A Dynamic Structural Model of Contraceptive Use and Employment Sector Choice for Women in Indonesia

Transcription

1 A Dynamic Structural Model of Contraceptive Use and Employment Sector Choice for Women in Indonesia -Uma Radhakrishnan Fourth Annual Research Conference on Population, Reproductive Health, and Economic Development Cape Town, 2010

2 Research Outline Develop a dynamic structural model to investigate the impact of the Indonesian family planning program on labor force participation and contraception choices of women Estimate model using simulated maximum likelihood techniques with Indonesia Family Life Survey 1(IFLS 1) data for the period Use exogenous variation in timing of introduction of 3 types of family planning clinics for identification 2

3 Indonesian Context Total Fertility Rate in Indonesia, to Notes: This figure is from Mize (2006). Source: Contraceptive Prevalence Survey (1987) and Indonesian Demographic and Health Survey (1991, 1994, 1997, ) 3

4 Indonesian Context Contraceptive Prevalence Rate in Indonesia, Notes: This figure is from Mize (2006). 4

5 Indonesian Context Labor Force Participation Rates by Gender in Rural and Urban Areas: Indonesia in 1971, 1980, and 1990 Male Urban Rural Total Female Urban Rural Total Both Sexes Notes: This table is quoted from Manning (1998). Source: CBS, Population Censuses, 1980 and

6 Women, Child Care, and Informal Sector What remains unchanged is that women continue to hold primary child care responsibility Large fractions of working women are employed in the informal sector characterized by: Flexible timing Easy entry and exit Proximity to residence Compatibility between work and family responsibilities (especially child care) 6

7 Research Motivation Very little investigation of the impact of the family planning program on women s labor force participation and wages To understand the compatibility between work and family responsibility, especially child care provision as women make joint contraceptive method and employment sector choices Structural model allows me to conduct policy experiments 7

8 Indonesian Family Planning Program Introduced in late 1960s Family planning program was introduced as part of five-year development plans Initial geographic expansion Phase (6 provinces including Java and Bali) Phase (10 provinces belonging to Outer Islands 1) Phase (remaining provinces) 8

9 Indonesian Family Planning Program: Geographic Expansion Phase 1 Phase 2 Phase 3 9

10 Indonesian Family Planning Program: Changing Nature Initially followed a clinic-based approach Community Health Centers (Puskemas) Failed to reach a large group of target women Community-based approach was first established in mid 1970s Key idea was to use existing institutions to promote family planning Family Planning Distribution Points (PKKBD) Village Integrated Health Posts (Posyandu) 10

11 Literature Review Impact of family planning programs on fertility and socio-economic outcomes Goldin and Katz (2002); Miller (2005); Joshi and Schultz (2007) Female labor force participation in developing countries Jaffe and Azumi (1960); Tiefenthaler (1994) Modeling contraceptive behavior Carro and Mira (2002) Joint modeling of employment and fertility decisions Hotz and Miller (1988); Francesconi (2002) 11

12 Contribution of this Research Distinguish between formal and informal sectors of employment Allow joint contraception and employment choices to understand link the between family responsibility and employment Endogenize wage rates so that sector-specific experience impacts wages, and this in turn affects cost of having a child Allow uncertainty in fertility control 12

13 Economic Model I develop a finite horizon, discrete choice dynamic structural model in which married women in each period choose both method of contraception and sector of employment to maximize their expected discounted life-time utility function 13

14 Model Marriage and education are treated as exogenous Choose a sector of employment, k=1, formal sector k=2,informal sector k=3, not working Choose a contraceptive method, m=1, modern method m=2, traditional method m=3, not using contraception k o t m m t 14

15 Classification of Choices Sector of Employment Formal: self-employed with permanent workers, government employees, private employees Informal: self-employed, self-employed with temporary workers, family workers Contraceptive Methods Modern: implants, IUD, condoms, pills, injections Traditional: rhythm, withdrawal, traditional herbs 15

16 Motivation to Control Fertility and Sector of Employment Motivation to control fertility depends on compatibility between raising children and employment sector Motivation to control fertility can be inferred by method of contraception used While making contraceptive decisions, a woman considers the trade-off between costs of having a child and the benefits from having one 16

17 Utility Function Expected discounted life-time utility function: c t T F t A E[ 0 ( ct qkmt kmt )] t A - consumption (pecuniary component) q kmt - nonpecuniary component - choice-specific time shock kmt 0 17

18 Budget Constraint c [ w w Y P m P N ] t k t h t m t m t n t w t k w t h Y t P m P n - sharing rule parameter - wage earnings of the woman in sector k - husband s wage - unearned income of husband and wife - price of contraception used - average expenditure on a child 18

