I conducted two surveys through a Facebook application. One of the topics were anonymity and aggression. The methods that i followed were a mixture of Opportunity sampling (post in fan pages and group walls etc.) and Snowball sampling (provided the users that took part in the survey at the end to invite their friends). Random sampling is really hard on Facebook due to people being reluctant to respond to pms that come from unknown(to them) people.
How easy is to publish a research were the sampling is not random? Do you know of any journals that do publish opportunity sampling research?
If you look at the literature, you will find that almost all journal articles are based on convenience samples. Sampling is a highly misunderstood issue in social science research. There is no doubt that random sampling would be the ideal case, but most of the time it is virtually impossible to achieve. First you need to consider what your population is. In your case it seems to be Facebook users. Theoretically a random sample would be fairly easy to achieve, but since Facebook does not cooperate it is practically impossible. Are your results worthless? Definitely not! This is mainly an issue of generalizability and huge differences between scholarly research and market research exist. Contrary to what many researchers believe, generalizability is LESS important in scholarly research. Most likely you investigate relationships between variables, which are more robust than characteristics of e.g. demographic data. Furthermore, academic research has various paths to generalization, e.g. theoretical generalization or replication studies. If your study is based on theory, a convenience sample might provide sufficient results. Read the paper from Blair and Zinkhan (2006) for further details ("Nonresponse and Generalizability in Academic Research, Journal of the Academy of Marketing Science, Vol. 34, Iss. 1, pp. 4-7).
If you conduct an experiment, the main feature is not the random sample, but the random assignment of subjects to different treatments. Many people even in academia do not understand that difference.
If you look at the literature, you will find that almost all journal articles are based on convenience samples. Sampling is a highly misunderstood issue in social science research. There is no doubt that random sampling would be the ideal case, but most of the time it is virtually impossible to achieve. First you need to consider what your population is. In your case it seems to be Facebook users. Theoretically a random sample would be fairly easy to achieve, but since Facebook does not cooperate it is practically impossible. Are your results worthless? Definitely not! This is mainly an issue of generalizability and huge differences between scholarly research and market research exist. Contrary to what many researchers believe, generalizability is LESS important in scholarly research. Most likely you investigate relationships between variables, which are more robust than characteristics of e.g. demographic data. Furthermore, academic research has various paths to generalization, e.g. theoretical generalization or replication studies. If your study is based on theory, a convenience sample might provide sufficient results. Read the paper from Blair and Zinkhan (2006) for further details ("Nonresponse and Generalizability in Academic Research, Journal of the Academy of Marketing Science, Vol. 34, Iss. 1, pp. 4-7).
If you conduct an experiment, the main feature is not the random sample, but the random assignment of subjects to different treatments. Many people even in academia do not understand that difference.
First of all i have to say a big WOW and thank you for the detailed response! I already downloaded the paper and its on the top of my todo list.
Indeed in my case the subjects were assigned randomly under the different anonymity states. My experiment was based in the theory that basically more anonymity produces more aggression but i studied the actual specifics between different anonymity states.
So far i got one journal saying to me that i didn't sample the population, this were the exact words, and i am guessing they meant that my sample was not random.
According to my opinion since there are different groups that get different treatment and people are randomly assigned to these groups there is no reason not to think that in general the whole facebook population would behave in a similar way.
I was quite puzzled when i received this answer from the journal (only from the editor, the paper never actually went for a blind peer-review) and i fear that it might have been an excuse as well.
Horst answer is very good, but I think what the reviewer is trying to tell you is that sampling friends has a known bias. Research into social networking has shown that individuals with ties (social relationships) will frequently have the save views on many topics. Question did you define the population you were going to survey. Not that I'm an expert in sampling (is anyone) but the best sample is a representative sample. A representative sample is that type of sample used in political elections. In a election survey there are more question to establish where you fit in the population than what the poll is about. Then the raw data is adjusted to deal with over and under sampling. For example, a US Presendial poll requires 400 to 1200 population. A random sample is a set of randomly selected subject having the characteristics described in the population definition. Here sample sizes are what ever you want, the more the better, but you need to explain how the subjects surveyed fit into the model, and account for any bias (over/under sampling of a characteristic) . Different pubs have there own protocol, you could as the editor for guidance.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Human-Computer Interaction