Overall, to obtain more reliable results and to guarantee that

the minimum sample sizes are drawn to analyze the data carefully.

It's better to don at the straighter level

because that would make your inter sample correction procedure much more efficient.

Because you are dividing up your entire population..

Finally, when you are trying to calculate the sample mean, its variance,

or if you want to calculate the confidence interval of the sample mean.

For stratified random sampling,

you essentially use the same procedure which you used for simple random sampling.

However, you need to rerate the sample units based on the stratum sizes.

And then calculate all the statistical measures.

That is the mean, the variance, both strata.

Similarly you can do the mean and variance of the mean of the sample.

And that's how you can get all the statistical measures.

However, how to determine this optimal stratum sizes when you are trying to do

stratified random sampling.

Usually the strata size will depend on your total sample size, your stratum size.

Also it depends on the variation of the observations within each stratum.

Final element which you have to consider when doing a stratified random sampling

is what do you do after the stratification?

Usually in order to improve the representativeness of the sample

after collecting the data.

You probably need to reweight the sample elements which you draw from each data

using the same weight as in traditional stratification.

Second, you have to use variables to compute the weights observed in the data.

The next type of probability sampling

which you are going to consider is cluster sampling.

Sometimes there's reason to draw groups or clusters and

observe units within the sample cluster.

This is a little bit different from stratified sampling.

Because you are not necessarily dividing up your entire population into strata and

drawing from every strata.

Here the idea is that you, again,

divide up your entire population into groups or clusters.

But on these, select certain clusters which are important for your analysis.

These clusters can be city blocks, counties, households, or firms.

Why do we need to do this?

Mainly because you don't want to look at the entire population or

each and every strata.

But you want to look at certain focused clusters which are important for

your analysis.

This will definitely reduce your cost of interviewing.

But at the same time it's more convenient when calculating or

when coming up with a sampling frame.

Because if there's certain elements missing from your sample

you can collect data from these clusters.

So overall, these three types of

probability sampling procedures are very important when you want to run

a very formal statistical analysis based on your secondary data.

In the next video we are going to consider other aspects of secondary data.

Mainly the different data types.

How do you run analysis on that data

using different methods like one way tables, two way tables.

Then correlation analysis, motivational analysis,

as well as regressional analysis.

Which are probably the most important aspects of a marketing research study.

Thank you. [MUSIC]