19 Wage Equation k t w w G t X O O G t X t L 1 O t 1 2 O t 1 t k w i L 1 2 k k 0 k1 k 2 k 3 t k 4 t 1 k5 t 1 t - education - age - provincial minimum wage rates - experience in formal sector - experience in informal sector - wage error - unobserved ability i w 19

20 Nonpecuniary Utility Number of births Age of youngest child Duration dependence Birth spacing Birth in the previous period Interactions of choices with exogenous characteristics such as age, religion, location, access to contraceptives Several other interaction terms 20

21 Birth Probability Function F ( t, m, M, ) m, t 1 t t f t m t m M t m f i m m - age - method of contraceptive in period t - duration for which the method was used - unobserved fecundity i 21

22 Data Indonesia Family Life Survey 1(IFLS 1), 1993 Covers 13 provinces (321 Enumeration Areas) and 83% of the population Retrospective panel Individual and family level data on employment, income, education, migration, contraception use, and fertility Community level data that can be linked to individual and household level data 22

23 IFLS 1 Provinces 23

24 Variables used in Empirical Analysis N=2,067 Age at time of marriage Urban Muslim Age Education Birth spacing Number of children Age of youngest child Gives birth Province Wages Unearned Income Method of contraception Sector of Employment Duration in formal sector Duration in informal sector Duration not working Duration using modern methods Duration using traditional methods Duration not using contraceptives Enumeration Area Source: IFLS 1 24

25 Descriptive Statistics Variable Mean Standard Deviation Sample of 2067 Women Age at time of marriage* Urban Muslim Sample of 20,707 woman-year observations Age* Number of children Age of youngest child* Gives birth Duration in formal sector* Duration in informal sector* Duration not working* Duration using modern methods* Duration using traditional methods* Duration not using contraceptives* Notes: * denotes unit of measurement is Year. Source: IFLS 1 25

26 Joint Choices and Identification Model joint contraception and employment decisions Unobserved heterogeneity may drive both decisions leading to biased estimates Use exogenous variation in timing of introduction of 3 different types of family planning clinics and exogenous variation in minimum wages rates for identification 26

27 Estimation Outline Solve the dynamic programming problem Use representative people to reduce computational cost (Brien, Lillard, and Stern (2006)) Estimate the birth probability function and wage equation outside the structural model. After solving the dynamic programming problem and estimating parameters of wage and birth function, using data on observed choices and state variables, estimate parameters of utility function and budget constraint using simulated maximum likelihood techniques. 27

28 Birth Probability Function-Probit Model Variables Estimates Age of Woman * (0.0021) Lagged Modern Method * (0.0574) Lagged Traditional Method * (0.1262) Duration of Modern method * Lagged Modern Method Duration of Traditional Method * Lagged Traditional Method * (0.0182) (0.0468) Constant (0.0508) 28

29 Main Results from Structural Estimation 1:Contraception Covariates Modern Method Traditional Method Age of Woman Positive Positive Muslim Positive Negative More than primary Education Positive Negative Posyandu Positive Negative Puskemas Positive Negative PKKBD Positive Positive Age of Youngest Child Negative Negative Birth in Previous Period Negative Negative Number of Births Positive Positive Duration Positive Positive Urban Negative Positive 29

30 Main Results from Structural Estimation 2: Employment Covariates Formal Sector Informal Sector Age of Woman Positive Positive Muslim Negative Positive Number of births Negative Positive Birth in Previous Period Negative Positive Age of Youngest Child Positive Positive Duration Positive Negative Urban Positive Negative 30

31 Main Results from Structural Estimation 3:Choice-Independent Covariates (Not choice Dependent) Utility from birth when age>35 Change Negative Utility from birth spacing Positive Utility from birth Positive Utility from number of births Positive 31

32 Policy Simulations Decrease cost of using contraceptives. Improvement in quality of family planning services such as reduced wait times Reduction in price of contraceptives Reduction in distance to clinics Increase utility experienced by working mothers Reduction in cost of child care Flexible timings in formal sector employment Better working condition in informal sector 32

33 Punch Line! Estimates indicate that : Informal sector jobs offer greater compatibility between work and child care. Women with more kids and young kids derive more utility from working in the informal sector. Access to modern methods of contraception (presence of clinics) reduces cost of using modern methods. Choice of contraception method and employment sector vary by exogenous characteristics such as religion, urban or rural residence, Muslim, presence of clinics and age of woman. Number of kids, birth spacing, and age of youngest child play an important role in impacting choices made by married women. 33

