The reading I've done basically says that it is not possible to make inferences about a wider population when using non probability sampling. However there is obviously still a lot of value in non-probability sampling, apart from reduced cost. For example, a lot of market research is derived from online panels. Academics also use non-probability sampling, I remember a study that we came across last semester in HMS 771 where Imma referred us to a paper called " How are high school studies related to performance in a psychology degree". The criticism was that this paper used a convenience sample; students from University of Woolongong. My comment is that it's interesting that a theoretically flawed sampling method can still produce valuable information. It would be interesting to know if there is a theoretical view that explains this.
Cluster Designs These designs seem to be predominantly used in face to face interviewing where the geographical dispersion of respondents is an issue. Are there any instances where one would use a cluster design in either telephone or online interviewing?
-------------------------------
Agree on the usefulness of studies using non-probability samples. I know that much of the academic research in areas that I have worked in are by and large convenience samples. Pragmatic considerations dictate that as the only option for most people doing student projects where dependence on subject pools and friends (and friends of friends) completing surveys is routine. Although it is often used as a criticism (e.g., psychology is the science of first year undergraduate students)....I figure that that is where not overstating your findings and the importance of replication come into play. Sometimes the large representative probability samples are guilty of other sins (like reporting significant 0.09 correlations from their huge sample as if they had any practical relevance).
re: the cluster designs, not something I am hugely familiar with, but I got the impression they're also useful when you want to draw a representative sample but don't have an appropriate frame. So I guess then you could use telephone of online interviewing in cluster designs if your main problem was the sampling frame and you needed to winnow down to a known frame?
----------------------------
Graham: "…it's interesting that a theoretically flawed sampling method can still produce valuable information. It would be interesting to know if there is a theoretical view that explains this."
Even with probability sampling there will be some sampling error. Probability sampling is preferred because you can estimate the sampling error. Sometimes it is simply impossible to obtain random samples. You will not be able to estimate sampling error in this case, and that's why, as Kate mentioned, "importance of replication comes into play." Replication and confirmation of finding in other studies will give you greater confidence in the generalizability of the results.
Graham: "Cluster Designs These designs seem to be predominantly used in face to face interviewing where the geographical dispersion of respondents is an issue. Are there any instances where one would use a cluster design in either telephone or online interviewing?"
Kate: "cluster designs …they're also useful when you want to draw a representative sample but don't have an appropriate frame. So I guess then you could use telephone of online interviewing in cluster designs …"
Spot on. Cluster designs are used to overcome the problem of the lack of sampling frame. After you obtain a sample using cluster design you can use any type of data collection methods, including telephone and online methods. For example, suppose you want to survey tennis players who participate in social tennis competitions in Melbourne. There is no available sampling frame of those players. However, a list of all tennis clubs in Melbourne is available. You can contact clubs and ask their permission to contact members via telephone or e-mail.
No comments:
Post a Comment