In Monte Carlo simulations, it is a commonly used procedure to generate synthetic data based on a large survey (e.g. a microcensus) first. These synthetic data is then used as universe/population for the simulation study. Out of this universe, samples are drawn repeatedly and estimates of interest are evaluated (e.g. sampling designs or imputation methods).

Examples for this procedure can be found below:

https://www.uni-trier.de/fileadmin/fb4/projekte/SurveyStatisticsNet/Ameli_Delivrables/AMELI-WP6-D6.1-310311.pdf http://www.tandfonline.com/doi/full/10.1080/02664763.2013.859237

I am wondering if it is possible to skip this generation of synthetic data and use real data as universe. To be specific: I have a large data set with 300,000 observations. Is it possible to use this data set as universe for my Monte Carlo simulation and draw repeatedly a sample, of e.g. 20,000 observations, out of this universe? Or do I need to generate a larger synthetic data set first?

The goal of the Monte Carlo simulation study is an evaluation of several imputation methods based on different missing data mechanisms. I want to make conclusions, which of the imputation methods are best for this survey.

Any advice or references are appreciated!

  • 2
    $\begingroup$ Could you clarify how what you are describing differs from bootstrap? $\endgroup$ – Tim Mar 18 '16 at 13:12
  • $\begingroup$ I was under the impression that bootstrapping is mainly used for variance estimation of the mean for example. Would you say that the principle of bootstrapping can also be applied to more complex contexts, like in my case the comparison of different imputation methods under different missing mechanisms? Do you maybe know any references, in which a bootstrap gets used under such conditions? Thank you very much for your help! $\endgroup$ – JSP Mar 18 '16 at 18:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.