The Stata Journal: Promoting communications on statistics and Stata

In this article, I present ivtreatreg, a command for fitting four different binary treatment models with and without heterogeneous average treatment effects under selection-on-unobservables (that is, treatment endogeneity). Depending on the model specified by the user, ivtreatreg provides consistent estimation of average treatment effects by using instrumental-variables estimators and a generalized two-step Heckman selection model. The added value of this new command is that it allows for generalization of the regression approach typically used in standard program evaluation by assuming heterogeneous response to treatment. It also serves as a sort of toolbox for conducting joint comparisons of different treatment methods, thus readily permitting checks on the robustness of results.

For Markov regime-switching models, a nonstandard test statistic must be used to test for the possible presence of multiple regimes. Carter and Steigerwald (2013, Journal of Econometric Methods 2: 25–34) derive the analytic steps needed to implement the Markov regime-switching test proposed by Cho and White (2007, Econometrica 75: 1671–1720). We summarize the implementation steps and address the computational issues that arise. We then introduce a new command to compute regime-switching critical values, rscv, and present it in the context of empirical research.

The analysis of multinomial data often includes the following question of interest: Is a particular category the most populous (that is, does it have the largest probability)? Berry (2001, Journal of Statistical Planning and Inference 99: 175–182) developed a likelihood-ratio test for assessing the evidence for the existence of a unique most probable category. Nettleton (2009, Journal of the American Statistical Association 104: 1052–1059) developed a likelihood-ratio test for testing whether a particular category was most probable, showed that the test was an example of an intersection-union test, and proposed other intersection-union tests for testing whether a particular category was most probable. He extended his likelihood-ratio test to the existence of a unique most probable category and showed that his test was equivalent to the test developed by Berry (2001, Journal of Statistical Planning and Inference 99: 175–182). Nettleton (2009, Journal of the American Statistical Association 104: 1052–1059) showed that the likelihood ratio for identifying a unique most probable cell could be viewed as a union-intersection test. The purpose of this article is to survey different methods and present a command, cellsupremacy, for the analysis of multinomial data as it pertains to identifying the significantly most probable category; the article also presents a command for sample-size calculations and power analyses, power cellsupremacy, that is useful for planning multinomial data studies.

In this article, we show how to implement merger simulation in Stata as a postestimation command, that is, after estimating an aggregate nested logit demand system with a linear regression model. We also show how to implement merger simulation when the demand parameters are not estimated but instead calibrated to be consistent with outside information on average price elasticities and profit margins. We allow for a variety of extensions, including the role of (marginal) cost savings, remedies (divestiture), and conduct different from Bertrand–Nash behavior.

Reweighting is a popular statistical technique to deal with inference in the presence of a nonrandom sample, and various reweighting estimators have been proposed in the literature. This article presents the user-written command treatrew, which implements reweighting on the propensity-score estimator as proposed by Rosenbaum and Rubin (1983, Biometrika 70: 41–55) in their seminal article. The main contribution of this command lies in providing analytical standard errors for the average treatment effects in the whole population, in the subpopulation of the treated, and in that of the untreated. Standard errors are calculated using the approximation suggested by Wooldridge (2010, 920–930, Econometric Analysis of Cross Section and Panel Data [MIT Press]), but bootstrapped standard errors can also be easily computed. Because an implementation of this estimator with analytic standard errors and nonnormalized weights is missing in Stata, this article and the accompanying ado-file aim to provide the community with an easy-to-use method for reweighting on the propensity-score. The estimator proves to be a valuable tool for estimating average treatment effects under selection on observables.

We present motivation and new commands for modeling count data. While our focus is to present new commands for estimating count data, we also discuss generalized binomial regression and present the zero-inflated versions of each model.

In many observational studies, the treatment may not be binary or categorical but rather continuous, so the focus is on estimating a continuous dose– response function. In this article, we propose a set of programs that semiparametrically estimate the dose–response function of a continuous treatment under the unconfoundedness assumption. We focus on kernel methods and penalized spline models and use generalized propensity-score methods under continuous treatment regimes for covariate adjustment. Our programs use generalized linear models to estimate the generalized propensity score, allowing users to choose between alternative parametric assumptions. They also allow users to impose a common support condition and evaluate the balance of the covariates using various approaches. We illustrate our routines by estimating the effect of the prize amount on subsequent labor earnings for Massachusetts lottery winners, using data collected by Imbens, Rubin, and Sacerdote (2001, American Economic Review, 778–794).

