahai, Ali Niknejad. This homework is due October 19, 2015, at Noon.

Review of Probability Distributions. CS1538: Introduction to Simulations

Chapter 6 Techniques of Integration

CHAPTER 1 PROGRAM OF MATRICES

2 b. , a. area is S= 2π xds. Again, understand where these formulas came from (pages ).

Joint distribution. Joint distribution. Marginal distributions. Joint distribution

Temperature influence compensation in microbolometer detector for image quality enhancement

Introduction to Algebra - Part 2

Robust Predictions in Games with Incomplete Information

Calculus Module C21. Areas by Integration. Copyright This publication The Northern Alberta Institute of Technology All Rights Reserved.

Sufficient condition on noise correlations for scalable quantum computing

Continuous Random Variable X:

Homework Assignment 3 Solution Set

Exam 2, Mathematics 4701, Section ETY6 6:05 pm 7:40 pm, March 31, 2016, IH-1105 Instructor: Attila Máté 1

The Shortest Confidence Interval for the Mean of a Normal Distribution

An Alternative Approach to Estimating the Bounds of the Denominators of Egyptian Fractions

Chapter 1: Logarithmic functions and indices

The Trapezoidal Rule

Probability Distributions for Gradient Directions in Uncertain 3D Scalar Fields

Math 113 Exam 2 Practice

Physics 116C Solution of inhomogeneous ordinary differential equations using Green s functions

The Minimum Label Spanning Tree Problem: Illustrating the Utility of Genetic Algorithms

Farey Fractions. Rickard Fernström. U.U.D.M. Project Report 2017:24. Department of Mathematics Uppsala University

Lecture Notes No. 10

Chapter 5 : Continuous Random Variables

5. (±±) Λ = fw j w is string of even lengthg [ 00 = f11,00g 7. (11 [ 00)± Λ = fw j w egins with either 11 or 00g 8. (0 [ ffl)1 Λ = 01 Λ [ 1 Λ 9.

Intermediate Math Circles Wednesday, November 14, 2018 Finite Automata II. Nickolas Rollick a b b. a b 4

Review of Calculus, cont d

Tutorial 4. b a. h(f) = a b a ln 1. b a dx = ln(b a) nats = log(b a) bits. = ln λ + 1 nats. = log e λ bits. = ln 1 2 ln λ + 1. nats. = ln 2e. bits.

The practical version

1.9 C 2 inner variations

ORDER REDUCTION USING POLE CLUSTERING AND FACTOR DIVISION METHOD

Chapter 7 Notes, Stewart 8e. 7.1 Integration by Parts Trigonometric Integrals Evaluating sin m x cos n (x) dx...

Reversals of Signal-Posterior Monotonicity for Any Bounded Prior

Properties of Integrals, Indefinite Integrals. Goals: Definition of the Definite Integral Integral Calculations using Antiderivatives

CS103B Handout 18 Winter 2007 February 28, 2007 Finite Automata

How can we approximate the area of a region in the plane? What is an interpretation of the area under the graph of a velocity function?

Transcription:

Technicl Appendix: Childhood Fmily Structure nd Schooling Outcomes: Evidence for Germny Mrco Frncesconi* Stephen P. Jenkins Thoms Siedler Universy of Essex nd Universy of Essex Universy of Essex Instute for Fiscl Studies nd DIW Berlin nd DIW Berlin Decemer 008

