Data on twins or on other types of family structures (for example, nuclear families, siblings, cousins) can be used to estimate the proportion of variability in observed traits (or phenotypes) that is due to genes. The models are essentially multivariate regression models with residual covariance structures dictated by Mendelian genetics. Usually, specialized software for structural equation modeling is used. However, the required covariance structures can also be produced using mixed models and by specifying an appropriate design matrix for the random part of the model. Stata’s xtmixed command can then be used to estimate the models. For binary phenotypes, such as diabetes, the appropriate probit models can be estimated using gllamm.

Additional information
germany10_rabe-hesketh.pdf
germany10_rabe-hesketh.zip

Matching, especially in its propensity-score flavors, has become an extremely popular evaluation method. Matching is, in fact, the best-available method for selecting a matched (or reweighted) comparison group that looks like the treatment group of interest.

In this talk, I will introduce matching methods within the general problem of causal inference, highlight their strengths and weaknesses, and offer a brief overview of different matching estimators. Using psmatch2, I will then step through a practical example in Stata that is based on real data. I will then show how to implement some of these estimators, as well as highlight a number of implementational issues.

Additional information
germany10_sianesi.pdf
germany10_sianesi_materials.zip

Methods for causal inference and the estimation of treatment effects have received much attention in recent years. Most of the methodological and applied work focuses on the identification of so-called average treatment effects, possibly restricted to the treated or the untreated. However, treatment effects may vary (hence the averaging), and it can be interesting to analyze the patterns of effect heterogeneity. In this talk, I will present a new command called hte that is used for heterogeneous treatment-effect analysis in Stata. hte first constructs balanced propensity-score strata and, within each stratum, estimates the average treatment effect. hte then tests for a linear trend in effects across the strata. The stratum-specific treatment effects and the estimated linear trend are displayed in a two-way graph. hte results from joint work with Jennie E. Brand (UCLA) and Yu Xie (University of Michigan).

Additional information
germany10_jann.pdf

Fixed-effects regression is considered a powerful method for estimating causal effects with survey data. However, in the linear model, the conventional technique of time-demeaning does not yield consistent estimates of the parameters when unobserved heterogeneity is not time-constant. Jeffrey M. Wooldridge (2002, Econometric Analysis of Cross Section and Panel Data [MIT Press], 317–322) derived a general model for the situation where unobserved and observed characteristics of individuals interact to produce the outcome. The fixed-effects model with individual constants and slopes (FEIS) is a remedy for coefficients that are biased due to, for example, maturation or learning where unobserved traits affect individual growth curves differently for treated and controls.

The Stata xtfeis command implements the FEIS estimator in Mata, allowing for individual constants and (potentially many) slopes. Without specifying slope variables, the model collapses to the conventional model estimated by xtreg, fe that accounts for individual constants only. xtfeis implements standard errors that are robust to serial correlation or heteroskedasticity of unknown form. Estimates of the slope parameters are available optionally. The command requires panel data with at least J + 1 observations per unit, where J is the number of individual-specific slope variables (usually, but not necessarily, also including the individual-specific constant). I will present results for the effect of marriage on male wages based on real data (GSOEP and NLSY) to demonstrate the practical relevance of the method. I will use simulation results to assess robustness of the estimator to autocorrelation, measurement error, and misspecification of functional form.

Additional information
germany10_ludwig.pdf

Respondent-driven sampling (RDS) is a sampling technique typically employed for hard-to-reach populations (for example, homeless people, people with AIDS, immigrants). Briefly, initial seed respondents recruit additional respondents from their network of friends. The recruiting process repeats iteratively, thereby forming long referral chains. It is crucial to obtain estimates of respondents’ network sizes (for example, the number of friends with the characteristic of interest). RDS shares some similarities with snowball sampling, but the theoretical foundation for inference using RDS samples is much stronger. We will give a brief overview of this technique and introduce a new user-written Stata command for RDS.

Additional information
germany10_schonlau_liebau.ppt

Bill Gould, president of StataCorp and head of development, talks about Stata.