34 Conclusion Investigate the expansion of Indonesian family planning program on employment and contraceptive choices of married women, while recognizing the interdependency of these choices. Access to contraceptives increases the likelihood of participating in the formal sector, conditional on adequate job creation rates in formal sector. Although outcomes are observed at the individual level, it has implications for the economy as a whole: participation of women in labor force increases per capita income and this translates into economic growth. 34

35 THANK YOU! 35

36 BACK UP SLIDES 36

37 Migration Non-random selection in migration Migration to obtain contraception will bias estimates Modeling migration will not reduce bias unless it is exogenous Assume that women lives in EA in all sample years as found in 1993, IFLS 1. 37

38 Nonpecuniary utility function k m k ' k q o X m X o m kmt 1km t 1t 2km t 2t 3 t 4 k k m m o O m M 1( t 35) n b n N o k k k o n N o r o N b n b n o i i b n N n r m m' m t k 5 t t 1 6 t t 1 7 t 8 t t t t k 9 t t 10 t t 11 t t 12 t 13 t t 14 t t t 15 t t t 16 t 17 t o 38

39 Exogenous Variation in Minimum Wage Rates Exogenous variation in minimum wage rates in the different provinces over time is used to identify parameters related to employment choices. Minimum wage rate is set by Ministry of Manpower based on recommendation of governors in different provinces. Internal and external pressures unrelated to local economic conditions in setting of minimum wage rates Reasonable to assume variation in minimum wage rates does not impact contraceptive choices. 39

40 Minimum Wag Real Minimum Wages in IFLS 1 provinces Real Regional Minimum Wage in Indonesia, (Rupiah/month) 3,000 2,500 2,000 1,500 1, Year North Sumatra West Sumatra South Sumatra Lampung Jakarta West Java Central Java Yogyakarta East Java Bali South Sulawesi South Kalimantan West Nusa Tenggara Notes: Real Wages are in 2000 Indonesian Rupiah. Source: Minimum Wage data was obtained from Arup Suryahadi and David Newhouse. 40

41 Solving the Dynamic Programming Problem Solution involves obtaining value function for each person for each point in the state space for a given set of parameters. Backward recursion. For t<t*, value function at each point in the state space is the sum of current utility plus the discounted value of the expected best choice next period. Backward recursion continues until t=0. Large state space makes it computationally expensive to evaluate the value function at every point. 41

42 Reducing Computational Cost Impose upper bounds on several state variables (age of youngest child, birth spacing, and number of births). Interpolation Model unobserved heterogeneity using Heckman and Singer (1984) approach 42

43 Representative People Individuals differ by the following exogenous characteristics: religion location presence of 3 types of clinic 4 types of people with respect to unobserved heterogeneity This results in 2*2*2 3 *4=128 representative people Use interpolation method in Brien, Lillard, and Stern(2006) 43

44 Birth Probability Function Estimate this is as a Probit, where the dependent variable is 1 if a birth occurs and 0 otherwise. Use parameters from Probit regression to obtain probability of birth in the structural model after conditioning on method used, duration of use, age of the women, and unobserved fecundity level. 44

45 Wage Equation Estimate outside structural model to reduce computational costs. Two-stage method in Heckman(1979) is used to correct for selection bias, as wages are observed only for working women. 45

46 Likelihood Contribution Likelihood contribution of a woman who is not working in period t is: Pr(, t, S( t)) i { k, m} exp[ V ( d kmt ( S( t), t, i ) / ] exp[ V ( dlzt ( S( t), t, i ) / ] { l, z} d kmt 1 2 df( ) dg( ) t t Likelihood contribution of a woman working in the informal sector in period t is: Pr(, t, S( t), w ) i exp[ V ( d kmt ( S( t), t, i ) / ] exp[ V ( dlzt ( S( t), t, i ) / ] l z t t t t { k, m} {, } d kmt g( w w ) df( ) 46

47 Why Structural Model? Enables policy simulation According to Professor Stern: Adds discipline to modeling and estimation, and makes it easier to talk about the model and the economics in it 47

48 Geographic Expansion of Indonesian Family Planning Program Phase 1 provinces - West Java, Jakarta, Central Java, East Java, Yogyakarta, and Bali Phase 2 provinces - Aceh, North Sumatra, West Sumatra, South Sumatra, Lampung, North Sulawesi, South Sulawesi, South Kalimantan, West Kalimanatan, and West Nusa Tenggara Phase 3 provinces-riau, Jambi, Bengkulu, East Nusa Tenggara, Central Kalimantan, East Kalimantan, Central Sulawesi, South East Sulawesi, Maluku, Irian Jaya, and East Timor. 48

