An Introduction to R: Monte Carlo Simulation - PowerPoint Presentation

Slide1

An Introduction to R: Monte Carlo Simulation

MWERA 2012

Emily A. Price, MS

Marsha Lewis, MPA

Dr

. Gordon P. BrooksSlide2

Objectives and/or GoalsThree main parts

Data generation in R

Basic Monte Carlo programming (e.g. loops)

Running simulations (e.g., investigating Type I errors)Slide3

Why Use Monte Carlo Methods?

According

to Mooney (1997) Monte Carlo simulations are useful

to

Make inferences

when weak statistical theory exists for an estimator

Test null

hypotheses under a variety of plausible conditions

Assess the quality of an inference method

Assess the robustness of parametric inference to assumption violations

Compare estimator’s

properties Slide4

What are Monte Carlo Methods?

Experiments composed of random

numbers

to evaluate

mathematical

expressions (Gentle, 2003)

Empirically

determine the sampling distribution of a test statistic

Computer-based

methods for approximating values and properties of random

variables

(Braun

& Murdoch, 2007

)Slide5

Logic of Monte Carlo

Mooney (1997) presents five steps

Specify the pseudo-population in symbolic terms in such a way that it can be used to generate samples. That is, writing code to generate data in a specific manner.

Sample from the pseudo-population in ways that

reflect the

topic of interest

Calculate

θ

in a pseudo-sample and store it in a

vector

Repeat

steps 2 and 3

t

times where

t

is the number of

trials

Construct

a relative frequency distribution of resulting values which is a Monte Carlo estimate of the sampling distribution of under the conditions specified by the pseudo-population and the sampling proceduresSlide6

Practical Issues/ Considerations

What software to use?

How much time to run the simulation?

Reproducibility of results

Adequacy of random number generator Slide7

Why use R?

It’s FREE

It is a flexible language that can be controlled by the user

It uses a vector based approach

Depending on the package, there are built in commands which the user can access and minimize the amount of programming required for MC simulation

Make sure to load the require packages at the beginning of the session

R

community has a plethora of information:

help websites,

listservs

, textbooks, blogs

Manuals for R available at http://cran.r-project.org/manuals.htmlSlide8

Part 1: Data Generation

RNG and

setting seed

Purpose of the

seed is

to recovery results

Initialize all parameters of interest

Loops

Print results

Access outputSlide9

Generating a Single Random Variable

R has four parts: CDF, PDF,

Quantile

function and simulation procedure

dnorm

,

pnorm

,

qnorm

,

rnorm

respectively

r

norm

(

x,mean

=0,sd=1

)

runif

(20,min=2,max=5

)

Distributions:

normal, uniform,

poisson

, beta, gamma,

chisquare

,

weibull

, exponential Slide10

Try it, you’ll like it!

r

norm

(

x,mean

=0,sd=1

)

G

enerate

a normal distribution

of 50 values with

a mean of 50 and

sd

of

10

x <-

sample(1:2,20,TRUE,prob=c(1/2,1/2))

Generate data that mimics

rolling a

dieSlide11

Generating Correlated Data

X~Normal

(20, 5),

Y~Normal

(40, 10),

corr

(X,Y) =0.6

4 inputs

Sample size, mean, variance-covariance matrix, and method

3 methods of data generation

Eigenvalue (default), Singular

V

alue, and

C

holesky Slide12

Try it, you’ll like it!

rmvnorm

(n,

mean, sigma, method)

Generate data for 3 variables such that

X

--Normal (20, 5), Y-- Normal (40, 10),

Z

-- Normal (60,15

) and

Corr

(X,Y

) =0.6,

Corr

(X,Z) = 0.7,

Corr

(Y,Z)=0.8Slide13

Part 2: Basic MC Programming

Four steps (

Braun

&

Murdoch, 2007

)

Understand the problem

Work out a general idea how to solve it

Flow charts

Translate your general idea into a detailed

implementation

Turn the flowchart into code

Check: Does it work?Slide14

Programming Commands*

Loops

f

or,

i

f,

i

felse

, while

Statements

repeat, break, next

* We can’t cover all programming aspects but wanted to mention other commands

Slide15

Functions

They

are “self-contained units with a well-defined purpose” (Braun & Murdoch

, 2007

, p. 59)

Take

an input, do some

calculations,

and produce an output

In R, functions are objects and can be manipulated like other more common objects such as vectors, matrices, and lists.

R provides source code for its own functions

R allows you to write your own functionsSlide16

Part 3: Running SimulationsTrimmed mean sampling distribution

Replicating a

published Monte Carlo

study

in

R.

Zimmerman

, D. W. (2004). A note on preliminary tests of equality of variances.

British Journal of Mathematical and Statistical Psychology 57

,

173–181.Slide17

QuestionsThank

you for your timeSlide18

References

Braun, W. J., & Murdoch, D. J. (2007).

A first course in statistical programming with R

. New

York: Cambridge

University.

Gentle, J. E. (2003). Random number generation and Monte Carlo methods (2nd ed.). New York: Springer-

Verlag

.

Mooney, C. Z. (1997).

Monte Carlo simulation

(Sage University Paper series on

Quantitative Applications

in the Social Sciences, series no. 07-116). Thousand Oaks, CA: Sage

.

Zimmerman, D. W. (2004). A note on preliminary tests of equality of variances.

British Journal of Mathematical and Statistical Psychology 57

, 173–181.Slide19

Our code Slide20
Slide21
Slide22
Slide23
Slide24
Slide25