Hello,

G power analysis, free online tool. Just google it. 

Regards

Almost all of the formulae that you will find for sample size assume that the population is infinite. They also assume that you are addressing just one issue which is either a binary (Yes/No, Male/Female) response, or a linear response with a mean and standard deviation that you already know. Most of these formulae were designed for tasks like testing failure rates of light-bulbs.

None of these formulae is helpful if you have several topics that you want to deal with, and if you are using the usual questionnaire response formats, such as Agree-Disagree rating scales, and more than two categories.

Good survey results can be gained with sample sizes of 300 or so.

Instead of thinking about the formulae, think about the form of analysis you intend to do with your data.

If you want simple frequencies and averages for the whole sample then a sample of a couple of hundred is usually more than plenty.

If you especially want to examine and compare frequencies or averages among sub-groups within your sample, then you may want to ensure that there are at least 100 or so of each group.

If you intend to run crosstabulations (comparing joint frequencies between two categorical variables) then you'll want expected frequencies more than 20 in each cell. So, for example, if you have 4 ethnic groups, and 4 levels of health care, then you will have 4*4=16 cells. So you want at least 16*20=320 respondents, or 4*20=80 respondents from each ethnic group - whichever is greater.

As the standard sample-size formulae clearly show, all other things being equal, a larger sample size will deliver more "accurate" results. The formulae also show that, beyond a point, a very small improvement in accuracy costs an enormous increase in sample size.

Unfortunately, too many researchers focus on sample size because it's the one thing that can be measured with confidence. But there are many other issues, that cannot be measured easily, that have a far greater effect on the validity and accuracy of your survey results. Sampling frame (asking the right people), questionnaire format, response format, question wording, questionnaire length and complexity all can conspire to create bias in your results. And that bias can be far far greater than the errors created by differences in sample size.

If you haven't done extensive pretesting of your questionnaire, and your sampling frame is not well representative of your population of interest, then a larger sample size just gives you more confidence in a wrong answer. 

Okaro  -

In general, a stratified random sample can reduce sample size needs, but only at the aggregate level, not if you want to provide inferences for each stratum, which would instead be just subpopulations/categories rather than strata. 

A cluster sample/area sample actually increases sample size needs, but can help solve logistical/administrative problems. 

A pilot study can help you decide on your sample design, sample size needs, and test your questionnaire.

You are planning ahead, and that is very important.

I dealt with continuous data.  Continuous data and yes/no data are covered in sampling books such as 

 Cochran, W.G.(1977), Sampling Techniques, 3rd ed., John Wiley & Sons,

Blair, E. and Blair, J.(2015), Applied Survey Sampling, Sage Publications,

and

Lohr, S.L.(2010), Sampling: Design and Analysis, 2nd ed., Brooks/Cole. 

[I don't know what books to use for likert and other data, but I assume that the ideas of stratified random sampling, cluster (random) sampling, and pilot studies carry across to various data types, including qualitative data.]   

There are also model-based, and model-assisted design-based methods, which can be very useful, but I think they are likely out-of-scope here. 

Sample size calculators on the internet are likely for yes/no questions only, and for the worst case regarding proportions, and ignoring the finite population correction (fpc) factor.  Best if you look at a statistics book. The sample size needs depend greatly on sample design, and type of data, and where applicable, the standard deviation for a population (or residuals or random factors of residuals). 

If you are collecting data for scientific/statistical inference, and no design will give you a manageable sample size for the populations you have, then trying to make such a goal may cause an increase in nonsampling error that could reduce the quality of your results further. Remember that it is not just sampling error, but nonsampling error as well that you need consider, and not just variance but bias as well. 

Best wishes with your project - Jim

Hume Winzar provides you with a practical and apropriate response. Follow his advice. It is theoretically solid and practice oriented.

Good luck

There are so many as for example that suggested by Knaub. However, i Concour with Jose's advice for you to go for the suggestion given by Winzar as it is more practical. In fact i am right on my way to trying it on a large developing country sample. Goodluck to you.

Here is a simple online opensource tool that you can use and reference to determine the sample size of a given population against which a questionnaire can be administered.

http://www.surveysystem.com/sscalc.htm

I would also advocate looking at Power analysis estimation, which is not limited to binary comparisons (t-Tests). For example, if you are using correlations, you can determine the sample size that you need according to the magnitude of the correlation that you want to detect with a given degree of confidence.

Alternatively, if you are interested in something as basic as a t-Test, there are many sites that will give you projected sample sizes. (I'm not sure where the "lightbulb failure" reference comes from, since the formulas involved are simple algebra and will apply to the sample size for any Yes-No question.)

Thank You dear all.Your answers are as illuminating as they are interesting.

Okaro -

One-size-does-not-fit-all.  It could be a substantial waste of resources to assume some sample sizes without more information.  A pilot study may be very important, perhaps even more so if you do not know how statistical science works for sample and sample size selection.  It would make your guesses more reasonable, and also help work out logistical and other issues.  You might consider it anyway. 

Jim

Dear Okaro, the answer given by  Hume F. Winzar is absolutely perfect. That is exactly that I use to say to my students when they are asking me such a question. Really, sample size is important, but there are many other aspects of surveys (as Hume rigthly points out) that will contribute to the validity of your measures, analyses and to the overall external validity of your research.