Research Sample Size


Sometimes part of being a data scientist requires that you actually do act like a "scientist" (obviously).

In this post, we're going to have a look at a "not-so" popular subject of determining the right sample size that allows you to make a proper conclusion with respect to the population that you're interested in.

More often than not, people usually assume that a sample size needs to bear some proportional relationship to the size of the population from which it is drawn. This not necessarily be the case.
Rather, at some point, having more samples need not mean a greater accuracy in doing your analysis.

What this means is, you really don't need to gather as much samples as possible in order to come up with a reasonable conclusion that can be applied to the population at large.

The absolute size of a sample is much more important. The size is pretty much dependent on the variation in the population parameters under study and the amount of estimated precision that is required by you. Thus, sometimes having a sample of 400 is good enough, while in some other scenario, more than 3000 is needed. If the variation is small, perhaps 50 would suffice.

In its most basic form of calculating the sample size, we typically assume that the probability sampling is 0.5 and we assume infinite population. In this scenario; a sample of 100 drawn from a population of 5000 would have the same estimating precision as 100 drawn from a population of 200 million.

The main difficulty with this determining the size of the population variance. In a nutshell, the greater the dispersion of data, the more samples we need. For instance, in a survey taken to determine the exam score for a group of 100 students - if we manage to survey 10 students and all of them told us that they've scored 80%, chances are the other students in that class did score 80% as well. However if all 10 of them reported a score that is different from each other, we probably would need to gather more samples to derive any meaningful conclusion.

This brings us to the topic of precision. Since we can never get a sample that can accurately reflect the population at large, we must decide the level of precision that we need. Precision in this aspect can be measured by:

  1. The confidence interval
  2. Margin of error

References:
1. Business Research Methods

Comments

Post a Comment

Popular posts from this blog

HIVE: Both Left and Right Aliases Encountered in Join

Recently was stuck on this error while trying to do a JOIN in HIVE. Below is my SQL create table ma.internet_v2 as select A.msisdn, A.imsi, A.dt, A.start_time, A.end_time, A.url, A.ttl_connection_dur_ms, A.ttl_upload_bytes, A.ttl_download_bytes, A.ttl_cdr_cnt, coalesce(B.domain_desc,A.domain_desc) domain_desc, coalesce(B.subdomain_desc, A.subdomain_desc) subdomain_desc from db.internet A left outer join hfz_domain_name_mapping B on instr(A.url, B.url_pattern) > 0; HIVE then give me and error message "Both Left and Right Aliases Encountered in Join". After going through some mail threads it seems that HIVE can support certain types of join. Only equality joins, outer joins, and left semi joins are supported in Hive. Hive does not support join conditions that are not equality conditions as it is very difficult to express such conditions as a map/reduce job. A poster from  Stack Overflow  suggested that one should use WHERE instead of ON. Duly following his advice,
Keep reading

Changing Windows Account password using Outlook

http://theessentialexchange.com/blogs/michael/archive/2007/11/24/changing-user-passwords-with-outlook-2003-2007.aspx . 1) Call the helpdesk - the most likely scenario if you are not in a hosted environment 2) Use the control panel - most Hosted Exchange providers provide a function on a user's control panel enabling the user to change their password, using the functionality in item (3). Most large companies will implement a similar capability for their users. Small companies will likely need to implement item (3) directly or their helpdesk will do it (as in item 1). 3) Enable IISADMPWD - use KB 555071 to enable the use of a web page to allow changing the password of an active directory user.
Keep reading

A Basic Recipe for Machine Learning

Image
Ever since wrapping up the three Deep Learning courses by Andrew Ng I've been meaning to write down some of the gems that he's highlighted throughout the course. One of the nice ones that I felt needed to be written down is his general recipe to approaching a deep learning algorithm/model. I've basically summarized it in a flowchart below (because everybody loves a flowchart right?) Basic deep learning recipe What is bias and variance? The below diagram is the typical explanation that I'm sure most of us are used to.  Graphical illustration of bias and variance How can we know if we have high bias or high variance?  For high bias, we could take a look at the training set performance. A poor performance is an indicator of a poor model fit, and signals that we could try to apply a bigger network to get a better fit of the model. For high variance, we could take a look at the validation set (or dev set - as Andrew calls it) performance.
Keep reading