Hello everyone, and thanks in advance for your help!
I am currently running a survey with the company I work for as a project for my postgraduate dissertation.
The survey has been run through fb. Initially we created a facebook ad which appeared in the timeline of people who are resident in the Uk and are between 25-45 years old (the sample we are interested in).
What we did next was (in order to increase the responses) posting the survey to the company social media pages (fb and twitter). This gave the access to the survey to a wider audience which not necessarily matches with the profiles we are looking for, however the survey contains demographic questions through which we could screen only the groups we were interested in.
I am about to write the sample selection section of my thesis and I am a bit confused about how i can define the method utilised.
Can I say the method is Stratified random sapling?
Dear Jessica,
I don't think your sampling was stratified because you are only interested in one category of respondents rather than seeking to match the proportion of your sample in different categories to the same proportions in your population.
Unless you can specify exactly who had access to your advert I think you will have to call your sample self selected. This may raise issues of potential bias.
Hi Peter,
Thank you very much for your reply!
I did some research and maybe quota sampling is the right one?
Hello,
Good day,
I think Purposive sample / (Non-probability sampling) : The researcher chooses the sample based on who they think would be appropriate for the study ( in your research - resident in the Uk and are between 25-45 years old ).
Regards,
Dear Jessica,
As I understand, I will describe your sampling as follows:
Sampling procedure:
A pretested questionnaire (Appendix 1, provide the copy of the quetionnaire at the end) was posted on Facebook. Initially, all the respondents who were between 25-45 years and from UK formed the sample . Further to increase the sample size and to get sample from other than UK, the questionnaire was posted on company's social media pages. From the respondents, those who fulfilled the following criteria were included in the sample (you have to define) like:
---- Only completed questionnaires were included in the study.
---- Partially or incomplete questionnaires were excluded from the study.
--- Respondents between 25-45 years age are considered for the study.
----- All those who responded from 1-01-2017 to 31-12-2017 (that is to define period of survey) are included in the study.
All the respondents not fulfilling the criteria are excluded from the study.
So, you may mention all the criteria which are used in the selection of the sample, one by one.
This approach is used many times in case control studies where you have to define your cases and controls.
The problem is that you really do not have a random sample, and apparently you do not have regressor data on the entire population for 'prediction' either (right?), so without either of those characteristics, you have no way of estimating the accuracy of your results. You have not considered the representativeness, of your sample, or lack of representativeness, to be able to infer information such as the mean of the population.
Your sample may cover a large part of the population, but miss other parts entirely.
This is a common problem. Sorry.
I do not see anyway that you can do such a data collection and qualify it as a randomized design, nor are you going to have regressor data for the whole population, so there is no way to reliably infer from the sample you collect, to the population. However, I attended the seminar below, and I think that Mike Brick has some very good ideas for improving the situation. I will follow that link with a link to the report that Mike used as the basis for his presentation, and that will be followed with a couple of links to other work that he did which I am guessing might have something useful for you:
J. Michael Brick on Inference from Nonprobability Sampling:
"2014 WSS PRESIDENT’S INVITED SEMINAR,"
http://washstat.org/seminars/2014/20140326_brick.html
The committee report:
https://www.researchgate.net/publication/273561892_Summary_Report_of_the_AAPOR_Task_Force_on_Non-probability_Sampling
"The Future of Survey Sampling," Article in Public Opinion Quarterly 75(5):872-888 · December 2011
Impact Factor: 2.25 · DOI: 10.2307/41345915
https://www.researchgate.net/publication/261967521_The_Future_of_Survey_Sampling
"Beyond traditional survey taking: adapting to a changing world, Explorations in Non-Probability Sampling Using the Web," J. Michael Brick, Proceedings of Statistics Canada Symposium 2014
http://www.statcan.gc.ca/sites/default/files/media/14252-eng.pdf
Please note that even a census can have huge measurement error, and other nonsampling error, and a random sample, even stratified to do a better job of properly representing the population based on known auxiliary data, can still miss important parts of a population distribution, or subpopulations/strata, even perhaps greatly underestimating variance, so there is inaccuracy in anything you do, but your sample has a high unknown risk because it was collected in a way that is not designed for representativeness. I recently saw an abstract to someone's paper on 'big data,' and they noted that even a very large sample does not mean it can't be badly biased. As I think a mentor, Ken Brewer, once quoted his mentor, Ken Foreman, as saying: "There is no substitute for knowing your data." I take that to mean that you have to know all the different characteristics of your data (and properly represent them).
- Badges
- Science topic
- Similar topics
- Social Science
- Media
- Social Media