Identifiction Issues nd Estimtion Methods A centrl concern of this study is tht n estimted effect of childhood fmily structure on eduction my e spurious due to the mutul ssocion etween fmily structure nd children s schooling chievements nd some unmesured true cusl fctor. For exmple the ssocion etween hving experience of life in non-intct fmily nd lower eductionl ttinment my not e the result of fmily structure during childhood; differences in ttinment my simply reflect the chrcteristics of fmilies in which children of lone mothers re rought up. Our econometric strtegy is to pply numer of different techniques which plce different ssumptions on the dt in order to identify the effects of experiencing life in nonintct fmily during childhood. Ech of these techniques hs dvntges nd disdvntges. Although the methods differ they shre numer of common elements. For this reson in wht follows we strip down the corresponding sttisticl models in order to etter highlight the difference in their identifying restrictions. Siling Difference Model Let index fmilies nd i index young dults (or children). For convenience ssume tht the reltionship we estimte is (A.) S = β F + u where S is eduction F is vrile tht indictes childhood fmily structure (e.g. ever lived in non-intct fmily in the first 0 yers of life) nd u is rndom shock wh zero men. In eqution (A.) which for the moment excludes other determinnts of schooling β is the prmeter of interest. Consistent estimtion of β requires tht F e uncorrelted wh the disturnce term. u We investigte this issue using frmework suggested y Behrmn et l. (994) nd Rosenzweig nd Wolpin (995). Consider two-child fmily. For the i-th child in fmily wh siling k u cn e decomposed s follows: (A.) u = δ ε + δ ε k + ε + η In our empiricl nlysis the reltionship (A.) is expnded to include set of child- nd fmily-specific vriles tht my e fixed or time-vrying. In this formultion β is ssumed to e the sme for ll individuls. Arguly the effect of fmily structure is heterogeneous (i.e. some children might e etter off in non-intct fmily while others might e worse off). The siling difference pproch would pply even if one specifies rndom-coefficients model in which β = β + ζ nd E( ζ u ) = E( ζ X ) = 0. However my not e fesile to estimte such model ecuse repeted oservtions wh ech fmily re needed wheres most of the fmilies in our smple consist of only two or three silings.

whereε nd ε k re the endowments or ily of ech siling ε denotes the genetic endowments tht re common to oth children of fmily nd η is rndom shock tht is specific to i in inclusive of mesurement error in schooling. Endowments of oth silings re likely to e trnsmted cross genertions in Glton-type lw of herily (Becker nd Tomes 986): (A.) ε = ρε + ψ where ψ is child-specific idiosyncrtic disturnce wh zero men nd uncorrelted wh other unoservles (including ψ ik the corresponding rndom term for siling k). Assuming tht 0 ρ < implies tht endowments regress towrds the men cross genertions. Finlly the fmily structure vrile F is self function of unoserved vriles tht pertin to the fmily ( φ ) nd to the two silings ( μ nd μ k) : (A.4) F = γ μ + γ μ k + πφ + θ where θ is disturnce tht ffects F ut does not ffect S except indirectly through F. It is well known tht the prmeter β is not identified wh equtions (A.) (A.4) if π is not zero ρ is not zero nd if eher δ nd γ or δ nd γ re not zero even if orthogonly restrictions on the moments involving η ψ nd θ re imposed. Tht is β is estimted wh is if eqution (A.) is estimted cross individuls wh different vlues of fmily nd children s endowments. 4 The siling difference model estimtes β y compring eductionl outcomes mong silings ccording to whether they experienced life in non-intct fmily during childhood. In our two-child fmily cse the siling difference estimtor is computed from (A.5) ΔS = β ΔF + Δu where Δr = r r k for ny term r in eqution (A.5). The whin-fmily covrince etween fmily structure differences nd the disturnce term in (A.5) is thus given y (A.6) cov( ΔF Δu) = ( γ γ )( δ δ ) E( ΔμΔθ ) + ( γ γ ) E( ΔμΔη). Therefore sufficient condion for β to e identified is tht γ = γ ; tht is prents respond to their children s idiosyncrtic endowments eqully. A stronger condion would e to These would e the selection-on-oservles ssumptions which re relevnt for ll cross-sectionl estimtors including those sed on propensy score mtching methods. 4 This conclusion pplies to propensy score mtching estimtes since these too rely on the selection on oservles ssumption of the cross-sectionl model (A.) (A.4).

