If you are the only one sending it out (so others are not re-posting or sharing), then you are sampling everyone on those channels. The tricky bit is calculating the response rate.

Noha Fadi - First, you have to decide is your study quantitative or qualitative. If quantitative, I think you should not use snowball or convenient sampling method as these are non - probability sampling methods. Since you are using online survey, probability sampling methods such as random sampling, stratified sampling, systematic sampling, quota and purposive sampling methods are most suited to your case. However, I would recommend simple random sampling technique. In case you have a qualitative research design, convenience sample would fit in your area of study.

A follow-up phone call would help you ensure that the response is original. Otherwise, you cannot understand whether the response is given by the right person to whom the questionnaire was mailed. Better, make sure your sample is at least 20 per cent of the total population size. However, you need to calculate effect sizes and statistical power of each result and analyse the results based on the threshold values of effect size and power. Keep a record of the responses received. Best wishes...

As you've chosen both non-probability methods. You'll be experiencing both sampling error and relative bias. As per your required results, the snowball method will be a good fit as you'll taking referrals from initial subjects to additional subjects. But using this snowball method will reduce your likelihood to obtain a good cross-section of relative data, for this you should also consider using probability methods.

It will be a convenientcesampling method. Maam please do take care of high attrition rate & sampling error in this type of non-probability sampling technique. A SRS is a better solution for you to complete your observational epidemiological study.

You can use both probability and non-probability sampling techniques based on your study population and study objectives.

yes, we can use probability and non-probability sampling techniques in online survey.

Hi! If the exact number/statistics of sampling frame is not available in that case it would be better to use non probability sampling. Thanks.

You can check for purposive sampling procedure.

If you have the list of all emails you can send your questionnaire to, then you can randomly sample for participants, but if not I don't think there is any other feasible choice than drawing a convenience sample.

Critical question: Are the people completing the online survey: (A) from a well-defined population – one that can be enumerated (listed), or (B) simply people who found a website that interested them and agreed to complete the questionnaire?

In situation “A” you can send invitations to a sample of people from a sampling frame (e.g., a membership list). If the completion rate is adequate and the sample size is large enough, you can use the sample survey results to estimate population values reliably. Examples of a well-defined population might be: all students registered at a school, all employees in a company, all customers who patronized a company or agency in a given period of time, or all vehicles that were registered in a year.

If the answer is “B,” which is typical of most online surveys, there is no way to do a true sample survey, because no one knows the population, other than some subset of internet users who were attracted to the site and decided to participate in a survey.

Net: if the researcher invited specific people (respondents) to participate in a survey, they presumably had a list of potential respondents. In this case, they can do a sample survey and estimate population parameters with a known probabilities using classic statistical techniques. They can also do significance tests when they compare values for different subgroups in the population. On the other hand, if the survey respondents are simply people that found the site and took the survey, it is impossible to define the population, other than users of a specific site at a specific time who decided to answer the questions. That is not a very interesting group for most research purposes. The results might help the website owners improve their offering or entertain their users, but that is about all.

Many people in situation “B” try to fake a representative sample by weighting specific subgroups according to their frequency in a well-known population. For instance, if 50% of the people in the US are female and 50% are male (the percentages are actually 50.8% vs. 49.2%), but the pool of online respondents ended up being 80% male and 20% female, the analyst will weight each female response as if it represented four female respondents. Usually, these weighting schemes involve multiple demographic variables (e.g., gender, age, ethnicity, socioeconomic status, region, primary language) or categorization variables (e.g., type of service consumed, day of week or time of day). The fake results might look like the population distributions, but they are not representative; and the statistical methods normally used to estimate population parameters are inappropriate.

Most general access internet surveys represent convenience samples that cannot be generalized to any known population of interest. They are often amusing (do you like cats or dogs better?), but they have no bigger meaning. Do not be deceived by very large numbers of respondents… 25,000 people who took an online survey represent 25,000 people who took an online survey, and the results cannot be generalized to any known population, not even the users of that particular website (because the survey respondents are probably not a random sample of all visitors). Because online data collection is so cheap and fast, many academic researchers, particularly social psychologists, have embraced these non-representative methods. Harvard’s Project Implicit, which offers Implicit Association Tests, is a prime example. We have no idea how the results of these surveys and “experiments,” based on convenience samples, generalize to any meaningful population. In other words, the online IAT studies lack external validity (and the internal validity of the implicit association test has also been questioned too). In contrast, many customer satisfaction surveys, which are sent to customers after a service occasion or product purchase, meet the requirements for sample surveys. But note, the results can only be generalized to one firm’s customers, not all of an industry’s customers.

The way your project is described – “The survey will be distributed online via social media channels. We are targeting adolescents” – suggests situation “B.” The result will be a convenience sample that cannot be reliably generalized to the adolescent population. So, you have to decide, given that the planned online survey is easy but unrepresentative of all adolescents, is it better to have some information, however biased, rather than none at all? What would be the practical impact of misapplying bad estimates in public policy decisions?

Online survey usually has very poor response rate. Considering that you are targeting adolescents , the response rate is likely to be even lower ! It is better to use the snowball technique. Convenient sampling would bring desired result only if you have a very large collection of contact details of the target group.

Many thanks. I really appreciate all your replies.

social media surveys are tricky. if you really segment your audience and make sure only the people eligible to respond are in fact responding, then you can use whatever mean you deem fit (prob or non-prob). in this case, if you ensure respondent eligibility, using CSM (convenience) can be easier to undertake.

I won't recommend snowball because we had many issues (both bias and response error) when we used this technique in a previous study. we had to then go back and solve this issues in a different research to provide backup for initial survey. but this is more of a personal comment.