49 Descriptive Statistics: Education Education Percentage Primary Junior Secondary Senior Secondary College Source: IFLS 1 49

50 Estimates of Contraceptive Failure Rates in the United States Method Implant 2.8 Injectable 3.2 IUD 3.7 Pill 6.9 Diaphram 8.1 Male Condom 8.7 Withdrawal 18.8 Periodic Abstinence 19.8 Other 32.0 Failure Rate in 12 Months (Typical Use) Notes: Failure rate is the percentage of women who accidentally become pregnant as estimated in Tussell and Vaughn (1999) using 1995 National Survey of Family Growth in the United States. 50 Source: Quoted from Tussell and Vaughn (1999).

51 Distribution of sample Women by Province, 1993 Province Number of women Percentage North Sumatra West Sumatra South Sumatra Lampung DKI Jakarta West Java Central Java DI Yogyakarta East Java Bali West Nusa Tennegara South Kalimantan South Sulawesi Total Source: IFLS 1 51

52 Descriptive Statistics Variable Mean Standard Deviation Sample of 2067 Women Age at time of marriage* Urban Muslim Sample of 20,707 woman-year observations Age* Number of children Age of youngest child* Gives birth Duration in formal sector* Duration in informal sector* Duration not working* Duration using modern methods* Duration using traditional methods* Duration not using contraceptives* Notes: * denotes unit of measurement is Year. Source: IFLS 1 52

53 0 Number of PKKBD Number of Posyandu Number of Puskemas Number of Family Planning Clinics Introduced between Number of Posyandus' introduced between Number of Puskemas introduced between Year Year Number of PKKBDs introduced between Notes: Posyandu is Village Integrated Health Posts. Puskemas is Community Health Center. PKKBD is Family Planning Distribution Points. Source: IFLS Year 53

54 Identification of the Wage Structure Identification of the wage structure comes from covariation of wages and observables across the two sectors for similar occupations. 54

55 Identification of State Dependence State dependence is separately identified from unobserved heterogeneity by variation in choices made by individuals with similar observable characteristics who have experienced a certain state relative to individuals who have not experienced that state. 55

56 Exogenous Variation in Timing of Introduction of Posyandu Average Fertility by Birth Cohort and Timing of Introduction of Posyandu Year of Birth Before After Notes: Posyandu is Village Integrated Health Posts. After is for EAs where Posyandu was introduced after 1980 and Before is for EAs where Posyandu was introduced before Source: IFLS 1 56

57 Exogenous Variation in Timing of Introduction of PKKBD Average Fertility by Birth Cohort and Timing of Introduction of PKKBD Year of Birth Before After Notes: PKKBD is Family Planning Distribution Points. After is for EAs where PKKBD was introduced after 1980 and Before is for EAs where PKKBD was introduced before Source: IFLS 1 57

58 Exogenous Variation in Timing of Introduction of Puskemas Average Fertility by Birth Cohort and Timing of Introduction of Puskemas Year of BIrth Before After Notes: Puskemas is Community Health Center. After is for EAs where Puskemas was introduced after 1980 and Before is for EAs where Puskemas was introduced before Source: IFLS 1 58

59 Issues with Using Access to Family Planning Program as Instruments for Identification Outcomes of interest may be biased by non-random nature of program expansion. Correlation between timing of introduction and unobserved taste for fertility will lead to biased estimates 59

60 Identification (Utility Parameters) Parameters of utility function are identified by Data on choices and individual characteristics Variation in timing of introduction of different types of fertility clinics within each enumeration area and variation across enumeration areas over time in access to contraceptives Exogenous variation in local labor market conditions (real minimum wage rates) across provinces and over time 60

61 Identification (Wage equation) Coefficients of the wage equation are identified by covariation of observable characteristics and wages across individuals within a sector Variance of the wage error is identified by differences in wages across individuals in a sector in a given period conditional on observables 61

62 Identification of Unobserved Heterogeneity Variance of the unobserved preference heterogeneity is identified by persistence in choices made by individuals over time relative to individuals with same observables. Variance of unobserved ability is identified by persistent differences over time across individuals in wages conditional on observables. Variance of unobserved natural fecundity level is identified by variation in fertility across women conditional on observables and choices made. 62

63 More about the Model Choose 1 of 9 alternatives; denote d kmt =1, if sector k and method m are chosen in period t Decision making horizon is from A 0 to T*, but women live until T F, T F >T* 63

