The Stata Journal. Editor Nicholas J. Cox Department of Geography Durham University South Road Durham DH1 3LE UK

Transcription

1 The Stata Journal Editor H. Joseph Newton Department of Statistics Texas A&M University College Station, Texas ; fax Associate Editors Christopher F. Baum Boston College Nathaniel Beck New York University Rino Bellocco Karolinska Institutet, Sweden, and University of Milano-Bicocca, Italy Maarten L. Buis Tübingen University, Germany A. Colin Cameron University of California Davis Mario A. Cleves Univ. of Arkansas for Medical Sciences William D. Dupont Vanderbilt University David Epstein Columbia University Allan Gregory Queen s University James Hardin University of South Carolina Ben Jann University of Bern, Switzerland Stephen Jenkins London School of Economics and Political Science Ulrich Kohler WZB, Berlin Frauke Kreuter University of Maryland College Park Stata Press Editorial Manager Stata Press Copy Editors Editor Nicholas J. Cox Department of Geography Durham University South Road Durham DH1 3LE UK n.j.cox@stata-journal.com Peter A. Lachenbruch Oregon State University Jens Lauritsen Odense University Hospital Stanley Lemeshow Ohio State University J. Scott Long Indiana University Roger Newson Imperial College, London Austin Nichols Urban Institute, Washington DC Marcello Pagano Harvard School of Public Health Sophia Rabe-Hesketh University of California Berkeley J. Patrick Royston MRC Clinical Trials Unit, London Philip Ryan University of Adelaide Mark E. Schaffer Heriot-Watt University, Edinburgh Jeroen Weesie Utrecht University Nicholas J. G. Winter University of Virginia Jeffrey Wooldridge Michigan State University Lisa Gilmore Deirdre McClellan

2 The Stata Journal publishes reviewed papers together with shorter notes or comments, regular columns, book reviews, and other material of interest to Stata users. Examples of the types of papers include 1) expository papers that link the use of Stata commands or programs to associated principles, such as those that will serve as tutorials for users first encountering a new field of statistics or a major new technique; 2) papers that go beyond the Stata manual in explaining key features or uses of Stata that are of interest to intermediate or advanced users of Stata; 3) papers that discuss new commands or Stata programs of interest either to a wide spectrum of users (e.g., in data management or graphics) or to some large segment of Stata users (e.g., in survey statistics, survival analysis, panel analysis, or limited dependent variable modeling); 4) papers analyzing the statistical properties of new or existing estimators and tests in Stata; 5) papers that could be of interest or usefulness to researchers, especially in fields that are of practical importance but are not often included in texts or other journals, such as the use of Stata in managing datasets, especially large datasets, with advice from hard-won experience; and 6) papers of interest to those who teach, including Stata with topics such as extended examples of techniques and interpretation of results, simulations of statistical concepts, and overviews of subject areas. For more information on the Stata Journal, including information for authors, see the webpage The Stata Journal is indexed and abstracted in the following: CompuMath Citation Index R Current Contents/Social and Behavioral Sciences R RePEc: Research Papers in Economics Science Citation Index Expanded (also known as SciSearch R ) Scopus TM Social Sciences Citation Index R Copyright Statement: The Stata Journal and the contents of the supporting files (programs, datasets, and help files) are copyright c by StataCorp LP. The contents of the supporting files (programs, datasets, and help files) may be copied or reproduced by any means whatsoever, in whole or in part, as long as any copy or reproduction includes attribution to both (1) the author and (2) the Stata Journal. The articles appearing in the Stata Journal may be copied or reproduced as printed copies, in whole or in part, as long as any copy or reproduction includes attribution to both (1) the author and (2) the Stata Journal. Written permission must be obtained from StataCorp if you wish to make electronic copies of the insertions. This precludes placing electronic copies of the Stata Journal, in whole or in part, on publicly accessible websites, fileservers, or other locations where the copy may be accessed by anyone other than the subscriber. Users of any of the software, ideas, data, or other materials published in the Stata Journal or the supporting files understand that such use is made without warranty of any kind, by either the Stata Journal, the author, or StataCorp. In particular, there is no warranty of fitness of purpose or merchantability, nor for special, incidental, or consequential damages such as loss of profits. The purpose of the Stata Journal is to promote free communication among Stata users. The Stata Journal (ISSN X) is a publication of Stata Press. Stata, Mata, NetCourse, and Stata Press are registered trademarks of StataCorp LP.