In this article, we describe an implementation of a space-filling location-selection algorithm. The objective is to select a subset from a list of locations so that the spatial coverage of the locations by the selected subset is optimized according to a geometric criterion. Such an algorithm designed for geographical site selection is useful for determining a grid of points that “covers” a data matrix as needed in various nonparametric estimation procedures.

I describe algorithms for drawing from distributions using adaptive Markov chain Monte Carlo (MCMC) methods; I introduce a Mata function for performing adaptive MCMC, amcmc(); and I present a suite of functions, amcmc_*(), that allows an alternative implementation of adaptive MCMC. amcmc() and amcmc_*() can be used with models set up to work with Mata's moptimize( ) (see [M-5] moptimize( )) or optimize( ) (see [M-5] optimize( )) or with standalone functions. To show how the routines can be used in estimation problems, I give two examples of what Chernozhukov and Hong (2003, Journal of Econometrics 115: 293–346) refer to as quasi-Bayesian or Laplace-type estimators—simulation-based estimators using MCMC sampling. In the first example, I illustrate basic ideas and show how a simple linear model can be fit by simulation. In the next example, I describe simulation-based estimation of a censored quantile regression model following Powell (1986, Journal of Econometrics 32: 143–155); the discussion describes the workings of the command mcmccqreg. I also present an example of how the routines can be used to draw from distributions without a normalizing constant and used in Bayesian estimation of a mixed logit model. This discussion introduces the command bayesmixedlogit.

This command meets the need of a researcher who holds multiple data files in comma-separated value format differing by a period variable (for example, year or quarter) or by a cross-sectional variable (for example, country or firm) and must combine them into one Stata-format file.

We present a new Stata command, bmte (bias-minimizing treatment effects), that implements two new estimators proposed in Millimet and Tchernis (2013, Journal of Applied Econometrics 28: 982–1017) and designed to estimate the effect of treatment when selection on unobserved variables exists and appropriate exclusion restrictions are unavailable. In addition, the bmte command estimates treatment effects from several alternative estimators that also do not rely on exclusion restrictions for identification of the causal effects of the treatment, including the following: 1) Heckman's two-step estimator (1976, Annals of Economic and Social Measurement 5: 475–492; 1979, Econometrica 47: 153–161); 2) a control function approach outlined in Heckman, LaLonde, and Smith (1999, Handbook of Labor Economics 3: 1865–2097) and Navarro (2008, The New Palgrave Dictionary of Economics [Palgrave Macmillan]); and 3) a more recent estimator proposed by Klein and Vella (2009, Journal of Applied Econometrics 24: 735–762) that exploits heteroskedasticity for identification. By implementing two new estimators alongside preexisting estimators, the bmte command provides a picture of the average causal effects of the treatment across a variety of assumptions. We present an example application of the command following Millimet and Tchernis (2013, Journal of Applied Econometrics 28: 982–1017).

In this article, I introduce the new command xtpedroni, which implements the Pedroni (1999, Oxford Bulletin of Economics and Statistics 61: 653–670; 2004, Econometric Theory 20: 597–625) panel cointegration test and the Pedroni (2001, Review of Economics and Statistics 83: 727–731) group-mean panel-dynamic ordinary least-squares estimator. For nonstationary heterogeneous panels that are long (large T) and wide (large N), xtpedroni tests for cointegration among one or more regressors by using seven test statistics under the null of no cointegration, and it also estimates the cointegrating equation for each individual as well as the group mean of the panel. The test can include common time dummies and unbalanced panels.

Dropbox makes scholarly collaboration much easier because it allows scholars to share files across different computers. However, because the Dropbox directories have different pathnames for different users, sharing do-files can be complicated. In this article, I offer some tips on how to navigate pathnames in do-files when using Dropbox, and I present a command that automatically finds and changes to a user's Dropbox directory.

In this article, I review An Introduction to Stata for Health Researchers, Fourth Edition, by Svend Juul and Morten Frydenberg (2014 [Stata Press]).