64 Likelihood Equation Solution to individual s optimization problem provides the choice probabilities in the likelihood equation Sample likelihood equation is the product across individuals, time, and choices of the contributing probability corresponding to each alternative L(.) Pr( i, t, S( t)) dh( i ) i t 64

65 Women as Decision Makers The utility maximization problem can be considered as A two-stage benevolent dictator problem. Chiappori s collective approach 65

66 Descriptive Statistics: Education Education Percentage Primary Junior Secondary Senior Secondary College Source: IFLS 1 66

67 State Space and Value Function State space at time t is: S( t) ( ot 1, mt 1, Dt 1, rt 1, Nt 1, bt, t, t, t ) Value function at time t given state S(t) and unobserved heterogeneity is: V max[ V ( S( t), ),..., V ( S( t), )] t 1, 1, t i 3, 3, t i where Vk, m, t ( S( t), i ) U kmt ( S( t), i ) EVt 1( S( t 1), i S( t), dkmt 1) For A 0 <= t < T* T F t ' T* V * k, m, T U ' kmt ' ( S( t'), i ) t ' T * For T* <= t < T F 67

68 Nonstructural Estimation: Marginal Effects at Means for Select Independent Variables from Contraceptive Choice Multinomial Probit Variable Pr(Method = Modern) = dp/dx Gave Birth Last Period * (0.0103) Number of Children * (0.0056) Age of Youngest Child * (0.0029) Age (0.0012) Posyandu * (0.0129) Puskemas (0.0131) PKKBD * (0.0015) Pr(Method = Traditional) = dp/dx * (0.0011) (0.0007) * (0.0004) (0.0001) (0.0016) (0.0016) (0.0015) Pr(Method = Not Using Contraceptive s)= dp/dx * (0.0104) * (0.0057) * (0.0029) (0.0012) * (0.0129) (0.0132) (0.0115) Choice Notes: Last N=20,707. Year Method 1 denotes Modern Method, * Method 2 denotes * Traditional Method, and * Method 3 denotes Not Using Contraceptives. Standard Errors (0.0099) are in parenthesis. *implies (0.0021) statistical significance (0.0099) at 5%. Source: IFLS 1 68

69 Selection in Wage Equation Estimated outside structural model Wages are observed for women in sectors they actually work To correct for selection bias, I model selection process as multinomial logit Use Fournier and Girgand s (2004) user written SELMLOG program 69

70 Nonstructural Estimation: Marginal Effects at Means for Select Independent Variables from Employment Sector Multinomial Probit Variable Pr(Sector = Formal) = dp/dx Muslim * (0.0117) Urban * (0.0080) Gave Birth Last Period * (0.0081) Number of Children * (0.0037) Age of Youngest Child (0.0018) Age * (0.0009) Pr(Sector = Informal) = dp/dx * (0.0115) * (0.0078) (0.0099) (0.0033) * (0.0016) * (0.0008) Pr(Sector = Not Working) = dp/dx * (0.0144) * (0.0101) * (0.0116) * (0.0045) * (0.0023) * (0.0011) Choice Last Year )* * * (0.0107) (0.0100) (0.0114) Notes: N= 20,707. Sector 1 denotes Formal Sector, Sector 2 denotes Informal Sector, and Sector 3 denotes Not Working. Standard Errors are in parenthesis. *implies statistical significance at 5%. Source: IFLS 1 70

71 Imputation Why Missing data is a problem? -Missing data maybe correlated with person-specific characteristics -May lead to biased estimates Use Sequential Regression Multivariate Imputation (SRMI) algorithm. Imputations are performed outside structural model 71

72 Descriptive Statistics Distribution of Woman-Year Observations by Choices Made Choice Percentage Modern Method and Formal Sector 9.17 Modern Method and Informal Sector 7.44 Modern Method and Not Working Traditional Method and Formal Sector 0.96 Traditional Method and Informal Sector 0.55 Traditional Method and Not Working 1.80 No Contraceptives and Formal Sector No Contraceptives and Informal Sector No Contraceptives and Not Working Total 100 Source: IFLS 1 72

73 Sources of Exogenous Variation Community level data in IFLS 1 includes timing of introduction of 3 types of fertility clinics (access to contraceptives) in each enumeration area Community Health Centers or Puskemas (33% introduced after 1980) Family Planning Distribution Points or PKKBD (58% introduced after 1980) Village Integrated Health Posts or Posyandus (77% introduced after 1980) 73

74 Likelihood Optimization Numerical optimization routine Berndt, Hall, Hall, and Hausman(1974) algorithm Start with an initial guess of the parameter vector Compute the likelihood function and its derivatives Update guess until likelihood function is maximized 74