3 The Stata Journal (2011) 11, Number 2, pp Generating random samples from user-defined distributions Katarína Lukácsy Central European University Budapest, Hungary lukacsy Abstract. Generating random samples in Stata is very straightforward if the distribution drawn from is uniform or normal. With any other distribution, an inverse method can be used; but even in this case, the user is limited to the builtin functions. For any other distribution functions, their inverse must be derived analytically or numerical methods must be used if analytical derivation of the inverse function is tedious or impossible. In this article, I introduce a command that generates a random sample from any user-specified distribution function using numeric methods that make this command very generic. Keywords: st0229, rsample, random sample, user-defined distribution function, inverse method, Monte Carlo exercise 1 Introduction In statistics, a probability distribution identifies the probability of a random variable. If the random variable is discrete, it identifies the probability of each value of this variable. If the random variable is continuous, it defines the probability that this variable s value falls within a particular interval. The probability distribution describes the range of possible values that a random variable can attain, further referred to as the support interval, and the probability that the value of the random variable is within any (measurable) subset of that range. Random sampling refers to taking a number of independent observations from a probability distribution. Typically, the parameters of the probability distribution (often referred to as true parameters) are unknown, and the aim is to retrieve them using various estimation methods on the random sample generated from this probability distribution. For example, consider a normally distributed population with true mean µ and true variance σ 2. Assume that we take a sample of a given size from this population and calculate its mean and standard deviation these two statistics are called the sample mean and the sample standard deviation. Depending on the estimation method used, the sample size, and other factors, these two statistics can be shown to asymptotically converge to the true parameters of the original probability distribution and are thus good approximations of these values. c 2011 StataCorp LP st0229

4 300 Random samples The researcher may wish to evaluate the accuracy of the estimation method prior to using it on real data. In such a case, a random sample generated from a distribution with known true parameters can be used, and its estimated sample parameters can be compared with the true parameters to establish the accuracy of the estimation method. Taking our previous example, we can use a random sample generated from a normal distribution with parameters µ = 0 and σ 2 = 1; we can estimate its sample mean by taking the arithmetic average, and we can estimate its sample variance by taking the arithmetic average of the squared errors. If these two values are reasonably close enough to the true values of these parameters, we can conclude that our estimation method (arithmetic averaging) is accurate. In this article, I introduce a command that generates random samples from userspecified probability distribution functions with known parameters that can be used, among other applications, in such simulation exercises. 2 Probability distributions in Stata A probability distribution can take on any functional form as long as it fulfills certain conditions; namely, it must be nonnegative and it must integrate or sum to one. Some probability distributions have become so widely used that they have been given specific names. Two of the most important are the normal and the uniform distributions; among others are, for example, β, χ 2, F, γ, Bernoulli, Poisson, logarithmic, and lognormal distributions. For more information on basic statistics and random sampling, refer to Gentle (2003) and Lind, Marchal, and Wathen (2008). Several built-in random-number functions such as runiform(), rbeta(a,b), rbinomial(n,p), rchi2(df), etc., are available to generate a random sample in Stata 10 and later versions. However, if a sample is to be drawn from any other distribution function, an inverse cumulative distribution function method must be used to apply an inverse distribution function on a uniform random sample. This method builds on the fact that if x is a continuous random variable with cumulative distribution function F x, and if u = F x (x), then u has a uniform distribution on (0,1). It holds then that if u has a uniform distribution on (0,1) and if x is defined as x = Fx 1 (u), then x has cumulative distribution function F x. This essentially means equation u = F x (x) must be solved for x. For more information on the inverse cumulative distribution function method and its applications to simulation experiments, refer to Ulrich and Watson (1987) and Avramidis and Wilson (1994). To apply this method, a number of built-in inverse cumulative distributions are available, such as invibeta(a,b,p), invbinomial(n,k,p), invchi2(n,p), invf(n1,n2,p), invnormal(p), etc. They can be applied to generate random samples as outlined in the following two examples, which generate a normally and a χ 2 distributed seriesmy normal and my chi2, respectively: generate my normal = invnormal(runiform()) generate my chi2 = invchi2(n,runiform())

