Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

This is only for my high school senior project, so it doesn't have to be perfect. I'm doing a project on global warming, and I want to survey people for their opinions. I know that if I use a convenience sample of my classmates, I'll have a lot of bias. I was wondering if there was a site on the Internet where I could post my survey for random people to answer, so that I could get as close as possible to an SRS. If not, I'd be interested in some other suggestions for ways to carry out my survey.

share|improve this question
Fairly sure that, on the internet, there's a higher ratio of trolls to sane people than there is in the real world. That might bias your results. Readers of large blogs, such as Pharyngula or Andrew Bolt's blog, have been known to crash polls by overloading them, and/or totally bias the results. It's pretty common that someone will put up a poll relating to the atheist/creationist flamewar, and within a day it will go from 100:20 in favour to 300:2000 against, due to Pharyngula readers. Be prepared. – naught101 Aug 7 '12 at 3:17
See self-selection bias – Glen_b Jul 10 '13 at 11:08
up vote 9 down vote accepted

For a high school project it will probably be difficult to gain a representative sample with your limited time and budget. I'd be somewhat skeptical an online vendor can achieve a random sample of anything. If I had to do it I think I would use the phone book and either call or go to the addresses to distribute the survey in your community. Although this isn't perfect (likley under-represents younger people, minorities, and people of lower socio-economic status) it makes the project more feasible in your situation (as opposed to conducting stratified sampling based on geographic regions). Both of these solutions are local samples, and would not get you a national sample. An agency I work for samples the local community based on parcel addresses and apartment listings obtained from the postal office (these sources of information are not always public though, and would require more front end work than opening up a phone book).

idclark is right and the General Social Survey has several variables addressing public opinion on global warming. You could either analyze that data directly (as it is open to the public) or if you are required to construct your own survey you could mimic the GSS's questions and then see how close your sample appears to be compared to the GSS. At the moment I am having a much easier time searching the GSS variable lists from the ICPSR archive than I am from the actual GSS website. You can actually conduct basic data analysis online at ICPSR of the GSS data (you can basically do frequencies and cross-tabulations). If all you need are frequencies this circumvents the need to use stat software such as SPSS or Stata.

Getting a representative sample is difficult not only because true populations which to construct your sample are sometimes hard to come by, but just as importantly not everybody you do sample responds to the survey. I wouldn't be fixated on a national sample either, I'm sure you can construct just as interesting questions to answer by estimating public opinion of the local community, school mates, teachers, etc. What I find interesting is in comparison (like say comparing your school teachers opinions to those of the students). You could also conduct an experiment via surveys (like say manipulate how you present arguments for or against global warming and see how people differ in their opinions of global warming).

share|improve this answer
Thanks you so much for the suggestions. – George Dec 15 '10 at 1:43

Your best bet is mechanical turk, It will cost you some money, but not much. If your questions are short and can be done in ~ 1 min you can easily charge 10 cents or so per answer, so if you want 50 responses it will cost only $5. You can get the data formatted in .xml, which you can import into R or even (shudder) Excel. The people you sample will be much more varied than you'd get in a high school class. And you can ask for demographic info (age, gender, nationality, etc), as questions in your survey. Also making the survey is easy, just adjust amazon's templates.

See for more issues with the biases you mention, the WEIRD study. Also see why people at mech turk participate:

share|improve this answer
Unless OP wants to primarily survey folks in India and Mexico, it's unlikely that mechanical Turk participants will form a representative sample of the demographic OP is looking for... – Cerin Aug 31 '15 at 15:23

I admire your ambition! To answer your question directly, I'm not familiar with any site. Perhaps a message board you're familiar with? Has your instructor addressed the issue of bias? It may be enough that you recognize the issue of bias in your sample and speak to how this will affect your survey. Most properly randomized nationally representative surveys have an N of somewhere between 1800 and 2300. The GSS probably has a few questions regarding global warming. However, this will be inaccessible if you're not familiar with SPSS or Stata. I know others on here can help you better than I in regards to weighting responses to correct for bias.

share|improve this answer
If you want to open an SPSS file without paying for software check out: PSPP: or read.spss in R… – Jeromy Anglim Dec 15 '10 at 4:23
@Jeromy: I had trouble opening an SPSS sav file in the past, but was unable to solve them via PSPP. Here's my blog post about it: – Galit Shmueli Dec 15 '10 at 4:59
@Galit Your post is from September 2008. It might be worth trying again. Alternatively, use the read.spss() function in R. – Jeromy Anglim Dec 15 '10 at 5:07
@jeromy Ah yes, i completely forgot! I haven't used PSPP, but I use read.dta several times a day. thanks for the reminder. – idclark Dec 15 '10 at 5:31
I have heard that spss.get from the Hmisc package should do a good job as well (potentially better than read.spss). – Henrik Jan 22 '12 at 17:07

a sample representative of population cannot be obtained through internet as you will only get people interested in answering your survey online, which will give you a biased sample.

share|improve this answer
This statement is only true if you assume that the other methods of survey recruitment contain less bias than an all internet based sampling approach. The fact is that all sampling methods contain some amount of bias, and it is incumbent upon the researcher to be cognizant of what those biases may be and to account for them to the extent possible. Where accounting for that bias is not possible, documenting the limitations of your findings is critical. There is research that shows internet based sampling methods induce less or equal bias than other "traditional" methods for certain surveys. – Chase Dec 14 '10 at 14:01
OK. interesting. I guess also that it depends on the type of survey you want to do. – RockScience Dec 15 '10 at 3:07

Marketing people are using self-selected samples in online-surveys all the time, so they probably have methods for cleaning their data. Sadly, I don't have a good pointer right now to look for their methods.

share|improve this answer

Just a thought, but it might make sense to implement an inexpensive Google Ad Words campaign where at least the participants would be coming directly from Google search and you can somewhat control some sort of stratification. Of course, it is never possible to get a true sample.

Along these lines, there is some work being done nowadays to adjust for selection bias of Web Surveys via propensity scores. This is being done by some of the majors, including Harris.

-Ralph Winters

share|improve this answer

protected by chl Apr 28 '14 at 11:49

Thank you for your interest in this question. Because it has attracted low-quality or spam answers that had to be removed, posting an answer now requires 10 reputation on this site (the association bonus does not count).

Would you like to answer one of these unanswered questions instead?

Not the answer you're looking for? Browse other questions tagged or ask your own question.