My topic is Policy for Home and School Relationships in a Multi-Cultural Community: The Optimization of Culturally-Relevant Learning in a private, urban secondary school in Antigua.
I have surveyed 25 - 30 students and 10 teachers. 5 of these teachers have been interviewed. The plan is to survey 50 parents but the response so far is 20. I need to start my analysis of the data by next week. What is a reasonable number I could use for reliability and validity so that the information gleaned is sufficient or adequate?
The sample size should be determined based on your research question or hypothesis and consequently the statistical tests which you will use.
Usually researchers consider in their research the possibility of low response, thus extend the time of data collection. If you cant reach the target size, then you have to mention this as a limitation
It is difficult for anyone in Researchgate to tell you what sample size will be enough for your dissertation. The best answer you can get is from your Advisor.
Cafille:
I did a similar research for my doctoral dissertation and had the same survey response issue. The view of my committee was that at least 35% response rate was a minimum acceptable figure in keeping with the literature on survey research. I have attached the links that I hope would be useful to your research.
many thanks,
Debra
Article Workforce Education and Development Curriculum Responsivenes...
Data Curriculum Inclusiveness Challenge: Responding to Multicultu...
Hi Cafille
Sample size for surveys should be based on the required degree of accuracy, not validity, reliability or a figure plucked out of the air by a doctoral committee. If for example, the target population (in your case parents) you wanted to generalize to was 50, and you required +/- 10% accuracy with 95% confidence when 50% of the parents respond "yes" to a question, you would need to obtain 34 completed questionnaires back from the 50 parents. If you wished to generalize to all similar parents in your country, with the number in the target population in the thousands, you would need responses from 95 parents for the same degree of accuracy.
The sample size should be determined based on your research question or hypothesis and consequently the statistical tests which you will use.
Usually researchers consider in their research the possibility of low response, thus extend the time of data collection. If you cant reach the target size, then you have to mention this as a limitation
It is difficult for anyone in Researchgate to tell you what sample size will be enough for your dissertation. The best answer you can get is from your Advisor.
Cafille -
Your problem is worse than one of number of respondents, but actually what subcategories do those who do respond actually represent. For example, if a few people are upset about something, then those may be about the only ones who might respond. They would not 'represent' your population, and results from them would be biased.
Even in the work I did on continuous data for official statistics where responses from the energy industry were mandatory, the standard deviations of populations, from which sample size estimations typically depend, were sometimes artificially inflated by nonsampling error. For you, the low response rate greatly impacts your results. A larger sample size may often just seem to confirm your highly biased results, until you approach a sample the size of the census, which would likely be unaffordable and have substantial nonsampling error anyway.
Your biggest concern is to figure out who the responders really represent, and to try to get others, not represented by them, to respond.
I suggest that you research the term "response propensity" group. These are a type of stratification. In stratified random sampling, you sample from groups randomly. But here, a little more like poststratification, you need to try to counter the fact that you did not use randomization, as much as possible, by grouping data by likelihood of response. - Another option is regression modeling, but you don't appear to have the kind of data necessary for that. - But even for qualitative data, you would want to avoid drawing conclusions from a biased set of respondents, so you need to discover whatever might influence responders, and group their responses accordingly. The worst problem would be if you have no respondents to represent a given 'category' of responders. Trying to collect more data then, and just settling for any respondents, will likely just give you more from categories/strata/response propensity groups that you already have, just then appearing to confirm your biased results. This, I think, is the biggest problem with high nonresponse: You cannot count on nonrespondents to be "missing at random." They probably aren't. At least in quantitative analysis, missing in some way not at random is called "nonignorable nonresponse," not that any kind of nonresponse should just be "ignored," but that special term is applied.
Best wishes - Jim
- Badges
- Science method
- Similar topics
- Mathematical Sciences
- Statistics
- Applied Statistics
- Survey Design