Correction For Selection Bias Research Articles

Neural tube defects in Beijing-Tianjin area of China. Urban-rural distribution and some other epidemiological characteristics.

Records in the obstetric wards and nurseries of 18 hospitals were reviewed and studied epidemiologically, covering about 210,000 deliveries and 1000 cases of neural tube defects (NTD). All live and still births occurred in the period 1970-84. Following the NTD classification used by Koch (1984), our case series consisted of anencephalus 50.3%, spina bifida 44.2%, and NTD with other system defects 5.5%. Overall NTD prevalence rate at birth was 4 per thousand, obviously higher than those in most other countries. Marked urban-rural differences in NTD prevalence rate at birth were observed. The rate in rural areas after correction for selection bias was still as high as 7.25 per thousand. As compared with cities and suburbs, the relative risk of NTD born to mothers in rural regions was 2.4. The male-to-female ratios were much less than 1 for various kinds of NTD. Looking at the effect of maternal age and birth order simultaneously in 12 different settings, it can be seen that prevalence of NTD was moderately higher among second and third births to women aged less than or equal to 24 and those greater than or equal to 35 years of age.

Self-employment vs wage employment: Estimation of earnings functions in LDCs

This study incorporates occupational choice into a model of wage determination in wage and self-employed activities. The self-selection model emphasizes managerial ability and capital market imperfections in determining whether an individual chooses between wage and self-employment. Using Heckman's correction for sample selection bias, the results show no bias in the OLS estimates but find that the two groups have different earnings structure. The results for employees confirm the robustness of the human capital model but the model explains only a small fraction of the variation of self-employed earnings.

Relationships among estimators of triangular econometric models

Estimators for triangular models where one endogenous variable may be dichotomous are compared. The selection bias correction method utilized in the simultaneous probit model is applied to standard simultaneous equations models. The resulting estimator is shown to be equivalent to 2SLS in some cases and provides a Hausman (1978) test of endogeneity. Also, other comparisons of estimators are made.

Child Health, Breast-Feeding, and Survival in Malaysia: A Random-Effects Logit Approach

Abstract Evidence from clinical studies suggests that children who are healthier at birth are more likely to be breast-fed, raising the possibility of selection bias in the estimation of the effect of breast-feeding itself on subsequent survival. Several maximum likelihood models are estimated to assess the severity of selection bias. A child's health enters the models as an unobserved variable within the framework developed by Heckman and Singer. Birth weight is treated as an indicator for unmeasured variables; that is, the “mixing distribution” for the unobservables shifts with the level of birth weight. Results indicate that the direct influence of breast-feeding on survival remains of overwhelming importance even after corrections for selection bias are made.

Child health, breast-feeding, and survival in Malaysia: a random-effects logit approach.

This study examines the possibility that estimation of the effect of breast-feeding on infant survival is affected by selection bias, in that children who are healthier at birth may be more likely to be breast-fed. Data are from the 1976 Malaysian Family Life Survey. "Ordinary logit models for breast-feeding and survival are estimated, and the results suggest that selection is indeed present. For example, children of higher birth weight appear to be more likely to be breast-fed and likely to survive. In addition, weight at birth and the duration of breast-feeding appear to be linked." Using birth weight as an indicator for the child's health, the authors conclude that "the direct influence of breast-feeding on survival remains of overwhelming importance even after corrections for selection bias are made."

Sample selection bias with multiple selection rules: An application to student aid grants

This paper extends the sample selection bias correction procedure, developed by James J. Heckman, to the case of multiple selection rules. Estimators are presented for independent selection rules and a consistent estimator for the variance-covariance matrix is derived. The methodology is applied to the problem of estimating the financial aid available to U.S. highschool students for college education. In order to be observed in receipt of financial aid, a student has to be both enrolled in college and to have had an expected value of aid in excess of the costs of applying. Both of these sources of selection bias were found to be important.

Estimating the Determinants of Employee Performance

Employers often wish to know whether the factors used in selecting employees do in fact allow them to choose the most qualified applicants. Because the performance of those not chosen is not observed, sample-selection bias is a likely problem in any attempt to validate employee-selection criteria. With minor modifications, the recently-developed techniques for dealing with sample-selection problems can be used in this context. Using data on applicants for first-line supervisory positions and ratings of on-the-job performance of those hired, ordinary least squares estimates of the determinants of performance are compared with maximum-likelihood estimates which correct for selection bias. The correction for selection bias produces some appreciable improvements in some variables' coefficients, though the corrected estimates remain insignificant at conventional levels. Differences in the firm's stated and actual hiring criteria are also noted.

A Least Squares Correction for Selectivity Bias

WHEN ESTIMATING REGRESSION MODELS it is very nearly always assumed that the sample is random. The recent literature has begun to deal with the problems which arise when estimating a regression model with samples which may not be random. The most general case in which one only has access to a single nonrandom sample has not been addressed since it is a very imposing problem. The case which has been addressed starts with a random sample but considers the problem of missing values for the dependent variable of a regression. If the determination of which values are to be observed is related to the unobservable error term in the regression, then methods such as ordinary least squares are in general inappropriate. By constructing a joint model which represents both the regression model to be estimated and the process determining when the dependent variable is to be observed, some progress can be made towards taking into account nonrandomness for the observed values of the dependent variable. The actual techniques employed fall into two rough groups, full information maximum likelihood models, and limited information methods which are more easily estimated. In the full information category are two methods. One model combines the probit and the normal regression models, and the other combines the Tobit or limited dependent variable model with the normal regression model. The form of the probit regression model is

2. Asteroid Compositional Types and their Distributions

A sample of 359 minor planets with available colorimetric, spectropnotometric, thermal-radiometric, and/or polarimetrie data are classified into broadly defined compositional types C (carbonaceous), S (silicaceous), M (metal-rich), E (metal-free; enstatite?), O (ordinary-chondritic?), T (Trojan; unidentified composition), and U (unclassifiable; none of the above).The small asteroids in Mars- or earth-crossing orbits are almost invariably of type S or O. For the main belt between 2.2 and 3.5 AU., distributions of the various types over diameter and orbital parameters are derived with corrections for observational selection bias. For 560 main-belt asteroids with diameters >50 km, 76% are of type C, 16% of type S, 5% M, and 3% of other types. The S objects become progressively less common with distance. The shapes of the diammeter-frequency relations for primitive (C) and evolved (S + M) types are statistically indistinguishable, both showing a change of slope at 160-km diameter.While large asteroids avoid the Kirkwood gaps more strongly than small ones, we find no significant gap-related anomalies in the relative frequencies of compositional types. Thus most of the observational basis for models in which primitive and evolved asteroids respond differently to collisional evolution has been removed.

Properties of Diagnostic Data Distributions

In applications of statistical methods to medical diagnosis, information on patients' diseases and symptoms is collected and the resulting data-base is used to diagnose new patients. The data-structure is complicated by a number of factors, two of which are examined here: selection bias and unstable population. Under reasonable conditions, no correction for selection bias is required when assessing probabilities for diseases based on symptom information, and it is suggested that these "diagnostic distributions" should form the principal object of study. Transformation of these distributions under changing population structure is considered and shown to take on a simple form in many situations. It is argued that the prevailing paradigm of diagnostic statistics, which concentrates on incidence of symptoms for given disease, is largely inappropriate and should be replaced by an emphasis on diagnostic distributions. The generalized logistic model is seen to fit naturally into the new framework.