Dear Experts
I am conducting a study related to academic institutions and all the data has been collected through internet. Respondents are from different countries e.g. 110 from country A, 90 from country B, 90 from country C, 60 from country D, 50 from country E etc. How should I justify my data collection as there is difference in number of responses among different countries?
Thanks
Hi, kindly peruse this paper: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3884902/pdf/nihms526903.pdf
Good luck!
You don't create new justification for decisions post-hoc, you must justify them before carrying out. After the fact you should assess whether your original justifications were correct.
A big concern with data collection strategies is the risk of bias. What steps did you take to estimate bias? Apart from country is there other demographic variables to assess for bias (age, gender, ethnicity, wealth etc)? If you haven't built your data collection to assess potential bias then you can't prove a lack of bias or correct for it.
One option is to randomly resample your larger groups to create a more balanced group size. If you have demographic data you can use it to guide random resampling to create unbiased groups for comparisons.
As the data has demographic variables from different countries ,i think creating strata s on each variable base and then comparing them across a single parameter should be good.
But this should fulfill your objectives.
Dear researcher,
To judge the representativeness of the respondents, you need to identify the population of your study and state all the procedures that have been taken to arrive to these respondents (sampling frame, sample type, sample size …etc.).
Good luck with your research
Dear Jawad
It is very important that in the "methods" section you justify this study design (which does not seem like something that just happened). Two possibilities in data management are 1) Compare behaviors between different countries, in that case the concern would be the size of the samples, 2) Add all the subsamples into one, but it would have to be justified.
In any case, your sampling is not random probabilistic, so you should be very careful in the conclusions you extract. In general, you should consider your study as exploratory.
Greetings,
Jorge
Thank you all
Mr. James, Yes I have detailed demographic information of each respondent e.g. age, gender, education level, country etc.
Depending on the countries you have surveyed, you may need to account for sample bias created by the fact that only those with internet access/literacy are able to participate, which does not necessarily give a good representation of the overall populations of those countries; only the a representation of internet users in that country. This is just a caveat that you must admit and your justification may be a lack of funding, time or access to these countries in order to utilise other survey methods. In addition, your response rate is difficult to indicate unless the participants were specifically targeted (ie. emailed the survey).
An important element of justifying your data collection is the response rate and potential reasons for it being lower in certain countries (language, literacy, interest in topic, incentives etc, % population of university students from each country.).
If you are looking to aggregate the populations from different countries, just check to see if the country variable in independent of whatever you are trying to evaluate.
As this is a post data collectcion problem , while analysiing you can adopt two methdos 1. use a popoulataion based responses for instance , a 36% desired response came the same can be used for represeing as per populataion , other use a LQAS method and from all teh sample take 19 for each country using a SRS sampling and on 19th Run LQAS golden rue
The difference in the number of responses is not an important issue. The samples would be more efficient with equal numbers, but I would rather have, say, one group with 90 and one group with 60 than two groups with 60. The more important issues are: (1) samples of 100 or less aren't very reliable if the full populatio ,is large and (2) there is likely to be bias in terms of who chose to respond and who did not. You might consider weighting your data to be more representative of the full population.
- Badges
- Science topic