Perhaps this is not an either/or question (the results either do generalize or they do not). Are the thousands of experiments (including the strong designs with random assignment) based on random samples? That is rarely, if ever, the case. Can any conclusions be made about people based on experiments? I think the answer is yes, especially when we see a similar result across experiments in different ecological settings with different kinds of people. Have any good theoretical generalizations come from research not based on random samples? Think about Piaget. Could the study discussed here be viewed as one study needing to be "replicated" with other people, but it is still a study that can provide some tentative and useful working knowledge? Did you expect to generalize from a single empirical research study (other than in polls) in your past research? If yes, who did you generalize to? All humans? All adults in the US? All members of our excellent educational research methodology list? Please provide an example for us to discuss. Hopefully the authors of the study are collecting demographic information at the end of their questionnaire and will use that information to inform their readers of the characteristics of their sample. Then intelligent readers can engage in making what Robert Stake called naturalistic generalization, comparing their own/local people to the sample and then making cautious generalizations based on similarity; that is, the reader rather than the researcher might make any appropriate generalizations. Stake's idea was similar to Don Campbell's concept of proximal similarity, or generalizing based on similarity. Related to this discussion, do you ever use p values on empirical (including experimental) data based on convenience samples? Or do you only use p values when you have a random sample from a delimited population? I am not saying here what I think is "correct" on this issue; I'm just asking the question. Getting back to the idea of asking everyone to participate in the internet study, could the EDRESMETH group be considered a population and those that don't participate non respondents? The response rate would then apply to the population rather than to the sample. Would p values and inferential statistics be needed, if one assumes he or she is working with population data (with less than a 100% response rate)? 

I used to be a stickler about random sampling, and I still am if one is claiming to use a single sample to provide statistical estimates of parameters in a particular population. However, I think there are additional varieties and degrees of generalizations. 

Again, I LOVE random samples! I just don't think that the lack of a random sample necessarily invalidates a study (unless, again, the author claims to make a statistical generalization from a single sample to a population). 

In short, I think the answer to the original question (to generalize or not) requires a more nuanced answer. 

I also think we should be a little kinder to those asking us to participate in their research. Let's shift our discussion to something like this: "Can one generalize from internet research?"

Cheers!

Burke Johnson