Running simulations in R

[noetsi]

I have wanted to do this for a long time. But the package that does this in SAS (PROC IML) is too expensive. I was wondering, and remembering that I am a beginner in both R and simulations, what a good R package would be to do this.

 

[Dason]

There might be packages for specific examples but in general you don't need packages to do simulations unless you're simulating from a specific distribution that isn't provide in base R.

 

[noetsi]

Do you mean there are functions that tell the base r to simulate data given certain assumptions?

 

[Buckeye]

I've tried to do this sort of thing before. I found this article which gives a basic linear model example: https://aosmith.rbind.io/2018/01/09/simulate-simulate-part1/

Generally, I always find my way back to sample size calculations. Or simulating an effect size of some sort. There's really no cookbook for it. So, I google until I find something useful.

 

[hlsmith]

Base R = the basic R environment without installing packages.

Do you know how to do this in SAS? If so it is just the same principles.

There are some packages that can create synthetic data comparable to your own data frame, which is interesting, but not necessary. I will post an interest toy example I stumble upon a couple of weeks ago.

 

[fed2]

you can also submit R code in proc iml with 'rsubmit'. probably its killer app is outsourcing to free software..........

 

[hlsmith]

Also, @spunky teaches a class on data simulation and love his copulas.

 

[spunky]

Also, @spunky teaches a class on data simulation and love his copulas.

Copulas are love
Copulas are life

 

[Dason]

Copulas are meh

 

[noetsi]

Base R = the basic R environment without installing packages.

Do you know how to do this in SAS? If so it is just the same principles.

There are some packages that can create synthetic data comparable to your own data frame, which is interesting, but not necessary. I will post an interest toy example I stumble upon a couple of weeks ago.

You need Proc IML to do it in SAS. I have an awesome book for that. But I will never have Proc iml

 

[noetsi]

Copulas are love
Copulas are life

They are way beyond my expertise, interest, or intelligence. I am trying to do basic simulation. Say I think there is an issue with a method (or more likely others do). Then I could test the various arguments with a simulation.

That is what I would like to be able to do.

 

[Dason]

Yeah so you just need to simulate some data using rnorm or whatever function you want to create samples from the distributions of interest.

 

[noetsi]

My concern with that is many simulation studies set a series of limits on the data, for example it has to be in a specific range, not just there be a specific distribution. But as I go through the R code it probably allows you to do that. When I find an example of what I mean I will post it.

 

[Dason]

They might just use a uniform distribution within that specified range to simulate from that then.

 

[hlsmith]

Yeah with triangle dist I think you put the min and max in. Also, with say ~N, you can just use the variance value to target a certain data range.

You just need to find an example of get your feet wet!

 

[jamesmartinn]

I echo @Dason's remarks about not really needing any specific packages to do simulations in R. I understand the pain points of PROC IML, but R overcomes this kind of by default. I really prefer building simulations from the ground up as it also provides you a chance to think through (or reaffirm) your understanding of methodology and it's components.

If you're looking for a package/suite to help with more structured simulations out of the box, maybe this can be of some help:
https://cran.r-project.org/web/packages/SimDesign/index.html