5 K. Lukácsy 301 However, a user may wish to draw a random sample from a distribution that is not built in Stata. In such a case, the inverse distribution function must be algebraically derived so that the inverse cumulative distribution function method can be used. But sometimes, the inverse of a function cannot be expressed by a formula. For example, if f is the function f(x) = x + sin(x), then f is a one-to-one function and therefore possesses an inverse function f 1. However, there is no simple formula for this inverse, because the equation y = x + sin(x) cannot be solved algebraically for x. In this case, numerical methods must be applied to match the values of x and y. 3 The rsample command The syntax of the rsample command is rsample pdf function [, left(#) right(#) bins(#) size(#) plot(#) ] This command generates the random variables into a new variable called rsample. pdf function is required. It is a string that specifies the probability distribution function that the random sample is to be drawn from. It must be formulated in terms of x, for example, exp(-x^2), although no x variable needs to exist prior to command execution. Two properties must normally hold for the probability distribution functions: 1. They must be nonnegative on the whole support interval of the random variable. 2. They must sum or integrate to 1. The second condition does not have to be fulfilled in this case because the rsample command calculates what the pdf function sums or integrates up to and uses that value as a rescaling factor. This way, a rescaled pdf function is used that always integrates or sums to 1. The first condition, however, must be fulfilled. If it is violated, an error message appears that reminds the user to supply a nonnegative pdf function. The rescaling constant always appears on the screen. left(#) and right(#) specify the support interval, that is, all values that the random variable can attain. left() must be specified to be less than right(). If these values are not specified, they take on default values of -2 and 2, respectively. bins(#) specifies the number of bins into which the support interval is split for the purposes of the algorithm. Essentially, it allows the user to specify the precision of the algorithm. If this value is not specified, it takes on a default value of However, if this value (whether defined by the user or its default) exceeds the total number of observations (whether defined by the corresponding optional parameter or by the size of the existing Stata file), it is automatically set to equal one-fifth of the total number of observations.

6 302 Random samples size(#) specifies the number of observations to be generated. The default is size(5000). If this value is specified even though the command is executed in an existing Stata session with a nonzero number of observations, then this value is not used and the original number of observations is preserved. plot(#) allows the user to choose whether results will be presented graphically in a histogram or whether no graphical view is needed. It is a binary option; that is, it can only take on values 1 and 0, where 1 is for yes and 0 is for no. The default is plot(1). 4 Examples In this section, I provide several specific examples with various pdf functions and various optional parameters specified. The graphical representations of the last three examples are presented in figures 1, 2, and 3 in the form of histograms of the generated random values. These three examples were generated from normal, lognormal, and Laplace distribution functions, respectively.. rsample exp(-x^2), left(-2.5) right(2.5). rsample exp(-abs(x)), left(-4) right(4) bins(500). rsample exp(-abs(x)), left(-4) right(4) size(500). rsample exp(-abs(x)), plot(0). rsample exp(-log(x)^2), left(0.1) right(6) bins(300) size(3000) plot(0). rsample exp(-x^2). rsample exp(-log(x)^2), left(0.1) right(6). rsample exp(-abs(x)), left(-4) right(4) Density values random sample theoretical pdf Figure 1. Random sample from a normal distribution function

7 K. Lukácsy 303 Density values random sample theoretical pdf Figure 2. Random sample from a lognormal distribution function Density values random sample theoretical pdf Figure 3. Random sample from a Laplace distribution function 5 References Avramidis, A. N., and J. R. Wilson A flexible method for estimating inverse distribution functions in simulation experiments. ORSA Journal on Computing 6: Gentle, J. E Random Number Generation and Monte Carlo Methods. 2nd ed. New York: Springer. Lind, D. A., W. G. Marchal, and S. A. Wathen Basic Statistics for Business and Economics. 6th ed. New York: McGraw Hill.

8 304 Random samples Ulrich, G., and L. T. Watson A method for computer generation of variates from arbitrary continuous distributions. SIAM Journal on Scientific Computing 8: About the author Katarína Lukácsy works as a statistical analyst for a large multinational company primarily focusing on improving the relevance of search results and related issues. Currently, she is finalizing her PhD dissertation on price-setting mechanisms from statistical, econometrical, and theoretical viewpoints.