We are living in a Big Data world. A massive amount of data is generated every day, and we are trying to crunch the data to derive useful information out of it. It is one of the reasons that fueled the growth of Advanced Analytics or Data Science. The machine learning and statistical methods are highly benefitted when we supply them with the right volume of data. However, we can develop good models even with reasonable datasets. The trick here is a proper data sampling technique.

Sampling is nothing more than a subset of data. Sampling methods can be broadly classified into two types:

1. Probability Sampling: In this technique, the samples are chosen at random, and each sample has a known probability of being selected. There are different kinds of probability sampling techniques and will be discussed in this article.

2. Non-probability Sampling: In this technique, the odds of observations being selected is not defined and selection is carried out by the subjective judgment of the researchers.

Probability Sampling using R

Simple Random Sampling

A simple random sample is generated by a design, which warrants that each subgroup of the population of size n has an equal probability of being picked as the sample. In R, we can draw a random sample of size 10 from the numbers 1 to 1000, using the sample function sample (1:1000, 10) . If we try this multiple times, we might get different results. This is correct behavior, however, sometimes we might need the results to be reproducible. In that case, we need to set the seed value. If we set the seed value, the random number sequence is set to a known value. In R, the random number generator is not truly random but pseudo-random as it is generated by an algorithm that returns the same pseudo-random sequence whenever we start seed for the process. We can use the set.seed () function to specify the seed value. The value can be any numeric value.

Another important argument in the sample() is the replace parameter. By default, it is set to FALSE. This means the first number picked for sampling will not be picked up again in the sample. When we set it to TRUE, the sample elements might repeat in the sample. This is also known as sampling with replacement.

Sometimes, we might need to use sample function to simulate a poll or survey. For example, we plan to design an experiment for a survey where we plan to know whether people like our products. The two responses that we are allowed are "Yes" or "No". Before sending out the surveys to 500 people, we prepare a demo analysis for which there is a need to generate a sample with a response ratio of 9:1. To simulate this we can do the following:

sample(1:20000, 500, replace = TRUE, prob = c(0.90, 0.10))

Stratified Sampling

In stratified sampling, the data population is divided into groups, and samples are taken from each group. The partitioning of the population into groups is called strate, and simple random sample for each group is called stratum.

In R, we can generate stratified sampling using various methods. One of the ways is shown below:

library(dplyr)
set.seed(1)
sample_iris <- iris %>%
  group_by(Species) %>%
  sample_n(10)

There are other sampling methods, like cluster sampling, systematic sampling and multistage sampling which are not widely used in the industries. In R, you can explore about this from the package survey.