Menu ENN Search
Language: English Français

Détermination de la taille de l'échantillon

This question was posted the Assessment forum area and has 2 replies. You can also reply via email – be sure to leave the subject unchanged.

» Post a reply

Anonymous 2548

General Manager

Normal user

27 Nov 2017, 14:33

Greetings,
I am about to begin a research on dietary intake of micro nutrients in children aged 6-59 months in two distinct ago-ecological areas in Ethiopia. Could you please suggest the type of soft ware for sample size determination & analysis of such type of study? Previous studies estimate deficiency of 89.5 and 23.6 in the two regions respectively.

Thank you

Bradley A. Woodruff

Self-employed

Technical expert

28 Nov 2017, 20:48

Dear Anonymous 2548:
There are a zillion websites which will help you with sample size calculations. In addition, there are many free programs you can download to do such calculations off-line. I like G-Power which can be downloaded from: http://www.gpower.hhu.de/en.html. Of course, with any statistical program, including those which provide sample size estimates, you must first understand the principles behind the calculations before asking the computer to do the calculations.

Gudrun Stallkamp

Welthungerhilfe (WHH)

Normal user

29 Nov 2017, 06:01

Hi there,
always good to keep in mind that you should also determine if cluster sampling is the choice of your sampling design or if you go for any other type of sampling.
For cluster sampling you'll have to consider the design effect (DEff). Using some of the sample size calculators online will give you the SRS sample size but they may not be able to incorporate the DEff. So you need to check if they have a provision for this or, if not, quickly do this manually (multiply by the factor you have determined suitable for your situation).
Also try to always be a little bit conservative (rounding up, rather than down), not all software or pages prompt you to do this, they often just give you a raw n considering SRS. While it's unethical to go to way too many respondents than what your sample size warrants (wasting time of more people than what's required), it is equally unethical to do a survey that samples too few respondents for a robust and defendable statement about your finding. In the worst case scenario the latter situation would render the results of your survey unusable (and that would have been a colossal waste of time for ALL respondents).
You can do quite a bit with the ENA for SMART software - that is if your sampling design fits what the software is made for. It helps with the n for a single prevalence estimate with a certain CI width, with or without DEff, and optionally you can calculate a number of HH you have to visit (e.g. if you're interested in kids in HH). It's not so straight forward if you need an n for comparing two time points or estimates though you can do this with roughly if you play with the CIs.
Alternatively, choosing the right formula for your design from a good stats text book and using, e.g., Excel, works very well, too.
It also helps to write down and have a table for your different sample size scenarios. Not all of the ones that you like from a technical point of view may fit the logistical, time, human, financial resources available. In that way and with a comment column, you can demonstrate you've done your homework, and justify why you chose a particular n scenario over another.
Best of luck!

Back to top

» Post a reply