Sample size depends on statistical significance, demand of the study, and convenience of collecting sample using survey. You may see more returns in online survey than off-line survey, if the question paper is simple to understand, interesting to read, and takes relatively lesser time. Best wishes.

However, there is no evidence-based findings in response to your question but somehow, following studies had profoundly collected larger sample size than regular paper based due to cost reduction and greater diversity of the population.

http://onlinelibrary.wiley.com/doi/10.1111/j.1083-6101.2007.00381.x/full

Article Electronic Cigarettes As a Smoking-Cessation Tool:: Results ...

Article ‘Vaping’ profiles and preferences: An online survey of elect...

As Dr Rashid points out there is no inherent difference in the statistical analysis of on-line data versus other collection media. However, your response rate, the proportion of those who respond to a request may vary.

However, if you wish to generalise from the survey, you might want to rethink your sample size. If you have 30 respondents from a population of anything over 300 the margin for error, at the 95% confidence level, is around (+/-)10-17%, depending on the distribution of responses.   That margin for error may be acceptable for your purposes but it is not very useful for most survey uses.You need at least 150 responses to reduce the margin for error to something reasonable.

If on the other hand you are taking a qualitative approach and do not wish to generalise, then 30 is a perfectly valid number.  With that number, assuming you engage people from across the range of your population, you are likely to have picked up most of the common experiences or perspectives. 

As others have said, the approach to determining a sample size does not depend on whether the sample is collected online.

But a sample of only 30 will have very low "power" which would make it almost impossible to detect any statistically significant effects. It would also be difficult to generate a useful predicted value, such as a mean score for the population, due to the very wide confidence intervals associated with such a small sample.

The subject depends on the sample size. However, larger sample sizes are preferred to validate the statistical indexes whether the approach was paper based or online .

Regards

SM Najim

Dear Tze

   Theoretically these may be same, however, in online survey, the response rate varies. Moreover, true randomisation may also be challenged.

Moreover for people perception should be assess with high number of sample size, as it has wide dispersion.

Regards

30 is also very small once you start looking at sub groups such as through cross tabulation eg binary outcome and two sexes even with perfectly balanced data you would end up with 30 / ( 2* 2) that is about 7   in each subgroup the 95% confidence interval with a n of 7 are large!

Hello folks!

This is really a very, very interesting question and something that has been calling my attention a long time ago. Related to it I have no a precise answer, but I have another doubt that I would like to share with all of you, if you do not mind.

According to the answers above, maybe 30 respondents can be considered a small sample. However, it is quite understandable that such consideration can be apllied to samples with people, specially presuming that in several cases with some additional efforts the study can get more than 30 respondents.

About this, I have been facing some difficulties related to collect data from companies, and in this kind of study my unit of analysis is not a person, but the company (my experience here in Brazil is that has been very hard to collect data from companies). In this case, in particular, should we have the same consideration about the sample size, I mean, should we consider 30 respondents companies as a very small sample size, even if I do not intend to generalize the results?

Thank you for reading my answer, and at the same time my question.

All the best!

F. Vieira

Francisco, getting companies to respond to surveys is extremely difficult.  Online is even more difficult . Even then, I think 30 is rather a small sample size even for companies as it is difficult to generalize from the findings. . However, for some topics, the total population is small.

The sample size depends on the population size, degree of variance expected in the response and the level of significance you need to attach to your study. 

Any way 30 is a small sample. if it is a qualitative study take sample size as 5 times the number of questions in the questionnaire (as a thumb rule suggested by researchers) as a minimum sample. 

Essentially, for a survey, an oversimplified 'rule of thumb' is that, the more respondents in a sample the better in order to more closely reflect the population of interest; generally, the size and structure of your sample will depend on your research context, your research objective, your proposed data analysis---and your resources for implementing the research such as time, labour,  perhaps cost, etc.

Indeed, perhaps the best first response to 'how many?' for a sample, is a series of further questions regarding what the researcher wants to do in a project.

30 might be okay for one purpose, but far too small for another.

Francisco.  I have had some issues with surveying organisations and you may find my experience relevant.  I needed to think very carefully about who I was actually surveying. 

Are you surveying individuals within different companies, or are you surveying the companies themselves?  If the latter, which is what I think you are trying to do, how do you know that those who respond represent the company?  Do they indeed represent a company view and is that what you want? What can you do to elicit the views that represent the population of interest?

If you wish to survey individuals within different companies, your sample frame then is not the companies and you have to consider how to sample them. 

Dear Tze

I would follow the literature and treat exactly as if it were a regular survey.

You could check the Timothy Hinkin article "A Brief Tutorial on the Development of Measures for Use in Survey Questionnaires".

More specifically about the pretest phase Schriesheim et al.  (1993) defend that a sample of 65 is enough. In other hand Anderson & Gerbing (1991) defend that two samples of 20 are recommended. But other authors recommend that items and respondents should be proportional. For example Rummel (1970) recommends 4 respondents for every item. Schwab (1980) defends that 10 respondents are recommended per item. 

Additionally, you could proceed the KMO test and check the results to know if your sample is adequate. For example Malhotra (2006) recommends that anything above .6 indicate the adequacy of the sample.

Give Hinkin article a try!

Best Regards

Good luck in your research.

Yves

I concur with the previous answers but nevertheless the sample needs to match the research population and the geographical area to be covered. For example, if your population is found in different geographical areas, you need more units of analysis to cover all relevant variables. (Own view no specific reference)

Good luck

The same factors that affect decisions about sample size in other types of surveys apply to online surveys.  In addition, however, internet surveys do pose special issues with obtaining representative samples and dealing with undercoverage due to access issues that need to be addressed.

I am puzzled by the suggestions that the number of respondents should be linked to the number of items in the questionnaire.  I can't see any particular reason for such an approach, especially for qualitative research.  I could see a possible argument for doing so in pre-testing but even so it would either very quickly get out of hand, or it could result too few respondents to adequately test the questions.  I am obviously missing something. 

Would those who suggested that the number of respondents be linked to the number of items in a questionnaire, please explain the logic?  (I think the question is relevant to this thread, but if people think otherwise, I will start a new question.)

I am similarly puzzled; I can only see this applying somewhat in a time-constrained project.