ssume tht children s endowments do not ffect prents ehviour (fmily structure) or lterntively there re no intrfmily responses (i.e. γ = γ 0). = 5 Before-After Comprisons nd Qusi-Experiments To see the ssumptions needed for identifiction in this cse we modify our empiricl frmework slightly nd explicly llow prents ehviour nd child unoservles to differ over time so tht: (A.7) F F + F nd u u + u wh the suscripts nd indicting some time period efore nd fter specific event occurs (e.g. deth of prent nd the introduction of divorce lw reform). Equtions (A.) nd (A.4) respectively ecome: (A.') u = δ ε + ε + ε + η τ τ τ τ nd (A.4') F = γ μ + μ + πφ + θ τ τ τ τ for τ =. A strightforwrd efore-fter comprison of outcomes for the sme individul requires repeted informtion on S which is prolemtic when schooling is mesured in term of highest eductionl ttinment. Bering this in mind nd imposing orthogonly restrictions on ll moments involving η τ ψ nd θ τ (A.) (A.') (A.) nd (A.4') will imply (A.8) fixed-effects estimtor sed on cov( Δ F Δu) = γ δ σ γ δ σ + γ δ σ γ δ σ where σ cov( p q) for p q = ε ε μ nd μ. Notice tht β cnnot e identified even pq = if we ssume tht the correltions etween ε s nd μ s re the sme efore nd fter the chnge of interest. Identifiction insted cn e gurnteed if there re no intrfmily responses (i.e. γ = γ = 0) or if child endowments re not regime specific (i.e. δ = δ = 0). The ssumption tht δ δ = 0 is perhps more credile when some exogenous = events re tken s instruments (e.g. the pssge of divorce lw regultions) rther thn others such s remrrige or prentl deth since specific relistions of ε (nd 5 If the fmily lloctes schooling so s to reinforce endowment differences etween silings then γ > 0 nd γ < 0; if insted the fmily compenstes for child-specific endowment differentils then γ < 0 nd γ > 0 (see Berhmn et l. 994). The ssumption of no intrfmily responses ws imposed y Rosenzweig nd Wolpin (99) nd Currie nd Cole (99) to estimte the determinnts of irth outcomes. 4

expecttions out ε ) my ultimtely led to such events. Even when those more credile exogenous circumstnces pply one concern is tht the regultions my e endogenous in the sense tht there my e trends in eductionl ttinments of children of divorced prents tht re correlted wh the introduction of specific divorce lw (Gruer 004). To illustrte this point let denote dummy vrile tht is equl to if individul i s prents divorced during his/her childhood nd 0 otherwise nd let q e the time period in which the divorce lw reform occurred. Suppose our outcome of interest the following specifiction (A.9) S 6 D i = α + α D α α D t α t q βd t q) + ξ 0 i + ( + i) + 4I( ) + ii( S is determined y wh E( ξ D t) = 0 where E ( ) is the expecttion opertor nd the term I( w) is function indicting tht the event w occurs (i.e. the post-reform period) so tht in eqution (A.9) β is our prmeter of interest. The prmeters α nd α reflect two different time trends in eductionl chievement for children in intct fmilies nd children in divorced fmilies respectively while α 4 reflects common ump in S from the time the reform is introduced onwrds. From (A.9) difference-in-difference estimte of β is given y β + α ( k + k' ) where k + k' represents the verge numer of clendr periods (sy yers) etween the post-reform nd pre-reform period oservtions in the smple. 7 Unless α = 0 this is clerly ised estimte of β. The is rises precisely ecuse the time evolution of S differs etween children in intct fmilies nd children of divorced prents. Wh this pproch therefore modelling group-specific trends will e crucil. 6 The fmily suscript hs een dropped for convenience. 7 Wh the fmily structure vrile nd the residuls in (A.9) following the sme structures s those specified in (A.') nd (A.4') necessry condion for identifiction of β is s efore tht child endowments re independent of the reform (or tht there re no intrfmily responses). 5